Fleig NSP Final A

Collective Decisions on
Conditional Topics
An Empirical Study of the Impact of
Nonseparable Preferences
DISSERTATION
ZUR ERLANGUNG DES AKADEMISCHEN GRADES

DOCTOR RERUM POLITICARUM
AN DER
FAKULTÄT FÜR WIRTSCHAFTS- UND SOZIALWISSENSCHAFTEN

DER RUPRECHT-KARLS-UNIVERSITÄT HEIDELBERG
vorgelegt von
Andreas Fleig
Heidelberg, Juni 2013

DEAN
Prof. Dr. Jürgen Eichberger, Department of Economics, University of Heidelberg
EXAMINER
Prof. Dr. Daniel Finke, Department of Political Science, University of Heidelberg
Prof. Christoph Vanberg, Ph.D., Department of Economics, University of Heidelberg
Date of Disputation: 04.10.2013
FINANCIAL SUPPORT
The completion of this work was promoted by the Baden-Württemberg Foundation and
the Completion grant of the Graduate College of the University of Heidelberg.
Scientific contributions
As part of my research over the last three years the following scientific contributions have
evolved:
CONFERENCE PRESENTATIONS
• Fleig, Andreas. 2011a. The Merits of Adding Complexity: Conditional Preferences

in Spatial Models of EU Politics. Presented at the 6th ECPR general conference at
the University of Iceland, Reykjavik.
• Fleig, Andreas. 2011b. Pooling vs. Delegation: Collective decision-making under

uncertainty. Presented at the Annual Meeting of the working group for action and
decision theory of the German association for political science (DVPW), University
of Kiel.
• Fleig, Andreas. 2012. Motivated by the Process: How decision-making procedures

shape the motivation function under majority rule. Presented at the 3rd Thurgau ex-
perimental economics meeting (theem) at the University of Konstanz, Kreuzlingen
(CH).
ARTICLES
• Finke, Daniel and Andreas Fleig. 2013. “The Merits of Adding Complexity: Non-
Separable Preferences in Spatial Models of EU Politics.” Journal of Theoretical Politics
25:547-576.
WORKING PAPERS
• Finke, Daniel and Andreas Fleig. “Identifying Social Preferences in Majority Deci-
sions.” mimeo.
• Fleig, Andreas and Daniel Finke. “Delegation, Uncertainty and Social Preferences
in Majority Decisions?” mimeo.
These contributions are also part of my dissertation. I indicated this in the according
chapters and sections by corresponding references.
Abstract
Analytical politics investigates collective decision-making in political systems. Such vot-

ing behavior in groups takes place in parliaments, committees or the board of local foot-
ball clubs. It is a frequent object of study for theoretical as well as empirical analysis.
Previous contributions have demonstrated well the stabilizing effect of procedural rules,
such as agenda-setting or multi-chamber systems, for collective decisions. These rules are
applied in many institutions, such as the European Parliament or the German Bundestag.
Their main purpose is to ensure reliable policy.
Previous work continually used the restrictive assumption of separable preferences. This
assumption implies that different aspects of a question do not influence each other. The
limited validity of this hypothesis is apparent even in everyday situations. For example,
the enjoyment of a delicious meal depends on the combination of food and drink. When
choosing between fish and venison for dinner you also have to consider the question of
which sort of wine to have with the meal; white with fish, and red with venison. This form
of interdependence also occurs in legislation. For example, the savings determined in the
Greek budget influence the preferences of the German public for financial assistance to
Greece.
The assumption of separable preferences is therefore in the critical focus of theoretical
research. This literature discusses the impacts of and solutions to nonseparable prefer-
ences in detail. The analysis suggests an increased complexity for every decision-making
process affected by nonseparable preferences. This complexity leads to difficulties in the
operationalization of nonseparable preferences and is one of the reasons that there are
too few empirical examinations. In addition, the stabilizing properties of institutional
arrangements identified under the assumption of separable preferences are in question.
The goal of this study is to close this gap between theory and empiricism.
I investigate nonseparable preferences by conducting a laboratory experiment, which al-
lows comprehensive environmental control. This facilitates the operationalization of non-
separable preferences. First, I prove the relevance of nonseparable preferences for analyt-
ical research on social interaction. The experiment is therefore completed by empirical
case studies. Next, I investigate the effects of nonseparable preferences on collective and
individual decision-making in the laboratory. Finally, I assess my contribution with re-
spect to current research in social science and discuss possibilities to more accurately
model of human behavior.
The dissertation is structured as follows. I start in chapter 1 with the presentation of
my research question and design. In chapter 2 the concept of nonseparable preferences
is further clarified by means of exemplary case studies. It also discusses the theoretical
foundations of nonseparable preferences. My hypotheses are specified along common
concepts used in the literature. Based on empirical data the relevance of nonseparable
preferences for political science research is demonstrated in chapter 3. Next, chapter 4
presents the design of the laboratory experiment. The effects of nonseparable preferences
on collective decision-making are examined in chapter 5. Subsequently, determinants for
the motivation function of individuals are scrutinized in chapter 6. In chapter 7 I report
the results of the post-experiment survey. All findings are evaluated in chapter 8, where I
focus on detailing their usefulness to future research on human behavior. Finally, chapter
9 summarizes the study and lists possibilities to further expand research in this area.
Zusammenfassung
Die Analytische Politikwissenschaft befasst sich mit kollektivem Entscheidungsverhalten

im Kontext politischer Systeme. Solche Gruppenentscheidungen betreffen Abstimmun-
gen in Parlamenten, Ausschüssen oder dem Vorstand des örtlichen Fussballklubs. Sie
sind ein häufiger Untersuchungsgegenstand theoretischer sowie empirischer Analysen
anhand derer die stabilisierende Wirkung fester Verfahrensregeln wie Agendasetzung
oder eines Mehrkammernsystems für kollektive Entscheidungen mehrfach nachgewie-
sen wurde. Diese Regelungen finden in zahlreichen Institutionen wie dem Europäischen
Parlament oder dem Deutschen Bundestag Verwendung. Ihre zentrale Aufgabe ist die
Sicherstellung verlässlicher Politikentscheidungen.
Die bisherigen politikwissenschaftlichen Beiträge verwendeten fortwährend die restrik-
tive Annahme separabler Präferenzen. Diese besagt, dass einzelne Aspekte einer Frage
sich nicht gegenseitig beeinflussen. Die limitierte Validität dieser These zeigt sich schon
in alltäglichen Situationen. Der Genuss eines guten Essens hängt von der Kombination
von Speisen und Getränken ab; die Wahl zwischen Rot- und Weißwein ist nicht von der
Entscheidung für ein Fisch- oder Wildgericht zu trennen. Diese Form von Abhängigkei-
ten tritt auch in Gesetzesvorhaben auf. So beeinflussen die beschlossenen Einsparungen
im griechischen Haushalt die Bereitschaft der deutschen Öffentlichkeit zu einer finanzi-
ellen Unterstützung Griechenlands.
Die Annahme separabler Präferenzen steht deshalb im kritischen Fokus der theoretischen
Forschung, welche eingehend Auswirkungen von und Lösungskonzepte für Nichtsepa-
rabilität diskutiert. Diese Analysen weisen auf eine gesteigerte Komplexität im Entschei-
dungsfindungsprozess hin. Die resultierenden Schwierigkeiten bei der Operationalisie-
rung sind zum einen der Grund für die fehlenden empirischen Untersuchungen. Zum
anderen stellen sie die unter der Annahme separabler Präferenzen identifizierten stabili-
sierenden Eigenschaften institutioneller Regelungen in Frage. Das Ziel meiner Arbeit ist
es, die hier herrschende Lücke zwischen Theorie und Empirie zu schließen.
Dies erfolgt anhand eines Laborexperiments, welches durch eine umfassende Modifikati-
on der Rahmenbedingungen die Operationalisierung von Nichtseparabilität ermöglicht.
In einem ersten Schritt weise ich die Relevanz des Konzepts der Nichtseparabilität für die
analytische Erforschung sozialer Interaktion nach. Hier wird das Experiment mittels em-
pirischer Fallstudien komplettiert. Anschließend untersuche ich die Auswirkungen von
Nichtseparabilität auf kollektives und individuelles Entscheidungsverhalten im Labor.
Zum Abschluss ordne ich meinen Beitrag in den Kontext aktueller sozialwissenschaftli-
cher Forschung ein und diskutiere die Forderungen nach einer realistischeren Beschrei-
bung menschlichen Handelns.
Kapitel 1 beginnt mit der Darlegung der Forschungsfrage und des Forschungsdesigns.
In Kapitel 2 wird mit Hilfe exemplarischer Fallstudien das Konzept nichtseparabler Prä-
ferenzen weiter präzisiert. In diesem Kapitel erläuterte ich außerdem die theoretischen
Grundlagen sowie die konkrete Operationalisierung von Nichtseparabilität und formu-
liere meine Hypothesen über die Folgen ihrer Nichtberücksichtigung. Anhand empiri-
scher Daten wird in Kapitel 3 die Relevanz von Nichtseparabilität für die politikwissen-
schaftliche Forschung aufgezeigt. Anschließend stellt Kapitel 4 das Design des Laborex-
periments vor. Die Auswirkungen von Nichtseparabilität auf kollektive Entscheidungs-
findung werden in Kapitel 5 untersucht. Darauf aufbauend werden die Determinanten
der Motivationsfunktion von Individuen in Kapitel 6 analysiert. Kapitel 7 diskutiert die
Ergebnisse des post-experimentellen Fragebogens. Die gewonnen Erkenntnisse werden
in Kapitel 8 bewertet. Dabei liegt der Fokus auf dem Nutzen für die zukünftige For-
schung über menschliches Verhalten. Schließlich fasst Kapitel 9 die Studie zusammen
und nennt mögliche Erweiterungen des Forschungsbereichs.
Contents
List of Tables vii
List of Figures ix
List of Abbreviations xi
1 Introduction 1
1.1 Research question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Previous research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Focus of the present study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.1 Research design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Why a laboratory experiment? . . . . . . . . . . . . . . . . . . . . . 19
1.4 Outline of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2 What are nonseparable preferences? 23

2.1 The general principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Introductory examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.1 The European Union regulation on chemicals . . . . . . . . . . . . 26
2.2.2 The privatization debate on Deutsche Bahn . . . . . . . . . . . . . . 27
2.2.3 The reform of the German Bundeswehr . . . . . . . . . . . . . . . . 29
2.2.4 The future of the German pension system . . . . . . . . . . . . . . . 30
2.2.5 The construction of a new convention center in Heidelberg . . . . . 31
2.3 Do delegation, decentralization and specialization help? . . . . . . . . . . 33
2.4 Theoretical foundations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4.2 Using spatial models . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.3 How to operationalize nonseparable preferences . . . . . . . . . . . 43
2.4.4 The magnitude of nonseparability . . . . . . . . . . . . . . . . . . . 47
2.4.5 The case of reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4.6 How to measure nonseparable preferences . . . . . . . . . . . . . . 50
2.5 The impact of nonseparable preferences . . . . . . . . . . . . . . . . . . . . 51
3 The merits and costs of incorporating nonseparable preferences 53

3.1 The literature on legislative models of decision-making . . . . . . . . . . . 54
3.1.1 Unconstrained bargaining models . . . . . . . . . . . . . . . . . . . 55
i
3.1.2 Constrained bargaining models . . . . . . . . . . . . . . . . . . . . . 55
3.1.3 Agenda-setting models . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Data on decision-making in the European Union . . . . . . . . . . . . . . . 57
3.3 The effects of misspecifying preferences . . . . . . . . . . . . . . . . . . . . 59
3.3.1 Illustrative case study I . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.2 Illustrative case study II . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3.3 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4 The extent of nonseparable preferences in EU law-making . . . . . . . . . 67
3.5 The magnitude of nonseparability . . . . . . . . . . . . . . . . . . . . . . . 69
3.6 The nonseparability in individual votes . . . . . . . . . . . . . . . . . . . . 74
3.7 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4 The Experiment 79
4.1 Experimental research in political science . . . . . . . . . . . . . . . . . . . 79
4.1.1 The rise of the experimental method . . . . . . . . . . . . . . . . . . 80
4.1.2 The first experiments on collective decision-making . . . . . . . . . 83
4.2 Experimental design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2.1 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.2.2 Anonymity and non-communication . . . . . . . . . . . . . . . . . 86
4.2.3 Payoff table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.2.4 Multiple tables and their characteristics . . . . . . . . . . . . . . . . 89
4.2.5 Multiple rounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.6 Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.2.7 Two procedures per session . . . . . . . . . . . . . . . . . . . . . . . 102
4.2.8 Payment mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.2.9 The subjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.3 Implications for the empirical analysis . . . . . . . . . . . . . . . . . . . . . 108
4.3.1 Individual and collective level . . . . . . . . . . . . . . . . . . . . . 109
4.3.2 Statistical independence of observations . . . . . . . . . . . . . . . . 110
4.3.3 Rather qualitative than quantitative findings . . . . . . . . . . . . . 110
4.3.4 Responsibility for the results . . . . . . . . . . . . . . . . . . . . . . 111
4.3.5 Summary of key characteristics . . . . . . . . . . . . . . . . . . . . . 112
4.4 Descriptive information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5 The influence of institutional rules on collective decisions 117

5.1 Reliability of the experimental design . . . . . . . . . . . . . . . . . . . . . 118
5.2 Theoretical benchmark predictions . . . . . . . . . . . . . . . . . . . . . . . 121
5.2.1 Derivation of the credible core . . . . . . . . . . . . . . . . . . . . . 123
5.2.2 Distribution of the equilibrium solution over the payoff tables . . . 127
5.3 How to judge a collective decision . . . . . . . . . . . . . . . . . . . . . . . 128
5.3.1 Decisions costs and welfare effects . . . . . . . . . . . . . . . . . . . 128
ii
5.3.2 Hypotheses about the treatment effect . . . . . . . . . . . . . . . . . 132
5.3.3 Operationalization of the statistical measures . . . . . . . . . . . . . 134
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.4.1 Selection of the core . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
5.4.2 Social welfare allocation and stability . . . . . . . . . . . . . . . . . 140
5.4.3 Distribution of wealth and approval rate . . . . . . . . . . . . . . . 144
5.4.4 Decision-making efficiency . . . . . . . . . . . . . . . . . . . . . . . 146
5.4.5 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.5 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6 The determinants of individual choices 151

6.1 The literature on behavioral patterns of individual decision-making . . . . 152
6.1.1 Other-regarding preferences . . . . . . . . . . . . . . . . . . . . . . 153
6.1.2 Social preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
6.1.3 What makes people social? . . . . . . . . . . . . . . . . . . . . . . . 157
6.1.4 Reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.1.5 Risk aversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
6.2 Sincere and sophisticated voting . . . . . . . . . . . . . . . . . . . . . . . . 162
6.3 Statistical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.3.1 The robustness of self-interest . . . . . . . . . . . . . . . . . . . . . . 167
6.3.2 Random utility model . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.3.3 Mixture stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.3.4 Model specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
6.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.4.1 Random utility mixture model estimates . . . . . . . . . . . . . . . 177
6.4.2 The influence of co-variates . . . . . . . . . . . . . . . . . . . . . . . 184
6.5 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
7 Post-experiment survey 193

7.1 General comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
7.2 Binary questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.3 Qualitative content analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.4 Word scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.4.1 The latent dimension of discrepancy . . . . . . . . . . . . . . . . . . 201
7.4.2 Word weights and fixed-effects . . . . . . . . . . . . . . . . . . . . . 202
7.5 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8 Complex reality and limited models 207

8.1 The necessary effort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
8.2 The concept of abstraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.3 A more realistic view of human behavior . . . . . . . . . . . . . . . . . . . 214
8.4 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
iii
9 Conclusion and outlook 223
9.1 Final summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
9.2 Future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
9.2.1 Crowdsourcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
9.2.2 Going to the field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
9.2.3 Communication and interaction . . . . . . . . . . . . . . . . . . . . 232
Bibliography 239
Appendix 301
A.1 Software used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
A.2 Weighted Euclidean distance including nonseparable preferences . . . . . 303
A.3 Structure of the DEU field reports and expert interviews . . . . . . . . . . 305
A.4 Three-dimensional contour plots . . . . . . . . . . . . . . . . . . . . . . . . 306
A.5 The extent of nonseparable preferences in the DEU data set . . . . . . . . . 310
A.6 The magnitude of nonseparability at the proposal level . . . . . . . . . . . 313
A.7 Experimental instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
A.8 All payoff tables used in the experiment . . . . . . . . . . . . . . . . . . . . 326
A.9 Payoff table characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
A.10 Frequency of use of each payoff table . . . . . . . . . . . . . . . . . . . . . . 329
A.11 Data structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
A.12 Trend analysis of the collective results . . . . . . . . . . . . . . . . . . . . . 330
A.13 Gini coefficients of the collective results . . . . . . . . . . . . . . . . . . . . 332
A.14 Standard experimental games used to measure social preferences . . . . . 333
A.15 Predicted probabilities for core solutions at different levels of rationality . 334
A.16 Coding guide for the qualitative content analysis . . . . . . . . . . . . . . . 355
A.17 References of experiments conducted online and in the laboratory . . . . . 358
iv
v
vi
List of Tables
2.1 Separable and nonseparable preference orders . . . . . . . . . . . . . . . . 25
3.1 Extent of nonseparability in EU law-making . . . . . . . . . . . . . . . . . 67

3.2 Model comparison of mean average error on issue and proposal level . . . 68
3.3 Nonseparability effect on relative model performance . . . . . . . . . . . . 69
3.4 Model hit rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.1 Treatment characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.2 Characteristics of the different decision situations . . . . . . . . . . . . . . 102
4.3 Design characteristics and their impact on the empirical analysis . . . . . 113
4.4 Data structure at the session level . . . . . . . . . . . . . . . . . . . . . . . . 114
4.5 Data structure according to the order of decision rules . . . . . . . . . . . . 114
4.6 Data structure according to constant-sum and non-constant-sum table . . 115
4.7 Data structure according to the order of decision rules and constant-sum
or non-constant-sum table . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.1 Two-way error components model . . . . . . . . . . . . . . . . . . . . . . . 120

5.2 Selection of the core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.3 Data structure according to the existence of an equilibrium alternative . . 128
5.4 Hypotheses for the treatment effect . . . . . . . . . . . . . . . . . . . . . . . 134
5.5 Core performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.6 Trend analysis of core performance . . . . . . . . . . . . . . . . . . . . . . . 140
5.7 Social welfare allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.8 Social welfare allocation of core alternatives . . . . . . . . . . . . . . . . . . 142
5.9 Stability within a voting procedure . . . . . . . . . . . . . . . . . . . . . . . 142
5.10 Stability across voting procedures . . . . . . . . . . . . . . . . . . . . . . . . 143
5.11 Standard deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.12 Standard deviation of core alternatives . . . . . . . . . . . . . . . . . . . . . 145
5.13 Approval rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.14 Decision-making efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.15 Robustness of treatment effects . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.1 Co-variates for the random utility mixture model . . . . . . . . . . . . . . 175

6.2 Model estimates under pooling . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.3 Model estimates of the second stage under sequential delegation . . . . . 181
6.4 Model estimates of the first stage under sequential delegation . . . . . . . 182
6.5 Model estimates under sequential delegation separated for beliefs . . . . . 184
6.6 Model estimates under simultaneous delegation . . . . . . . . . . . . . . . 185
6.7 Model estimates including co-variates . . . . . . . . . . . . . . . . . . . . . 187
7.1 Results of the binary questionnaire . . . . . . . . . . . . . . . . . . . . . . . 195

7.2 Number of different criteria a subject considered . . . . . . . . . . . . . . . 196
7.3 Correlation coefficient between different criteria . . . . . . . . . . . . . . . 196
vii
7.4 Free-input field “Please describe your decision rule with your own words?” 197
7.5 Free-input field “Which players did you focus on?” . . . . . . . . . . . . . 198
7.6 Free-input field “Which criterion of an alternative did you consider?” . . . 198
7.7 Free-input field “What changed when the decision-making procedure split
up?” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.8 Free-input field “What changed when the decision-making procedure split
up?”separated by treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.9 Words with the highest impact factor . . . . . . . . . . . . . . . . . . . . . . 203
7.10 None decisive terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8.1 Methods of data collection for investigating nonseparability . . . . . . . . 211
9.1 Pros and cons of internet experiments . . . . . . . . . . . . . . . . . . . . . 231
A.2 Structure of the DEU expert interviews . . . . . . . . . . . . . . . . . . . . 305

A.3 The extent of nonseparable preferences in the DEU data set . . . . . . . . . 310
A.5 Payoff tables used in the experiment . . . . . . . . . . . . . . . . . . . . . . 326
A.6 Payoff table characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
A.7 Frequency of use of each payoff table . . . . . . . . . . . . . . . . . . . . . . 329
A.8 Data structure according to decision rule and payoff table properties . . . 330
A.9 Trend analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
A.10 Gini coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
A.11 Experimental games to measure social preferences . . . . . . . . . . . . . . 333
A.15 Coding guide for the free input fields . . . . . . . . . . . . . . . . . . . . . 355
viii
List of Figures
2.1 A simple spatial model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.2 Indifference curves for separable preferences . . . . . . . . . . . . . . . . . 44
2.3 Indifference curves for nonseparable preferences . . . . . . . . . . . . . . . 45
2.4 Ridge lines for nonseparable preferences . . . . . . . . . . . . . . . . . . . . 46
2.5 Indifference curves for non-reciprocal nonseparable preferences . . . . . . 49
3.1 Model predictions for COM1999/163 with separable preferences . . . . . 61

3.2 Model predictions for COM1999/163 with nonseparable preferences . . . 62
3.3 Model predictions for COM1999/582 with separable preferences . . . . . 64
3.4 Model predictions for COM1999/582 with nonseparable preferences . . . 65
3.5 Mean average error per issue at different levels of nonseparability . . . . . 71
3.6 Comparison of model’s predictive accuracy at the issue level . . . . . . . . 72
4.1 Payoff scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4.2 Example of a payoff table . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.1 Example of the determination of the core alternative . . . . . . . . . . . . . 126
6.1 Robustness of the core alternative . . . . . . . . . . . . . . . . . . . . . . . . 169
7.1 WORDFISH : latent discrepancy dimension . . . . . . . . . . . . . . . . . . . 201

7.2 WORDFISH : word-weights vs. word fixed effects . . . . . . . . . . . . . . . 202
A.1 Three-dimensional contour plots for Council regulation COM1999/163 . . 306

A.2 Three-dimensional contour plots for Council regulation COM1999/582 . . 308
A.3 Mean average error at different levels of nonseparability and model’s pre-
dictive accuracy at the proposal level . . . . . . . . . . . . . . . . . . . . . . 313
A.4 General instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
A.5 Pooling instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
A.6 Simultaneous delegation instructions . . . . . . . . . . . . . . . . . . . . . . 320
A.7 Sequential delegation instructions . . . . . . . . . . . . . . . . . . . . . . . 323
A.8 Predicted probabilities under pooling . . . . . . . . . . . . . . . . . . . . . 334
A.9 Predicted probabilities under simultaneous delegation . . . . . . . . . . . 341
A.10 Predicted probabilities under sequential delegation . . . . . . . . . . . . . 348
ix
x
List of Abbreviations
θ An actor’s unconditional ideal poin
A Matrix of saliences
a Single element of the matrix of saliences
ACP African, Caribbean and Pacific Group of States
AIC Akaike information criterion, a measure of the relative goodness of fit of a

statistical model (Akaike 1973)
ALDE Alliance of Liberals and Democrats for Europe
AMT Amazon Mechanical Turk
APM Accumulated payoff mechanism
BW German Federal Armed Forces (translation of Deutsche Bundeswehr)
CI Confidence intervals
CP-nets Conditional preference networks
CRRA Constant relative risk aversion
D Somers’ D, an ordinal measure of association introduced by Somers (1962)
d Metric distance
D+ A measure for advantageous inequality in the inequality averversion model

of Fehr and Schmidt (1999)
D- A measure for disadvantageous inequality in the inequality averversion model

of Fehr and Schmidt (1999)
DB Deutsche Bahn
DEU Decision-making in the European Union
DGP Data-generating process
DVPW German association for political science (translation of Deutsche Vereinigung

für Politische Wissenschaft)
ECHA European chemicals agency
ECPR European consortium for political research
EDE Experimenter demand effect
EP European Parliament
xi
EPP-ED European People’s Party-European Democrats
ERC Equity, reciprocity and competition
ESA Economic Science Association
ESM European stability mechanism
FBNE Feedback Nash equilibrium
Gini Gini coefficient, a measure of statistical dispersion (Gini, 1912)
HSP Heterogeneous social preferences
i Index of actor
IIA Independence of irrelevant alternatives
iid Independent and identically distributed
INET Institute for new economic thinking
IP Ideal position of an actor in a spatial model
IPO Initial public offer
MAE Mean average error
MCMC Monte Carlo Markov Chain
MEP Member of the European parliament
MPNE Markov-perfect Nash equilibrium
N Number of observations
n Number
Nash product Maximand of the Nash Bargaining Solution (Nash, 1950)
NBS Nash bargaining solution
NSP Nonseparable preferences
OLNE Open-loop Nash equilibrium
ORSEE Online recruitment system for economic experiments
OSLA One-step look-ahead algorithm
P Probability
POL Pooling; a specific voting procedure in my laboratory experiment
QMV Qualified majority voting
QRE Quantal response equilibrium
r Minkowski r-metric
RCV Roll-call vote
xii
REACH The European Union regulation on registration, evaluation, authorisation
and restriction of chemicals
RMDSM Random multiple decision selection mechanisms
RRPM Random round payoff mechanism
RUMM Random-utility mixture model
S&K Sauermann and Kaiser (2010), I adopt their general laboratory environment
for my experimental design
SCF Social choice function
SD Standard deviation
SE Standard error
SEQ Sequential delegation; a specific voting procedure in my laboratory experi-

ment
SIM Simultaneous delegation; a specific voting procedure in my laboratory ex-

periment
SPE Subgame-perfect equilibria
SQ Status quo
SQP Sequential quadratic programming
SSC Statistical software components
SSS-V Sensation Seeking Scale V
SVO Social value orientation
theem Thurgau experimental economics meeting
tm Text mining
U Utility level
U.S. United States of America
w Weight of spatial dimension
WED Weighted Euclidean distance
WEIRD Western, educated, industrialized, rich and democratic
WTO World Trade Organization
z-Tree Zurich Toolbox for Readymade Economic Experiments
xiii
xiv
1 Introduction
The debate surrounding a budget for a political program is a matter of everyday poli-
tics. The preferences of the politicians deciding on the size of the budget depend on the
politically determined aim of the program; e.g., who is entitled to a subsidy or who is ex-
empt from a certain regulation. If the purposes of both the program and politics match,
one favors the allocation of more money. If they contradict each other, one aims to keep
the financial support low. A relationship which is as obvious as it is banal. A current
and concrete example of such conditionality is not hard to find: the austerity measures
determined in the Greek budget affected the willingness of the German public to offer
financial assistance to Greece during the last two years.1
Such dependencies are not restricted to a financial dimension. The passing of exploitation
rights from a principal to an agent most certainly depends on the compliance of inten-
tions between the two. The principle of subsidiarity is applied more often if the lower
administrative level is politically in line with the higher level of administration. Pref-
erences for the distribution of voting rights are connected to the presence of comrades
and competitors in the electorate. Formulating the actual question of a referendum (e.g.,
adding or omitting parts) exerts a lot of influence on the public vote.
When political scientists analytically describe such conditionalities with theoretical mod-
els, even simple compounds can very quickly become complicated. Therefore, the indi-
vidual components are often examined separately, i.e., the result of one decision is in-
vestigated without considering the outcome of other decisions. Possible interactions are
denied. This simplification entails the risk of obtaining incorrect results and drawing the
wrong conclusions.
Analytical politics incorporates conditionalities as nonseparable preferences (henceforth
NSP) into its models. The most basic and simple definition for NSP is that the preferred
choice of an actor on one issue changes with the result of a second issue: how much one
wants to spend on a policy program depends upon exactly to what degree it supports
one’s own policy-making intentions.
Survey and referendum research relies on the measurement of individual preferences. It
is essential to measure them correctly and precisely. The current trend suggests that these
fields will become even more important at the local as well at the federal level in the fu-
ture. Beedham (1993, p.5) identified a development in the nature of democracy indicating
1 Source: Public poll by the Forsa survey research institute. The survey was conducted on September 15th
and 16th 2011 and the number of respondents was 1002 German citizens (Forsa, 2011).
1
1 Introduction
“a shift from representative democracy to direct democracy”,2 i.e., towards more citizen
participation and public voting.3 It is inevitable that a survey or referendum enables its
participants to express their preferences in an undistorted, clear and complete manner. A
single binary vote may not be enough if the topic in question contains interrelated issues.
This would risk mistaking the respondents’ conditional response as their genuine first
preference (Lacy, 2001a). Thus, not the genuine response is captured but only the reply
adjusted to another circumstance. Conclusions based on such results are inaccurate and
misleading.
In my study, I empirically analyze whether and, if it is the case, how the nonseparability
of preferences influences decision-making processes. This enables me to identify pro-
ceedings which avoid the biased results otherwise obtained. My contribution does not
refer to a single empirical case or specific data set. Rather, I draw attention to the phe-
nomenon of NSP and gather contributions from various fields to synthesize a single and
consistent assessment. This lays the coherent foundation for research on NSP, which has
until now been lacking.
The following sections present an overview of my study. I start by explaining my research
question in more detail in Sec. 1.1. Next, I discuss previous research in Sec. 1.2 in order
to point out the research gap which I aim to fill. This section also concisely addresses
the areas of scientific research which might gain from applying my results. I follow these
findings up with a delimitation of the focus of the study in Sec. 1.3 in which I describe
my research design and substantiate the use of a laboratory experiment. The last section
presents a further outline of the study.
1.1 Research question
Political science deals with a broad variety of issues. These range from everyday deci-
sions to simple judgments and complex problems. In the realm of politics decisions are
usually made in a group and not individually.4 Such collective decision-making takes
place in parliaments, committees, or the board of local football clubs.5 Analytical pol-
itics aims to identify the determinants for decision-making by means of a formalized
2 Morris (1999, p. 23) characterized this development as “shift from representational (Madisonian) to direct
(Jeffersonian) democracy” and as the “fundamental paradigm that dominates our politics”.
3 Cf. Dalton et al. (2001) for a summary of feasible institutional designs in the continuum between direct and
representative democracy. In general, the question whether political decisions should be taken rather in
parliament or by referendum is controversial. For example, proponents of more direct democratic partic-
ipation repeatedly point out the endemic irresponsibility of much too low estimated costs in government
procurement (Merkel, 2010). Yet, opponents insist on the non-negotiable democratic legitimacy of par-
liament decisions which are more reliable and robust against short-term mood changes and propaganda.
Cf. McConnachie (2000) for a discussion of the advantages and disadvantages of a public vote and Fiorino
and Ricciuti (2007) on the determinants of direct democracy.
4 Miller (1997, p. 1181) pointed out that the “mediation of groups distinguishes how individuals participate
in political settings from the way that they participate in markets.”

5 In the tradition of Buchanan and Tullock (1962) this means “ordinary politics” which operates within the
already established “constitutional stage” (cf. Buchanan, 1986).
2
1.1 Research question
approach6 . This includes multiple criteria such as the wishes and beliefs of individuals
(e.g., Dyer et al., 1992; Wallenius et al., 2008) and institutional settings (e.g., Peters, 2000).
Collective decision-making was and is a frequent object of study for theoretical as well
as empirical analyses (Kirsch, 2004).7 Previous contributions have successfully demon-
strated the influence and stabilizing effect of procedural rules, such as agenda-setting,
majority voting, or multi-chamber systems that structure the legislative process (e.g.,
Bottom et al., 2000; Miller and Hammond, 1990).8 These rules are applied in many in-
stitutions such as the European Parliament or the German Bundestag. Their central pur-
pose is to ensure reliable policy-making (Tsebelis, 2002) and prevent infinite debates (cf.
McKelvey’s Chaos Theorem, McKelvey, 1976; Shepsle and Cox, 2007).
Individual and collective decisions are mostly more complex than simple ’yes or no’ ques-
tions.9 Imagine the preparation of an exquisite dinner. The decision of which wine is the
most appropriate hinges on the main course one has selected. The overall success of the
dinner depends on the combination of drink and food: with fish I prefer white wine; if
venison is served I opt for red wine. Another example that illustrates this complexity is
that of two neighboring villages which both face decisions on development plans of their
municipality. The first village plans to use certain areas of farm land to build homes and
for industrial purposes. Unfortunately, the second village is planning to build a vast in-
cineration plant on fields adjacent to this neighborhood. Obviously, the latter plan might
reduce the value of land if the first village chooses the wrong fields. Again, the preferred
choice in one decision depends on the outcome of another.10
Highly complex and multidimensional legal matters are typical of the field of politics.11
Conflicts on legislative proposals can usually be subdivided into multiple, interrelated
issues (e.g., von Prittwitz, 2007; Tsebelis, 1994). As a consequence, legislators’ prefer-
ences on one issue may depend on the (expected) outcomes of other issues (Hinich and
Munger, 1997). One well-known case is the conditionality in legislators’ preferences, as
discussed above, on the characteristics of and budget for a political program. In fact, a
“large [amount of] literature in economics demonstrates that people cannot be assumed
to have separable preferences on public spending issues” (Lacy and Niou, 2000, p. 8).
Many contributions to research in political science agree that NSP exist (e.g., Brams et al.,
1997; Lacy, 2001a). They occur in combinations of different subjects, various fields, vari-
6 Cf. Morton (1999, Chap. 2) for an excellent comparison of formal as well as non-formal models.
7 Cf. Kirsch (2004) for a comprehensive overview on the fields of new political economy and institutional
analysis. He investigated collective decision-making in political systems in general. More specifically,
Schofield (1996) discussed social choice theory in economic and political decisions.
8 Frohlich and Oppenheimer (2007, p. 363) discussed an alternative solution to reach stability by introducing
a “culturally accepted conception of justice within a utility function.” Although this seems promising, I
will follow the standard response of restricted procedures respectively preferences.
9 Braumöller (2003) argued that theories which posit complex causation, or multiple causal paths, pervade
the study of politics. He offered an extensive selection of examples for various fields of political science.
10 Sec. 2.2 provides a multitude of further examples.
11 Krehbiel et al. (2004, p. 251) pointed out that “social choice theoretic results clearly demonstrate the ana-
lytical significance of dimensionality.”
3
1 Introduction
ant questions, a variety of individual or collective actors, etc. Many theorists have dis-
cussed the possible impacts and consequences in detail (e.g., Denzau and Mackay, 1981;
Kramer, 1972). Nevertheless, nearly all empirical analyses assume for reasons of sim-
plicity that preferences are separable (e.g., Blundell, 1999; Schneider et al., 2010). This
demanding assumption requires that “preferences on every issue and set of issues are
independent of - or, can be separated from - the outcomes of other issues” (Lacy, 2001b,
p. 240); an assessment that seems rather daring in many situations.
Neglecting conditionalities also outlasts recent improvements. Political scientists have

made rapid progress in measuring policy positions, conducting expert interviews (e.g.,
Benoit et al., 2005; Thomson et al., 2006), computer-aided text analysis (e.g., Laver et al.,
2003; King and Hopkins, 2007) or dynamic ideal point estimation (e.g., Jones et al., 2009;
Martin and Quinn, 2002). This recent work has increasingly tried to include information
on preferential distributions as well as the collection of actor-specific salience12 . This is
a clear improvement compared to focusing solely on an actor’s first preference.13 How-
ever, the existing methods build on the assumption of separable preferences. Thus, an
important aspect remains unspecified.
I do not argue that every research project should implement nonseparability. But one
ought to investigate its possibilities. If one does not take them into account, this aspect
of methodology should be addressed. Previous theoretical analyses have identified an
increased complexity for every decision-making process affected by NSP (cf. Enelow and
Hinich, 1984). This complexity leads to difficulties in model operationalization and is the
reason this topic has only seen little empirical examination. Moreover, it questions the
stabilizing properties of institutional arrangements identified under the assumption of
separable preferences. This has implications even for the most basic principles of institu-
tional analysis.
The objective of my study is to close this gap between theoretical and empirical research.
Therefore, I analyze the influence of NSP on individual and collective decision-making
empirically. Firstly, I investigate whether the existing theoretical considerations on NSP
and their impact are correct. Secondly, I look for additional, yet unknown, patterns to
supplement further arguments to the discussion on nonseparability. This complements
traditional research as “almost all existing work in analytical political theory takes a be-
havioral perspective, seeking to explain how people act [and] why societies take the polit-
ical decisions they do” (Hinich and Munger, 1997, p. i). It allows me to draw conclusions
about the impact of NSP on actors’ behavior and, therefore, on how decision-making
processes in politics can be better understood.
12 The concept of salience in theoretical models denotes the importance an actor attaches to the issue in
question (Hinich and Munger, 1997).
13 Looking beyond political science, Russ (2011, p. 211) summarized contemporary contributions in the dis-
ciplines of biology, psychology, finance and economics with the mnemonic “complexity science has come
into being.”
4
1.2 Previous research
The basic principle of nonseparability states that if a connection exists between individ-
ual parts of an overarching theme, those parts should not be separated. Far from being
an innovative concept, the idea of nonseparability has existed for quite some time. It has
also been discussed for its relevance in analytical political research (e.g., Strom, 1990). In
this section I review the previous contributions to and applications of the nonseparability
principle.14 My interest was not to focus only on theoretical work, but empirical contri-
butions have been rare. Rather than just referring to existing work, I will also point out
the missing elements of NSP and include possible improvements. I argue that investigat-
ing NSP can enhance scientific research in many subjects: from referendum and survey
research, to organizational theory, and institutional analysis. By mentioning the potential
fields of application so prominent at this point, I hope to contribute to the benefit of fu-
ture research in analytical politics. The consequences of neglecting nonseparability differ
from field to field, but in general NSP may distort results and lead to wrong conclusions.
The field which has explored most explicitly the issue of nonseparability is that of survey
research. This holds for its measurement as well as for its explicit designation. There-
fore, I start my literature review with contributions in this field. The bulk of literature
on nonseparability, however, has focused on its potential for strategic manipulation (e.g.,
Holt and Anderson, 1999). This subject comprises innate political fields such as admin-
istrative decisions, international relations, elections and referendums. After discussing
this work, I look at the existing (empirical) research on the interplay of procedural rules
and NSP. This covers the fields of organizational politics as well as institutional analysis.
The inherent logic of nonseparability is clearly linked to central topics of those subjects,
e.g., delegation or decentralization. Both concepts are concerned with the separation or
unification of decision-making competences. But nonseparability is not only relevant in
institutions, it exists within preferences. Therefore, I complete the review by discussing
(mainly theoretical) contributions on the modeling of utility functions of individual as
well as collective actors.
SURVEY RESEARCH
Surveys rely heavily on the measurement of individual preferences, and it is essential

to measure them correctly and precisely. The problem of conditional preferences is dis-
cussed in this field in the context of question order effects (Schuman and Presser, 1996).15
These effects show that the answers to identical questions vary systematically because
of the way in which they are ordered. Most contributions look into the effect of rear-
ranging questions. For example, Malecki and Gabel (2007) investigated a change in the
14 Chap. 2 elaborately discusses the concept of nonseparability including a multitude of examples. In partic-
ular, Sec. 2.4 explores the theoretical foundations.
15 Question order effects are separated into “part-whole” and “whole-whole” effects. The first refers to
questions with super- and sup-ordinate parts (e.g., “Should taxes be increased?” - “On what should the
surplus be spent”) and the second to questions on the same level (Willits and Ke, 1995).
5
1 Introduction
Eurobarometer survey16 and Siminski (2008) pointed out that question order effects are
not necessarily limited to single questions. Instead, “question order has the potential
to cause systematic positive or negative bias on responses to all questions in a battery”
(Siminski, 2008, p. 477). The literature explains this by default with framing, consistency
and salience effects (Schuman and Presser, 1996, p. 12ff).17
Lacy (2001a,b) was the first to discuss NSP in relation to surveys. The measurement of
separable preferences can be limited to the actors’ first preference. Measuring nonsepa-
rability requires an evaluation of preferences according to several specifications, making
this a disproportionately more complex task (cf. Sec. 2.4.3). Nevertheless, Lacy (2001a,b)
investigated nonseparability in respondents’ preferences in the United States (U.S.) when
answering survey questions. He demonstrated that respondents’ preferences with regard
to income taxation depended on crime prevention policies; preferences on environmen-
tal pollution depended on environmental regulation; preferences on defense spending
depended on social spending; preferences on immigration policy depended on the con-
stitutional status of English being the only official language in the U.S.; and so on. Thus,
NSP exist in real world political issues18 , and they might affect how people vote, form
judgments or make decisions. The studies clearly showed that common survey ques-
tions about first preferences alone are insufficient when dealing with NSP. Disregarding
conditional preferences at the individual level increases the risk of mistaking a respon-
dent’s conditional response for the genuine preference. This leads to misconceptions at
the aggregate level of analysis.
Bonica (2012) focused on fiscal preferences using an interactive questionnaire. Respon-

dents were asked to adjust single issues spending levels within an overall fixed budget.
the results showed that the preferences on security spending correlate with self-reported
ideology. This is in line with the findings of Lacy (2001a) concerning the interplay of
preferences on defense and social spending.
As the contributions discussed so far agree that NSP exist, they also tried to measure
them. Survey questions are suitable for this task, but NSP are far from being an accepted
standard in this field. Too little is known about their impact or importance. It is also
necessary to look into research in other areas and to bring the respective findings together.
STRATEGIC MANIPULATION
Most of the theoretical literature on nonseparability has focused on the potential for
strategic manipulation when separating or combining the decisions over issues which
influence each other (e.g., Ordeshook, 1986). Already Schwartz (1977) set up a general
16 The Eurobarometer is a regularly and biannual survey conducted on behalf of the European Commis-
sion. It reports public opinions across the EU member states and is accessible at http://ec.europa.eu/
public_opinion/index_en.htm.
17 Cf. Druckman (2004) for an assessment of political conditions (e.g., elite competition, deliberation) under
which framing effects occur.

18 The contribution of Lacy (2001b) clarifies also that dependencies are very well possible across policy fields.
6
model to investigate the separation of a bundle of positions into single issues. The model
focused on compromises across issues in which “vote trading (log rolling) unseparates
separate issues [by] combining positions on distinct issues to form single legislative pack-
ages” (Schwartz, 1977, p. 999). He concluded that vote trading can have advantageous as
well as disadvantageous influence in the attempt to achieve a Pareto-efficient outcome.19
In the case of NSP, and in contrast to a trade, the preference of an actor in one issue
changes with the result of a second issue. Probably the most obvious situation in which
nonseparability occurs is a dependency of spending preferences. Hinich and Munger
(1997) discussed the consequences for two or more public policies when they are subject
to the same budget constraint. Given this constraint, the amount of money allocated
to one policy program reduces the overall funds still available. Thus, the allocation to
one program influences the spending preferences over all remaining policy programs.
Whether they increase or decrease depends on whether the programs are substitutes or
complements.
Linking single components is by far not limited to budget negotiations. Morgan (1984,
1994) applied bargaining theory to international crises and argued that issue linkage
plays an important role in the bargaining process. He identified the incorrect specification
of such interconnections as one possible reason for the theory‘s explanatory shortcom-
ings. In his studies the focus was on three aspects, which so far have not been included
sufficiently: the relative salience of issues, the relative power of actors, “and the possibil-
ity that the linkage issue raises other issues with which it is inversely related” (Morgan,
1990, p. 329). As this conditionality was not accounted for in the original framework, he
judged its setup as “underdeveloped” (Morgan, 1990, p. 312).
In the context of economic sanctions and military coercion, Lacy and Niou (1998a) sup-
plemented the theory of issue linkage. The authors “develop a game-theoretic model of
disputes on multiple issues where disputants may have either separable or nonseparable
preferences” (Lacy and Niou, 1998a, p. 2).20
These various theoretical models of strategic manipulation prove that researchers are
aware of the possible influence of NSP. However, their consideration is far from state-of-
the-art in any field. At least they provide a solid basis for future empirical tests. I hope
that my own study encourages this.
ELECTIONS AND REFERENDUMS
The separation, combination or concatenation of decisions is an important aspect of all

democratic decision-making procedures, e.g., when elections and referendums ask for
19 An outcome is defined as Pareto-efficient or Pareto-optimal if no other alternative is Pareto-preferred
to it. That is, no other alternative can make at least one individual better off without making another
individual worse off (cf. Dougherty and Edward, 2012, p. 656).
20 Interestingly, Lacy and Niou (1998a) did not even regard the inclusion of NSP as their crucial contribution.
Instead, they argued that the “key feature of the model is that states have incomplete information about
each other’s preferences” (Lacy and Niou, 1998a, p. 2).
7
1 Introduction
the electorate’s opinion. It is inevitable that they enable voters to express their preferences
in an undistorted, clear and complete manner.
Benoit and Kornhauser (1991) took a look at assembly elections in which the electorate
had to vote for single candidates. The authors pointed out that, in fact, citizens’ pref-
erences concerning the constitution of the assemblies as a whole, and not the platforms
of the candidates themselves, are essential. To justify the assumption of separable pref-
erences it is therefore necessary to investigate assembly-based and not candidate-based
preferences. For this purpose the authors introduced the term “simply voting”, which
extends the idea of sincere voting, with the assembly level in mind. Yet, they acknowl-
edged that even then “seat-by-seat procedures are efficient or neutral only under extreme
conditions” (Benoit and Kornhauser, 2007, p. 1).
Somewhat more sophisticated are election models for double-member districts. Here,
Lacy and Niou (1998b) investigated electoral equilibria under the assumption of NSP and
simultaneous voting. The inclusion of NSP creates multiple equilibria and favors candi-
dates with extreme positions. These findings were quite different from the then-standard
results and models (e.g., Cox, 1984). The authors emphasized that many researchers rec-
ognize the pivotal role of the separability assumption, but that until their contribution it
was “imposed on preferences in nearly all of the formal models of elections and legisla-
tures” (Lacy and Niou, 1998b, p. 91).
In general, polls are well suited to clarify the impact of interrelated issues. Thus, a simple
and single binary vote (whether in an election, referendum or opinion poll) may not be
enough if the topic in question contains interrelated issues. This would risk mistaking a
conditional response for a genuine first preference (Lacy, 2001a). Conclusions based on
such results are biased and inaccurate.
But do such relations really exist? Carrubba and Singh (2004) argued that individuals
can have quite sophisticated policy preferences incorporating multiple layers. Using the
example of a European Union common defense force, the authors emphasized that ig-
noring “correlated preferences” leads to falsely-specified empirical models (cf. Carrubba
and Singh, 2004, propositions 2 and 3, p. 221). In addition, these secondary effects may
also explain volatility in public opinion polls.21
Often, a referendum asks its participants for assessments on several issues simultane-
ously. Brams et al. (1997) argued that such a voting procedure prevents the complete and
sincere representation of voters’ preference ordering and, ultimately, can cause represen-
tation bias. Their work highlighted the importance of procedure in dealing with NSP
and suggested alternative aggregation and voting rules. The authors discussed approval
aggregation, split aggregation, Borda Count and approval voting. All these rules offer
21 Cf. Zaller and Feldman (1992) for a more comprehensive discussion whether surveys actually reveal pref-
erences and on “methodological artifacts” as question order effects, semantic priming, consistency ef-
fects, framing effects, etc.
8
the participants more options for expressing their preferences:22 approval aggregation
counts abstention at the same time as a yes and a no vote; split aggregation divides ev-
ery abstention evenly across all alternatives; Borda Count asks every voter for all desired
combinations of issues and awards the number of votes according to the listed order;
approval voting enables citizens to vote for all acceptable combinations.23
Lacy and Niou (1994, 2000) had similar research objectives when looking into referen-
dums on multiple binary issues. They were concerned that majority rule may select out-
comes inferior to the Condorcet winning set when voters have NSP across the issues.
The choice may have even been a Condorcet loser or Pareto dominated by other sets.24
Like Brams et al. (1997) they proposed alternative voting procedures like set-wise vot-
ing, issue-by-issue voting and vote-trading. These mechanisms prevent the selection of
inferior outcomes and are facilitated by legislatures.25 This leaves a usual referendums
with the ambiguous characterization that it maximizes “the quantity of participants in
democratic decision-making but minimizes the quality of participation” (Lacy and Niou,
2000, p. 1).
These insights into referendums were contested by allegations that the undesirable out-
comes involve contrived voting situations that would be unlikely to occur in actual elec-
tions. Using computer simulations, Hodge and Schwallier (2006) responded to this accu-
sation by investigating the desirability of referendum outcomes. This robustness check
also validated the negative effects of unaccounted NSP in randomly generated elections.
In addition, the simulations provided support for the claim that certain alternative voting
methods can more accurately reflect the will of the electorate.
ORGANIZATIONAL POLITICS
Previously, I discussed the strategic separating of issues; now, I turn to analyzing decision-
making structures. Both fields are strongly connected as they ask the same question: what
influence is dedicated to the process of decision-making? When organizational units are
confronted with decisions on jurisdiction or subsidiarity they are dealt with by delega-
tion, decentralization or specialization (e.g., Shy, 1996). Each organizational form implies
the separation of decision-making with respect to either a spatial, temporal, hierarchical
or administrative dimension (Bendor and Meirowitz, 2004).
22 None of the rules was designed to deal with the problem of NSP. Rather, they intend to provide a better
opinion assessment of decision-making with multiple alternatives. Yet, nonseparability links otherwise
separate issues thereby generating a plurality of alternatives. For example, in the case of the aforemen-
tioned exquisite dinner the combination of the binary decision for each wine (white or red) and main
course (fish or venison) results in four possible variants of which one is the result. Thus, the voting
schemes can be helpful in the identification of NSP.
23 Hodge (2006) offered a more elaborate overview on alternative voting rules and discusses (non)separable
preference orders in the broader context of combinatorial questions.

24 Such outcomes are called “multiple-election paradoxes” (Xia et al., 2011).
25 Lang and Xia (2009) agreed that to allow voters to express their full preferences on the set of all combina-
tions of values may prevent multiple-election paradoxes. But this is practically impossible as soon as the
number of issues or the size of the domains increases above a few units. By relaxing separability restric-
tions, they identified a practicable compromise. The authors therefore applied a sequential composition
of local voting rules related to the setting of conditional preference networks (CP-nets).
9
1 Introduction
Much of the literature in the field of organizational theory investigates these organiza-
tional tools. Their benefits, implications and requirements are discussed and known in
detail (cf. also Sec. 2.3). The impact of each decision rule is assessed with respect to the
output performance of the respective institution. This search for the optimal organiza-
tional structure is important, e.g., for public administration, the industrial sector and
research institutions (e.g., Olsen, 2007).26 If the separation involves decisions over multi-
dimensional issues characterized by conditional dependencies, this leads to unwelcome
side effects and a loss of efficiency.
A special topic is the growing field of organizational politics, which at its very core is
interested in the effects of political tactics (coalition building, sanctions, ingratiation, etc.)
on organizational functioning (for a comprehensive overview cf. Ferris and Hochwarter,
2011).27 Its two most common approaches are the investigation of the effect of polit-
ical behavior on organizational as well as individual attributes (e.g., work processing
and career success) and to comprehend the reactions of individuals to perceptions of or-
ganizational politics (cf. Witt et al., 2000, p. 42). While the first aims for a quantitative
(objective) research implementation the second views responses as based on an individ-
ual’s (subjective) perception of reality (Lewin, 1936).28 As it holds true for organizational
theory in general, the assessment of behavior must be based on an accurate specification
of dependencies. Otherwise, the conclusions reached are distorted.
INSTITUTIONAL ANALYSIS
Institutional analysis is closely related to organizational theory. Many contributions have

long investigated the trade-offs inherent in every institutional reform debate. Tensions
between stability, efficiency and transparency are at the very heart of every negotiation.
Prominent examples are the federal reform of Germany (e.g., Burkhart, 2009) as well as
the reform of European treaties (e.g., Finke, 2009). Consequences of reallocating agenda
control or decisions powers must be evaluated correctly, otherwise, an assessment of
reforms achieved is not possible.29 Therefore, interdependencies between different levels
of government or political fields must be captured accurately.
26 For the field of politics Moe (1990, p. 39) pointed out that it is “the structural means by which political
winners pursue their own interest”.
27 Vredenburgh and Maurer (1984) provided an introduction into the field of organizational politics by re-
viewing previous concepts. Incorporating them into a coherent process framework for understanding
individual and group political activity the authors synthesized the following definition: “Organizational
politics (a) is undertaken by individuals or interest groups to influence directly or indirectly target indi-
viduals, role, or groups towards the actor’s personal goals, generally in opposition to other goals, (b) con-
sists of goals or means either not positively sanctioned by an organization’s formal design or positively
sanctioned by unofficial norms, and (c) is objective and subjective in nature, involving real organizational
events as well as perceptual attributions” (Vredenburgh and Maurer, 1984, p. 50).
28 Witt et al. (2000) scrutinized the impact of participation in decision-making: does it matter for job sat-
isfaction i) if decisions are reached in consensus with the supervisor or ii) if orders are given without
contestation. Here, satisfaction depends “on a judgment by the individuals as to whether a perceived
behavior is within parameters of sanctioned behavior” (Witt et al., 2000, p. 343) or not. Such judgments
are highly subjective.
29 Shepsle and Weingast (1984, p. 49) demonstrated “that institutional arrangements, specifically mecha-
nisms of agenda construction, impose constraints on majority outcomes.”
10
Interestingly, there are fields of research closely considering dependencies within their
field as well as with adjacent areas. Those are not necessarily denoted with the term
“nonseparability” but research considers the relationships explicitly. The current dis-
cussion on the future of the German pension system may serve as a first example of a
field with strong interdependencies, and of a policy area characterized by various re-
form initiatives. In this area, the campaign for the parliamentary elections in Germany
in September 2013 brought about a new wave of reform proposals. While they comprise
a variety of regulatory and legal aspects, Börsch-Supan and Gasche (2013) conclude that
all propositions are nevertheless just “moving injustice from one social security system
into another” and illustrate this by a tax-financed basic pension which burdens labor
costs and reduces employment. The strong ties between the individual components of
the German welfare state make changing a single part very difficult.30 This points out
that (many) political decisions influence adjacent areas.
If one takes a closer look, the federal system of Germany also serves well as a more in-
depth case.31 The intertwining, i.e., the interdependence, of national and federal powers
in Germany is well-known and well-researched (Wachendorfer-Schmidt, 2005, p. 11). Its
disentanglement was expected to increase the efficiency as well as the transparency of
politics (Zohlnhöfer, 2008, 2009). The reform aimed at key areas of the political order,
such as legislative power distribution, division of labor within the administration, budget
responsibilities, etc. The reallocation of agenda control was another key part, and related
power struggles were the main reason for the delays and difficulties of the reform (Benz,
2005).
Interdependencies in political systems are not per se good or bad (cf. Wachendorfer-
Schmidt, 2005, Chap. 6, p. 377ff), but they lead to certain behaviors that can be assessed
properly only if one knows of these dependencies and takes them into account (e.g.,
Ganghof and Manow, 2005). This also holds true for reforms, e.g., when the possible
consequences of reallocating legal responsibilities are evaluated. A correct assessment
is not possible without considering the structure of the system. Therefore, interdepen-
dencies between different levels of government or political fields must be captured to
evaluate the mechanisms of politics.
Empirical contributions in the field of institutional analysis concerned with nonsepara-
bility focused on changes in the policy outcome when conditional decisions are split up
by institutional requirement (e.g., separation of power or regional jurisdiction). These
institutional rules range from agenda-control over decision competencies to the general
institutional environment. To that point, nonseparability is just one of many factors in-
fluencing the final outcome as it relates to the overall structure (McCubbins et al., 1989)
or the so called “structure-induced equilibria” (Shepsle and Weingast, 2004). Yet, insti-
30 For a comprehensive overview on the development and current status of the German welfare state
cf. Schmidt (2012).
31 For an up to date overview on general theories about multilevel systems cf. van Deemen (2009) and for a
more international perspective cf. Kaiser et al. (2013).
11
1 Introduction
tutional rules are far from unbiased; they provoke strategic behavior hampering equilib-
rium solutions (e.g., van Deemen, 2009).
A well-known characteristic of U.S. politics is the principle of divided government (Fio-
rina, 1992b). Smith et al. (1999) analyzed voting behavior in the U.S. and focused on the
question of whether the voting decision for the president is dependent on the current leg-
islative majority. Using national and statewide surveys, Craig et al. (2006) had the same
objective. While no study found a decisive answer, the strength of the effect depended
on voters’ level of information (Fiorina, 1992a), i.e., if they knew the current and probable
future majority in the House of Representatives and the Senate.
Strategic voting is an essential part of all modern electoral systems, and the respective
tactical moves may include very sophisticated considerations. Shikano (2009, p. 271) pro-
posed for Germany’s national “mixed-member electoral system” an interaction effect be-
tween voting decisions and party competition.32 He argued that expectations concerning
the national level proportional representation influence the electorate to vote strategically
in the plurality tier. Since the relevant variables were not available through survey data,
he used computer simulations. These simulations were most appropriate in explaining
short-term linkage between the two tiers.
Thiem (2009) investigated agenda-setting and roll-call vote (RCV) requests in the Euro-
pean Parliament (EP). The EP offers a unique case, as a member of the European Par-
liament (MEP) is considered to be an agent to their national party and their European
group simultaneously. Both bodies are able to discipline their MEPs. The author argues
that there are theoretical reasons to question the connection between strong European
group leadership and unity among the factions in roll-call votes. Because disciplining
abilities of the European groups are limited, party unity is more likely a condition for
and not the result of a RCV. Thus, the role of national parties may, so far, have been un-
derestimated. This also implies that RCV in the EP serves mostly as a signaling device
(towards national constituencies) and is not constituted for disciplining purposes.
These contributions show that the influence of the institutional structure or single organi-
zational peculiarities can be misinterpreted if conditionalities are not assessed correctly.
Behavior due to a specific dependency may be attributed to an aspect which is not at all
involved or responsible.
ANALYTICAL POLITICS
Conditionality is not only relevant in combination with one institutional arrangement or

specific agenda-setting33 . Analytical politics covers a wide field of topics in which it aims
to analyze and explain political behavior (cf. Hinich, 2008). It focuses on the identification
of empirical regularities.
32 Cf. Soberg and Wattenberg (2001) for a comprehensive overview on mixed-member electoral systems and
their characteristics.
33 Wilson (2007) provided an extensive overview on the topic of agenda-setting in political science and eco-
nomics.
12
The concept of abstraction is a key aspect of all analytical research as it transforms com-
plex (policy) processes into manageable models. Those models serve as “stylizations
meant to approximate in very crude fashion some real situation” (Shepsle and Bonchek,
1997, p. 9) and incorporate preferences of the actors involved, in the form of utility func-
tions. NSP alter individuals’ utility function, exhibiting effects in all configurations. This
must be considered at an early stage when setting up analytical models containing the
specified institutional arrangement of interest.
A well-known and often used tool in the field of analytical politics is spatial modeling
(e.g., Davis et al., 1970; Krehbiel, 1988). The operationalization of spatial models consists
of formulating indifference curves (for a more detailed discussion cf. Sec. 2.4.2). These
curves rely on correctly specified utility functions, i.e., the correct assessment of indi-
viduals’ preferences.34 If the curves are inaccurate, the results obtained from the spatial
analysis are implausible. Different models can neither be compared, nor crucial aspects
of (political) behavior identified.
In general, models in the realm of analytical politics describe individual actors and their
behavior. But this does not mean that each actor must actually be an individual person
in the real world. Theoretical models allow incorporating administrative bodies, compa-
nies, legal persons or societies as well. The models abstract from the type of actor and
operationalize both individual and collective protagonists as unitary actors (e.g., Baron
and Ferejohn, 1989). Here, Le Breton and Sen (1999) made a first step by discussing social
choice functions35 (SCF) which were “separable only with respect to the elements of some
partition of the set of components and these partitions vary across individuals” (Le Bre-
ton and Sen, 1999, p. 605). Thus, the simplifying separability assumption was abandoned,
at least in part.
Over thirty years ago, Stiglitz (1972) examined the importance of legislation on bankrupt-
cies, take-overs, and the financial policies of companies. He showed that the real-term
decisions of a firm (i.e., the manufacturing sector) are not separable from its financial de-
cisions which depend on its debt-equity ratio. Previous studies had claimed that, if there
is no chance of bankruptcy, financial policy has no effect on the value of the firm, im-
plying that there is no optimal debt-equity ratio. Thus, financial regulation may not only
affect a company’s legal status but also its every-day decisions. During the current global
financial crisis, this insight is more important than ever. It highlights the importance of
regulators in financial markets and their scope of action (cf. Doina and Nicolae, 2011).
Finke (2009) argued that governmental preferences with regard to expanding the EU’s ju-
risdiction depend on the decision rules applicable inside the EU’s decision-making bod-
34 While recent contributions increasingly paid attention to preferential distributions and actor-specific char-
acteristics (e.g., salience, Benoit et al., 2005; Slapin and Procksch, 2010) they stick with separable prefer-
ences as default.
35 Le Breton and Sen (1999, p. 605) defined a SCF as “a mapping which associates a social alternative with
every profile of individual preferences defined over a set of social alternatives. The value of a SCF at any
profile is to be viewed as the ’optimal’ or most ’desirable’ outcome in that state of the world.”
13
1 Introduction
ies and vice versa. Whereas some governments preferred further integration only if they
were to maintain their veto power (e.g., unanimity rules in the Council of Ministers),
others preferred further integration only if decision-making procedures were to become
more efficient (e.g., lower voting thresholds). The underlying logic can be summarized
as public goods inhibiting economies of scale, set against individual policy losses due to
heterogeneity. The author tested his hypothesis by applying a two-stage item response
model to data on governmental reform positions.36
Analytical models rely on the implementation of individual utility functions, much in

the same way as surveys rely on the measurement of individual preferences. Its un-
derlying pattern matters for the assessment of behavior. In both cases, only the correct
operationalization establishes the nonnegotiable and fundamental condition for proper
scientific research. Thus, NSP represent not a specific obstacle of a single field of research.
This is also evident in the wide-ranging fields covered in this summary.
1.3 Focus of the present study
In short, this study faces a trade-off. Every contribution to analytical research must strike
a balance between scientific interest and existing resources, between including more de-
tails versus keeping modeling within manageable limits. I focus on the negligence of
NSP. More precisely, I ask whether analytical research should override them and keep
the analysis simple but potentially biased, or include nonseparability together with its
complex modeling requirements.
Whereas the theoretical concept of NSP is relatively old, empirical tests thereof are rare.
Except for the literature discussed in Sec. 1.2, almost all research projects use the simpli-
fying assumption of separable preferences as their default. This may lead to distorted
results and invalid conclusions. Of course, many studies which exclude NSP may be
right in doing so. Yet, the nearly unambiguous exclusion from analytical political re-
search can hardly be justified just because it is traditional to do so.37 In other words, I
do not argue that every research project should engage in measuring or operationalizing
nonseparability. This depends in each case on the specific research question and design.
But scientific research standards can only be maintained when the role and relevance of
NSP have been duly addressed.
Analytical politics dismantles challenging complex political phenomena into feasible com-
ponents, using models as “internally consistent bodies of theory that describe human be-
havior” (Hinich and Munger, 1997, p. 3). Applying these models enables the analysis of
36 Shu (2003) investigated a related field when looking into NSP in public opinion polls and referendums on
European integration.
37 Benoit and Laver (2007c, p. 31) made a similar statement to another quasi-standard of political science
research, the usage of Euclidean utility functions when modeling political decision-making. I will come
back to that discussion in Sec. 2.4.3.
14
intricate questions by reducing complexity through abstraction from reality. If one uti-
lizes this research paradigm and accepts the general principle, the degree of abstraction
still remains controversial. What details can be ignored, which features omitted, which
simplifying assumptions can be made without losing validity? In general, this depends
on whether or not the aspect in question is of significance for the findings of analytical
political science research. My analysis reveals the necessary requirements and knowl-
edge gained with respect to NSP. This enables me to make the significance assessment
based on a sound foundation.
It is far beyond the scope of a single study to look into every individual sub-field of
analytical politics. I restrict myself to a single but central aspect, the formulation of the-
oretical models, which form the very core of analytical politics. Here, their purpose is
to explain collective decisions. Yet, all collective action rests upon individual behavior.
Therefore, the models implement human behavior in a computable way; it goes without
saying that it is essential to model the underlying behavioral concepts properly. Only
then can the results of the empirical analysis be obtained in an intelligible and reliable
manner. Ganghof and Manow (2005, p. 11) pointed out that in addition to the disclo-
sure of behavioral assumptions “the focus [of rational choice] rests on analytically distin-
guishable mechanisms and systematic institutional effects.” To uncover such patterns it
is important to formulate the underlying function, i.e., the formal model, punctiliously.
But the unambiguous assumption of separable preferences, usually without taking other
possibilities into account, is far from appropriate.
The relevance of my argument to consider NSP can be summarized by two key ques-
tions. Firstly, whether and, if so, how much the inference of analytical research suffers
when falsely assuming separable preferences? For example, are policy recommendations
about which decision process should be used (e.g., open vs. closed rule) still valid? Sec-
ondly, whether and, if so, how nonseparability affects individual and collective behav-
ior? Preferences determine the action of every individual. Yet, behavior is also shaped
by institutions (Plott, 1979); perhaps more in the domain of politics than anywhere else.38
Thus, what influence does nonseparability exert under different institutional arrange-
ments?39 To answer these questions, I will use both field and laboratory data, and exam-
ine collective as well as individual behavior.
Initially, I intend to establish the relevance of NSP. For this, a review of previous research
will answer the question of whether the extra expense of considering nonseparability is
worthwhile. I have chosen the field of legislative decision-making, where conflicts over
law proposals are frequently conceptualized as being multidimensional (e.g., Thomson
38 The separation of powers is one of the most fundamental concepts of political theory (de Groot-van
Leeuwen and Rombouts, 2010).
39 Institutions affect behavior in a variety of ways (cf. Shepsle and Bonchek, 1997). For example, democracy
encourages cooperation (Dal Bo et al., 2010), procedural justice determines the evaluation of a given
outcome (Lind, 1988) and information aggregation is strongly related to the applied voting rule (Morton
and Williams, 1999). My investigation will therefore also look into the relationship of nonseparability
with such further aspects.
15
1 Introduction
et al., 2006), but studies in this field rarely engage in a discussion of the potential interre-
lation between issues with respect to actors’ preferences. I therefore begin my empirical
investigation by examining the impact of NSP on the performance of several models of
legislative decision-making. Using empirical data, I endeavor to determine the relevance
of neglected nonseparability. The computations also demonstrate the problematic data
collection at the individual level in the field.
In a second step, I bring the discussion a more comprehensive level. I am interested in
understanding the implications of NSP on individual and collective decision-making in
general. The focus is on universal patterns. Thus, I refrain from using observational data
from only one specific environment. Instead, I turn to the general and universal frame-
work of laboratory experiments.40 My experiment compares individual and collective
votes regarding a problem, characterized by NSP. The records offer insights into the dy-
namics of the decision-making process (cf. Kocher and Sutter, 2005). The decisions are
reached in the group at large or in situations in which the group is split. Comparing
those, I review existing hypotheses and look for still unknown patterns.
This approach underlines very well the interdisciplinary character of this study. On the
one hand, the use of laboratory experiments and rational choice are far more common in
economics (cf. Sec. 4.1). On the other hand, the juxtaposition of procedures is represen-
tative of political science (e.g., Steunenberg, 1994; Gailmard, 2009). Reviewing the liter-
ature of the two fields, Mertins (2008, p. 27ff) concluded that there had been rather little
research on the dynamics of procedures in economics. That may sound harsh, but it just
depends on the field to which field one attributes such excellent contributions as, e.g., the
work of Elinor Ostrom (most prominent Ostrom, 1990).41 Nevertheless, at least compared
to political science the research in “economics has devoted little attention to whether the
type of decision maker [group or individual] matters for economic decisions” (Kocher
and Sutter, 2005, p. 200).
Finally, I bring the different findings together. The consequences for analytical research
dealing with NSP depend on two aspects. Firstly, the effort necessary to incorporate
NSP must be assessed. Secondly, researchers must have an understanding of how much
knowledge is to be gained by expanding their analysis. The final part of this study eval-
uates these aspects and weighs them against each other.
1.3.1 Research design
This study applies rational choice theory. After all, this theory and its assumptions are
the reason for my research in the first place. Models of rational choice commonly em-
ploy utility functions based on preferences to describe human behavior. To specify such
40 This is straight in line with Plott (1991, p. 902) who argued that “laboratory methodology involves a shift
from a focus on particular economies as they are found in the wild to a focus on general theories, models
and principles that govern the behavior of economies”.
41 For an overview of her social science masterpiece cf. Nutzinger (2010).
16
functions requires the application of simplifying assumptions, and a delimitation of the

object of study. Thus, the question - as well as the investigation - of the nonseparability
assumption is inherent to rational choice. Most importantly, current contributions em-
phasize institutional factors and cognitive aspects of behavior explicitly (e.g., Lane and
Ersson, 2000; Simon, 1982).
Rational choice theory aims to explain social behavior, while its objects of study are indi-
viduals (Shepsle and Bonchek, 1997, Chap. 2). This is based on the insight that all social
relations must be derived ultimately from individual decisions (Braun, 2013, p. 164). This
methodological individualism has its limitations, as the aggregated collective preferences
do not necessarily correspond to individual preferences (Coleman, 1990). However, ra-
tional choice focuses on the preferences and beliefs of individuals, and assumes that they
behave rationally and in their self-interest. This means they choose between available
alternatives by ranking them according to their preferences and beliefs. For this purpose,
the approach utilizes “the formal precision instruments of micro-economics, decision the-
ory and game theory” (Holzinger, 2009, p. 544). In order to act rationally, each alternative
is assigned a relative value. Applying a “first generation” version (Braun, 2013, p. 164), it
is possible to make the right choice, even in very complex situations.42
I agree with Morton (1999, p. 24) that “empirical analysis has, can, and should be used
to empirically evaluate formal models in political science.” In this field, rational choice
theory is a methodically disciplining access. Empirical research that uses this approach is
compelled to disclose its basic assumptions about actor behavior. This enables a thorough
analysis of the logical framework and prevents logical fallacies (cf. Bueno de Mesquita,
2004). Schotter (2006, p. 500) argued that “the combination of rational choice models and
empirical support is the only way to gather knowledge in the social sciences”. But ratio-
nal choice theory has often been criticized.43 Rather than providing an in-depth literature
review, I follow up with the main criticisms of the theory: to much attention is given to the
formation of preferences, while institutional contexts are largely ignored. Furthermore, a
series of implausible assumptions and unrealistic constraints shows limited understand-
ing of the human individual (cf. Green and Shapiro, 1994; Taylor, 2006). Nevertheless, it
has achieved wide acceptance in economics in recent history (Becker, 1976; Franz, 2004),
even if it is still very controversial in sociology and political science (cf. Kirchgässner,
2008; Opp, 2004; Vanberg, 2002).44
One particular point of criticism is the traditional assumption that a person can at any rate
order their preferences amongst a variety of alternatives, and chose the best option. The
42 Simon (1993) referred to this degree of complete rationality as “Göttlichkeitsmodell” (i.e., model of divin-
ity). Here, four criteria of rationality must be fulfilled (von Neumann and Morgenstern, 1944): complete-
ness, transitivity, continuity, and independence.
43 Gilboa (2010) offered insights to the rational choice paradigm in general. He also provided an overview
of its development due to criticism of the economic nature of the approach.

44 Holzinger (2009) concluded that in addition to the difference between disciplines a difference between the
German and international approach exists. She demonstrated this by going through the single sub-fields
of political science such as political theory, international relations, etc.
17
1 Introduction
presumption that a person can keep track of all the possibilities to choose from is quite
strong (Sen, 1997).45 The first fundamental contributions (e.g., Arrow, 1950; Niskanen,
1971; Olson, 1965) are today an integral part of political science literature.46 But follow-
ing its strong presumptions, these early models could not explain altruistic or collective
behavior within the rational paradigm. Fortunately, since the 1950s and 1960s a growing
amount of research has established a much broader framework (e.g., Kahneman et al.,
1982; Sen, 1995).47 While some years ago Frohlich and Oppenheimer (1996, p. 118) argued
that “concern for fairness [are] particular problematic for economic theory”, nowadays,
models can incorporate social concerns (e.g., Ockenfels, 1999). The parsimonious setups
were continually expanded, which made rational choice into a much more hard-headed
tool when investigating behavior and decision-making (Wandling, 2012).
Consideration of psychological research has distinctly improved the realism of analytical

research (e.g., Quattrone and Tversky, 1988). Researchers now are able to incorporate
into their theories the idea that “it is sometimes misleading to conceptualize people as at-
tempting to maximize well-defined, coherent, or stable preferences” (Rabin, 1998, p. 1).48
In this context, Schotter (2006, p. 500) concluded that “rational choice theory does not pro-
vide us with the truth about human behavior but rather with a very compelling platform
with which we can seek the truth.” Undeniably, rational choice theory is neither always
the proper nor the only possible theoretical framework for scientific research.49 But ac-
cording to Schotter (2006, p. 507), it is the “theory that offers us a platform to understand
what is failing in our assumptions and to change those elements without throwing the
rest of theory away.” Step by step, the theory became more realistic and less idealized by
incorporating previously omitted details (Ariely, 2009).
NSP have been disregarded for a long time even though an astonishing amount of work
has been done in two strands of analytical research. Recently, political scientists have
begun to focus on the issues of causal complexity.50 These studies have improved the
capacity to empirically test for conditional effects (respectively “causation”, cf. Brady,
2008; Gerring, 2005), and to disentangle the sufficient and necessary conditions for the
45 According to Sen (1977), a person who depicts these abilities would have to be completely rational. But,
even then an individuals’ rational or egoistic action does not per se lead to a general equilibrium. Still,
this rationality means that the person would not reveal inconsistencies in their choice behavior. On the
other hand, they would not be able to consider the total variety of alternatives carefully, thus, leading to
the conclusion that the “purely economic man is indeed close to being a social moron” (Sen, 1977, p. 336).
46 Braun (2013) discussed these first contributions to rational choice theory in detail.
47 Mitchell (1999) reviewed in detail the then valid tenets and methods of political science from 1950 to 1970.
48 In this study I will discuss current approaches which take into account that individuals are not purely
selfish (for a literature overview cf. Sobel, 2005). This development is not entirely contrary to the view
of classical economists. Following Smith (1759, p. 1) how “selfish soever man may be supposed, there
are evidently some principles in his nature, which interest him in the fortunes of others, and render their
happiness necessary to him, though he derives nothing from it, except the pleasure of seeing it.”
49 Concerning alternative methods, King et al. (1994) offered a comprehensive volume on scientific inference
in qualitative research.
50 Pearl (2009) discussed the paradigmatic shift that was necessary to enable the causal analysis of multi-
variate data. In short, he concluded that the data or distribution of data alone is not enough. Causal
questions require some knowledge of the data-generating process (DGP).
18
occurrence of political events (e.g., Braumöller and Görtz, 2000; Brambor et al., 2006; Ra-
gin, 2006). At the same time, increasing progress has been made to operationalize spatial
models of legislative politics (e.g., Laver and Schilperoord, 2007; Martin and Quinn, 2002;
Slapin and Proksch, 2008).
The improvements in both fields laid the foundation for enabling the examination of
nonseparability in an exact and precise manner. By no means should these contributions
now be ignored. Rather, future research should be adjusted to include the consideration
of NSP. This follows Ostrom (2006, p. 8) who advocated that “one does not just toss out a
theory that has proven valuable in many settings because it does not work well in others.
Many efforts to broaden the theory are well underway, and it will continue to be usefully
employed to address many interesting questions in competitive situations.”
1.3.2 Why a laboratory experiment?
I investigate the effect of nonseparability on individual and collective decision-making

using experimental methods. Krehbiel (1986, p. 547) already summarized the obstacles
in assessing collective decisions: preferences are difficult to measure, institutions are re-
markably complex, and strategies are highly diverse. The measurement of NSP is es-
pecially challenging (cf. Sec. 2.4) and the same is true for operationalizing it. But in an
experimental environment the preferences of the subjects can be induced correspond-
ing to the research question through monetary payments (McDermott, 2002, p. 326). The
participants are remunerated for their participation and their performance in the experi-
ment. The amount of the payment depends on their decisions made during the session.
By defining the payoffs the experimenter obtains control over the incentive structure of
the participants.51 It is possible to set up the (nonseparable) decision problem in the way
most applicable to the research interest (Webster and Sell, 2007, Chap. 9). In other words,
I expose the participants to a decision problem affected by nonseparability and observe
their reactions. This saves me from starting an onerous search for such situations.
Experiments enable the analysis of each step of decision-making through monitoring ap-
proaches (Ordeshook and Winer, 1980, p. 730). This is often not possible using observa-
tional data, which yields only the outcome of a decision-making process (Schnapp et al.,
2009). Therefore, the behavior of every individual that takes part in a collective decision-
making can be scrutinized. The possible juxtaposition of individual and collective lev-
els allows the disentanglement of additional research questions (Halfpenny and Taylor,
1973).
Going into more detail, I use a computer-based laboratory experiment. Compared to a
field experiment, the high degree of environmental control enables me to secure a high
51 The assumption that preferences can be induced by financial incentives follows the axiom of “nonsatia-
tion”. This axiom states that individual utility is a monotonically increasing function of the monetary
reward (Smith, 1976; Smith and Walker, 1993). In other words, the participants will act according to the
given earning opportunities.
19
1 Introduction
level of internal and construct validity (Morton and Williams, 2010, p. 192ff). This em-
phasis on validity is important, because NSP may appear to many people as quite an
abstract feature. Furthermore, collective decision-making involves interaction between
subjects (e.g., Butler and Camerer, 2005, p. 12ff). The laboratory facilitates the control of
communication (either allow no communication, or prescribe how to communicate) and
the necessary assurance that all subjects are taking part at the same time (Cason and Mui,
2007).52 Because of these characteristics, laboratory experiments have matured over the
last two decades and achieved recognition as an approved method for the validation of
theoretical models (Palfrey, 2009).
The laboratory allows me to induce NSP in a controlled environment, but comes with
the usual trade-off between internal and external validity (Morton and Williams, 2006,
p. 8ff). The main disadvantage of experiments is their low external validity, specifically
the poor generalizability of the results.53 The approach should thus not be used for single
case studies, but for “theory development, testing and refinement” (McDermott, 2002,
p. 341).54 This matches the aim of my research, as I scrutinize the effects of wrongly
specified utility functions. The main concern of my experiment is to examine existing
but untested hypotheses empirically (cf. Sec. 2.5). In addition, the excellent monitoring
opportunities enable the search for behavioral patterns so far not discussed in relation
with nonseparability. This supplements prominently the consideration of their relevance.
Already Smith (1973, p. 3) pointed out that the “concepts of utility are not only basic
to microeconomics theory; they are also basic to experimental methodology in general.”
Thus, rational choice theory and the experimental method complement each other. Fur-
thermore, I can rely on a rich tradition of experiments (cf. Sec. 4.1.2) as well as other re-
search designed to study collective decision-making in general, and committee decisions
in particular (e.g., Barbera et al., 2005; Huitt, 1954; Keiding and Peleg, 2001). My work
is part of a body of contemporary political science research to which the contribution of
experimental methods is still growing (cf. Morton and Williams, 2010).
1.4 Outline of the study
The study is divided into three main parts. The first part clarifies the concept of NSP
and emphasizes its relevance. In Chap. 2 various short examples and simple case studies
demonstrate the existence of nonseparability in everyday situations. The chapter also
covers extensively the (previous) theoretical modeling of NSP. In addition, I add further
52 In addition, it is much easier to observe which person actually participates. In online surveys always an
element of uncertainty remains about who has completed the questionnaire (cf. Eckel and Wilson, 2006).
53 Morton and Williams (2010) pointed out that experimental findings need to be confirmed by additional
contributions. For example, further experiments with variations in design, subject pool and treatment.
54 Schram (2005) provided an in depth discussion of the artificiality criticism on laboratory findings. He
pointed out that it is necessary to distinguish between various goals of experimental contributions and
that, e.g., theory testing is affected by this in a completely different way than theory development.
20
1.4 Outline of the study
arguments to the theoretical concept that will be scrutinized empirically in later chapters.
Chap. 3 presents concrete evidence for the significance of NSP. Here, I conduct an empir-
ical study on the influence of NSP on (spatial) models of legislative decision-making. As
a data base I use the widely known DEU project (Thomson et al., 2006). The chapter
exemplifies the shortcomings of traditional research methods for analyzing nonsepara-
bility. As empirical research on NSP is so far still sparse, it proves the necessity of further
investigations.
In the second part, I investigate the impact of NSP on individual and collective decision-
making within the framework of different institutional arrangements. As the measure-
ment of separable preferences itself is hard enough, to include conditional dependencies
is beyond the manageable scope of traditional research techniques. Thus, the four chap-
ters of this part focus on a laboratory experiment. Chap. 4 lays out the experimental
design in detail. It also contains a review of laboratory research in political science, and
places the current experiment into that context. The respective sections elaborately de-
scribe the design decisions made as well as their implementation. Chap. 5 presents the
results of the experiment on the aggregate level. I investigate the influence nonsepara-
bility has on the collective outcome. The results are then further scrutinized in Chap. 6,
where I turn to the individual level. I analyze the single votes of every committee mem-
ber to identify the determinants for the individual choices. Subsequently, Chap. 7 reports
the results of a post-experiment survey. This adds insights from the participants’ percep-
tion of the experiment. The questionnaire surveys the subjects about their decision rules
and important characteristics of alternatives they have chosen.
The third part concludes my study and is concerned with the consequences for scientific
research when encountering NSP. It links the previous theoretical arguments and the em-
pirical results. The rational choice framework prescribes ambitious requirements when
condensing a complex reality into theoretical models. Those are even more demanding
when dealing with nonseparability, but still the capabilities of models are limited. I dis-
cuss this interrelation in detail in Chap. 8. The chapter merges my various findings and
offers guidelines for whether and how NSP should be taken into account. It also places
my contribution into the current state of scientific research. Finally, Chap. 9 concludes
the study with a summary of the previous chapters and an overview of possible future
research.55
55 Overall,
a multitude of computations was necessary to achieve this varied research design. Sec. A.1 pro-
vides an overview of all software used in this study and offers a short description of their application.
All programs are listed including their version numbers and relevant functions.
21
1 Introduction
22
2 What are nonseparable preferences?
In Chap. 1 I introduce the central idea of NSP. Based on this lead-in, the following sec-
tions put nonseparability into a broader context. This is necessary to substantiate its
significance. Many readers might now expect a highly technical discussion of a generally
rare phenomenon; however, NSP are quite the opposite. They are not at all rare. That is
the first point I make in this chapter.
In the following I illustrate plainly the underlying idea of NSP. For this purpose culinary
examples, as in Sec. 2.1, are widely popular (Lacy, 2003). Moving from hypothetical to
empirical examples I depict short case studies verifying the nonseparable character of ac-
tual political problems in Sec. 2.2. Those include various fields of political science, such as
public opinion surveys or legislation procedures. The illustrative examples were chosen
because of their high public salience (in Germany) and the different administrative levels
of government (European, federal, state and local).56 All these instances show that NSP
are highly relevant in the context of organizational structures, i.e., decisions about delega-
tion, decentralization or specialization. I therefore summarize the implications for these
administrative instruments separately in Sec. 2.3. Next, Sec. 2.4 provides an overview of
the various theoretical contributions to the concept of NSP and looks thoroughly into
its theoretical foundations. Finally, Sec. 2.5 brings the different arguments together, and
points out the consequences of neglecting NSP. Those are tested empirically in the fol-
lowing chapters.
2.1 The general principle
A short and simple definition of NSP reads as follows: An actor’s preferences over two
issues are considered nonseparable if the outcome on one issue alters their preference
ordering over another issue and vice versa (Enelow and Hinich, 1984, p. 18ff). Also, in
case each issue is decided separately, an actors’ utility depends on the combination of
both decisions. While this shortcut is rather brief, there is a lot more to be said about
NSP. They are not far away from decisions we make every day.57
56 Admittedly, one reason was also their local proximity to my affiliation at the University of Heidelberg.
57 Dixitet al. (2009) argued that people constantly interact. In their view they “play games of strategy all the
time” (Dixit et al., 2009, p. 3). Here, the term strategy refers to situations which “are distinguished from
individual decision-making situations by the presence of significant interactions among players” (Dixit
et al., 2009, p. 41). Such real life interactions may well be affected by NSP.
23
Beginning with such an everyday decision, imagine three couples spending their holi-
days together in a rented apartment. One evening, the couples decide to have dinner
together. How will they organize their meal? One option is that all six go shopping to-
gether. This, however, would entail a time consuming exercise in deliberation inside the
mall. Hence, they will resort to delegation, which leaves several options. For example,
one of them could be appointed chef. Depending on her or his skills, this may or may
not be an excellent deal for the remaining five. Another option is that one couple buys
the food for the main course, the second is responsible for the drinks, and the third cou-
ple is responsible for the dessert. Or maybe the men could get the food and the women
the drinks. The problem is that the participants’ utility depends on the combination of
drinks and food. Imagine that all shopping activity takes place simultaneously (to maxi-
mize the time spent together at the beach). Assuming that communication is impossible
(the thick walls of the mall are blocking the mobile network), both shopping delegations
face a high level of uncertainty with respect to the others’ choice. In a worst case they
may end up with sweet white wine to venison. How does each of the delegates handle
this uncertainty? Given the structure of the “game” (both delegations make their deci-
sions by simple majority), even the knowledge of the others’ preference does not solve
the underlying uncertainty problem, i.e., there is no equilibrium in pure strategies.
The daily commute to work is another common case to exemplify NSP.58 Imagine that as
a commuter you have two options, going by car or going by bike. In general you prefer
going by car, as it is more convenient and faster. This is your everyday first preference.
Yet, due to an oil crisis, gasoline gets rationed. The gas station in the neighborhood is
already completely out of gas. Now, of course, the bicycle is the preferred means of
transport. Otherwise you would end up sitting in a “dry” car. In this example, gasoline
and car are complements. Therefore, they ought not to be separated.
Moving the set up closer to a political environment, consider a case in which the Ministry
of Transport decides over several alternative routes for a new railway line connecting two
cities. At the same time, the environmental administration discusses a reform of its “Bio-
diversity Act”. The exact provision of this act will affect the calculation of expenses for
each alternative railway line (and thus the Ministry’s preferences). Add to this constel-
lation a construction authority planning to re-regulate the security standards of railway
tunnels. Under these circumstances the cost-benefit calculation for the competing routes
is highly uncertain. Other examples include the dependency of preferences for delegating
competencies to an agency on the distribution of power in the agency’s executive board.
Another scenario to consider involves preferences to liberalize certain industries that may
be conditional upon regulatory safeguards and financial compensations intended to pro-
tect the national champion against the adverse effects of competition.
As mentioned before, there are various fields in which one can find dependencies. This
prevalence leads to the question how rare NSP are at all? Lacy and Niou (1994, 2000) con-
58 This example is borrowed from Ted Bergstrom and his lecture on separable preferences (Bergstrom, 2011).
24
2.1 The general principle
cluded that the separability, and not nonseparability, of preferences may be considered
a rather restrictive assumption.59 Thus, NSP should resemble the standard assumption.
Consider the choice over two dichotomous issues. Theoretically, we could encounter 4
different issue combinations resulting in 24 possible preference orderings. Of these, only
2
8 can be derived by assuming separable preferences, excluding 3 of possible preference
orderings. This is shown in Tab. 2.160 .
Table 2.1: Separable and nonseparable preference orders

EXPLANATORY NOTE
The table lists the possible preference orders for two dichotomous issues. A rule of thumb allows us to determine if an
order is separable or not. For two dichotomous issues a preference order is separable if and only if the most desired and
the most disliked issue combination form exact opposites as, e.g., YY and NN or YN and NY.
Separable orders Nonseparable orders
1. YY > YN > NY > NN 1. YY > NN > NY > YN

2. YY > NY > YN > NN 2. YY > NN > YN > NY
3. NY > NN > YY > YN 3. YY > NY > NN > NY
4. NY > YY > NN > YN 4. YY > YN > NN > NY
5. NN > YN > NY > YY 5. NY > YN > NN > YY
6. NN > NY > YN > YY 6. NY > YN > YY > NN
7. YN > NN > YY > NY 7. NY > NN > YN > YY
8. YN > YY > NN > NY 8. NY > YY > YN > NN
9. NN > YY > NY > NY
10. NN > YY > YN > NY
11. NN > NY > YY > NY
12. NN > YN > YY > NY
13. YN > NY > NN > YY
14. YN > NY > YY > NN
15. YN > NN > NY > YY
16. YN > YY > NY > NN
In the general case, we encounter 2 p issue combinations and (2p)! possible preference or-
ders for p dichotomous issues. Remarkably, the percentage of orders assuming separable
preferences already falls below 1% if p ≥ 3 (Brams et al., 1997, p. 5).61
In addition to arguing that NSP are not rare but widespread, a clear definition also has
to explain what NSP are not. Nonseparability should not be confused with elementary
cause-effect relationships, in which an exogenous alteration causes an adjustment of a re-
lated issue. For example, the reduction of a budget leads to less spending. With nonsepa-
rability, the restrictions (e.g., the budget constraint) and the resulting adaptation (e.g., the
spending decisions) are endogenously determined. They need not necessarily be made
by the same actor, but they are part of the same decision-making process.
59 Bradley et al. (2005) looked for differences in the preference ordering of subjects when alternating between
discrete and continuous space. In both cases separability implies for multidimensional alternatives that
the preference orderings are independent of the outcomes on all other criteria. The authors argued that
in economics and when using spatial models “it is natural to assume that alternative sets are continu-
ous. But in multiple elections they are naturally discrete; in simultaneous referendums, they are binary”
(Bradley et al., 2005, p. 335). They stated that in continuous context separability is a very strong property
whereas in the discrete case it imposes much weaker restrictions on an ordering. However, common set
operations which preserve separability (cf. Gorman’s theorems, Gorman, 1968) never apply in discrete
space.
60 This tabular illustration is based on Brams et al. (1997, p. 30, table 1).
61 3 dichotomous issues can be combined in 8 possible ways: 111, 110, 101, 100, 011, 010, 001 and 000. These
8 combinations can be arranged in 8! = 40320 orders of which only 384 can be derived by assuming
separable preferences.
25
2.2 Introductory examples
The following examples clarify NSP further. Moving from hypothetical to empirical
cases, I choose short studies verifying the occurrence of NSP in actual “real-world” pol-
itics. Please note that these examples serve an illustrative purpose. None of them re-
sembles an all-encompassing case study. The background research has been conducted
thoughtfully, but the main goal was to highlight the prevalent nonseparabilities. Thus, I
include literature for further reading in all studies.62 The examples discuss various levels
of government, and range from the European over the national and federal to the local
level of politics.
The multitude of examples illustrates the commonness of NSP. In addition, they also
highlight specific aspects of the nonseparability concept in concrete terms. I start at the
European level, presenting a first case documenting political dispute characterized by
NSP. This illustrates plainly the multidimensionality of legislative issues, which has to be
considered when assessing the corresponding policy decisions. The same holds for the
second example, which is located in the national realm of politics. In my report I focus
on the privatization debate on Deutsche Bahn, specifically on two sub-questions in order
to make their interrelation clear; i.e., the starting point for nonseparability. These first
two instances clarify the relevant reasoning and preference dependencies when encoun-
tering NSP in the domain of politics. The third example illustrates the consequences of
badly coordinated policy decisions. More precisely, the lack of agreement on interrelated
issues leads to less than optimal social results, as in the reform process of the German
Bundeswehr. This pattern is a very important aspect, as it introduces the interrelation of
institutional structure with nonseparability discussed in Sec. 2.3. A prominent part of my
study is devoted to the question of how one can assess and measure NSP (cf. Sec. 2.4.6).
The final two examples take on this problem. A survey on the future of the German
pension system reported (by mistake) NSP for a large share of the German population.
Looking more closely at this study, I disclose the problems of a premature assessment.
To further strengthen the connection to reality, I investigate an actual referendum con-
ducted in Heidelberg. This local politics example highlights potential problems when
looking for interrelations between issues through public polls.
2.2.1 The European Union regulation on chemicals
In its 6th Term (2004-2009) the EP enacted the regulation for registration, evaluation, au-
thorization and restriction of chemicals (COM(2003) 644, further REACH). The regulation
sets out the obligations of the industrial sector to assess the risk of chemicals. The indus-
trial sector is obligated to limit the consumption of perilous substances and to substitute
them with less hazardous alternatives if possible. Also, consumers must be allowed to
62 As all studies use secondary data I include also references to the original sources of the data.
26
access the appropriate information. The changes and improvements made must then be
documented for revision by the European chemicals agency (ECHA).63
The protocol of the original EP debate on the regulation (which took place at the 17th
November 2005) offers a detailed insight into a political dispute characterized by NSP.64
The discussion of REACH had been highly ideologically driven right from the begin-
ning65 and remains a controversial issue (DBT, 2012). The ideological dimension in par-
ticular made it difficult to reach an agreement. In addition, many single components of
the regulation led to one of the most complicated legislative processes ever.66
Firstly, the parliament had to decide what chemicals would be included in the regula-
tion. Secondly, it had to decide on how extensive the necessary documentation was to
be? The protocols of the parliamentary debate and associated committees state the va-
riety of issues that needed to be considered (e.g., Mann, 2005): environmental protec-
tion, consumer protection, competitiveness of industrial sectors, privacy and patent law,
sponsorship of research and development, cost estimations, extent of information gath-
ering and disclosure, etc. These issues are interrelated in many ways. They inhibit var-
ious trade-offs between different objectives as environmental protection vs. commodity
prices, cost absorption by firms vs. competitiveness of the affected industries, disclosure
of trade secret vs. information rights of consumers, etc. It is no wonder the regulation led
to a lengthy and controversial debate. The final compromise was seen by many MEPs as
a very well-balanced scheme. It would achieve a durable “balance on this delicate, com-
plex and controversial decision" (Statement of Lord Bach speaking for the UK presidency,
EP, 2005)
This example of a parliamentary debate emphasizes the possible complexity of legislative
decision-making. The dependencies led to an incredibly diverse consultation process.
Research investigating the field of legislation has to consider NSP in case such multidi-
mensional law proposals are analyzed. It is incorrect to only examine single issues. This
would pose a threat to the inference one draws from empirical results, which would be
biased, for example, when identifying decisive aspects of the debate or evaluating how
important the various aspects were to the different political groups.
2.2.2 The privatization debate on Deutsche Bahn
Another simple example for conditionality of political preferences is the privatization

debate on Deutsche Bahn (DB). The DB was founded in 1994 and is the largest rail trans-
63 For more information on REACH please consult the information archive of the “Directorate-general of
the European commission for enterprise and industry” at http://ec.europa.eu/enterprise/sectors/
chemicals/reach/how-it-works/index_en.htm (accessed September 2, 2012).
64 The protocol is accessible at http://www.europarl.europa.eu/sides/getdoc.do?pubref=-//ep//text+
im-press+20051116ipr02381+0+doc+xml+v0//en (accessed September 2, 2012).

65 Cf. statement in the EP (2005) of Alexander Graf Lambsdorff from the alliance of Liberals and Democrats
for Europe (ALDE), MEP of the 6th EP and deputy chairman of the German Free Democratic Party.
66 Cf. statement in the EP (2005) of Thomas Mann from the European People’s Party-European Democrats
(EPP-ED), MEP of the 6th EP and REACH Rapporteur of the Employment and Social Committee.
27
port and railway infrastructure company in Central Europe. It is structured as a group

of companies for specific sectors such as, e.g., passenger transport, railway infrastructure
and logistics (Radke, 2011). Until today, the Federal Republic of Germany retains all the
shares. Since 2006, there has been an ongoing discussion about potentially privatizing the
company (Brunnhuber, 2006). The possibilities discussed ranged from a sale of shares of
the existing (vertically) integrated group, to a complete sale of the privatized transport
companies after the separation of the transport and rail network. Most economic studies
called for a separation of railway infrastructure and railway operations. A special report
of the Monopolies Commission advocated that the railway infrastructure constitutes a
natural monopoly and should therefore remain in state ownership (Basedow et al., 2006).
On the other hand, the full privatization of railway transport is economically desirable.
Bogart (2009, 2010) and Lalive and Schmutzler (2008) underlined that competition is the
decisive factor for growth, quality and reliability in passenger transport by rail. It is ex-
actly this competitive aspect that would be lost if the initial public offer (IPO) were to
yield a vertically integrated enterprise. By further retaining the railway network, the DB
would possess numerous discriminatory possibilities (Basedow et al., 2006, p. 27). Oppo-
nents of the IPO referred to the bad example of railway privatization in the UK and the
massive “GB rail efficiency gap” (McNulty, 2011, p. 72).
In 2007, the government of Christian and Social Democrats had agreed on a concept
called “holding model”: The DB would be preserved as an integrated group, and yet
investors would be allowed to participate. Yet, this would only apply to the transport and
logistics sector of the companies. However, this plan was abruptly stopped by the global
financial crisis (DPA, 2011b). At first this was only meant to be a temporary measure. But
it currently appears as if the IPO is postponed into the far future. Tedious technical issues,
a lack of reserve vehicles and weather problems have also hampered the IPO intentions
(DPA, 2011a).
While the debate has not yet found a satisfactory end, it is still very well suited to clarify
NSP. One aspect in the political debate focused on privatization in the strict sense, i.e., the
sale of state shares of the corporation through an IPO. This is an idea strongly rejected
by the political left-wing (i.e., the parties The Left and The Greens). Others associated
with this decision on privatization the question of how to handle a separation of railway
infrastructure and railway operations. The Liberals and Christian Democrats supported
the IPO of the DB only if the separation of infrastructure and transportation would secure
sufficient competition.67 The Social Democrats, however, preferred privatization only if
the DB would continue to exist as an integrated group.68
The point where NSP come into play is marked by the expression “only if ”. So, we ob-
serve no NSP for the left-wing parties. They rejected privatization of the DB no matter
67 Press release of the Liberal parties’ presidium (FDP, 2007): “The separation of network and transport is
essential.”
68 Cf. the party program of the Social Democrats for the general election 2009.
28
what. Yet, the phrase “only if ” makes the conditional binding of the two components
very clear for the other parties. While the Social Democrats rejected privatization if the
DB were to be split up, they advocated the IPO if the DB were to be preserved as inte-
grated group. Here, the first preference on one issue (privatization) changes depending
on the outcome of another issue (separation). This corresponds perfectly with the def-
inition of NSP by Enelow and Hinich (1984, p. 18ff). Finding a solution to this kind of
political task is not possible by only focusing on one of the aspects. The next example
shows the results if politicians attempt it nevertheless.
2.2.3 The reform of the German Bundeswehr
In early 2010, the German Minister of Defense, Karl-Theodor zu Guttenberg, initiated an

analysis to identify the strengths and weaknesses of the German Federal Armed Forces
(Bundeswehr, further BW). The study was carried out by a commission, which recom-
mended a comprehensive restructuring of the armed forces. The proposed reform should
adapt the defense resources to Germany’s current and future security challenges (Weise
et al., 2010). In October 2010, zu Guttenberg presented to the public different models of
how the future structure of the BW could look like. Two main features dominated the
proposed reform: the abolition of the compulsory military service, and the closure of
various military sites throughout Germany.
Compulsory military service has been impregnable for a long time in Germany. It was
seen as an important link between civil society and the military (Frevert, 2001). Yet the
commission concluded that, to meet the current and future tasks of the BW, a more profes-
sional and specialized military with highly trained and equipped troops was necessary.
This, however, could not be achieved in a compulsory military service of just six months.
While this decision is a politically highly interesting topic, the focus of this introductory
example lies on the second important feature of the reform, the closure of military sites.
Military bases are frequently found in rural areas. Often, these sites have a long tradition,
and are soundly connected with their region nearby. They are an important source for
income, tax revenues, job opportunities, etc. If a base is closed the whole region also suf-
fers and loses an important economic pillar (Rudolf, 2011). When soldiers, workers and
their families move away, even small retailers, pubs and supermarkets feel the change.
In 2011, it was announced that the BW should be reduced from around 250,000 to 180,000
soldiers and that 31 of 400 military sites in Germany would be closed. The affected mu-
nicipalities feared economic losses up to millions, depopulation and vacancy (Juettner
et al., 2011). Therefore, it is not surprising that many federal states and municipalities
called for compensation and assistance packages for the abandoned areas (Beck, 2011).
The relationships between local municipalities and the army have always been a give and
take. Many communities undertook investment in infrastructure projects to satisfy and
retain their military sites. This is precisely where the problem occurred. Many regions
29
had been overconfident that their bases would not be abandoned. Under this presump-
tion of preservation, considerable expenditures have been made; over the last 5 years
preceding the reform € 160 million had been spent on infrastructure projects at military
sites which are destined to be closed (Bundesministerium der Verteidigung, 2012).69 Ex-
pecting a troop withdrawal, the local authorities would not have made these investments
but preferred to consolidate their budgets.
One important but simple example is the 5,000-resident city Kusel. The military site near
the city is one of five sites in the Rhineland-Palatinate which will be closed. Nearly 1,200
soldiers will move away. The decision hit the local policy makers by surprise. Especially,
as in recent considerations of the army reform even strengthening the site had been dis-
cussed. At the moment of the decision, civil investments on the military site took place,
a new gymnasium and administrative buildings were being built (Juettner et al., 2011).
A reason the coordination failed might be the varying distribution of preferences and con-
ditionalities. On the one hand, the municipalities had NSP about infrastructure expen-
ditures and the continuation of the facilities (investment only on condition of preserved
existence). On the other hand, the federal government was probably not at all interested
in infrastructure expenses, as the reform in itself was already highly complicated and
demanding (future troop size and composition, restructuring due to the abolition of the
compulsory military service).
This example shows the difficulties that arise when decision-making is divided on prob-
lems which are related to each other. The lack of coordination between infrastructure
investments at the local and defense policy reforms on the state level led to avoidable
cost. It is not necessarily the case that the decision-making bodies are situated at differ-
ent levels of government. It could also happen between two different ministries or two
neighboring cities which do not coordinate. So, they end up which less than optimal
outcomes, like investments in the infrastructure at locations shortly to be abandoned.
2.2.4 The future of the German pension system
I have already argued a few times that NSP are often ignored although they are actually
commonplace. Ironically, NSP are sometimes identified in instances in which they do not
really exist. This is due to a misunderstanding of their basic principle. Based on a current
survey study I clarify this misconception. This instance is an example of how not to deal
with NSP.
On August 29th and 30th 2012, the survey institute TNS-Emnid surveyed 1001 German
citizens for their views on the future of the German pension system. The results caught
the attention of the media quickly, as the survey followed a recent and broad discussion
on the peril of old age poverty. The debate started after a statement of the German Federal
69 Even though this may be a relatively small amount compared to the overall € 31.55 billion in the German
defense budget (Deutscher Bundestag, 2011), this is an immense sum of money.
30
Minister of Labour and Social Affairs, Ursula von der Leyen, on a prospective lowering of
the pension level to around 43% of the depositor’s nominal wage (von der Leyen, 2012).
Media reports on the survey said that a majority of 51% of Germans citizens was will-
ing to accept the proposed lowering. Only 39% were in favor of increasing the pension
contributions in order to prevent the lower pension level. At the same time, 32% of the re-
spondents were in favor of supporting particularly low pensions with taxes, in return for
lowering the contributions to the pension fund. Another 19% stated that future retirees
should make provisions for themselves by investing in private pension insurance.70
I do not question the reported numbers, but I do point out a certain aspect of nonsepa-
rability in the coverage of the survey. In (nearly) all news reports it was stated that the
lowering was accepted in return for the tax means subsidies (e.g., Christians, 2012). This
implies a conditional relationship, as the lowering is conditioned on the subsidies. The
two simple but separate questions “Do you support or reject the lowering of the pension
level?” and “Do you support or reject to subsidize particularly low pensions?” are not
suited to identify such a trade-off. They only tell us just how many respondents support
a lower pension level, and how many people support tax subsidies.
No kind of conditionality or trade-off can be identified in this way. I do not argue that
such a trade-off is absent; in fact, I honestly do not know, and would probably also agree
that an interrelation is likely (to a certain degree). The point I wish to make is that these
questions cannot tell us whether or not the German public has such considerations. A
correct survey question, aiming to identify a possible nonseparability between these is-
sues, would read as “Assuming that particularly low pensions are increased with tax
means, do you support or reject the lowering of the pension level?”
This example shows that NSP and surveys have a “special” relationship. On the one
hand, surveys possess the ability to identify even highly complex preference relations.
This holds at least in theory, and it has yet to be proven empirically. Often, nonseparabil-
ity is avoided because it takes time and entails complex requirements. On the other hand,
not accounting for NSP in surveys risks mistaking a respondent’s conditional response
for the genuine preference, and vice versa (Lacy, 2001a). In contrast to this example,
Sec. 2.4.6 clarifies how a survey investigates NSP properly.
2.2.5 The construction of a new convention center in Heidelberg
This example looks at a referendum at the local level of politics. In 2010, the citizens of
Heidelberg were called to vote in a referendum on the construction of a new convention
center, respectively, the extension of the existing city hall with an annex to serve as con-
70 Fora more comprehensive review of the survey and the public discussion on the reform of the Ger-
man pension system in general please consult the Focus money online dossier at http://www.focus.
de/schlagwoerter/themen/r/rentenniveau (accessed December 11, 2012).
31
vention center.71 Remarkably, the referendum was due because more than 18,000 citizens
signed a petition that called for a public vote on the extension of the city hall. The refer-
endum then took place on 25th July 2010 and with a two-thirds majority (by a turnout of
38.9%) the extension was rejected.72
As often in such cases, this public vote preceded a lengthy and difficult discussion. Plans
to build a new convention center near the Heidelberg central station have existed since
the 1990s. One after another, four official tenders (1996, 2000, 2004 and 2006) provided
no viable concept. After all plans had failed, including those involving private investors,
the Heidelberg City Council decided in 2008 to change the location. Instead of building a
new convention center near the central station, it decided to modify the existing city hall
by an extension. The two locations are about 3 km away from each other. The existing
city hall was built from 1901 to 1903 and its Renaissance architecture fits very well into
the historic old town of Heidelberg. Yet, its capacity (max. 3500 people) and alignment
could no longer keep up with the demands of modern convention centers.73
The issues in the debate were controversial. Proponents of the expansion highlighted the
fact that a conference center in the old town would be a unique selling point for Hei-
delberg. In many cities, such centers are located near train stations or outside the inner
cities. A building located in the historic old town would result in a big boost to meetings,
conferences and tourism in general. In addition, an extension would also be much less
costly than constructing of a new building from scratch at a different location.74 Oppo-
nents of the expansion pointed to the increase in tourism, too. Heidelberg’s old town,
they claimed, would not be able to cope with more tourists. Additional tourism and traf-
fic would be harmful and difficult to manage because of the poor urban transport links.
This is especially true in comparison to the location near the train station and its optimal
infrastructure connections. So far, this discussion seems like a simple one-dimensional
question. The key reason why this referendum is mentioned here is the following: the op-
ponents had a second objection, the architectural design of the extension. They claimed
it would distort the historic facade of the town hall and disfigure the homogeneous im-
pressions and decades-old skyline of the old town center.75
The referendum consisted of a single question, which read as “Should the city of Heidel-
berg build an addition or new building for an expanded convention center at the location
71 Homepage of the architectural design competition: http://www.heidelberg.de/servlet/PB/menu/
1198771/index.html (accessed September 19, 2012).
72 Official election archive of the city of Heidelberg: http://www.heidelberg.de/servlet/PB/menu/
1208720/index.html (accessed September 27, 2012).

73 The City Journal Heidelberg provided a detailed online dossier on the referendum at http://ww2.
heidelberg.de/stadtblatt-online/index.php?artikel_id=3744&bf= (accessed September 20, 2012).

It is still accessible and contains information on the extension plans, the different positions of the local
parties in the city council, statements from various citizen groups, etc. The city administration also offers
a comprehensive summary of the events at http://www.heidelberg.de/servlet/PB/menu/1125805/
index.html (accessed January 7, 2013).
74 Cf. representative for the proponents the then acting mayor Würzner (2010).
75 Cf. representative for the opponents Lask (2010).
32
2.3 Do delegation, decentralization and specialization help?
of the city hall?” The possible answers were ’yes’ and ’no’. The architecture of the build-
ing was not part of the vote, even though it had just been the design of the new structure
which had brought a whole new level of controversy into the discussion (cf. Möslinger,
2010). The referendum contained no possibility for people who wanted a convention cen-
ter at this location with a different design to express their preferences unambiguously. If
participants chose “yes”, they approved location and design. If they chose “no”, they
rejected both. Alternative designs, an important part of the public debate, were not part
of the referendum.
Again, this points out that holding a referendum alone is not enough. To increase demo-
cratic legitimacy the details of the implementation must also be considered (Lacy and
Niou, 2000). This problem is in line with the challenges Lacy and Niou (1994, 2000) iden-
tified for referendums. They argued that if the topic in question contains interrelated
issues, a single dichotomous vote may not be enough to enable all participants to express
their preferences.
The previous sections demonstrate that the concept of NSP is linked to complex decision
situations. Multidimensionality is obvious, but the specific dependencies between the
issues contributes to the intricacy as well. Even collective decision-making on simple
issues can be a costly, i.e., time-consuming and nerve-racking, process; let alone reaching
an agreement on complex policy proposals.76 This raises the question of whether one can
reduce this complexity and circumvent the requirements of dealing with nonseparability.
THE INSTRUMENTS OF ORGANIZATIONAL THEORY
The ubiquitous remedies to reduce complexity and inefficiency found in organizational

theory are delegation, decentralization and specialization.77 Kaiser (2007b, p. 128) points
out that representative democracies use the delegation of powers of attorney a lot to
enable collective action at all.78 Yet, highly decentralized and specialized policy-making
processes must be accompanied by some powerful central coordination or regulatory
authority.79 Otherwise decisions are made under uncertainty, bearing the risk of less
than optimal and inefficient outcomes.
The main purposes of delegating powers in the firm, the bureaucracy or the political sys-
tem at large are the reduction of transaction costs (e.g., Epstein and O’Halloran, 1999)
and information retrieval (e.g., Chongwoo and In-Uck, 2011; Gautier and Paolini, 2007).
76 Already Simon (1962) proved that complexity occurs in a large array of topics as, e.g., social, biological,
physical and symbolic systems.
77 These organizational tools correspond well with the idea of “polycentric systems” in solving collective-
action problems (cf. Ostrom, 1999, 2008).

78 Cf. Strom et al. (2003) for an overview on delegation patterns in western parliamentary democracies.
79 For a formal model and empirical test of administrative procedures cf. Epstein and O’Halloran (1996).
33
Departments and committees are manned by specialists who know the expected effects
of competing policy alternatives as well as the key players in their area of expertise (e.g.,
Dewan and Hortala-Vallve, 2009; Gilligan and Krehbiel, 1990). Hence, delegation and
specialization supposedly lead to faster decision processes and superior policies. Decen-
tralization is justified by similar arguments and applied when the locals are believed to
hold superior information on the effects of competing policies. The question concerning
the optimal vertical and horizontal division of competences can be found in the litera-
ture on federalism (e.g., Buchanan and Tullock, 1962; Oates, 2005), the theory of the firm
(e.g., Tirole, 1988; Williamson, 1975)80 , the division of labor within government and par-
liament (e.g., Gailmard, 2009; Gilligan and Krehbiel, 1987) and intergovernmental as well
as international agreements (e.g., Koremenos, 2008).81
Delegation and decentralization bear potential problems. Firstly, the specialized agents
may follow their own policy agenda and misrepresent the preferences of their principals
(e.g., the voters). The danger of bureaucratic drift increases with information asymmetry
and conflicting interests between principals and agents (e.g., McCubbins and Schwartz,
1984; McCubbins et al., 1987). Secondly, delegation in general and decentralization in
particular may forestall economies of scale and scope (e.g., Weingast, 2009). With re-
spect to the production of policies, the former is prevalent in the case of decentralization:
Why should each federal state have its own army? Why should each county regulate
solar thermal heating systems individually? By contrast, economies of scope are more
of an issue with respect to the horizontal allocation of competences: Why should the en-
vironmental authorities, who know all the effects of existing industrial filters, leave the
regulation of the respective sector to the Ministry of Industry?
Under the assumption of separable preferences, the application of delegation, decentral-
ization and specialization has proven to reduce costs (Epstein and O’Halloran, 1999) and
complexity effectively (e.g., Severinov, 2008; Tsebelis, 1994). They are approved and of-
ten used organizational measures. Yet, this tells us nothing about the underlying driv-
ing forces of complex decision-making. Each measure causes the separation of decision-
making with respect to either a spatial, temporal, hierarchical or administrative dimen-
sion (cf. Bendor et al., 2000; Bendor and Meirowitz, 2004). This contradicts the basic tenet
of nonseparability stating that interlinked parts should not be separated. The small case
study on the reform process of the German Bundeswehr (cf. Sec. 2.2.3) provides a first
insight into what happens otherwise. It clearly describes the consequences of badly co-
ordinated decisions. In addition to this specific sample, I am looking for general patterns
concerning institutional structures and nonseparability. The question is, e.g., what hap-
pens if single decisions of a problem affected by NSP are delegated.
80 Cf. Grant (1996) for a comprehensive knowledge-based theory of the firm. He looked particular into hier-
archy and the distribution of decision-making authority.
81 Koremenos (2008) investigated “when, what, and why do states choose to delegate”. She found that “the
presence of a complex problem increases the probability of external delegation [to a third party outside
of the agreement] by nearly forty percent and internal delegation [to a collective formed by the members
of the agreement themselves] by twenty-one percent” (Koremenos, 2008, p. 169).
34
THE INTERPLAY OF INSTITUTIONS AND PREFERENCES
Recently, behavioral economists started to scrutinize the effect of the institutional struc-
ture (e.g., Ambrose and Schminke, 2003; Wilson and Eckel, 2011) on the motivations of
individual choices (e.g., Bandiera et al., 2005; Bolton et al., 2005). In this context, institu-
tions comprise decision-making rules, proceedings as well as algorithms (Kröning and
Strichman, 2008). The investigations are not solely restricted to simple economic games.
Ehrhart et al. (2007) examined the impact on the setting-up of a budget, a common pol-
icy issue. The results are not decisive but point to the relevance of the decision-making
environment as “preferences may be sensitive to the choice process” (Sen, 1997, p. 745)
itself. This is of central interest for research in the field of empirical institutional analysis,
which seeks to determine these “causal mechanisms, i.e., the way how institutions affect
behavior structuring” (Kaiser, 2007b, p. 120).82
My research is based on the interplay of decision-making procedures and preferences

of the decision-makers. Already Plott (1979, 1991) argued that the interaction of pref-
erences and institutions determines policy outcomes. The preferences are channeled
through institutions which condition actors’ behavior and combine them to collective
choices. A change in either preferences or institutions, while the other remains constant,
might change outcomes, but need not necessarily do so. This allegory is referred to as
the “fundamental equation of politics” (Hinich and Munger, 1997, p. 17) and depicted in
Eqn. 2.1.83
pre f erences ~ institution = outcomes (2.1)
The equation is rather self-explanatory. Outcomes merely refer to the result of a decision-
making process, whether it is what to cook for dinner, which movie to go to, or whether
or not to build a new city hall. In other words, this means the societal consequences
of decisions made (Lane and Ersson, 2000). Preferences are straightforward too, as they
resemble what a person prefers in a given set of choices (Shepsle and Bonchek, 1997).
The most simple form is “given by a binary relation over the set of options” (Barrett and
Salles, 2006, p. 1). People may differ in their preferences in terms of direction, salience
and over time (Ehmke et al., 2005). Complementary concepts as, e.g., endogenous pref-
erences can also be incorporated (cf. Grendstad and Selle, 1995). This supplements the
82 For empirical examples cf. Diermeier and Gailmard (2006) who showed that social preferences depend on
the decision-making process, or Sobel (2005, p. 392), who summarized “context-dependent preferences”
as to permit “the strategic context to determine the nature of individual preferences.”
83 The importance of this fundamental principle is aptly illustrated by the blog “Rule 22” maintained by
Ragusa et al. (2012). The name refers to a standing rule of the United States Senate most commonly
associated with the filibuster. The authors are motivated by “the intention to forward the insights into
the public debate” and analyze political events through a general institutional perspective. The blog
entry “Things Institutionalists Know that You Should: Plott’s Equation” (http://rule22.wordpress.
com/2011/09/22/things-institutionalists-know-that-you-should-plotts-equation, posted on
September 22, 2011 by Nate Birkhead) emphasized the fundamental role of Plott’s equation along various
examples.
35
standard approach, which assumes “that people harbor a stable, well-defined, and dis-
cernible order of preferences” (Simon et al., 2004, p. 331).
The term most difficult to circumscribe is “institution”. Of course, the word itself is
common-place and often refers to any formal structure or organization. Yet, according
to North (1990, p. 79ff), a distinction between institution and organization is pertinent.
North classified organizations as the endogenous best response of humans to their in-
stitutional environment. On the other hand, he defined institutions as “the constraints
that humans devise to shape human interaction” (North, 1990, p. 3). Thus, “institutions
are the rules of the game” (North, 1997); they are set first. Within these rules, an actor’s
behavior creates organizations.
A central question is where these rules come from. In this context, Hall and Taylor (1996)
provided an excellent overview of the different research paradigms which have appeared
within the “new institutionalism”: rational choice, historical and sociological. Rational
choice institutionalism has developed the most “precise conception of the relationship
between institutions and behavior” (Hall and Taylor, 1996, p. 950). Yet, this results from
a highly simplified image of man (Opp, 2004). It has provided major scientific contri-
butions to political research, but has only limited explanation power in explaining the
emergence of institutions (Hall and Taylor, 1996, p. 952). Historical and sociological insti-
tutionalism provides more insights through their focus on existing power relations and
how they shape further developments. In sum, Hall and Taylor (1996) do not argue in
favor of one approach but rather “for greater interchange among them” (Hall and Taylor,
1996, p. 955). This implies for the emergence of institutions that they are inherited, or
else developed, by actors when the need arises. This fits with North (1990, p. 3) and his
argument for “humanly devised” institutions.
Ostrom (1986) also focused on the multiple usage of the term institution. Reformulat-
ing Eqn. 2.1 to an equation system as in Eqn. 2.2, she argued that we “need to address
questions concerning the origin and change of rule configurations in use. How do indi-
viduals evolve a particular rule configuration? What factors affect the likelihood of their
following a set of rules? What affects the enforcement of rules?” (Ostrom, 1986, p. 22)
action structure ~ decision model = outcomes

(2.2)
rules ~ physical laws ~ behavioral laws = action structure
In Eqn. 2.2 rules (i.e., institutions) determine, in combination with physical and behav-
ioral laws, the feasible action structure of actors. The interaction of this structure with the
decision model then produces the resulting outcomes. An example of such an extended
system was provided by Lane (2000). He argued that the most important action structure
is “the market”, which offers rules for interaction of economic and political interests.
A broad strand of literature agrees that interests and institutions affect social outcomes.
Krehbiel (1986, p. 544) considered political outcomes as products of decision-makers’
36
preferences and institutional features. Looking into legislative committees he described

committee strategies as “determined by preferences and institutions” (Krehbiel, 1986,
p. 555). Current contributions also refer to this research framework. Thomson et al. (2006,
p. 9) classified it as “one of the major preoccupations of modern political science: the in-
terplay between institutions and preferences in determining policy outcomes.” Schofield
(2008, p. 2) stated that social choice theory in its very core “seeks to understand the con-
nection between individual preferences, institutional rules and outcomes.”
An unsettled matter concerns the relative importance of preferences and institutions for
the resulting outcomes. Dowding and King (1995, p. 7) advocated the dominant role
of institutions as, “generally speaking, the institutions of politics provide a larger part
of the explanation than do preferences”. To make such a strong and general claim is
controversial, as it extends to a large variety of decision problems and contexts. Yet,
“most of the new institutionalism still leaves a role for preferences to play” (Stoll, 2013,
p. 7) and prefers to emphasize the importance of institutions in relation to preferences.
Here, Riker (1980, p. 20) predicated strongly that social research “cannot study simply
tastes and values, but must study institutions as well” and defined institutions as “rules
about behavior, especially about making decisions” (Riker, 1982b, p. 4).
Various previous contributions have focused on the influence of procedural rules and
demonstrated their stabilizing effect (e.g., Shepsle, 1979). Without institutions, any mul-
tidimensional decision problem is inherently unstable (McKelvey, 1976). The field of
comparative politics features a vast literature on the variance in institutional arrange-
ments across countries. Well known fields of research are regime types, electoral and
party systems, the structures of government, civil-military relations, corporatism, social
cleavages, etc. (for an overview cf. Lijphart, 1971, 1999). Only one prominent example in
the field of analytical politics is the so called “structure-induced equilibria” (Shepsle and
Weingast, 2004), which takes the structuring capabilities of institutions as given.
In recent studies, “institutions are more and more regarded as result (or equilibrium so-
lution) from strategic action” (Kaiser, 2007b, p. 124).84 This highlights that “preferences
matter, too” (Stoll, 2004, p. 4). Turning the focus back from the institutional structure
alone, Lane (2000) argued that interests are important, and that institutionalism over-
states the role of procedural rules. Stoll (2004) reviewed the ways in which preferences
interact with political institutions to shape party systems. Overall, she acknowledged
that “political outcomes are a function of both political institutions [...] and the prefer-
ences of the citizenry” (Stoll, 2004, p. 1). Yet, she emphasized that “institutions do not -
and cannot - tell the whole story: preferences have work to do as well.” (Stoll, 2004, p. 2)
As with institutions, views on preferences changed. The current state of research consid-
ers preferences not as exogenously fixed, as they were in classic rational choice theory
84 When assessing the degree of risk sharing and redistribution in different federal fiscal constitutions, Pers-
son and Tabellini (1992) aptly illustrated how different forms of interaction (e.g., voting or bargaining)
lead to different institutional frameworks.
37
(Simon et al., 2004). They “rather are reconstructed in the course of decision making”
(Simon et al., 2004, p. 335). March and Olsen (2006, p. 689) emphasized that rules are fol-
lowed if they are judged “appropriate [...] in a specific type of situation.” To incorporate
these insights in Eqn. 2.1 leads to Eqn. 2.3. The outcome is still defined by preferences and
institutions. However, it also accounts for the possibility of a “feedback loop” between
preferences and institutions.85
 
→
pre f erences institution = outcome (2.3)
←
Holyoak and Simon (1999) found that in legal decision-making shifts in one task can
trigger according alterations in subsequent tasks involving similar underlying issues. In
addition to the pure existence of such shifts, it is important to search for regular patterns.
Looking into legislative bargaining, Miller and Vanberg (2014, p. 18) gathered evidence
indicating that the “unanimity rule motivates subjects to be more ’bullish’ in their bar-
gaining behavior”. This is no single or isolated finding; performing a literature review on
the effects of procedure on social interaction, Mertins (2008, p. 10) identified that process
fairness affects fairness perceptions. She concluded that “clear evidence for procedures in
influencing human decision-making exists” (Mertins, 2008, p. 31). Following this insight,
the current state of research moves away from classic but static approaches (Wandling,
2012).86
Including dynamics is complex (and difficult to operationalize) but closer to reality; such
a dynamic is found when variations of the procedural rule lead to a change in preferences.
Nonseparability adds a new element to the interplay of institutions and preferences. The
separation of decision-making competences, or defining one part of a decision, causes
an adjustment of the remaining (conditional) preferences.87 Any organizational mea-
sure, whether it is delegation, decentralization or specialization, implies a multilevel or
multistage decision (Strom, 1990, p. 107ff). Thus, it causes the separation of decisions
belonging to one overarching question. Yet, this violates the basic tenet of separability,
stating that interlinked parts should not be separated. This emphasizes again the com-
monness of NSP, and highlights that delegation, decentralization or specialization are no
solutions to, but rather part of, the unanswered (empirical) question about the influence
of nonseparability.
The next section looks into the theoretical aspects of nonseparability and focuses on the
question of how to operationalize NSP. In the subsequent part, I use the framework out-
lined in compound with the argument that NSP are highly relevant in the context of
85 This is different from classical multiattribute decision theory (e.g., Edwards and Newman, 1982; Keeney
and Raiffa, 1976) as preferences cannot be added up or weighted accordingly to specific environments.
86 Lichbach (2001) juxtaposed rationalist, culturalist, and structuralist research approaches against each
other. The comparison clarified that these approaches cannot go their traditionally separate ways any-
more. They must submit themselves to the mutual exchange of criticism in order to remain valuable.
87 Cf. Sengul et al. (2012) for a discussion on the strategic aspect of delegation.
38
2.4 Theoretical foundations
organizational structures and explicate common arguments on the effect of NSP with re-
spect to organizational measures of delegating and sequencing. Those are not completely
new and have existed in the theoretical literature for quite some time, but their empirical
verification is still pending.
This section discusses the theoretical foundations for analyzing NSP. As the concept is
deep-seated in the theoretical literature (for an overview cf. Strom, 1990), it combines
existing knowledge of different fields into one comprehensive discourse. I start the the-
oretical discussion in Sec. 2.4.1 by stating basic definitions. Next, I discuss in Sec. 2.4.2
the usage and operation of spatial models in analytical politics. Spatial theory serves
as main methodological tool for analyzing preferences in general (Shepsle and Bonchek,
1997). Following this approach, I illustrate in Sec. 2.4.3 the implementation of NSP into a
simple spatial model. Sec. 2.4.4 discusses briefly the magnitude of NSP in such a model,
and Sec. 2.4.5 introduces a specific extension to the operationalization with respect to the
realm of politics. Finally, as every model evaluation must rest on empirical data, Sec. 2.4.6
exemplifies how NSP can be measured.
2.4.1 Basic definitions
Collective decisions on conditional topics are the subject of my research. I have, so far,
already talked a lot about ’conditionality’, but skipped a detailed discussion of the term
’collective’. A collective decision obviously refers to a group of individuals. In general,
such a group is not limited to any size. However, my laboratory experiment will be
restricted to small groups because of the space provided in the lab. Political decision-
making to a large extent takes place in committees. I regard a committee as a small to
medium-sized group of people who have institutionalized interactions. Following the
work of Black (1958, 1991), and in the tradition of rational choice theory, the focus of this
study lies on voting procedures and results in such committees.88
While it is relatively easy to define an individual decision, this is a more complex en-
deavor in the case of a collective decision. In this study, a collective decision represents
the coordination of actions within a group of relevant actors. More precisely, each partic-
ipant of the decision-making group chooses one option out of a set of options, and rejects
all others (Pritzlaff, 2006, p. 208). The decisions of every involved individual engage with
each other on the collective level and form the collective decision of the group.
88 Cf. Nullmeierand Pritzlaff (2009) for an alternative approach and a more specific look into the concrete
deliberation of committee meetings and their internal decision processes.
39
2.4.2 Using spatial models
A key component when analyzing preferences in the rational choice tradition are spatial
models. Their underlying assumption states geometric dependence among modeled ob-
jects. Fundamental to spatial models is the work of Hotelling (1929), Lerner and Singer
(1937) and Smithies (1941). While these authors focused on economics and the competi-
tion between companies, the contributions of Downs (1957) and Black (1958) established
spatial models in analytical political science, and Davis and Hinich (1966, 1967, 1968) laid
the groundwork for spatial theory of voting.89 Recently, increased efforts and progress
have been made to operationalize spatial models of legislative politics (e.g., Benoit and
Laver, 2007c; Laver and Schofield, 1998; Slapin and Proksch, 2008).
To operationalize spatial models in politics, one assumes that political agents have ra-
tional preferences over a predetermined decision space (cf. Schofield, 2008). Thus, the
actors have a clear and unambiguous preference for the realization of each dimension of
the space. The point in space where the first preferences for every dimension intersect
is called the ideal position (IP) of an actor. Each possible individual policy can be repre-
sented as point in the decision space (Humphreys and Laver, 2010). For each such point
the utility level U of every actor i can be determined as the distance between this point
and their IP; i.e., how far for a specific policy is away from the most preferred option.
Fig. 2.1 shows a two-dimensional policy space (X,Y) including the IP of three actors and
the status quo (SQ) as the currently enacted policy. Every policy closer to an actor’s IP
raises the utility of this actor and is therefore preferred to the SQ. Colomer (1999) called
this the “preferred-to-set concept”. It comprises all possible policy outcomes an actor
prefers over the current SQ. The left figure shows cycles drawn from each actor’s IP
through the SQ. These cycles represent the actor’s indifference curves. At every point on
the curve the utility level of the respective actor is the same, thus they are indifferent
when choosing between points on the same curve. The form of circles is the result of my
assumption in the left figure of both dimensions being equally important to all actors.
If one dimension were to be more important to an actor the curve would take the form
of an ellipsis. This happens in the figure to the right. A differences in salience of the
issues stretches the indifference curves in the dimension of the lesser important issue.
This means that for actor 1 the issue Y is more important, and, for actor 2, X outranks
Y. Independent of any salience, every point within the curve exhibits a higher and ev-
ery point outside the curve a lower utility. In the two-dimensional policy space exist
infinitely many indifference curves, one passes through each possible combination of X
and Y.
The darker areas in Fig. 2.1 resemble policies (X,Y) which two of the three actors would
prefer to the SQ, i.e., the indifference curves of two individuals overlap. These areas
are called winset as they allow a majority of actors to improve from the current situation
89 Cf. Davis et al. (1970) for a non-technical introduction to (spatial) models of social choice.
40
Figure 2.1: A simple spatial model

EXPLANATORY NOTE
The figures show a two-dimensional policy space (X,Y), the SQ and IP of three actors. The IP of an actor represents her
most preferred policy (or point of the policy-space), which is given by the combination of the two dimensions. Cycles or
ellipses around the IP indicate an actor’s indifference curves.
Circular indifference curves Elliptic indifference curves
(assuming simple majority is the necessary threshold). The winset of the SQ is the set of
outcomes that can defeat the SQ in a pairwise comparison, i.e., the set of policies that can
replace the existing one.
A related concept is the concept of the core. A core is the set of points with an empty
winset; i.e., the points that cannot be outdone by any other point under the corresponding
decision rule (Tsebelis and Garrett, 2001, p. 21ff). When looking at collective decisions
it is important that the necessary majority threshold is reached. In Fig. 2.1 no policy
exists for which all three actors are better off, i.e., no winset exists under unanimity rule.
Thus, whether the actors will deviate from the current SQ or not depends on the majority
necessary for altering the enacted policy. Importantly, the difference in salience in the
right figure influences the winset. Here actors 1 and 2 have no common ground for an
improvement from the SQ.
This example is formalized as follows: For each point in a two-dimensional decision
space (X,Y), the utility level of every actor i can be determined as the utility at their ideal
position (U IP ) reduced for the distance of the policy in question to it. Accounting for the
dimension-specific differences in salience this results in
 2  21
Ui ( x, y) = U IP,i − wx,i × ( IPx,i − X )2 + wy,i × IPy,i − Y

(2.4)
In Eqn. 2.4, the distance between policy and IP in each dimension is weighted with an
actor-specific weight wi . The underlying logic is that if one issue is very important to an
actor, then a deviation from their first preference in this dimension has a stronger utility-
reducing effect. Eqn. 2.4 makes one additional (and fundamental) assumption I have not
41
discussed so far. The equation represents the distances in terms of Euclidean space. The
Euclidean metric determines the distance between two points as the root of the sum of
squares of differences of the individual dimensions (Hinich and Munger, 1997, p. 76ff).
Before applying these concepts to the case of NSP there are some remarks to be made. Us-
ing utility in political science (like many other empirical tools) can be attributed to influ-
ence from economics (Webster and Sell, 2007). Thus, it was not developed for “political”
questions and should henceforth be scrutinized for its suitability. The main application
for utility space in economics is dealing with the allocation or distribution of goods: to
identify goods as substitutes or complements, to investigate their (marginal) exchange
rate, etc. In politics we mostly deal with policy programs instead of goods. Nevertheless,
the use of spatial models and Euclidean space is standard practice in political science.
The work of Tsebelis (2002) is just one prominent example.
In the literature, nonseparability has been prominently discussed with respect to spend-
ing preferences over public policies which are subject to the same budget (Hinich and
Munger, 1997, p. 60ff). Given an overall budget constraint, the amount of money allo-
cated to one policy program influences the spending preferences for all remaining pro-
grams. Milyo (2000a,b) showed that the Euclidean metric cannot correctly represent ac-
tors’ utility functions over such a multidimensional budget-rivalry problem. Accord-
ing to Benoit and Laver (2007c, p. 31), this finding further supported those who criti-
cize the convention of using the Euclidean metric to model political decision-making.
Humphreys and Laver (2010, p.14) argued that “metric assumptions for models of policy-
based political decision making are under-researched, and which distance metric - if any
- is appropriate for modeling human political preferences remains an open question.”
Subsequently, I take this criticism very seriously but do not deviate from the conven-
tional Euclidean space.90
The main reason I use Euclidean space is the perception of differences by humans. Shep-
ard (1991) studied the perception of (dis)similarity by individuals in a multidimensional
setting using the “Minkowski r-metric” in an experiment. In terms of this metric a dis-
tance d between the values i and j of dimension x is calculated as
  1r
dij = ∑  xik − x jk 
 r
(2.5)
k
In his analysis Shepard (1991) iteratively changed r. His goal was to identify the best pos-
sible match between the perception of distance by the subjects (determined from their
behavior during the experiment) and the distance measure based on the metric. It is
noteworthy that the two most prominent metrics can be represented by Eqn. 2.5. The
90 Only few experimental studies apply other metrics; and if they do, they do not usually find any differences
due to the specific metric used (e.g., Berl et al., 1976).
42
city-block metric91 for r = 1 and the Euclidean metric for r = 2. Looking for the most
appropriate metric for different choice problems Shepard (1991) found the smallest devia-
tion for the city-block metric using separable dimensions. On the contrary, the Euclidean
metric produced its best predictions dealing with nonseparable dimensions.92 Thus, Eu-
clidean space represents the appropriate metric for my investigation of nonseparability.93
2.4.3 How to operationalize nonseparable preferences
By using two-dimensional spatial utility functions, NSP can nicely be illustrated (cf.
Hinich and Munger, 1997). The canonical equation for a spatial loss utility function or
weighted Euclidean distance (WED) in a d-dimensional policy space is

WED (θ, x ) = (θ − x )T A (θ − x ) (2.6)
 unconditional ideal point, x=(x1 ,x2 ,...,xd ) de-

in which θ=(θ1 ,θ2 ,...,θd ) describes an actor’s
a11 ... a1d
 . .
. . ...  describes the importance the

scribes the policy in question and A =  ..
 
ad1 ... add
actor attaches i) to each dimension (main diagonal) and ii) to the interaction between the
dimensions. Thus, the utility of an individual is determined by the weighted distance
between an actor’s ideal point and the enacted policy.94
If we assume that actors preferences are separable across dimension, it holds that aij = 0
unless i = j. In other words, a person’s “preferences on every issue and set of issues
are independent of - or, can be separated from - the outcomes of other issues” (Lacy,
2001a, p. 132). This leads to a utility function represented by a simple “additive model”
(Strom, 1990, p. 57) in which both dimensions can be evaluated individually. The devia-
tions from an actor’s ideal position can be calculated separately for each dimension and
then summed up in the end. Going back to the culinary example of the two couples in-
troduced above it implies that an actor’s preference ordering over the disposable wines is
independent of the meal and vice versa. This leads to Eqn. 2.7, which elucidates that the
pleasure of food and drink is each separately formed for themselves and only assembled
into one expression in the end (cf. Eqn. 2.4).
 2
adrinks (θ drinks − xdrinks )2 + a f ood θ f ood − x f ood

WED (θ, x ) = (2.7)
91 The city-block metric determines the distance between points as the sum of the absolute differences of the
individual dimensions (Torgerson, 1952, 1967, p. 38ff).
92 Interestingly, Soto and Wasserman (2010) reached the same conclusion when looking at animal behavior.
93 Most literature looking into the perception of separable dimensions finds that city-block metric provided
a far better description of psychological distance relationships (for an overview cf. Lockhead and Pomer-
antz, 1991). However, allowing for two-dimensional responses of subjects Nosofsky (1985) found that
the Euclidean metric provided a far better description of psychological distance relationships.
94 This complex mathematical equation rests on the same basic principles discussed in Sec. 2.4.2 and Fig. 2.1.
43
In Eqn. 2.7, θ describes the initial (unconditional) first preferences of an actor concerning
either food or drinks, a depicts their respective salience and x the actually chosen selec-
tion. With this information it is possible to calculate an actors’ “culinary utility” when
having dinner. Apparently the pleasure is very dependent on the difference between
initial preference and current selection.
Moving closer to a political environment, Fig. 2.2 illustrates the spatial representation of a
decision problem for a member of the Euro-group on two separable issues. The member
is voting on the budget for the European stability mechanism (ESM) and the required
level of fiscal reform the Euro-group demands for its assistance.
Figure 2.2: Indifference curves for separable preferences

EXPLANATORY NOTE
The figure depicts the unconditional ideal point and corresponding indifference curves for a member of the Euro-group
who prefers a medium high budget for the ESM and medium strict fiscal requirements. The salience of issue 1 (budget
for the ESM) equals 0.7 (a11 = 0.7) and the salience of issue 2 (required level of fiscal reform) equals 0.3 (a22 = 0.3). As one
issue is more important, the indifference curves have the form of an ellipsis. I assume separable preferences by setting
a12 = a21 = 0. The winset originating from the SQ covers the complete issue space.
In contrast, with conditionality the utility in one dimension depends on the outcome
of the second dimension. Any aij ̸= 0 implements such nonseparability into the loss
utility function (Eqn. 2.6). If both decisions are made together, the actor will still follow
the unconditional ideal point. Yet, this no longer holds if the two decisions are made
separately and the first decision does not match the unconditional ideal position. Then,
the actor reacts by adjusting the preferences of the other topic. This adjustment resembles
an actor’s conditional ideal point (Enelow and Hinich, 1984, p. 18).
Nonseparability may be positive (rectified) or negative (directed opposite). A plain ex-

ample for negative NSP is the simultaneous decision over the spending for two projects
(X,Y), assuming that actors preferences are subject to either an explicit or implicit bud-
getary constraint. Then, they will lower their spending preference for project Y once con-
fronted with the decision to spend more on project X than they had originally preferred
(Hinich and Munger, 1997). However, the projects might just as well be complementary
44
and, thus, interrelated by positive nonseparability. For example, consider investments

into railway networks and trains. Actors would enhance their spending preference for
the infrastructure once confronted with a decision to invest more into trains than they
had originally hoped for. Fig. 2.3 shows the resulting indifference curves for these kinds
of NSP. In this graph I use again the narrative of a Euro-group member voting on the
ESM budget and level of fiscal reform.
Figure 2.3: Indifference curves for nonseparable preferences

EXPLANATORY NOTE
The figures depict the unconditional ideal point and corresponding indifference curves for a member of the Euro-group
who prefers a medium high budget for the ESM and medium strict fiscal requirements. The salience of issue 1 (budget
for the ESM) equals 0.7 (a11 = 0.7) and the salience of issue 2 (required level of fiscal reform) equal 0.3 (a22 = 0.3). As one
issue is more important, the indifference curves have the form of an ellipsis. I assume NSP by setting a12 = a21 ̸= 0. As a
result we end up with tilted or slanted indifference curves around the ideal position of the actor (cf. Hinich and Munger,
1997; Strom, 1990).
The left figure depicts positive nonseparability with a12 = a21 = 0.8. Here, stricter rules lead to a preference for a higher
budget, and vice versa. The winset of the SQ does no cover the complete issue space. The right figure depicts negative
nonseparability which is implemented by a12 = a21 = −0.8. Here, stricter rules lead to a preference for a lower budget,
and vice versa. The indifference curves stay tilted, but resulting from the position of the SQ in the lower left corner, the
winset covers the complete issue space again.
positive nonseparability negative nonseparability
Another way of depicting NSP is using ridge lines. Those show “alternatives most pre-
ferred on one dimension for any given alternative in the other dimension.” (Strom, 1990,
p. 108) Thus, they describe the conditional ideal positions of the actors according to differ-
ent outcomes of a previous decision on another issue (Denzau and Mackay, 1981). With
separable preferences the preferred alternative of a dimension is not depending on other
outcomes. Thus, ridge lines run only vertically or horizontally. With NSP the ridge lines
are tilted. Using the two-dimensional ESM example it is easy to exemplify the calculation
of the conditional ideal position. Knowing the decision made on the budget of the ESM
B∗ the conditional preferred level of fiscal reform R| B∗ is calculated as
 
a12
R| B∗ = θ R − ( B∗ − θ B ) (2.8)
a22
45
Here, θ B and θ R describe the initial unconditional ideal position of the actor, and ( aa22
12
)
sets the amount of NSP ( a12 ) between issue 1 and issue 2 in relation to the salience of
the second issue ( a22 ), i.e., the amount of fiscal reform. The computation for multidimen-
sional space is analogous (Enelow and Hinich, 1984, p. 18). In Fig. 2.4 the ridge line for a
member of the Euro-group with positive NSP are depicted.
Figure 2.4: Ridge lines for nonseparable preferences

EXPLANATORY NOTE
The figure shows the unconditional ideal point for a member of the Euro-group who prefers a medium high budget for
the ESM and medium strict fiscal requirements. It also shows the actor’s ridge line (dotted). As the actor has positive NSP
(a12 = a21 > 0) a higher budget implies stricter fiscal reforms. In the figure the budget for the ESM is chosen first and
fixed at B∗ . This budget exceeds the initially preferred budget of the actor. Therefore, they adjust their preference with
respect to the second issue (indicated by the arrow). The required level of fiscal reform is indicated as R| B∗ . The change
in the preferred level of fiscal reform leads to the members’ (new) conditional ideal point.
In sum, as with the ideal position of an actor or the salience of the issues, also the de-
gree and direction of conditionality influences the outcome of decision-making. This is
represented by the different shapes of the indifference curves and the varying emergent
winsets. Therefore, the utility function is supplemented by an “interaction” term. Us-
ing the culinary example to illustrate the effect this interaction only represents how well
or badly the choices for wine and dish go together. This is shown in Eqn. 2.9 where d
indicates the drinks and f the food served.
 2 
ad (θd − xd )2 + a f θ f − x f + ad f + a f d θ f − x f (θ d − xd ) (2.9)
  
WED (θ, x ) =
So far, two questions are still unanswered. Firstly, what upper and lower bounds ex-
ist for the magnitude of nonseparability? And secondly, does the dependency occur in
both directions, so does aij = a ji always hold? The two following sections answer these
questions.
46
2.4.4 The magnitude of nonseparability
This section briefly discusses the possible range of values of NSP. While the determina-
tion of a lower bound for NSP is trivial, a general definition for an upper bound is not
so easy. The lower bound of 0 equals just the scenario of separable preferences where no
conditionality between different issues exists. On the other hand, if such dependencies
exist, how strong can they be? Like with salience, the weight attached to a dimension or
to a combination of dimensions, is actor-specific. A proper definition for an upper bound
should take this into account. Hinich and Munger (1997) assumed for each combination
of two dimensions that the upper ceiling for nonseparability meets the criteria of
aii × a jj − aij × a ji > 0. (2.10)

 
This assumption guarantees that the interaction effect between the dimensions aij × a ji
 
cannot outweigh the non-interaction part aii × a jj . In spatial terms an actor’s indiffer-
ence curves keep the form of an ellipsis (Hinich and Munger, 1997, p. 218). With respect
to Eqn. 2.9 the criteria ensures that the term under the square root is non-negative. Ob-
viously, assumption Eqn. 2.10 is driven by the choice of the Euclidean metric.95 From a
substantial perspective this upper bound does not define an extreme case of NSP. More
precisely, there is no ex-ante reason why the interaction term should be less valuable to
the actor than the two issue-specific terms.
2.4.5 The case of reciprocity
In the realm of politics the allocation of resources often depends on the correspondence of
political will. For example, the transfer of decision-making authority to an autonomous
government agency is often conditional upon the expected policy outcome. Yet, this
does not necessarily apply in reverse. The conditionality in legislators’ preferences over
the characteristics of and the allocation of resources to a political program is another,
although the most prominent, example. In order to model such non-reciprocity appro-
priately Finke and Fleig (2013) proposed a simple modification to the standard Euclidean
utility function. More precisely, this implied an extension of the concept of nonseparabil-
ity with respect to a) the direction and b) the reciprocity of the effect.
Many legal proposals feature a budget as well as a policy dimension. In these cases it
appears straightforward to assume that legislators’ spending preferences depend on the
characteristics of the corresponding policies. For example, legislators’ preferences for an
agency’s budget will be more generous if this agency pursues their political interests.
Similarly, members of the IMF’s Board of Governors make their decision over the size of
a loan dependent on the fiscal reforms pursued by the remitting government. Whereas
these statements may appear uncontroversial, political scientists tended to ignore this
95 The empirical analyses in Chap. 3 exemplifies the practical necessity of this condition.
47
conditionality, because often the political decisions over the budget for, and the charac-
teristics of, a certain policy took place simultaneously.
Commonly, the direction of NSP is assumed to be either positive or negative, as shown

in Fig. 2.3. The figure discussed positive as well as negative NSP using the example of
a member of the Euro-group deciding on the budget of the ESM and the required level
of fiscal reform. Reciprocity assumes that nonseparability goes both ways. No matter
which of the two issues is settled first, it will be followed by an adjustment in an actor’s
preference on the remaining issue, unless the outcome is equal to the unconditional ideal
position (cf. the concept of ridge lines and Fig. 2.4).
This might not always be a useful assumption. For example, consider the case where one
issue deals with the policy of a political program and another issue deals with the budget
for this particular program. The spending preferences may be conditionally dependent
on the policies pursued. However, the policy preferences may remain unaffected by the
allocated level of spending. A political conviction stays, even if the funding provided for
the favored policy changes. In terms of the ESM example this implies that the Euro-group
member prefers a certain level of domestic fiscal reforms (i) which does not depend on
the funding volume of the ESM; but the preferences for the size of the ESM’s budget (j)
most certainly depend on the fiscal reforms required for being eligible to an ESM loan.
Only if the preferred policies are implemented the member is willing to vote in favor of
a high budget.
Unfortunately, the idea of non-reciprocal nonseparability cannot be accommodated by

simply assuming aij = 0 and | a ji | > 0.96 In the standard, reciprocal version of non-
 
separability (Eqn. 2.9) the sign of the interaction term (θ i − xi ) θ j − x j depends on both
 
dimensions. Assuming a positive sign for aij + a ji , reciprocity causes an increase in the
loss utility U(xij ) if either θi > xi and θ j < x j or vice versa. Thus, in the reciprocal case
the member responds to certain policies with an increase and to certain policies with a
decrease of their preferred budget.
Yet, in the case of non-reciprocal nonseparability a deviation from the preferred poli-
cies (whether too strict or too lax fiscal reform requirements) always leads to a reduced
budget preference. In other words, if the member cannot get their desired policies imple-
mented, their willingness to pay is diminished. The further the policy diverges from the
members’ unconditional ideal position, the smaller the most preferred budget becomes.
This results from the absolute distance between an actors preferred and enacted policy in
 
the interaction term, implemented as |θ i − xi | θ j − x j . The sign of the interaction term
is now exclusively determined by the budget dimension j, whereas the size of the effect
is determined by the policy dimension i. Simply put, divergent spending is always a bad
thing, but just how bad a thing it is depends on whether or not an actor approves of the
96 The same argument was made by an unpublished anonymous manuscript (Anonymous, 2010), which was
assigned to me for reviewing, by introducing “partially nonseparable preferences” as well as “mutually

nonseparable preferences”.
48
policy being funded. Eqn. 2.11 implements this type of non-reciprocal conditionality into
a standard WED.97
 2
aii (θ i − xi )2 + aij + a ji θ j − x j |θ i − xi | + a jj θ j − x j
   
WED (θ, x ) = (2.11)
Fig. 2.5 illustrates the corresponding indifference curves using the example of the Euro-
group member one last time. It is easily shown that this member’s tolerance for over-
spending depends on the fiscal requirements, i.e., the conditions which a state has to
fulfill for being eligible to an ESM loan. If the conditions are either too strict or too lax,
the board members’ willingness to grant more money than their unconditional spending
preference is limited. In fact, if the fiscal requirements are too lax they would prefer the
SQ even if the financial volume equals their unconditional spending preference, and vice
versa. In other words, the combination of both issues is relevant for the decision of the
board member.
Figure 2.5: Indifference curves for non-reciprocal nonseparable preferences

EXPLANATORY NOTE
The figure depicts the unconditional ideal point and corresponding indifference curves for a member of the Euro-group
who prefers a medium high budget for the ESM and medium strict fiscal requirements. The salience of issue 1 (budget for
the ESM) equals 0.7 (a11 = 0.7) and the salience of issue 2 (required level of fiscal reform) equals 0.3 (a22 = 0.3). NSP are
implemented by setting a12 = a21 = 0.8. I assume non-reciprocal nonseparability, i.e., I apply Eqn. 2.11 instead of Eqn. 2.9
for calculating the indifference curves. As a result, they and the winset of the SQ are “heart”-shaped.
In politics one encounters this type of non-reciprocal NSP in different ways. Either a
political actor decides over policies and corresponding budgets, or over the delegation of
competences to a supranational authority and the division of power within this authority,
or over the amount of regulation in combination with the extent of exceptions for affected
(befriended) actors, etc. In all cases, the nonseparability is non-reciprocal in the sense that
the allocation of competences, budget, etc. depends on the policy it will be used for, but
not vice versa. In other words, the policy dimension defines the characteristics of the
97 In Sec. A.2 I proof the derivation of the WED including non-reciprocal nonseparability (Eqn. 2.11).
49
political will, whereas the allocation dimension assigns the means with which this policy
is to be enacted.
2.4.6 How to measure nonseparable preferences
The empirical estimation of NSP requires evaluating actors’ utility function at more than
one point of the policy space, i.e., one needs more information than just the ideal position
or the first preference. This is not a trivial task when collecting data. For example, in
studies using expert interviews for data collection this would increase the length of the
interview. The experts would have to provide hypothetical evaluations of actors’ utility
function at several values. Typical questions would read as: “Assume government Z is
confronted with a policy x which is unchangeable. How would this affect government
Z’s preference over the size of the budget y for that policy program?”
Lacy (2001b, p. 253ff) used the following arrangement of questions to identify NSP in a
survey. First, respondents were asked for their unconditional first preference on an issue,
e.g., “Do you think the state should increase in-come taxes, cut income taxes, or keep
income taxes where they are now?” Second, respondents were asked for their uncondi-
tional first preference on another issue, e.g., “Do you think the state should spend more
money to fight crime, less money to fight crime, or continue spending the same as it does
now?” Finally, to identify NSP, the survey asked for the conditional response of subjects
with follow-up questions: “If the state significantly cut income taxes, then would you
want the state to spend more money to fight crime, less money to fight crime, or continue
spending the same as it does now?”; “If the state significantly increased income taxes,
then would you want the state to spend more money to fight crime, less money to fight
crime, or continue spending the same as it does now?”; etc. This way of questioning had
to be repeated in all combinations, and also in reversed order.
It is important to understand that this approach is a significant improvement for measur-

ing NSP compared to Hansen (1998) who looked for trade-offs across tax and spending
issues. The questions read as e.g., “Do you favor cuts in spending on national defense in
order to increase spending on domestic programs like education, and highways?”; “Do
you favor an increase in the federal budget deficit in order to increase spending on do-
mestic pro-grams like Medicare, education, and highways?” These questions ask for a
respondent’s preference on two issues simultaneously. Thus, it is not clear if a respon-
dent answers “no” because they do not want an increase in spending, or they do want to
increase spending, but just by other means. Lacy (2001b, p. 242) therefore classified these
types of questions as “double-barreled”.
An innovative survey methodology was put forward by Bonica (2012). He analyzed

the relationship between fiscal preferences and ideology using an interactive question-
naire. Baseline respondents were presented with the President’s requested budget. Sub-
sequently, they were asked to adjust the spending levels of the single categories with
50
2.5 The impact of nonseparable preferences
respect to their personal preferences. The “interactive” set-up incorporated trade-offs be-
tween spending dimensions. The author found that the preferences on security spending
correlate with self-reported ideology. In addition, he argued that public goods can be
divided into rival and non-rival government goods and services.
In summary, it is possible to measure NSP using surveys. Yet, it is time consuming,

expensive and complex. The type of questions changes from concrete to hypothetical.
Such theoretical questioning might very soon overstretch the nerves of any interviewee
expert or layman. For that reason the correct estimation of nonseparability from real
world data is a very costly and difficult undertaking. Many contributions shy away from
these efforts and just make the simplifying assumption of separable preferences.
2.5 The impact of nonseparable preferences
As shown in this chapter, NSP alter theoretical expectations in three ways. Firstly, ignor-
ing the elements in the secondary diagonal of A (cf. Eqn. 2.6) causes a misspecification
of actors’ utility functions. While this leaves the identification of the unconditional ideal
point (or first preference) untouched, the utility assigned to the remaining alternatives
will be falsely specified. As a consequence the evaluation of decision-making models
may be biased. I investigate this argument in Chap. 3 by analyzing competing models of
EU legislative politics and the consequences of NSP for their comparison.
ARGUMENT 1 (misspecification): Neglected nonseparable preferences pro-

duce misspecified utility functions. This leads to distorted results of models
relying on these functions.
Secondly, a strand of theoretical literature focuses on the effects of separating or com-

bining decisions over multidimensional policies characterized by NSP (cf. Sec. 1.2). More
specifically, the focus of these contributions was on the effect on the policy outcome when
separating decision over nonseparable issues. Such a separation assumes different deci-
sion makers or decision-making rules for each dimension. To put it more generally, the
preferences of an actor depend on exogenous factors. Well known examples are multi-
level government, multistage decision-making process, etc. (Strom, 1990, p. 107ff). The
purpose of this delegation or division of labor is to allow individual decision makers to
invest more time, effort, expertise, etc. In theory, this should improve the quality of the
results. Yet, the separation increases the complexity of the decision-making structure,
thereby weakening the link between individuals’ action sets and the foreseeable conse-
quences. Also, people are subject to cognitive limitations (Stanovich and West, 2000). It
is unrealistic to assume that humans are always able to take into account all restrictions
and implications of a decision for other topics. Thus, the effect of NSP depends heavily
on the extent and level of information an actor has (cf. Denzau and Mackay, 1981), i.e.,
the prevalent level of uncertainty (Bendor and Meirowitz, 2004).
51
Looking at committee decisions on multiple issues Ordeshook (1986, p. 252) stated that
with nonseparability “the order in which the committee votes on the issues can affect the
final outcome.” If decisions are taken sequentially,98 nonseparability works in favor of the
individuals who decide first (Strom, 1990, p. 61ff). Anticipating the effect of their choice
on the other players’ preference orderings concerning the subsequent issues, they can
realize higher gains when compared to simultaneous decisions (cf. Strom, 1990, p. 122ff).
ARGUMENT 2 (first-mover advantage): If a decision which contains nonsep-

arable preferences is taken separately and sequentially, this leads to a first-
mover advantage.
Thirdly, decisions on two dimensions may be taken separately but simultaneously. In this
case each of the two actors must anticipate what the other might choose. The situation
gets worse when each dimension is subject to collective decision-making. One example
is the case of a referendum on a set of interrelated issues (Brams et al., 1997).99 It is
important to understand that NSP are different from ordinary logrolling, the political
exchange over several issues or trade across law proposal (cf. König and Junge, 2009).
In contrast to a trade, the preference of an actor on one issue changes with the result of
a second issue. Thus, a change in one dimension leads through the conditionality, to a
re-adjustment (of preferences) in the other dimension (Strom, 1990, p. 57).
ARGUMENT 3 (sub-optimality): If a decision which contains nonseparable

preferences is taken separately and simultaneously, this leads to a sub-optimal
outcome, as coordination is difficult.
Arguments 2 and 3 are analyzed in Chap. 5 and Chap. 6. Using a laboratory experiment
I studied a collective group decision in comparison a situation in which the group is split
into two delegations. According to the arguments, I varied the level of uncertainty as
well as the voting sequence, i.e., I used sequential and simultaneous decision-making.
So far, the formulations are general in nature. They address the influence of NSP as a
matter of principle. When explaining the design of my experiment and depicting its
specific properties, I formulate concrete hypotheses for the impact of nonseparability.
Those arise from the interplay of the arguments discussed, and the experimental design.
These hypotheses are directly related to statistical measures. Their relationship to the
arguments is best understood as a meta-level (general argument) and sub-level (case-
specific hypothesis) order.
98 Krehbiel (1986, p. 542) argued that “legislative decision making is fundamentally sequential”. The impor-
tance of sequential voting was investigated by Xia et al. (2011). Looking at multi-issue domains as an
extensive-form game they proved several multiple-election paradoxes in strategic sequential voting and
noted that changing the order of the issues cannot completely prevent such paradoxes.
99 This can nicely be illustrated with the example of constructing a new town hall (cf. Sec. 2.1). A corre-
sponding referendum would ask “Should we build it at all (yes/no), where (old town/at the river) and
by which architecture (modern/classic)”. At the ballot box voters preference ordering depends on the
outcome of each single question which, unfortunately, they cannot know. So, people may prefer a new
city hall, but only in a modern design and outside the old town.
52
3 The merits and costs of incorporating
nonseparable preferences
To operationalize NSP is a demanding task. The additional effort can only be justified if
the resulting benefit is correspondingly large. This consideration has to be made every
time utility functions are specified. This chapter offers such an assessment of merits
and costs. The assumption of separable preferences has been applied throughout the
empirical research on EU legislative politics. The following sections now incorporate
NSP into spatial models of legislative decision-making. Contrasting previous work with
my results provides an answer to my first key question, whether and, if so, how much
the inference of analytical research suffers when falsely assuming separable preferences.
While legislative models are not the general focus of my research, they serve well as
examples showing the consequences of neglecting NSP. Of course, the assessment of one
empirical case is only a sample. Chap. 8 picks up again the discussion of merits and costs
in a more general context.
The remainder of this chapter is structured as follows. Sec. 3.1 discusses the literature on
models of legislative decision-making. Such models enable researchers to predict the out-
come of legislative processes. These predictions are formalized in term of winset, core,
winning coalition, etc. (cf. Sec. 2.4). Using these models it is also possible to evaluate the
observed decisions of legislative bodies with respect to certain criteria. I categorize them
into three groups according to their different levels of constraints: i) unconstrained bar-
gaining models, ii) constrained bargaining models and iii) agenda-setting models. Next,
I review the “Decision-making in the European Union” (further DEU) project (Thomson
et al., 2006) in Sec. 3.2. The project’s data set serves as basis for my empirical analysis.
The effects on legislative decision-making models, when misspecifying preferences by
omitting nonseparability, are explained in Sec. 3.3. Here, I include two exemplifying case
studies. I start the empirical analysis in Sec. 3.4 by investigating the spread of NSP in the
DEU data. Next, in Sec. 3.5 the magnitude of NSP is estimated by means of simulation
techniques. Sec. 3.6 turns from the collective outcome towards the individual votes in the
European Council. Finally, I summarize the finding of this chapter in Sec. 3.7.
The main empirical results of this chapter are also published in “The Merits of Adding
Complexity: Conditional Preferences in Spatial Models of EU Politics” (Finke and Fleig,
2013). This article investigated the impact of including NSP into spatial models of an-
alytical politics on comparisons of legislative models. It challenged the ambiguous as-
53
3 The merits and costs of incorporating nonseparable preferences
sumption of separable preferences which had been applied throughout previous empiri-
cal research. The goal of this chapter is similar, thus Sec. 3.4 and Sec. 3.5 are largely asso-
ciated with Finke and Fleig (2013). Compared with the publication this chapter contains
additional information: the overview of legislative decision-making models has been ex-
tended (Sec. 3.1), supplementary data is discussed (Sec. 3.2), an additional case study is
included (Sec. 3.3.2) and the argument for the use of modeling at the individual level is
assessed (Sec. 3.6). These extensions enable a more detailed review and comprehensive
analysis.
3.1 The literature on legislative models of decision-making
All legislative models of decision-making are built on a common foundation (Ferejohn

and Fiorina, 1975); to evaluating the models information is necessary about the insti-
tutional environment, the preferences of the involved actors as well as their relative
salience, the discussed issues, the localization of the SQ and the outcomes in policy space.
Mostly, this information is collected through surveys or expert interviews. Starting from
this common ground the different models emphasize various aspects in the decision-
making process. This also leads to different solution and equilibria concepts.
• Dynamic exchange models: These models implement an indefinite sequence of

bilateral tradings between actors. Through these (bilateral) interactions a model
reaches its solution by gradual adjustments away from its initial position in recip-
rocal concessions. If there are no more partners for bilateral exchange found, the
system is in equilibrium and the resource allocation (policy) at this point is the out-
come (Coleman, 1990).100
• Negotiation models: These models combine multilateral negotiation with institu-

tional constraints. To account for a variety of actors the models include a weighting
variable assigning the actors different degrees of influence (Schneider et al., 2006).
These models emphasize strongly the importance of institutional constraints. Typi-
cal model solution concepts are the winset, subgame-perfect equilibria or the Nash
bargaining solution (NBS). Those enable the multilateral negotiation to reach a sta-
ble equilibrium (Hug, 2009). Otherwise the chaos theorem would predict unstable
multiple equilibria (McKelvey, 1976; Richards, 1994).
• Procedural models: These models focus on the legislative process. Important as-
pects are the sequence in which different institutions are involved as well as who
possesses agenda control. They aim to assess to what extend upstream decision-
makers can influence the actions (and scope of actions) downstream. As typical
solution concept these models use backward induction.101
100 A first model of resource exchange goes back to Goode (1971).
101 For a comprehensive overview on procedural models cf. Steunenberg and Selck (2006, p. 54ff.).
54
3.1 The literature on legislative models of decision-making
I continue by presenting the most popular implementations of these legislative models

on EU decision-making.102 These are unconstrained bargaining models (Sec. 3.1.1, imple-
menting dynamic exchange), constrained bargaining models (Sec. 3.1.2, implementing
negotiation) and agenda-setting models (Sec. 3.1.3, implementing procedural aspects).103
All models are based on the principles of spatial modeling (cf. Sec. 2.4.2).
3.1.1 Unconstrained bargaining models
In his study of decision-making in the European Community Van den Bos (1991) pre-
sented a compromise model. This model assumes that a bargaining solution “takes all
positions of member states into account, weighting these by the resources a member state
can apply during the negotiation and the importance each attaches to the decision at
hand” (Van den Bos, 1991, p. 176). As a consequence the model’s prediction xC∗ results
from Eqn. 3.1, where si is the relative salience actor i attaches to the issue, vi describes
actor i’s relative bargaining power and θi is their unconditional ideal point.
∑in=1 si vi θi
xc∗ = (3.1)
∑in=1 si vi
Achen (2006) proved that the prediction of the compromise model is an approximation
of the asymmetric bargaining model with separable preferences if disagreement is un-
desirable and, therefore, not an option. This facilitates exchange and mutually adjust-
ments. The proof is of significant importance, because it allows simulating the effect of
NSP on the compromise model by applying standard spatial utility functions. The mul-
tidimensional extension of the compromise model has been termed “position exchange
model” (Arregui et al., 2006). Its prediction equals the multidimensional, unconstrained
and asymmetric bargaining model. The unconstrained bargaining model is the least con-
strained of all models. It neither considers the reference point, nor does it explicitly ana-
lyze the decision-making process.104
3.1.2 Constrained bargaining models
Using the DEU data set Schneider et al. (2010) calculated several versions of the sym-
metric and asymmetric bargaining model.105 Briefly summarized, this bargaining model
predicts that actors agree on the policy x within the winset to the SQ which maximizes
102 For a more encompassing summary of modeling EU law-making procedures cf. Thomson et al. (2006).
103 This excludes dynamic models such as the “expected utility model” (cf. Bueno de Mesquita, 2011; Stokman
and Bueno De Mesquita, 1994) which endogenize changes in actors’ negotiation position during the
process. Yet, the selection points out the connection to other areas of political research. For example,
“bargaining theories are omnipresent in studies on international relations” (König, 2013, p. 2) as well.
104 An applications to the EU can be found, e.g., in Stokman and Van Oosten (1994).
105 Baron and Ferejohn (1989) clarified the conditions under which the strategic implementation of the Stahl-
Rubinstein model (Rubinstein, 1982) and the axiomatic bargaining model (Nash, 1950) lead to identical
equilibrium predictions.
55
the product of their utility functions (further Nash product106 ). They assume the spatial
Euclidean utility function presented in Eqn. 2.4. As Schneider et al. (2010) used the as-
sumption of separable preferences Eqn. 3.2 presents the symmetric version of the model:
n
xnbs = max xϵΘ ∏ (ui ( x ) − ui (SQ)) (3.2)
i =1
The utility is defined as the difference between the distance of the actor’s ideal position
to the disagreement value (i.e., SQ) and the distance of the actor on the proposed policy.
The constrained bargaining model is more restrictive than the unconstrained bargaining
model in that it constrains the set of feasible outcomes to the winset. The latter is deter-
mined by the voting rules for each proposal, either qualified majority voting (QMV) or
unanimity in the Council.
3.1.3 Agenda-setting models
Agenda-setting models produce equilibrium predictions by solving the multistage EU

law-making game via backward induction. They are part of a broad rational choice liter-
ature, which regards the results of policy actions as a common result of preferences and
institutions (e.g., Ostrom, 1986; Shepsle and Weingast, 1995). Hörl and Warntjen (2005)
reviewed previous literature on legislative decision-making which use procedural spa-
tial models. The authors conceded that these models have considerably enhanced the
understanding of EU legislative decision-making. However, the article also explicitly
addressed criticism concerning central assumptions of the approach, in particular the
formation of preferences within legislative bodies.
Most prominent are the competing models by Tsebelis (1994), Moser (1997) and Steunen-
berg and Vught (1997). An important distinction between the models must be made with
respect to the significance of European Commission and European Parliament in the leg-
islative process under different institutional arrangements.107 Under the consultation
procedure108 the EP may approve, reject or amend a legislative proposal proposed by the
Commission. However, most of the authors agreed that the European Commission can
successfully set the agenda by placing its proposal inside the winset of the SQ and the
amendment proof set, i.e., the unanimity core. The powers of the EP in the co-decision
procedure are theoretically more contentious.109 This legislative arrangement devotes the
same weight to EP and European Council. Here, some authors claim that the EP enjoys
conditional agenda-setting powers, others argue in favor of keeping the Commission as
106 The Nash product is the maximand of the Nash Bargaining Solution (Nash, 1950).
107 Cf. Steunenberg (1994) for a comparison of decision-making in the European community under different
institutional arrangements. He examined the consultation procedure, the cooperation procedure as well
as the co-decision procedure.
108 Cf. http://www.europarl.europa.eu/aboutparliament/en/00c5e7159b/Consultation.html (accessed
April 28, 2013).

109 Cf. http://www.europarl.europa.eu/aboutparliament/en/0080a6d3d8.html (accessed April 28, 2013).
56
3.2 Data on decision-making in the European Union
agenda-setter, but successful amendments cannot violate the preference of the median
MEP (this is called “amendment proof” ). Subsequently, I follow the latter approach.
Hence, the agenda-setting model not only constrains the set of feasible solutions to the
winset, but furthermore includes the agenda-setting power of the Commission.
3.2 Data on decision-making in the European Union
The DEU project (Thomson et al., 2006) resembled an encompassing summary of model-
ing EU law-making procedures. This collaborative research project gathered information
on the (ideal) position and salience of national governments, the European Commission
and the EP on a total of 66 law-making proposals. To ensure a high level of salience (i.e.,
political relevance and contestation), the proposals were selected from the press archive
Agence d’Europe110 . The political conflict on these proposals revealed between two and
six issues, resulting in a total of 192 issues discussed in the Council between January 1999
and December 2000.
The ultimate purpose of the research project was a comparative evaluation of compet-
ing models of EU law-making (cf. Achen, 2006; Schneider et al., 2006). The tremendous
success of the research project manifested itself in more than 170 citations in articles and
books and a still growing amount of publications using the data set (for a recent overview
cf. Thomson, 2011). Motivated by this success a follow-up project has been carried out
which focuses on yet another 59 legal proposals initiated after the Eastern enlargement of
the European Union (henceforth DEUII, Thomson et al., 2012). The revised and expanded
DEUII data includes the described information for 125 legislative proposals comprising,
in total, 331 controversial issues of EU legislation between 1996 and 2008 (Thomson et al.,
2012, p. 604).
Looking more into the details, the DEU data contains information for 51 multidimen-
sional proposals which in total comprised 144 issues. They cover in equal share the Con-
sultation and the Co-decision procedure, as well as unanimity voting and QMV111 . The
DEUII (Thomson et al., 2012) extended this collection to 101 multidimensional proposals
containing 307 issues. The data shows clearly that (at least the contested) EU law-making
is characterized by multidimensional (i.e., complex) decision-making.
In the DEU and DEUII data all information (i.e., individual policy position and salience
of every member state for every issue) has been scaled between 0 and 100. The data in-
cludes information on the location of the SQ (enacted policy previous to the negotiations)
110 Homepage: http://www.agenceurope.com.
111 The QMV threshold refers always to the Treaty of Nice (Nice, 2001). Originally, the Treaty of Amsterdam
(Amsterdam, 1997) established the European Community and defined the voting rule for the European
Council of ministers (article 205 section 2). The Treaty of Nice (article 3) amended the weighting of votes
in the Council and the QMV threshold. It added the requirement that “the Member States constituting
the qualified majority represent at least 62 % of the total population of the Union” as well as “two-thirds
of the members” (Treaty of Nice, article 3, Sec. 1.b and 1.c).
57
and the final outcome. All the data has been gathered by expert interviews.112 Sec. A.3
provides a short overview on the questionnaire of these interviews. In the case of the
DEU data set I also had access to the documentation of the interviews: each proposal
was accompanied by a short field report, drafted by the responsible interviewer, which
contains a short description of the major issues, an assessment of the member states opin-
ions and the Commission’s intention when drafting the proposal. These reports proved
to be a very helpful starting point for the investigation into potential nonseparability.
The empirical analysis on the influence of NSP on legislative decision-making models

was restricted to data from the DEU project (as in Finke and Fleig, 2013) for two reasons.
Firstly, success and acceptance of the originally data set are undeniable (Thomson, 2011).
I have no doubt that the DEUII will develop to the same prominence, but this can only be
proven by time. Secondly, the investigation of NSP is difficult as the original interviews
did not ask for nonseparabilities. Based on the field reports, which document the inter-
views in detail, this problem could be solved at least to some extent. As I had only access
to the reports of the DEU data set I restricted my assessment of NSP accordingly.
Before going into the analysis of the data I take a closer look at the process of the data col-
lection. The most important aspect of empirical data is its reliability. The in-depth expert
interviews of the DEU project are recorded precisely. Also, the interviews have been con-
ducted across all proposals with the same level of thoroughness. But can interviewing
experts provide enough information to code a proposal? This includes the identification
of the single issues in the proposal as well as position and salience estimates for every
single member state in each issue. An alternative approach would have been to identify
the single issues of each proposal through hand coding.113 Appendix II of Thomson et al.
(2006) actually contrasted the expert judgments with information found in Council doc-
uments. Benoit and Laver (2007b) set the results of expert interviews and hand-coding
against each other and provided a discussion on the effects and consequences. The au-
thors concluded that no method is generally superior to the other. When considering
some caveats it is even possible to combine the two approaches.
Leuffen et al. (2012) compared different aggregation strategies for multiple sources in
small-n research. The authors focused on qualitative research strategies, e.g., process
tracing.114 Overall, weighted averages of multiple origins performed best if the varying
sources are independent. If independence cannot be assured the research should focus
on the most precise and accurate data, depending on the trustworthiness of the source.
With respect to the DEU project (nearly) all experts “had first-hand knowledge [...] and
were usually participants” (Thomson et al., 2006, p. 32) in the decision-making.115 The
112 For each proposal up to 5 different experts were interviewed.
113 This does not apply to the preferences of the Council members because “the written accounts of council
meeting often do not detail the positions initially favored by the actors” (Thomson et al., 2006, p. 31).
114 In general, also the mixing of qualitative and quantitative methods is possible. Here, Wolf (2010) provided
an excellent discussion (including application criteria) on triangulation.

115 Sec. 2.2 of Thomson et al. (2006) provides an overview on the criteria used in the expert selection process.
58
3.3 The effects of misspecifying preferences
project also offers information on every single person interviewed which secured “the
reliability and objectivity of measurement” (Leuffen et al., 2012, p. 3).
In general, the assumption of Euclidean space is not without problems (cf. Sec. 2.4). Yet,
when taking a closer look at the proposals and issues listed in the DEU data set ,I found
that they rarely represent the type of allocation-rivalry problem I describe in Sec. 2.4.2.
Therefore I proceed under the assumption of Euclidean space.
In this section I explain how the prevalent assumption of separable preferences may bias
a comparative evaluation of the aforementioned legislative models. The models differ
with respect to their different levels of constraints. NSP affects these constraints and
leads to heterogeneous model predictions. This is because constrained models have a
tendency to produce extreme predictions. For example, in terms of the DEU project, con-
sider the most extreme case in which only one country prefers the SQ. All others prefer
significant changes in the same direction. In the case of unanimity voting, this implies
that constrained models will predict the SQ to prevail. Another extreme scenario illus-
trates the reverse effect. Assume the SQ equals zero, all member states except one are
located at 50 and the one member state as well as the agenda-setter prefer full reform
(100). Following the agenda-setting model, the agenda-setter chooses a proposal clos-
est to its ideal point within the winset of the SQ and the amendment proof set i.e., the
unanimity core. As a result the model predicts full-scale reform.
The crucial question therefore is to what extent a misspecification of NSP endangers the
inference drawn from empirically testing spatial models of law-making. Such effects of
false specification may bias the evaluation of competing models of decision-making, as
the constrained models predictions hinge upon the correct specification of the winset and
the unanimity core. The following two sections present two more detailed case studies.116
Both plainly illustrate two points. Firstly, the assumption of (non)separable preferences
significantly alters the winset to the SQ. Secondly, the results of the model comparison
depend on the level of the assumed nonseparability.
3.3.1 Illustrative case study I
In May 1999, the Commission proposed a Council regulation for “closer dialogue with
the fishing industry and groups affected by the common fisheries policy”117 (further
116 Both case studies discuss a proposal contained in the original DEU data (Thomson et al., 2006). The
following specification of the issues is based on the corresponding field reports.
117 The Commission Proposal COM (1999) 382 - 1999/0163/CNS is accessible at http://ec.europa.eu/
prelex/detail_dossier_real.cfm?CL=de&DosId=147501 (accessed August 29, 2012).
59
COM1999/163). The Council had to decide by QMV; the role of the EP had been con-
sultative. The purpose of the proposal was to create a legal basis for supporting the
representation of fishery organizations at the EU level.
Within the Council, intergovernmental conflict arose along two issues.118 The first is-
sue concerned the extent to which national organizations should be represented in the
newly established Advisory Committee on Fisheries. Advisory Committees are widely
regarded as highly influential with respect to the EU’s tertiary legislation (König et al.,
2010). The set of possible solutions ranged from an exclusive representation of EU-level
organizations to a privileged representation of national organizations. The legal SQ pro-
vided for a mixture of national and EU-level organizations being represented. The second
issue arose over the kind of legal basis which should be created for the financial support
of fishery organizations to enable them to organize at the EU level. Three alternatives
were on the table: i) no financial support whatsoever for fishing industry organizations,
ii) the expenditure should be included in the optional “A”-part of the EU budget, iii) a
legal obligation should be created to provide sufficient funding for national organiza-
tions to participate in the Advisory Committee, and this post should be included in the
“B”-part of the EU budget. The SQ was optional financial support.
In the following I assumed that actors’ spending preferences are conditionally dependent
on the representation of interests in the committee to be created (cf. Sec. 2.4.5). Here, the
argument was that representation predetermines the advisory committee’s policy deci-
sions. Therefore, any deviation from member states’ ideal positions on the representation
issue reduced the expected satisfaction with committee decisions and caused a decrease
in their preferred level of financial support. In other words, I assumed non-reciprocal
nonseparability and therefore apply Eqn. 2.11.
Fig. 3.1 depicts member states’ unconditional ideal positions, the SQ ante, the winset for
QMV, the actual policy outcome and the model predictions for separable preferences.119
Member states with a powerful fishery lobby, such as Sweden, Portugal, Ireland, Spain,
the UK, the Netherlands and France, preferred maximal powers for their national or-
ganizations in the Advisory Committee. While most other member states preferred the
maintenance of the SQ, only Germany, the EP and the European Commission preferred
to completely exclude national organizations. Moreover, the German government was
alone in its pursuit off cutting the financial support to the Advisory Committee. While
France, the Netherlands and the UK preferred to maintain the present level of financial
118 In addition to the expert interviews the conflict is also well documented in Consilium (1999); this inter-
institutional dossier of the European Council shows the record of the discussion of the group “Internal
Fisheries Policy” on the commission proposal. It lists in detail the comments from member states (Con-
silium, 1999, Chap. II, p. 2ff) as, e.g., the “serious reservations [...] whether it is really necessary to adopt a
new regulation and new financial measures” (Consilium, 1999, p. 3) expressed by the German delegation.
119 The presentation of this simple two-issue case is also possible through a three-dimensional graph. This
form of presentation depicts the utility aggregation in form of “utility-hills” more pictorially. However,
the undulating contour blocks information. Therefore, I include the three-dimensional representation
only in Sec. A.4.
60
support, most other countries preferred some extension, with Austria, Belgium, Finland,
Greece, Luxembourg and Sweden demanding the highest level of financial support. Not
surprisingly, the European Commission (and the EP) preferred a new obligatory item in
its budget.120 The winset in Fig. 3.1 originates from the SQ, and multiple “mountains”
are visible within it. These result from the different possible winning coalition. Most
important are the locations of the model predictions, compared to the actual outcome.
Under the assumption of separable preferences the agenda-setting model performs best.
Figure 3.1: Model predictions for COM1999/163 with separable preferences

EXPLANATORY NOTE
The figure depicts the spatial representation of member states’ unconditional ideal points and model predictions for
the Council regulation COM1999/163. The two issues of the proposal are the extent of national representation and the
corresponding funding. The contour plots indicate the Nash product within the QMV winset of the SQ. Accordingly, the
thick lines demarcate the border of the winset. Utility gains are calculated on the basis of Eqn. 2.6.
Predictions for separable preferences: A (72.4 / 59.7), B (64.9 / 69.3) and C (59.5 / 71.8).
LABELS
A = prediction of the agenda-setting model; B = prediction of the unconstrained bargaining model; C = prediction of
the constrained bargaining model; AT = Austria; BE = Belgium; COM = European Commission; DE = Germany; DK =
Denmark; EL = Greece; EP = European Parliament; ES = Spain; FI = Finland; FR = France; IE = Ireland; IT = Italy; LU =
Luxembourg; NL = Netherlands; PT = Portugal; SE = Sweden; UK = United Kingdom.
Subsequently, I increased the level of nonseparability. Under the assumption of Euclidean

utility functions, Eqn. 2.10 provided an (empirically necessary) upper bound to the pos-
sible range of values. However, there is no ex-ante argument as to where in this range the
correct value may be located. Thus, I scaled the level of nonseparability as percentage of
the upper bound defined in Eqn. 2.10.
Fig. 3.2 shows the same information as before, but this time under different levels of
nonseparability. Overall, increasing nonseparability shifted the winset and the prediction
120 Thisdoes not necessarily contradict the view of the European Commission as “motor of integration”
(Bailer, 2006, p. 12).
61
towards the de facto outcome (100/50): compared to the SQ ante (35/50) this means no
change in policy but one with respect to the financial support. The shift from 0% to 50%
nonseparability increased the predictive accuracy of all three models. Unchanged, it was
still the agenda-setting model delivering by far the most accurate prediction. Given its
unconditional ideal position and its high salience on the issues of financial support, the
European Commission placed its proposal at the lower right border of the winset which,
thanks to the extreme German position, is also located within the Council’s amendment
proof set.
Figure 3.2: Model predictions for COM1999/163 with nonseparable preferences

EXPLANATORY NOTE
The figures depict the spatial representation of member states’ unconditional ideal points and model predictions for
the Council regulation COM1999/163. The two issues of the proposal are the extent of national representation and the
corresponding funding. The contour plots indicate the Nash product within the QMV winset to the SQ. Accordingly, the
thick lines demarcate the border of the winset. Utility gains are calculated on the basis of Eqn. 2.11.
Predictions for NSP, when the amount of NSP is determined by Eqn. 2.10 and amounts to ...
... 50% of the maximum, i.e., a12 + a21 = a11 +2 a22 . ... 100% of the maximum, i.e., a12 + a21 = a11 + a22 .
A (85.2 / 44.5); B (63.9 / 57.5); C (75.7 / 71.4). A (99.9 / 59.7); B (86.7 / 73.1); C (92.0 / 70.5).
LABELS
Under full-blown nonseparability, the agenda-setting model’s prediction proposal was

located very close to the Commission’s ideal point. At this level of nonseparability, the
constrained bargaining model proved to be most accurate. The example clarifies that
a different degree of nonseparability leads to different model predictions and different
conclusions when assessing model performances.
3.3.2 Illustrative case study II
In November 1999, the European Commission proposed a regulation amending the “com-
mon organisation of the market in bananas”121 (further COM1999/582). The purpose of
121 The Commission Proposal COM (1999) 582 - 1999/0235/CNS is accessible at http://ec.europa.eu/
prelex/detail_dossier_real.cfm?CL=en&DosId=153098 (accessed December 20, 2012).
62
the proposal was to modify the existing quota import regime for bananas, because suc-
cessive rulings from the World Trade Organization (WTO) found certain aspects of the
previous regime122 not to be in conformity with WTO standards (EC, 2000). As bananas
are a common commodity the proposal was important for many countries that export
bananas to the EU as well. Thus, in addition to the EP Ecuador and the U.S. participated
consultatively in the debate.123 I again chose a Council decision reached by QMV, be-
cause the minor blocking-possibilities of single states enable me to highlight the effect of
NSP clearer.
The Council discussed two controversial issues of this regulation in particular. The first
issue was the type of import regime that would be adopted. The set of discussed frame-
works ranged from i) keeping the SQ (which would violate the WTO ruling) ii) over a
simple increase of the current quotas iii) to a transformation of the system into a liberal
regime based on a flat tariff. In this continuum the SQ represents a protectionist quota
system. The second issue arose over a transitional period during which a tariff quota
would apply. The objective of the transitional regime was to enable banana-producing
regions to make appropriate adjustments to a freer market. While the general idea was
not contested its duration was. The different options asked for a transitional system un-
til i) the year 2000, ii) the year 2006, iii) the year 2010 or an unlimited system. The SQ
ante referred to unlimited usage as the date for a new import regime had to be set by the
proposal.
With respect to this proposal I assumed that actors’ preferences are conditionally de-
pendent. This judgment was reinforced by the corresponding field report of the DEU
project and the expert statement ascertaining that “the type of system and the timing
are intimately interrelated because the agreement on one may accept the agreement (or
disagreement) on the other one and vice-versa” (Thomson et al., 2002, p. 7). More specifi-
cally, the degree of liberalization determined how much transitional time was considered
necessary to adapt. Furthermore, to set a date for the new import system influenced
how much liberalization was deemed appropriate. In other words, I assumed mutually
positive and reciprocal nonseparability and apply Eqn. 2.9.
As in the previous illustrative case study I start by depicting in Fig. 3.3 member states’
unconditional ideal positions, the SQ ante, the winset under QMV, the actual policy out-
come and the model predictions for separable preferences.124 The graph shows that,
again, member states lined up according to interests of powerful national lobbies. Eu-
122 The previous banana import system was based on the Council Regulation (EEC) No 404/93, as amended
in 1998 by Regulation (EC) No 1637/98.
123 In addition to the expert interview the proposal is also well documented in Consilium (2000); this inter-
institutional dossier of the European Council explains in detail the proposal intentions, the amended
paragraphs and the “numerous close contacts with supplier countries and other interested parties” (Con-
silium, 2000, p. 2).
124 I also include for this case study a three-dimensional representation in Sec. A.4. This presentation de-
picts the utility aggregation in form of “utility-hills” more pictorially but the undulating contour blocks
information.
63
ropean producers of bananas, i.e., France, Spain and Portugal, preferred the current SQ
(0/100). It was similar in the case of Ireland and the UK. Both host large companies work-
ing in this sector.125 Italy also favored the SQ despite not producing any bananas. The
reason is that Italy preferred subsidies for bananas to diminish the corresponding (fixed)
EU budget to protect other vegetables of importance for Italy.
Figure 3.3: Model predictions for COM1999/582 with separable preferences

EXPLANATORY NOTE
The figure depicts the spatial representation of member states’ unconditional ideal points and model predictions for the
Council regulation COM1999/582. The two issues of the proposal are the import system to be adopted for the commerce
and distribution of bananas within the EU countries and the year in which the new system will take effect. The contour
plots indicate the Nash product within the QMV winset to the SQ. Accordingly, the thick lines demarcate the border of
the winset. Utility gains are calculated on the basis of Eqn. 2.6.
Predictions for separable preferences: A (24.1 / 68.3), B (11.8 / 62.4) and C (31.3 / 55.8).
LABELS
Denmark; EC = Ecuador; EL = Greece; EP = European Parliament; ES = Spain; FI = Finland; FR = France; IE = Ireland; IT
= Italy; LU = Luxembourg; NL = Netherlands; PT = Portugal; SE = Sweden; UK = United Kingdom; USA = United States
of America.
A rather large coalition of countries (with no economic relations to banana import) was
located at the other extreme, promoting a liberal regime in favor of cheap consumer
prices. The report also outlined the intentions of Ecuador and the U.S. which, not surpris-
ingly, preferred a liberalization of the current system. The commission was located at a
medium position on both issues, perhaps having positioned itself as a mediator between
the extreme coalitions. Only one winset originated from the SQ. The contrast between the
125 Inthe corresponding field report the interviewed expert clarified that in both countries important compa-
nies trade directly with ACP countries (African, Caribbean and Pacific Group of States).
64
two groups made it hard to reach an agreement at all. Under the assumption of separable
preferences the constrained bargaining model performed best.
In Fig. 3.4 I scaled the level of nonseparability as percentage of the upper bound defined
in Eqn. 2.10. The figure provides the same information as before under different levels of
nonseparability. Increasing the amount of NSP shifted the winset and model prediction
towards the de facto outcome (50/40) which means a moderately strong liberalization
within a medium transfer period. Interestingly, the outcome exactly matched the position
of the U.S. and the European Commission.126
Figure 3.4: Model predictions for COM1999/582 with nonseparable preferences

EXPLANATORY NOTE
The figures depict the spatial representation of member states’ unconditional ideal points and model predictions for the
Council regulation COM1999/582. The two issues of the proposal are the import system to be adopted for the commerce
and distribution of bananas within the EU countries and the year in which the new system will take effect. The contour
plots indicate the Nash product within the QMV winset to the SQ. Accordingly, the thick lines demarcate the border of
the winset. Utility gains are calculated on the basis of Eqn. 2.9.
Predictions for NSP, when the amount of NSP is determined by Eqn. 2.10 and amounts to ...
... 50% of the maximum, i.e., a12 + a21 = a11 +2 a22 . ... 100% of the maximum, i.e., a12 + a21 = a11 + a22 .
A (28.4 / 57.1); B (18.8 / 57.7); C (32.0 / 55.7). A (50.0 / 40.0); B (44.6 / 39.4); C (35.1 / 52.8).
LABELS
of America.
The shift from 0% to 50% nonseparability increased the predictive accuracy of the agenda-
setting and the unconstrained bargaining model. But still the constrained bargaining
model provided the most accurate prediction. The increase in nonseparability clearly re-
shaped and enlarged the winset of the QMV decision. This corresponds to the discussed
changes of indifference curves (Sec. 2.4.2) which become more and more tilted as NSP
increase. Under full nonseparability the winset extended further and was adjacent to the
126 The field report clarified that the depicted outcome represents the in 2006 valid agreement. Later on,
the system was further liberalized. France, Spain and Portugal gave up their resistance and a flat tariff
regime was introduced.
65
de facto outcome. Thus, the agenda-setting model predicted the outcome exactly. The
performance of the unconstrained bargaining model also further improved. Its predic-
tion was now very close to the outcome, in contrast to the constrained bargaining model
which now performed worst. This example further illustrates the effect of NSP on the
different legislative models.
3.3.3 Hypotheses
The aim of the DEU research project was a performance comparison of different legisla-
tive models. Such a comparison helps to determine the importance of processes and
institutions in relation to bargaining power and dynamic interactions. In the existing
literature the predictive accuracy of competing models is evaluated by comparing the
mean average error (MAE) at the issue or at the proposal level (e.g., Schneider et al., 2010;
Thomson, 2011).127 As an alternative, and to account for case-specific idiosyncrasy, pre-
dictive accuracy was also evaluated using pairwise comparison and so-called hit rates,
i.e., predictions within a predefined, narrow margin to the real outcome (cf. Achen, 2006).
The DEU project claimed two findings (Schneider et al., 2006, p. 303): firstly, uncon-
strained models (such as the unconstrained bargaining model) reveal a smaller MAE.
Secondly, the constrained models (e.g., the agenda-setting model) reveal a better hit rate.
Yet, as the DEU project did not look into possible conditionalities in their multidimen-
sional law proposals, the final conclusions of their research may be biased. The sensitiv-
ity to the correct specification of winset and core depends on correctly specified utility
functions. This brings me to the following hypotheses.
HYPOTHESIS 1: Falsely assuming separable preferences decreases the predic-

tive power of models and constrains the set of feasible outcomes.
HYPOTHESIS 2: Agenda-setting models are particularly vulnerable because

their predictions are placed at the boundaries of the falsely specified set of
feasible outcomes.
A ubiquitous assumption of separable preferences can lead to biased predictions and

wrong implications about the logic of policy-making. This directly follows from ARGU -
MENT 1 formulated in Sec. 2.5: “Neglected nonseparable preferences produce misspeci-
fied utility functions. This leads to distorted results of models relying on these functions.”
I argue that the conditionality of preferences should be taken into account. This holds for
the collection of such data as well as for the empirical analysis.
My analysis proceeded in three steps. Step one identified the occurrence and direction of
NSP in the DEU data set. Working on from there, step two used the coded NSP scheme
as independent variable for the model comparison. Thus, revealing the different impact
127 The MAE is calculated as the average distance between model prediction and the facto outcome across all
observations.
66
3.4 The extent of nonseparable preferences in EU law-making
of nonseparability on the models’ predictive accuracy. As the interviewers did not ask
for NSP it was not possible to encode distinct values for NSP.128 Thus, in step three I
simulated model predictions for different levels of NSP.
3.4 The extent of nonseparable preferences in EU law-making
In this section I investigate the extent of NSP in the DEU data set. Therefore, I filled
in the secondary diagonal of the matrix A (Eqn. 2.6) with regard to the existence and the
direction of NSP. Furthermore, I identified whether the effect is reciprocal or not. If so, the
nonseparability term had to be modified by using the absolute distance of the allocation
dimension as introduced in Eqn. 2.11.
The coding itself was conducted by three graduate students independently of one an-
other. All students received a thorough introduction into the issue of nonseparability.
The coding focused on three aspects: Firstly, it was identified whether or not for actors’
preferences over two issues nonseparability exists at all. Secondly, the direction of non-
separability (positive or negative) was added. Third, it was determined whether the NSP
are reciprocal or not, i.e., whether the issues can be classified to resemble a policy and an
allocation dimension. Inter-coder reliability was approximately 90% with respect to the
existence and direction of nonseparability and 75% with respect to the reciprocity.129
Tab. 3.1 summarizes the results of the coding efforts. NSP were discovered in half of the
multidimensional proposals and in slightly more than half of the issues.
Table 3.1: Extent of nonseparability in EU law-making

EXPLANATORY NOTE
The table lists the extent of nonseparability in the DEU data set (Thomson et al., 2006). The observations are structured
according to the number of issues within a proposal.
No. of Total no. of No. of proposals No. of issues affected

Total no. of proposals
dimensions issues affected by NSP by NSP
2 25 50 14 28
3 15 45 7 18
4 7 28 4 15
5 3 15 2 10
6 1 6 1 6
Total 51 144 28 77
These values do not imply, however, that half of the cells in the secondary diagonals
of matrix A were affected. For example, the proposal for a regulation of “Audiovisual
industry: development, distribution and promotion of works” (CNS/99276) aimed at
subsidizing certain enterprises in the audiovisual industry. Issue 1 was concerned with
128 However, the field reports contain information about the interrelation of issues. This information was
used to construct the NSP scheme in step 1.
129 As stated before, the main empirical results of this chapter are published in Finke and Fleig (2013). The
coding scheme resembles the empirical basis of this article. Accordingly, the correct coding of disputed
issues in the NSP scheme has been discussed and settled by the authors.
67
the amount of money allocated to the project, whereas issues 2, 3, 4 and 5 dealt with the
distribution of the money. Hence, NSP were suspected to exist between issue 1 and each
of the other four issues. Preferences between issues 2, 3, 4, and 5 were separable. Never-
theless, Tab. 3.1 reports all fives issues as being somehow affected by nonseparability.
Sec. A.5 provides a complete list of the NSP coding, including a short description of pro-
posals and issues. Admittedly, it is unfortunate that it is only possible to theorize on and
not prove the existence of NSP. However, in much the same way all previous empirical
studies using the DEU data could only assume but not prove separable preferences.
Next, the NSP scheme served as independent variable for the model comparison. The
dummy variable “NSP” equaled 1 if an issue had been affected by NSP, otherwise it
equaled 0. Tab. 3.2 compares the predictive accuracy of the three models introduced
above (Sec. 3.1). On average, it repeats the main findings of the DEU project. Mod-
els highly constrained by procedural assumption reveal a higher MAE than the uncon-
strained bargaining model. Following Tab. 3.2, this difference in the models predictive
accuracy does not depend on the existence of NSP. As the table contrasts findings on
issue and proposal level, I reveal the number of observations N in each case separately.
Table 3.2: Model comparison of mean average error on issue and proposal level
EXPLANATORY NOTE
The table lists the MAE on issue and proposal level for the different models. It separates predictions with and without
NSP. The standard deviation (SD) is shown in parentheses.
MEAN AVERAGE ERROR

without NSP with NSP
LEVEL OF ANALYSIS Issue Proposal Issue Proposal
N=63 N=23 N=81 N=28
24.47 29.67 24.57 28.67

Unconstrained bargaining model
(18.73) (19.01) (26.69) (17.73
30.66 28.99 28.22 31.59
Constrained bargaining model
(25.44) (24.89) (24.35) (25.44)
29.59 29.60 27.48 32.66
Agenda-setting model
(30.01) (30.88) (26.77) (32.48)
Reporting standard errors and confidence intervals (CI) has been criticized because the
policy positions in the DEU data have been measured using different scales (e.g., Achen,
2006). Hence, standard errors should be interpreted with caution. Moreover, Junge (2010)
argued that the error term should be modeled at the level of individuals’ utility function
because theoretical models themselves are based on the actions of individuals. Although
the merits of this approach seem clear, so far, it is not the accepted new modeling stan-
dard.130 Sec. 3.6 picks up this argument again and provide an overview and outlook on
voting models based on individual data.
The relative accuracy changed when comparing the three models pairwise. The regres-
sion results in Tab. 3.3 reveal that the bias resulting from NSP is significantly advanta-
geous to the unconstrained bargaining model. The regression models deployed the stan-
130 This assessment of the literature excludes those studies which use the DEU data to explain choices of
individual actors (e.g., Cross, 2012; König and Junge, 2009).
68
3.5 The magnitude of nonseparability
dard set of control variables used for model comparison in the DEU project (cf. Schneider
et al., 2010). This included two variables controlling for a potentially systematic measure-
ment error, namely scale and skewness of the preference distribution. The most powerful
predictor to explain the models’ different errors was a dummy for whether or not the SQ
has in fact been changed. As argued in Sec. 3.3.3, constrained models are very sensitive
to the veto power of individual member states. Accordingly, it comes as no surprise that
the unconstrained bargaining model was significantly more powerful in predicting pol-
icy change. Finally, the dummy for NSP turned out to be significant for a comparison
between the unconstrained bargaining model and the agenda-setting model, as well as
for a comparison between the unconstrained and constrained bargaining model.
Table 3.3: Nonseparability effect on relative model performance

EXPLANATORY NOTE
The table shows the effect of nonseparability on relative model performance at the issue level. Statistically significant
(two-tailed) at the 0.1 level *, at the 0.05 level ** and at the 0.01 level ***.
N = 144 RELATIVE MODEL PERFORMANCE

△Error: △Error: △Error:
unconstrained unconstrained constrained
bargaining model bargaining model bargaining model
- - -
constrained agenda-setting agenda-setting
bargaining model model model
Coef. SE Coef. SE Coef. SE

Scale (1 if dichotomous) -2.63 4.96 -11.77 7.68 -9.98 6.62
Number of Issues -1.43 1.49 0.11 2.31 1.32 1.96
Skewness of preference
-3.24* 1.66 1.88 2.57 5.12** 2.19
distribution
Voting rule (1 if QMV) 6.54* 3.73 8.17 5.78 1.62 4.91
Procedure (1 if
2.60 3.48 -1.18 5.19 -3.73 4.57
Co-decision)
Reform (1 if SQ
18.88*** 8.33 23.94*** 6.70 5.07 5.69
prevails)
NSP -6.76** 3.41 -9.35* 5.27 -2.59 4.48
Constant -3.35 5.94 -10.87 18.67 -7.52 7.81
Adjusted R2 0.16 0.10 0.02
So far, the evidence shows that nonseparability occurs in EU law-making (Tab. 3.1). It also
biases the comparison of constrained to unconstrained models if misspecified (Tab. 3.3).
The ultimate question is: is it possible to accommodate NSP to ensure a fair model com-
parison? In the most ideal case, the information for each cell in the secondary diagonal
of matrix A could be gathered via expert interviews, just as the values for salience in the
main diagonal. Of course this would increase the length of the interview, as the experts
would have to provide hypothetical evaluations of actors’ utility function at several val-
ues (as discussed in Sec. 2.4.3). But it is simply impossible to gather actor-specific data on
the strength and direction of any potential nonseparability in retrospect. And the inter-
viewers of the DEU project did not ask specifically for NSP when conducting their expert
69
interviews. This section will therefore discuss the simulation of model predictions for
different degrees of NSP.
Above, the elements in a secondary diagonal have been coded with respect to three crite-
ria (cf. Sec. A.5): the existence of nonseparability, its direction (positive or negative) and
whether the nonseparability is reciprocal or not. All this information was used to enhance
the simulation of different values of NSP. The strength of nonseparability was simulated
for values between 0 and the upper bound defined in Eqn. 2.10, i.e., aii × a jj − aij × a ji > 0.
This upper bound varied according to the relative salience each actor attached to the two
issues in question. To ensure the comparability of the results the simulated strength of
nonseparability was operationalized as a percentage of these maximal values.
This approach has several limitations. The simulations did not account for constella-
tions in which different actors’ preferences were characterized by either different levels
or different directions of nonseparability. Due to the discontinuity of the object function
it was almost impossible to employ typical optimization methods for constrained non-
linear optimization problems such as Sequential Quadratic Programming (SQP) in order
to identify the global optimum.131 A proper choice was the utilization of a derivative-
free global optimization algorithm. I implemented this algorithm in MATLAB (Version
7.12.0.635) based on the idea of Perttunen et al. (1993). Using MATLAB ’s “Global Opti-
mization Toolbox” (Version 3.2-R2011b)132 and its related extension “Genetic Algorithm
and Direct Search Toolbox”133 .
Fig. 3.5134 shows each model’s MAE evaluated at different levels of nonseparability. Over-
all, 40 of the 81 issues revealed NSP in a non-reciprocal relation. Accordingly, the figure
depicts the results for two sub-samples. Those 41 issues categorized with reciprocal NSP
reveal an improved predictive accuracy for levels of nonseparability up to approximately
65% of the maximum. This improvement was weakest for the unconstrained bargain-
ing model, whose predictive success was outperformed by the constrained bargaining
model for levels of nonseparability larger than approximately 45% of the maximum. For
this sub-sample the predictive accuracy of the agenda-setting model equaled the uncon-
strained bargaining model for levels of nonseparability larger than 50%. For the second
sub-sample of 40 issues which were affected by non-reciprocal NSP the results differ. As-
suming separable preferences all three models clearly performed worse when compared
to the other sub-sample. This finding suggests that modeling a non-reciprocal relation
between issues is an error-prone undertaking. Assuming higher levels of nonseparabil-
ity left the predictive accuracy of the bargaining as well as the unconstrained bargaining
model unchanged. Yet, the predictive accuracy of the agenda-setting model increased
131 Cf. Kröning and Strichman (2008) for a broader overview on procedures for automated verification and
reasoning, theorem-proving, compiler optimization and operations research.
132 Source: http://www.mathworks.de/products/global-optimization (accessed April 4, 2012).
133 The corresponding user’s guide (version 1) can be obtained from http://www.mathworks.com/help/
releases/R13sp2/pdf_doc/gads/gads_tb.pdf (accessed April 4, 2012).

134 The DEU data applies different measurement scales. Therefore CI, as used in the figures in this section,
should be interpreted with caution (Achen, 2006).
70
drastically. This increase became significant for levels of NSP larger than 90%, a level
at which the agenda-setting model clearly outperformed its two competitors. Assuming
higher levels of NSP the agenda-setting model successfully overcame the SQ bias.
Figure 3.5: Mean average error per issue at different levels of nonseparability
EXPLANATORY NOTE
The figure depicts each model’s MAE per issue at different levels of nonseparability and includes 90% CI. The graphs are
the result of a local polynomial smoother using an Epanechnikov kernel (Epanechnikov, 1969) with a degree of 0. It is best
interpreted as a moving average through the simulated data. The 41 issues categorized with reciprocal NSP are depicted
in the right and the 40 issues categorized with non-reciprocal NSP in the left figure.
mean average error (issue level)
percentage of max. NSP

LABELS
A = agenda-setting model; B = unconstrained bargaining model; C = constrained bargaining model.
Fig. 3.6 compares two models predictive accuracy by subtracting their predictive errors
for each individual issue (paired comparison). The main results resemble the findings of
Tab. 3.3 and Fig. 3.5.
In Sec. A.6 in Fig. A.3 I depict also each model’s MAE and the paired comparison at the
proposal level. The errors are standardized on the maximal error size, which depends on
each proposal’s dimensionality.135 These figures were placed in the appendix because the
main results resemble just the findings at the issue level.136 For low levels of nonsepa-
rability the unconstrained bargaining model outperformed its competitors. With respect
to the minority of 7 proposals which remained unaffected by any allocation dimension,
an increase in nonseparability up to approximately 80% caused an increase in all models’
predictive accuracy. With respect to the majority of 21 proposals affected by an allocation
dimension, the constrained bargaining model and the agenda-setting model gained in
predictive accuracy. More specifically, the agenda-setting model clearly outperformed its
competitors for higher levels of nonseparability.
135 The maximal error size increased by the factor of 100 for each additional dimension. This is due to the
scaling of all individual issues between 0 and 100 in the DEU and DEUII data.
136 In addition, the different spatial dimensionality prevents the application of the hit rate as an criterion for
the predictive power of the models at the proposal level.
71
Figure 3.6: Comparison of model’s predictive accuracy at the issue level

EXPLANATORY NOTE
The figure depicts models relative predictive accuracy. The graphs are the result of a local polynomial smoother using
an Epanechnikov kernel (Epanechnikov, 1969) with a degree of 0. It includes 90% CI and is best interpreted as a moving
average through the simulated data. The 41 issues categorized with reciprocal NSP are depicted in the right and the 40
issues categorized with non-reciprocal NSP in the left figure.
error model 1 - error model 2

LABELS
An alternative method used to evaluate the predictive accuracy of competing decision-

making models is the calculation of hit rates (Achen, 2006). The underlying idea is that
models should be able to provide forecasts with a reasonable degree of precision. Hence,
the quantity of interest is the percentage of cases in which a model predicts with an accu-
racy of less than 1%, 5% or 10% of the maximum divergence. Given the data at hand the
corresponding thresholds were 1 point, 5 points and 10 points on the 100 point scale.
Tab. 3.4 presents the corresponding numbers for the three models. Comparing the second
to the third column it is obvious that those cases for which the separability assumption
was suspected to be false, a lower hit rate at all three levels of precision is revealed.
Please keep in mind that the agenda-setting model predicts perfectly when the final posi-
tion equals the Commission’s position which, however, must be in the amendment proof
set. As a consequence, and starting from a very low number of correct predictions, the
two bargaining models showed a stronger increase in highly accurate hit rates when non-
separability was tuned in. This relative advantage vanished once the precision threshold
was altered from 1 per cent to 10 per cent. The test for differences in binomial proportions
showed that only the agenda-setting model revealed a statistically significant improve-
ment in the hit rates under all three thresholds.137 By contrast, the unconstrained and
137 The p100 − p0

corresponding test statistic is calculated as z = 
p0 × (1− p0 )
with n as the number of observations, p0 as
n
the hit rate under NSP = 0% and p100 as the hit rate under NSP =100% (cf. Sprinthall, 2011, Chap. 4 and
5). The null-hypothesis assumes no differences in proportions and is rejected at an 0.1 level for z ≥ 1.28.
72
constrained bargaining model revealed even lower hit rates at the 10 per cent threshold
once nonseparability had been accounted for.
Table 3.4: Model hit rates

EXPLANATORY NOTE
The table lists models’ hit rates at different levels on nonseparability. The z-Test is used to determine significant differences
in performance.
MODEL HIT RATES

Cases without NSP NSP = 0% NSP = 50% NSP =100% z-Test
comp. NSP 0 - 100%
DIVERGENCE <1%
Unconstrained
13.1 4.3 7.9 8.4 6.82
bargaining model
Constrained
16.9 7.9 11.1 15 6.52
bargaining model
Agenda-setting
20 13.7 15.1 16.9 2.10
model
DIVERGENCE <5%
Unconstrained
21.5 18.50 18.40 18.20 0.16
bargaining model
Constrained
24.3 18.10 19.70 22.30 2.09
bargaining model
Agenda-setting
26.9 19.10 20.90 23.80 2.20
model
DIVERGENCE <10%
Unconstrained
40.8 38.7 36.4 33.3 1.50
bargaining model
Constrained
39.7 36.1 35.3 35.1 0.28
bargaining model
Agenda-setting
33.1 27.4 29.4 33.4 1.98
model
A comparison of the unconstrained bargaining model and the agenda-setting model

showed that under the false assumption of separable preferences the former outper-
formed the latter by 38.7 to 27.4 per cent. However, under the assumption of a very
high level of nonseparability the agenda-setting Model delivered 33.4 per cent correct
predictions, whereas the performance of the unconstrained bargaining model dropped to
33.3 per cent. These results further supported the argument according to which falsely-
assumed separability gives a systematical advantage to unconstrained models.
In sum, the findings are threefold. Firstly, for those issues unaffected by reciprocal NSP an
increase in nonseparability up to approximately 70% improve the predictive accuracy of
all three models. Secondly, this improvement is strongest for the constrained and weakest
for the unconstrained bargaining model. Thirdly, a drastic increase in the predictive ac-
curacy of the agenda-setting model for issues affected by non-reciprocal NSP is observed.
The agenda-setting model emerges as the most powerful model for levels of nonsepara-
bility higher than 50%. Overall, in line with Finke and Fleig (2013) it is safe to conclude
that the advantage of unconstrained models reported in the literature (cf. Sec. 3.1) has
been (at least) reinforced by falsely assuming separable preferences. The simulation of
NSP diminishes the advantage, enables a fairer model comparison and underscores the
agenda-setting power of the European Commission.
73
3.6 The nonseparability in individual votes
This section takes up again the argument that models which are based on individual
decisions should also model the error term of utility functions at that level (cf. Junge,
2010). The modeled source of error affects predictions and performance of a model (cf.
Signorino, 1999). According to Signorino (2003), model error of strategically acting indi-
viduals can result from various types of uncertainty as bounded rationality of agents, ac-
tor’s private information and regressor error or strategic error in specification (Signorino
and Yilmaz, 2003).
This argument holds true for all theoretical models of decision-making and implies that
model predictions should not only be applied at the aggregate but also the individual
level. Thus, the model would not only predict the collective outcome but also deliver a
detailed record of each single actor voting in favor or against a proposal. The accuracy of
the single votes could then be assessed with respect to the implemented degree of non-
separability. In the ideal case this would be done actor-specific. Unfortunately, this step
was not feasible for my research as the DEU data did not contain enough observations for
an individual assessment. In the remainder of this section I discuss in detail the problems
I encountered. I also show what previous contributions have accomplished in this area
and detail some options for future research.
The problems with regard to an individualized analysis were twofold. Firstly, the expert
interviews of the DEU project used to assemble the NSP scheme contained, if any, only
general remarks on conditionality. To supplement this data by individual co-variates as
sector-specific net contributions to the EU budget only resulted in crude approximations.
Thus, the coding could only operationalize NSP for all actors the same way. Secondly, the
total number of usable observations in the DEU data set was too small. For the examina-
tion of individual voting behavior only majority decisions can be used as in unanimous
decisions a yes or no distinction cannot be made. In addition, these decisions must be
reached with certain level of disagreement. A majority decision taken unanimously con-
tains not more information as a unanimous decision in the first place.
Looking more into details, the small number of observations is not caused by lack of
accessibility as all voting records of the EU Council can be publicly obtained. Pursuant
to article 207 (3) of the Treaty of Amsterdam (Amsterdam, 1997) the results and expla-
nations of the EU Council legislative votes are made public. They are published as
Council minutes138 which summarize the discussion of the Council meetings and con-
tain the voting records as addenda (for a more detailed description cf. Hagemann and
De Clerck-Sachsse, 2007). In addition, the General Secretariat provides a monthly sum-
mary of Council acts (legislative as well as non-legislative), which includes the results
138 All Council minutes since 1999 can be accessed at http://consilium.europa.eu/documents/

legislative-transparency/council-minutes.
74
3.6 The nonseparability in individual votes
of all ballots, explanations of voting and general statements.139 Thus, secrecy is not the
reason for the insufficient amount of data. In fact, many previous studies used data
from the voting record of the EU Council of ministers. These studies analyzed the im-
pact of related EU institutions, different decision rules and party affiliation on Council
decisions (for an overview cf. Hagemann, 2007, p. 280-281). The contributions used var-
ious theoretical (e.g., spatial) models to approach the data. Hagemann (2007) addressed
the measurement of policy preferences in the Council of Ministers and which research is
suited to its voting data. The study focused on the type of data that is in fact available
from the Council. Assessing different analyzing methods a simulation-based Bayesian
Monte Carlo Markov Chain (MCMC) model proved to be most adequate. Thus, also the
appropriate theoretical decision model has already been determined.
However, most decisions of the EU Council are reached in a “culture of consensus”

(Hagemann and De Clerck-Sachsse, 2007, p. 3). This results in predominantly unanimous
decisions and a level of disagreement not sufficient for individual voting assessment.
König and Junge (2009) studied this unusual prevalence of consensus based on voting
preferences and logrolling opportunities of the member states on 48 Commission propos-
als. The authors found that even member states with veto power support Commission
initiatives they prefer less than the SQ. Such solutions are made possible by logrolling
across domain-specific or simultaneously decided proposals. They are also driven by the
Commission’s attempts to avoid a divided Council. Another facilitator of the consensus
decisions is identified by Mattila (2004), who investigated whether the Council is shaped
by cleavages based on national or EU-level factors. His findings suggest that the political
space of the EU is defined by two scales: a traditional left-right as well as an indepen-
dence versus integration dimension. Interestingly, he found that presiding countries take
the role of an arbitrator. Thus, the “culture of consensus” in the EU Council is based on
several pillars which all contribute to the large proportion of unanimity.
This homogeneity in final votes does not allow the evaluation of individual behavior.
Therefore, additional data must be considered and two extensions seem most promising.
Firstly, one could extend the focus of research to the whole negotiations process. Sec-
ondly, the administrative level of negotiations monitored could be extended. The next
three paragraphs discuss studies that incorporated such additional information.
Another possibility is to look at interventions during negotiations within the Council

of Ministers. Following this approach, Cross (2012) interpreted these interventions as
signaling policy positions and as attempts at influencing negotiations. The analysis is
based on the records of negotiations of the Council Secretariat. “These documents con-
tain detailed footnotes recording member states’ interventions over the course of nego-
tiations. These footnotes identify both the member states making interventions and also
the level of negotiation within the Council at which the interventions took place” (Cross,
139 Themonthly summaries of the EU Council meetings can be accessed at http://www.consilium.europa.
eu/documents/legislative-transparency/monthly-summaries-of-council-acts.
75
2012, p. 56). The results showed differences between member state intervention behavior
which are influenced by the structural characteristics of the policy space.
Hagemann and De Clerck-Sachsse (2007) analyzed the impact of the 2004 enlargement on
the internal working processes in the Council. Their data comprised formal statements
of Council members following the adoption of a proposal. “These formal statements
often consist of a country’s explicit disagreement or reservation with regard to a policy”
(Hagemann and De Clerck-Sachsse, 2007, p. 3) and are included in the Council minutes.
The authors claimed that they serve a signaling purpose to clarify disagreement with
either a single issue or the enacted proposal in general and interpreted them as “a way to
avoid policy gridlock through contested voting” (ibid.).
Negotiations are not restricted to the ministerial level alone.140 Haege (2007) investigated
the influence of national bureaucrats on the decision-making in the EU Council of Min-
isters. Yet, his findings showed that these national officials tend to only decide the less
salient and more complex proposals. Hayes-Renshaw et al. (2006) compared voting at
the ministerial and official level of the Council. In line with the previously discussed
studies the authors found that ministers generally endorse consensus, and that contesta-
tion occurs rather at the level of officials. Nearly half the dissent was expressed by single
members and most often the cases in question were concerned with “apparently quite
technical agricultural issues” (Hayes-Renshaw et al., 2006, p. 5).
To summarize, this section discussed three possible sources of information for an individ-
ualized analysis. The suggestion to include other negotiating levels is not possible, since
the encoding of NSP would be highly questionable in this case. The expert interviews re-
ferred only to ideal positions and salience assessments of member states observed at the
ministerial level. A feasible option would be to include either interventions expressed
during or formal statements made after the negotiations. Yet, of the proposals covered in
the DEU data and affected by NSP only 6 express the required dissent at the individual
level.141 These are too few observations to receive reliable evidence. Maybe the evalua-
tion of NSP on the individual level can be accomplished in future research by including
data from the DEU II project (Thomson et al., 2012). However, this data will not contain
individual-specific information of NSP. Thus, so far the analysis of individual voting is
desirable but not yet available.
3.7 Chapter summary
This chapter offered an assessment of the merits gained when considering NSP in an em-
pirical analysis. It also provided a first glimpse of the necessary expenses. For this pur-
140 Hayes-Renshaw and Wallace (2006) provided a comprehensive overview on the individual functions and
processes of the EU Council of ministers and its development over time.
141 The corresponding proposals are COD96112, CNS98092, CNS98347, CNS99138, CNS99163 and CNS99235
(cf. Sec. A.5).
76
3.7 Chapter summary
pose I used the field of legislative decision-making in the context of EU politics. Mod-
els used to assess decision-making rely on utility functions which have to be specified
accordingly to the policies in question. When such proposals consist of multiple dimen-
sions the (non)separability of single issues has to be taken into account. Otherwise the
utility functions are misspecified and the model produces inaccurate results. This is di-
rectly in line with ARGUMENT 1 (established in Sec. 2.5) and confirms HYPOTHESIS 1.
The analysis and coding of the prominent DEU data set provided two main findings:
Overall, 55.7% of the issues and 50% of the multidimensional proposals (which constitute
one-third of all proposals) in the data set are affected by NSP. Thus, it becomes clear that
NSP are in fact a widespread phenomenon in EU politics. Moreover, in the majority
of EU law proposals affected by nonseparability the effect is non-reciprocal. The results
suggest that the potential bias is most severe in those cases. Thus, the in Sec. 2.4 discussed
extension of the NSP concept with respect to reciprocity is strongly encouraged.
Next, using the NSP coding scheme I investigated if a comparison of various models’
predictive accuracy depends on the existence of NSP. In my analysis I found clear ev-
idence that the empirical evaluation of competing models of EU legislative politics is
biased. The bias arises from the different degrees of restrictions incorporated into the
models. These restrictions must rely on a correct specification of actors’ utility functions,
otherwise they are modeled incorrectly. This is particularly unsatisfactory for those re-
searchers interested in the effect of procedural aspects such as the allocation of agenda
power and voting rights. I conclude that this demonstrates the importance of a correct
model specification and confirms HYPOTHESIS 2.
This chapter answers my first key question; yes, neglecting the nonseparability of prefer-
ences poses a threat to the inference of the corresponding research. Applying simulation
techniques I demonstrate that overlooking NSP may have caused a substantial bias in
the empirical evaluation of competing models of EU legislative politics. While the effect
is not rectified for all models, increasing the level of nonseparability boosts the predic-
tive power of the agenda-setting model. Thus the potential gains of incorporating NSP
become clear. Using simulations may also provide a way out to the dilemma of opera-
tionalizing nonseparability, although this approach has its own limitations. For example,
it is only possible to theorize about and not prove the existence of NSP.
In this chapter I had to talk a lot about limitations of the applied research method. This
illustrates all too well that still many aspects are missing in the study of NSP, maybe
most evidently in the last paragraph of the analysis. Sec. 3.6 discusses the argument
that models based on individual decisions should also model the error term of utility
functions at that level. While the idea seems promising, an empirical implementation
was not feasible. The DEU data set contains not enough corresponding individual data.
One main reason is the low level of contestation in EU Council decisions. Although
some studies used additional information to operationalize interventions during Council
negotiations the reliability of this data has to be verified by further studies.
77
The limitations clearly demonstrate why I decided to continue my research on NSP with
a laboratory experiment. The high degree of environmental control enabled me to set
up the decision problem in the most applicable way, as laboratory experiments secure a
high level of internal and construct validity. This chapter discussed multiple models of
decision-making and devoted a lot of space for discussions on preference measurement
and the importance of various institutional rules. But despite the immense efforts put
into the DEU research project I was only able to “assume” NSP based on the expert in-
terviews. These assumptions could only be made in a general fashion as an actor-specific
assignment was not feasible. However, a laboratory experiment facilitates to induce NSP
for every actor and to observe every single individual decision.
78
4 The Experiment
I demonstrate very clearly the difficulties of operationalizing NSP in Chap. 3. In particu-

lar, I show the limitations of observational data in this context. Thus, using a laboratory
experiment is the logical choice for further investigation due to the excellent monitor-
ing capabilities and implementation options. The experimental setting can be configured
in detail according to the research interest (cf. Sec. 1.3.2). Thus, NSP are now ensured
(because induced) and the investigation can focus on the resulting effects.
The second part of my study consists of four chapters. In those I deal with my second
key question, whether and, if so, how nonseparability affects individual and collective
behavior. By contrasting several institutional arrangements directly I identify the conse-
quences when a decision is affected by NSP. This follows ARGUMENT 2 and ARGUMENT
3, which are concerned with either sequential or simultaneous decision-making. But I
also look for further behavioral patterns to supplement the nonseparability concept.
This chapter starts in Sec. 4.1 with a short overview of important contributions in exper-
imental laboratory research. I pay special attention to the topic of collective decision-
making. Next, I explain my experimental design. Sec. 4.2 describes in detail every part
of the experiment. The implications of the design choices for the empirical analysis are
summarized in Sec. 4.3. Subsequently, I present some descriptive information about the
experimental process in Sec. 4.4.
4.1 Experimental research in political science
The basis for laboratory experiments in social science was set in economics (Thye, 2007).142
The beginning of experimental work in political science is due to the outstanding work
of Plott (for an overview on Plotts’ work cf. Ortmann, 2003). Since then many excellent
contributions have opened the field for further work. Today researchers argue that “lab
experiments are a major source of knowledge in the social sciences” (Falk, 2009, p. 535).
When conducting experiments, political scientists were mainly concerned with the wel-
fare implications of decision rules like majority and unanimity voting (e.g., Miller and
142 Palfrey(2009) referred in particular to the work of Vernon Smith (cf. Smith, 1991, for an overview
on his work). A testament to his exceptional position as founding father of laboratory research is
the statement of the Nobel Prize Committee describing Smith’s contribution as “having established
laboratory experiments as a tool in empirical economic analysis” (2002 Nobel Prize Announcement,
http://almaz.com/nobel/2002-prizes.html).
79
4 The Experiment
Vanberg, 2014; Sauermann and Glasmann, 2011), majority voting at elections (e.g., Fed-
dersen et al., 2009; McKelvey and Ordeshook, 1982; Morton and Williams, 2011), sophisti-
cated behavior in plurality voting under majority rule (e.g., Felsenthal et al., 1988; Niemi
and Frank, 1985), bargaining under majority rule (e.g., Bianco et al., 2008; Diermeier and
Morton, 2005; Salant and Goodstein, 1990)), and committee decisions under majority rule
(e.g., Coleman and Ostrom, 2009; Fiorina and Plott, 1978).143
The focus on majority voting has two reasons. Firstly, majority voting is the most widely
spread decision rule for democratic government (Erlenmaier and Gersbach, 2001, p. 2).144
Secondly, following Rawls (1971), those procedures are considered fair that are acceptable
to everybody under the veil of ignorance. Tossing a coin is often accepted as fair proce-
dure because none of the conflicting parties is able to anticipate whether it will be on the
winning or on the losing side. Moreover, in infinitely recurring situations coin-flipping
produces an equal split, widely accepted to be the only fair division in constant-sum
games. Yet in the realm of politics, which is characterized by heterogeneous preferences
and binding, non-recurring decisions, neither coin-flipping nor random draws from a
lottery are frequently accepted as producing just outcomes.
Instead, social choice theory proves that simple majority voting comes closest to satisfy-
ing all criteria of a “fair procedure” (Fey, 2004; May, 1952). This holds even if it is notwith-
standing the conflict between procedural justice (Rawls, 1971) and welfare enhancement
inherent in majority voting (Arrow, 1950; Risse, 2004). The quality of majority voting
with respect to redistributive justice depends, among other things, on the distribution of
preferences. Here the advantages of induced, and thus known, preferences enable the
experimenter to conduct a detailed and clear cut investigation of the decision-making
process.145
4.1.1 The rise of the experimental method
There are a multitude of summaries on experimental research in political science, like

Kagel and Roth (1995), Croson (1999) or ?. In addition, newer introductions to politi-
cal science discuss the research contribution of experiments (e.g., Bernauer et al., 2009,
p. 91ff). I do not intend to repeat these volumes but instead focus on a small composition
of excellent contributions. Those strongly influenced my decision to use a lab experi-
ment and how to design it. The next sub-section deals in detail with the beginnings of
experiments in research on collective decision-making.
143 For an extensive review please consult the “Cambridge Handbook of Experimental Political Science”
(Druckman, 2011).
144 Any democracy is commonly associated with political equality and majority rule (Saunders, 2010a, p. 2).
Yet, most legislative processes in democracies do not use simple majority voting but rather a system of
checks and balances, division of power and representative elements (Beitz, 1990, Chap. 4, p. 60).
145 A central aspect of induced value theory (Smith, 1976) is that by defining the financial incentives the
same motivation can be assigned to all subjects. The common payoff structure controls for individual
differences (Morton and Williams, 2012, p. 17).
80
The experimental method has come a long way in political science.146 For most of the
time, experiments were considered inappropriate because this discipline was “limited by
the impossibility of experiment. Politics is an observational, not an experimental science”
(Lowell, 1910, p. 7). This does not mean that the benefits of the method have not been
appreciated. On the contrary, many contributions agreed that “the experimental method
is the most nearly ideal method for scientific explanation, but unfortunately it can only
rarely be used in political science because of practical and ethical impediments” (Lijphart,
1971, p. 683).
I will not deny that there were also some other perceptions in previous research. For
example, Campbell (1969, p. 409) argued that (experimental) administrators “should be
ready for an experimental approach to social reform, an approach in which we try out
new programs designed to cure specific social problems, in which we learn whether or
not these programs are effective, and in which we retain, imitate, modify, or discard
them on the basis of apparent effectiveness on the multiple imperfect criteria available.”
Burgess and Robinson (1969) pointed out that artificial confirmation of a hypothesis in
the laboratory (though always possible), was less likely than artificial dis-confirmation;
whereas the opposite holds for research in the “natural setting”. But these benevolent
statements captured by no means the majority opinion.
Even if experiments in social science may never achieve the same degree of “straightfor-
ward elegance” (Johnson, 2008, p. XI) as in natural science the critical perception changed
step by step.147 Factors which facilitated this development are the advances in computer
technology (e.g., lower costs and better networks)148 , growing research on causal rela-
tionships (e.g., Braumöller, 2003) and its difficulties with observational data (e.g., Jeffrey
et al., 2006), increasing interest into research questions on underlying assumptions about
the nature of political or institutional decision-making, etc.
“From nature to the lab” by Morton and Williams (2010) offered an extensive overview
on experimental research and its present state. It contains a multitude of detailed exam-
ples from a wide array of scientific research. With a focus on social science in general
the authors presented a distinctive approach on how to conduct experimental research.
The monograph covers laboratory, survey, and field experimentation. The design of my
experiment follows their approach by emphasizing causal inference.149 Many details of
my design apply to recommendations of this book, e.g., the subject recruiting as well as
the payment mechanism. The work of Morton and Williams (2010) constitutes a compre-
146 For an assessment of the history of the experimental method in economics cf. Plott (1991). The analogy of
the development in both disciplines is clearly observable, even if the current degree of acceptance differs
(McDermott, 2002).
147 In his magnificent collection of outstanding experiments in natural sciences Johnson (2008, p. XI) referred
to the logical simplicity of apparatus and analysis of these experiments as “beauty in the classical sense”.
148 Cf. Humphreys (2004) for a comprehensive discussion on computational methods and their influence on
scientific research.
149 Causal inference was thoroughly discussed by Shadish et al. (2002). The authors focused on theoretical
aspects but included also practical advice for implementing an (quasi-)experimental design.
81
4 The Experiment
hensive and detailed volume.150 I follow up with two smaller reviews on the history of
experiments in political science. Both of them are shorter and by no means as encom-
passing. However, both discuss a specific aspect of experimental research I would like to
highlight with respect to my own contribution.
Faas and Huber (2010) entitled the increasing application of laboratory experiments with
the expression “From Wallflower to Mainstream”. The article reviewed the state of exper-
imental research done in the fields of elections, public opinion, public goods, collective
action, social trust, legislative bargaining and decision-making. It also contained a more
detailed section on field experiments which studies voter mobilization. The authors ar-
gued that experimental research is on the rise, as an increasing number of related journal
articles, books, and conferences show us. Yet, they also noted that in particular in Ger-
man political science the use of experimental methods is still rather uncommon.151 As a
reason for this the authors emphasize the missing scientific training with respect to ex-
perimental methods in political science curriculum at German universities.152 Despite its
small-scale proliferation in Germany I choose the experimental approach because I agree
with the advantages of the experimental method discussed by the authors which clearly
demonstrate the added value of this method.153
Political science has often been influenced by associated fields of studies with regard to
research methods (Druckman and Lupia, 2006). McDermott (2002) compared the histor-
ical and cultural differences when conducting experiments in two neighboring scientific
fields, economics and social psychology (for an overview cf. Martin, 2008). Although
both disciplines use experiments widely, their approaches differ with respect to many
fundamental basics.154 Most prominent are the attitude towards deception or the mone-
tary payment of subjects. Bloomfield et al. (2009, p. 14ff) argued that “many experiments
in economics are not actually experiments - they are demonstrations.” A typical psychol-
ogy experiment manipulates a single variable while holding all other aspects constant.
Accordingly, Shadish et al. (2002, p. 7) characterized experiments as to “explore the ef-
fects of things that can be manipulated”. But in economics testing theories is also com-
mon, which can be done without any manipulation. As my research framework resem-
bles rational choice I followed the tradition of experimental economics.155 Thus, subjects
150 With Morton and Williams (2012) the authors also provided a shorter and apt summary of the usage of
experimental methods in political economics.
151 Of course, there are also attempts in Germany to promote the experimental method further. For example,
the working group for action and decision theory of the German association for political science devoted
its 2012 yearbook to it (cf. Bräuninger et al., 2012).
152 In addition, in Germany political science students usually do not complement their education with addi-
tional psychological, sociological or economic methods courses, in which they may learn experimental
procedures (Faas and Huber, 2010, p. 724).
153 I explain my decision to investigate NSP by conducting a laboratory experiment in detail in Sec. 1.3.2.
154 Cf. Croson (2005, p. 131) for a in depth comparison of experiments in economics and psychology with
respect to “incentives, context, subject pools, deception, experimental details and data analysis”.
155 Cf. Guala (2005) for a comprehensive discussion on the principles of experimental inference and the scope
and limitations of the laboratory in experimental economics. In addition, cf. Bergstrom and Miller (2000)
who focused on topics of microeconomics.
82
received a monetary payment (depending on their performance in the experiment) and

no crucial or necessary information was withheld.
All reviews have in common that they take a great deal of their space and devote it to ex-
plain the details of the experimental method. They define the basic principles of causality,
randomization and validity as well as discuss pros and cons of different types of exper-
iments conducted in either the laboratory, field or as a survey. Of course, all authors
have done experimental research themselves and thus can be expected to advertise this
method. But in every volume potential errors, traps, disadvantages and limitations of the
method are also discussed. Overall, I agree with the pointed appeal of Kinder and Palfrey
(1991): “An Experimental Political Science? Yes, an Experimental Political Science!” Yet,
not without adding that, while experimental methods have much to offer political sci-
ence research, no method can ever be suited to fit all research questions, however flexible
and universal the approach might be.156
4.1.2 The first experiments on collective decision-making
Collective voting behavior is and was an area of public choice characterized by a con-
siderable amount of disagreement (cf. Clinton and Meirowitz, 2004). This gave rise to
an extremely “fruitful area for experimental research” (Anderson and Holt, 2002, p. 4).
The beginnings go back to simple paper and pencil experiments which focused on the
effect of different decision rules. Here, Fiorina and Plott (1978) used a plain design and
put five-person committees in front of a blackboard and introduced a single point in the
two-dimensional space as the SQ. “Any subject could propose an amendment [...] to the
motion on the floor. If it passed [...] the amendment became the new motion on the floor
and the process continued” (Fiorina and Plott, 1978, p. 577). At any point of the experi-
ment a majority could end the debate. Thus, each “committee pushed a point around the
blackboard until a majority voted to quit and go home” (Fiorina and Plott, 1978, p. 577).
The canonical experiment of Fiorina and Plott (1978) has been advanced with respect
to alternative procedural rules and sequential decision-making by Kormendi and Plott
(1982) as well as Eavey and Miller (1984, 1995). All together, these contributions com-
pared for committee experiments “the predictions of a variety of models drawn from
Economics, Sociology, Political Science and Game Theory [...] to the experimental re-
sults” (Fiorina and Plott, 1978, p. 575). The later studies paid more attention to general
aspects of agenda control and information levels. Mostly, the committees were made up
of an odd number of participants and voted under simple majority rule on monetary
payoff schemes which resembled closely a “surface plot” as shown in Fig. 4.1.
These experiments found that equilibrium concepts such as the core predict the observed
behavior fairly well (Tullock, 1981), but are blurred by a level of uncertainty which varies
156 Mosesand Knutsen (2007) offered a comprehensive overview of methodology in social sciences. The vol-
ume emphasized the debate between positivist and constructivist approaches and stands representative
for the methodological pluralism of the discipline.
83
4 The Experiment
Figure 4.1: Payoff scheme

EXPLANATORY NOTE
The figure shows a payoff scheme handed out to every subject in the study of Fiorina and Plott (1978, p. 596). Their ex-
perimental design was straightforward. Each committee was asked to select a single point in a two-dimensional decision
space (x1 ,x2 ) on a blackboard by majority rule. Each subject was assigned a payoff function defined over ordered pairs of
the coordinate system. Those pairs represented the amount of money the subjects would receive if that point was chosen
as a collective outcome. From the scheme every subject could infer its payoff for each point of the decision space.
across designs. More advanced concepts for dealing with random errors in uncertain
environments included the selection set (Salant and Goodstein, 1990) and, more recently,
the quantal response equilibrium (QRE; McKelvey and Palfrey, 1995, 1998). If, however,
the core is empty as it would be the case in a multidimensional, simple majority setting
(Berl et al., 1976), the outcomes are located close to the center of the policy space, best
approximated by concepts such as the yolk (Plott, 1967) or the uncovered set (Bianco
et al., 2004; Fishburn, 1977).157 In my analysis I used a two-way random error component
model (cf. Sec. 5.1) as well as a random utility mixture model (cf. Sec. 6.3) to account for
uncertainty and errors in the estimations. This also included a probabilistic extension of
the core which enabled a far more accurate analysis then a deterministic approach.
4.2 Experimental design
My experiment falls into the category of investigating “Formal Theory” (Morton and
Williams, 2010, p. 145ff). It resembles a “stress test” (Schram, 2005, p. 234) for theories
of collective as well as individual decision-making (for an overview on such theories cf.
Davis and Holt, 1993). Morton and Williams (2010, p. 151) defined such a stress test as
“when the researcher chooses to allow for one or more of theoretical assumptions to be
violated or purposefully investigates situations in which the researcher is uncertain as to
whether theoretical assumptions hold”.
157 For a recent overview cf. Bianco et al. (2008).
84
I adapted the general experimental environment presented by Sauermann and Kaiser

(henceforth S&K, 2010). These authors conducted a series of laboratory experiments de-
signed to elicit the relevance of social preferences for majority decisions. They concluded
that fairness motivates majority decisions. In my experiment I studied decision-making
in committees under majority rule. Subjects voted simultaneously for one out of several
alternatives. If neither of the alternatives got a majority, a new ballot was held. This pro-
cess continued until one of the alternatives received a majority of votes which marked
the completion of the voting process. Subjects were provided with cardinal information,
i.e., they knew their own and other players’ payoffs and they were informed about other
players’ previous votes.
During the experiment I altered the procedural rules in a way that subjects were con-
fronted with separable as well as nonseparable decision situations. Between these modi-
fications I kept other factors as subjects’ preferences and information condition constant.
I was interested in the way subjects adjust to the nonseparability of the decision-making
situation and if the observed outcome varies with alternations of the procedural rule.
This included behavior at the collective as well as at the individual level. I favored a
more realistic framework and used multiple variations of preference distributions.
It is first and foremost important that subjects understand the game they are playing.
Chou et al. (2009) emphasized the importance of game-form recognition very distinctly.
Only under this precondition experimenters can set up an experimental design that is
highly applicable to the research interest. To secure the game-form recognition I also
conducted a pilot session as test run and a post-experiment survey (cf. Chap. 7).
4.2.1 Procedure
All sessions were conducted in the experimental laboratory managed by the Department
of Economics at the University of Heidelberg158 . Subjects were recruited using the ORSEE
software (Online recruitment system for economic experiments, Greiner, 2004) out of the
subject pool of the laboratory. I implemented my design using the z-Tree (Zurich Toolbox
for Readymade Economic Experiments) software developed by Urs Fischbacher (2007).
The participants entered the laboratory one by one and were assigned to cubicals by
drawing lots. The instructions, which are included in Sec. A.7159 , were handed out to the
subjects and read aloud by a supervisor.160 Questions were answered privately.
After that the subjects went through the rules on their screen again and had to answer
some questions correctly in order to prove their understanding of the upcoming game.
Special emphasis was placed on ensuring that participants understood the key aspects of
158 Homepage of the laboratory: http://www.uni-heidelberg.de/fakultaeten/wiso/awi/forschung/
awiexplab.html.
159 Please note that the instructions are only available in German.
160 The instructor was the same person throughout all the experimental sessions.
85
4 The Experiment
the experiment, such as their decision-making competences, the timing of the ballots, the
set of feasible alternatives, etc. Also, the basic features of the PC setting as, e.g., the screen
arrangement, the payoff table scheme and the entering field for their vote were part of
the test. The experiment did not start before all subjects had passed this mandatory test.
Next, the participants took part in multiple rounds of my voting experiment (cf.Sec. 4.2.5).
After the conclusion of all ballots the participants answered a post-experiment survey at
their PC workstation (cf. Chap. 7). Upon completion, they were asked to leave the room
and to stay in the waiting area. After everyone had gone out, the participants were indi-
vidually asked back into the experimental laboratory and received their payoffs.
4.2.2 Anonymity and non-communication
During the whole experiment, communication between the subjects was forbidden. Pre-
vious studies found that communication enforced the relevance of social norms. This was
evident, e.g., in form of “universalism” in Miller and Oppenheimer (1982) or “collective
resistance against transgression” in Cason and Mui (2007). Yet, also under an anony-
mous setting the participants not only maximize their individual payoffs (van de Kragt
et al., 1983). Moral motivations may be enhanced by communication, but they do not
dependent on it (Fehr et al., 2002a).161
In addition to (active) communication “silent identification” (e.g., the disclosure of name
and place of birth) has also been found to increase solidarity between subjects (e.g.,
Bohnet and Frey, 1999). More specifically, in their public good experiment Andreoni
and Petrie (2004) experienced less free riding and, if identification was combined with
information on others behavior, increasing contributions to a common public good.162 I
followed none of these approaches. My design excluded pre-play communication and
guarantees all participants absolute anonymity. Other players were indicated only as
Player 1, Player 2, Player 3, etc.
Anonymity also applied for the payment process. Payoffs were handed out when only
the receiving subject was in the room. This leaves only the experimenter as person the
subjects could have interacted with. This possibility for distortion is called experimenter
demand effect (EDE).163 Such an effect occurs when subjects change their behavior in
161 In their public good experiments van de Kragt et al. (1983) observed significant fewer realizations of the
public good when communication was not allowed. Nevertheless, subject’s contributions were still suf-
ficient in 65% of the trials.
162 The identification and monitoring of others is strongly associated with experimental research on group
identity. In both cases the decision of the subjects may (partly) be influenced by a feeling of solidarity on
the basis of a common ground. The difference is that group membership mostly rests on an “artificial”
aspects (e.g., which of two paintings they like best, Chen and Li, 2009, p. 436) and identification on
“true” personal characteristics (e.g., gender). Chen and Li (2009) provided a comprehensive overview
for literature on this topic.
163 Due to the architecture of the laboratory a “double-blind” implementation was not feasible. Such an
experimental procedure would allow neither the subjects of the experiment nor the persons conducting
the experiment to know the critical aspects of the experiment (Shuttleworth, 2008). Cf. Cox et al. (2008,
p. 18-20) for more information on the comparative advantages of single-blind and double-blind protocols.
86
an experiment due to cues about what constitutes appropriate behavior and would be
expected (or “demanded”) from them (Zizzo, 2010, p. 2). Barmettler et al. (2011) inves-
tigated this potential contamination of experimental data as many experiments do not
provide anonymity between experimenter and subjects. The authors varied the degree
of anonymity in common laboratory setups but found no significant EDE on subjects’
behavior. This is in line with Frank (1998) who tested if subjects care about an exper-
imenter’s welfare. Using ultimatum bargaining experiments he found no such effect.
Zizzo (2010) showed that EDE can be a problem when they are correlated with the true
experimental objectives; but he also admitted that “given the trade-offs implicit in de-
signing and running an experiment, researchers may decide to accept the risk of an EDE”
(Zizzo, 2010, p. 28). So far the literature provides enough evidence that the experimenter-
subject contact should constitute no distorting factor. However, a post-experimental sur-
vey gave me the opportunity to look for indications of EDE.
4.2.3 Payoff table
The number of points each player earns in case a given alternative was selected was
specified in neutrally labeled payoff tables as, shown in Fig. 4.2. Here, every participant
(player 1 to 6) is assigned a certain number of points in each case. A total of nine alter-
natives were available. These alternatives formed a combination of columns (letters A,
B and C) and rows (numbers 1, 2 and 3). Thus, A1, A2, A3, B1, B2, B3, C1, C2 and C3
constituted the nine possible alternatives.
Figure 4.2: Example of a payoff table

EXPLANATORY NOTE
The figure shows a payoff table as presented to the participants in my laboratory experiment. The two-dimensional
design enables the implementation of NSP when separating the issues. Then, a decision on each single issue influences
the preferences for the outcome of the other (Strom, 1990, p. 57). As long as no restrictions are imposed or (partial)
decisions are made, every individual prefers one of the given alternatives as their first choice (i.e., unconditional first
preference). For example, assuming rational and self-interested actors, this is { B2} for player 5 with 45 points (i.e., his
maximum score). By altering the options or fixing a partial decision an individual’s preference for the preferred outcome
changes also in the second dimension. For example, in case options { A} and { B} are excluded, player 5 is preferring row
1 (with a payoff of 26) instead of row 2 (now with a payoff of 12). This makes {C1} the conditional first preference.
A B C
Player 1 7 2 1
Player 2 28 8 41
Player 3 27 12 24
1
Player 4 22 44 4
Player 5 4 2 26
Player 6 6 9 20
Player 1 58 35 5
Player 2 2 11 3
Player 3 10 4 45
2
Player 4 19 1 8
Player 5 8 45 12
Player 6 11 10 28
Player 1 8 21 18
Player 2 10 32 23
Player 3 10 18 15
3
Player 4 17 13 11
Player 5 10 6 20
Player 6 32 17 4
87
4 The Experiment
I build my research on a solid foundation of previous experimental work on collective

decision-making (cf. Sec. 4.1). Contrary to these contributions the preferences were not
generated by monetary payment schemes or geometric representation (cf. Fig. 4.1) but
payoff tables. This followed two reasons. Firstly, the change is justified by the results
of the prior work which often were based on kind of “vague” statements: committee
decisions “continue to cluster around an alternative”164 (Fiorina and Plott, 1978, p. 579) or
“the influence observed to date conforms closely to that predicted by the core” (Kormendi
and Plott, 1982, p. 189). S&K (2010, p. 669) summarized these findings with “a general
tendency toward the core point”. I do not question the overall conclusions these authors
draw from their data, as their contributions are too elaborated and numerous. The core
seems to be a good approximation for committee behavior. Nevertheless, the usage of
payoff tables provided more clear-cut results than the spatial proximity information. An
alternative is selected or not, enabling an analysis with higher discriminatory power.
Secondly, the experimental data of Diermeier and Gailmard (2006) suggested that partic-
ipants evaluate endogenously and exogenously generated inequality differently.165 Ac-
cordingly, Mertins (2008) pointed out the effect of various procedures and the importance
of procedural fairness. Endogenous inequality results from decisions made by the partic-
ipants in the laboratory while exogenous inequality resorts to intentional modifications
the experimenter makes before the beginning of the experiment (e.g., giving one subject
an initial endowment of 50 while another subject receives only one of 5). Rabin (1993,
p. 1296) argued that “people’s notions of fairness are heavily influenced by the SQ and
other reference points.” The experimental studies on redistribution of Rutstroem and
Williams (2000) and Sutter (2002) found that subjects’ voting behavior depended sys-
tematically on their relative position to the SQ.166 Thus, in line with the design of S&K
(2010), but contrary to prior contributions167 , I did not specify a SQ ante (i.e., select one
alternative as default option) because it might bias subjects’ behavior (cf. Samuelson and
Zeckhauser, 1988).
In line with the experiments discussed in Sec. 4.1.2 the dimensions of the payoff table
were labeled with “neutral” numbers and letters. Lacy (2001b) showed that conditional
preferences exist for many policy areas. Therefore, framing168 could be used to induce
164 Continue to cluster stands for the frequent occurrence of outcomes in the near vicinity of an alternative.
165 Maier-Rigaud and Apesteguia (2003) identified another interesting “endogenously vs. exogenously” ef-
fect. The authors varied if a prisoner dilemma is automatically assigned to offering the subjects the pos-
sibility of choosing between two different representations of the same dilemma. They found significantly
more cooperation when the game could be chosen.
166 Colomer (2001, p. 10) stated in his evaluation of different voting rules that “the social efficiency of the out-
come is highly dependent on the SQ, the distribution of bargaining costs among voters, and the agenda
setter’s maneuvering.” Thus, the implemented SQ might not determine the final outcome alone but it
will exert significant influence.
167 For example, in the two-dimensional setting of Fiorina and Plott (1978, p. 577) “each committee began at
the point (200, 150). That is, the status quo in each experiment was the extreme northeast point in the
issue space”.
168 Framing occurs when different, but logically equivalent, phrases cause an individual to alter their prefer-
ences (Tversky and Kahneman, 1986).
88
conditionality (Tversky and Kahneman, 1981). however, this would bring about other
problems. First, as with salience, the strength and direction of NSP varies for every in-
dividual. The empirical analysis in Chap. 3 clearly shows how much the salience of a
single issue can vary across actors. In order to analyze the decision on a framed ques-
tion empirically, one would have to measure these individual values.169 In the context
of the laboratory this could have been achieved by conducting surveys on participants’
perception of the experimental decision problems. Nevertheless, the downside of such
an approach would be that the identification of individual salience and NSP would, even
in the best case, only become clear ex-post.
A second problem with using framing is that subjects themselves introduce a number
of latent dimensions into the decision the experimenter wants them to make (Halfpenny
and Taylor, 1973). As these dimensions are not homogeneous across subjects, it is not
possible to identify all of them even ex-post.170 To avoid these problems and because my
experiment was the first which focuses explicitly on the aspect of nonseparability, I chose
to apply a neutral context. This generic environment ensured the highest possible level
of control over the motivation of the subjects (cf. Morton and Williams, 2012, p. 22).171
4.2.4 Multiple tables and their characteristics
I used different payoff tables in the experiment. The aim was to observe decision-making
under a variety of problems. This resembles reality with its multitude of everyday deci-
sion situations better than the implementation of just one fixed payoff scheme. Sec. A.8
contains a list of all tables used in the experiment. The tables differ with respect to two
important characteristics: i) if an equilibrium prediction exists and ii) if the alternative’s
overall sum varies or stays constant within a table.172 These variations are not the treat-
ment of the experiment (cf. Sec. 4.2.6), but rather allowed me to study the treatment effect
under different constellations (cf. Sec. 5.4).
169 For example, the DEU research project (cf. Sec. 3.2) used in-depth expert interviews.
170 Halfpenny and Taylor (1973) conducted a series of committee experiments. In an attempt to add more
interest and realism to the experimental situation they told their subjects that they were facing a decision
for the location (inner city vs. suburb) and fittings of a new office block. From comments made by their
subjects during the pilot session “it became apparent that they were, in effect, introducing an unspecified
number of additional dimensions, e.g., whether to build in the Green Belt, or the availability of commuter
trains.” (Halfpenny and Taylor, 1973, p. 28) Thus, they choose to conduct their experiment with labeling
the dimensions “X” and “Y”.
171 In general, the question for using neutral or loaded instruction in either payoff tables or payment schemes
must depend on the goal of the research. Eckel and Grossman (1996) emphasized that the importance of
social and psychological factors can only be studied by abandoning abstraction (at least to some extent).
On the other hand, to assume the effect of loaded instructions as incontrovertible may be wrong. Quiet
some empirical studies found no effects even if “the underlying context is heavily loaded” (Abbink and
Hennig-Schmidt, 2006, p. 103). For further information on priming effects cf. Ortmann (2005) and Ortman
and Gigerenzer (2000).
172 The experiment inhibited a certain intrinsic aspect. In addition to theory driven hypotheses I was also
interested in emerging patterns of behavior I could not anticipate before. Such observations are much
more likely and substantiated when looking on a multitude of decision situations.
89
4 The Experiment
Following social choice theory (e.g., Fishburn, 1973, 1977) the appropriate baseline equi-
librium for the committee decision-making is the concept of the core (cf. Sec. 2.4.2).173
This is the same solution concept as in S&K (2010). Previous work suggests the valid-
ity of this approach, e.g., the experiments discussed in Sec. 4.1.2. The core alternative is
characterized as a unique cooperative solution under the assumption of rationally acting
individuals. More generally, the core is defined as the alternative for which no single
player or subgroup has an incentive to leave the coalition supporting it (Peleg and Sud-
hoelter, 2003, Chap. 3 and 12). Hence, the core alternative beats all other alternatives in
pair-wise comparison (analogous to Berl et al., 1976, p. 468).
The core is a comprehensible and simple concept. Nevertheless, it is not limited to a spe-
cific institutional rule. Miller et al. (1996) conducted experiments to test whether individ-
uals partitioned into two chambers using simple majority voting do, in fact, agree on the
“bicameral core”174 . They found that the chambers almost always chose the outcome the
formal theory predicted. This is a very useful insight as it allowed me to operationalize
different decision-making mechanisms together with their respective equilibrium con-
cept (cf. Sec. 4.2.6).
With a view to the applied stopping rule this choice may be disputable.175 Subjects voted
simultaneously, and if an alternative received a majority of votes this marked the com-
pletion of the voting process. There never was a pair-wise vote between two alternatives.
In fact, this comparison had to be made by the subjects while considering their decision.
This stopping rule was also used by S&K, (2010), but deviates from the procedures used
in previous studies (cf. Sec. 4.1.2). Yet, although those contributions do not agree on one
single best practice but cover a whole range of stopping rules as, e.g., motions to adjourn
(Fiorina and Plott, 1978; Bottom et al., 2000) or 5-times pairwise winner requirements
(Halfpenny and Taylor, 1973).
Even the underlying idea of my rule can be found in this wide variety. Kormendi and
Plott (1982) appointed one of their subjects’ convener with the sole ability to make pro-
posals the group could ratify. Other participants were limited to make suggestion for
proposals. While in their setting “the convener may propose any option he/she wants or
he/she can refuse to propose any option if he/she so desires” (Kormendi and Plott, 1982,
p. 192), another aspect is in accordance with my implementation. Once ratified (i.e., pro-
posed by the convener and accepted by a majority), an option it is no longer contestable
but immediately final.
173 Kramer (1972, p. 169) stated that “any n-person game, can be analyzed from either a cooperative or non-
cooperative point of view”. The decision for either depends on whether one judges “that the terms of the
original agreement can be enforced” (ibid.). I discuss this aspect in Sec. 5.2.1 in detail.
174 Miller et al. (1996, p. 87) defined the bicameral core as “the core, which will depend on the way in which
a given set of voters is divided into two chambers, a majority of each chamber being required to pass
legislation.”
175 In addition to the stopping rule other factor my influence the suitability of the core concept as well. McK-
elvey and Ordeshook (1981) analyzed spatial majority voting games. Their results suggest that the per-
formance of the core can be affected by several different aspects; e.g., the structure of the alternative space
or the dominance relation underlying the social ordering.
90
The main difference between my design and the previous contributions is the information
level of the subjects. My experiment is characterized by complete information of one’s
own and other subjects’ payoffs (cf. Sec. 4.2.3). This should enable the participants to
make the necessary considerations. Contrary to my approach, other designs (e.g., Fiorina
and Plott, 1978; Eavey and Miller, 1984) limited the actual level of information to, e.g., the
ordinal value of alternatives or the ideal positions but not the preference distribution of
other subjects (S&K, 2010, p. 669).
Previous experiments without a stopping rule have shown that once the choice had con-
verged on the core alternative a deviation is highly unlikely (S&K, 2010, p. 668ff). I re-
viewed these findings in a pre-test session of my experimental design. Here, the collective
decision was not immediately locked in after enough subjects of a group had agreed on
one alternative. Subjects were informed that the threshold was reached and which alter-
native constituted the collective decision, but they were also asked to vote again.176 In
this pre-test 18 participants were in the laboratory and were separated into three groups
of six players. The voting procedure of one round did not end until all three groups had
reached a collective agreement. Thus, some groups had to vote no less than five more
times. I conducted three rounds in which subjects voted in the group of six and further
three rounds in which subjects voted in groups of three players.177 Overall, I collected
18 collective decisions. In all rounds and under all procedures the first reached collective
decision was not altered. Admittedly, I found some variance in voting behavior across
groups; i.e., different groups agreed on different alternatives and subjects exposed to the
same decision problem voted differently. But not once was the first reached collective
decision revised.
In addition, work on voting under plurality rule has shown that strategic voting pushes
the collective decision towards a stable equilibrium (Feddersen, 1992). Here, many con-
tributions investigated party-positioning games (for an extensive review on this topic cf.
Coughlin, 1990a,b). Other models by Cox (1997) and Palfrey (1989) assumed nonstrate-
gic parties but strategic voters when looking at Duverger’s Law (Duverger, 1954). Palfrey
(1989) argued that as the size of the electorate increases the rational support for all par-
ties, except for the top two, goes asymptotically toward zero. Fey (1997) investigated
the “wasted vote” phenomenon and how voters can coordinate. He showed that in a
Bayesian game model of strategic voting non-convergence is possible, but this consti-
tutes only an extreme case.
The second main difference between the payoff tables lies in the alternative’s overall
sum. The experiment contained constant-sum as well as non-constant-sum tables. In a
constant-sum payoff table the overall sum of an alternative (the sum of points of all six
players) stays constant. All nine alternatives provide the identical amount of points. In
176 Subjects
were also still provided with information about other players’ previous votes.
177 Thesemodifications refer to voting procedures I use as treatment in my experiment. I discuss the proce-
dures and their differences in detail in Sec. 4.2.6.
91
4 The Experiment
non-constant-sum tables the alternative’s overall sum varies between alternatives. This
implemented two different decision situations for the participants.
In general, experiments can be classified according to their potential for conflict. This
is based on the overall experimental setting and its inherent dynamics. For example,
compared to a public good game the use of a bargaining setting implements a distinctly
higher level of conflict. Bargaining focuses not on achieving a goal together but on dis-
tributing a collective resource; a task that clearly comprises a certain amount of disagree-
ment. Hinich and Munger (1997, p. 7) stressed that such “disagreement tests collective
choice mechanisms; conflict strains the ties that gather a group of individuals into a so-
ciety.” My design resembled no standard bargaining procedure. In such a setting a first
actor proposes a distribution and a second actor either accepts or rejects the proposal (cf.
Kagel and Roth, 1995, Chap. 4). Yet, I followed a similar intention by using constant-sum
as well as non-constant-sum payoff tables.
A constant-sum table leaves the subjects with one single question: how should they split
the exogenously defined amount of points. Comparing two alternatives, every gain of
one player is mirrored by a loss of at least one other player. The focus rests on distribu-
tion and constrains the tactical opportunities of the subjects. Because of this limitation
“zero sum games can be regarded as the branch of game theory with the most solid the-
oretical foundations” (Palacios-Huerta and Volij, 2008, p. 3). Non-constant-sum tables
bring another aspect into play, the overall sum of points. This may be seen as a measure
for social welfare or an alternative’s effectiveness. This sum differs between the given
alternatives. Only the possible range for the overall sum is exogenously given. The com-
mittee decides about the exact amount endogenously. Thus, subjects have to deal with a
trade-off between distribution and allocation.178
This characteristic is best met by the concept of game harmony put forward by Zizzo and
Tan (2009).179 They described a subject’s perception of the experimental game as decisive
for their decision “how to play”. Game harmony is then “a generic game property that
describes how harmonious (non-conflictual) or disharmonious (conflictual) the interests
of players are, as embodied in the payoffs” (Zizzo and Tan, 2009, p. 3). Harmony mea-
sures can be calculated solely based on the payoff matrix. My two table set-ups were
located at the extremes of the game harmony scale.180 Constant sum-games are games
178 Not all findings correspond with this view. Bottom and Paese (1997, p. 1920) pointed out that even in bar-
gaining situations with the possibility for joint gains the “negotiators often behave as though their own
interests and their counterpart’s were completely opposed. Rather than engaging in information sharing,
exploration of options, and joint problem solving, negotiators exchange threats, make ultimatums, and
rely on other tactics that might be more appropriate in a zero-sum game.” Neale and Bazerman (1991,
p. 61) denoted this as the “myth of the fixed pie” where participants do not perceive a possible common
progress.
179 Contrary to my experiment the study by Zizzo and Tan (2009) focused on the relationship between game
harmony and cooperation in static games. In their experiment players move simultaneously and the
game is played only once.
180 Zizzo and Tan (2009, p. 4) pointed out that “in the coordination game there is perfect harmony of interests
between the players: the only problem is one, indeed, of coordination. In the constant-sum game, the
gain of a player is the loss of another, which means that there is perfect disharmony of interests.”
92
of pure disharmony, as every subject’s gain is another subject’s loss. On the other hand,
games with varying payoff-sums still require coordination. But cooperation even is in
the interest of the most self-interested subject.181
The payoff tables were assigned randomly in every experimental session. Therefore I had
to account for (statistical) effects which resulted out of the different usage of tables. With-
out this calibration the results would be biased. For example, consider the overall sum
of the alternatives. They range in payoff table 17 from 104 points to 145 points. Looking
at payoff table 19, the range extends from 83 points to 91 points. When interested in the
average number of points collected, I have to take the possible range into account. If one
were to use frequently table 17 and less often table 19 one would end up, everything else
equal, with a higher absolute average of points. An effect based solely on the payoff table
differences. Thus, I abstained from using absolute values when interpreting the results
in the empirical analysis but employed relative judgments. The specific normalization is
discussed in the corresponding section of Chap. 5.
4.2.5 Multiple rounds
Every experimental session lasted multiple rounds. In every round I varied two aspects.
Firstly, committee members were matched randomly out of 18 participants.182 This pre-
vented subjects from building a reputation between rounds.183 Secondly, the payoff ta-
bles were assigned to each round by a random draw. This draw took place in every
experimental session without replacement. Thus, every table could be played only once
during a session. Yet, no session lasted long enough to use all the tables.184 Both “ran-
dom” facts were stated to the subjects at the beginning of the experiment.
Subjects were instructed that the experiment would continue for several rounds and that
there was no predetermined upper limit to the number of rounds. Nevertheless, subjects
may have guessed the maximum number of rounds from the overall time schedule. The
recruiting email sent to all members of the subject pool, informed participants about how
long they would be required to stay. Still, subjects were not informed or aware of whether
a certain round was the final ballot of the experiment.
Since my experiment extends over several rounds I have to consider possible effects over
time. Following Crawford and Haller (1990) as well as Roth and Erev (1995) it is reason-
able to assume that subjects changed their voting patterns after they got acquainted to
181 Dixit et al. (2009, Sec. 2.B, p. 21ff) classified games in a similar fashion. They focus on if “player’s interests
[are] in total conflict, or is there some commonality?”
182 The matching followed the stranger matching procedure which resembles “a random matching, i.e., in
every period, the group is determined by the computer’s random generator” (z-Tree tutorial, source:
www.iew.uzh.ch/ztree/ztree21tutorial.pdf, version June 3, 2002).
183 As I excluded reputation building through my design I will not discuss its effects in detail. For compre-
hensive assessment please cf. Bolton and Ockenfels (2005) and Levin (2009, Chap. 4 and 13).
184 The longest session lasted for 2:45h and extended over 16 rounds.
93
4 The Experiment
the voting game.185 If this happens, the analysis would miss this behavior if only focused
on average values over all rounds. Yet, such a change of behavior is no axiomatic statue.
Herzberg and Wilson (1988) found no support of a change in behavior over rounds when
they conducted experiments on the proportion of sophisticated and sincere voting. Thus,
experience alone may not be enough to change voting behavior. Nevertheless, in my em-
pirical analysis I accounted for this by depicting not only average values but by testing
all data for possible trends over the rounds.186
4.2.6 Treatment
The experiment focused on problems related to economies of scope, namely the sepa-
ration of decision-making competences over policy proposals characterized by NSP.187
This was secured by the large extent of environment control in the laboratory which also
enables a high level of internal and construct validity (Morton and Williams, 2010, p. 192).
My payoff tables have a two-dimensional structure (cf. Fig. 4.2). However, this does not
automatically results in nonseparable decision problem. A selection of nine alternatives
from which one can be unambiguously selected may just as well be shown in either a
3 × 3 or 1 × 9 table. The form of presentation (alone) is not decisive188 because the payoff
tables are not assembled in a way that adjacent fields are related. In the spatial presen-
tation of payment schemes (cf. Fig. 4.1) the proximity of any point to the ideal position
is crucial; and for nearby points (per definition) similar values apply. Yet, in my payoff
table C2 needs not to be more alike to C3 then A1 is to C3. The spatial principle of conti-
nuity is not found in the tables. The necessary key element for nonseparability is the split
into multiple dimensions and the separation of the associated decisions.
Therefore, my treatment is a change of voting rules which the subjects had to follow.189
The intervention varied the decision-making with respect to two aspects. Firstly, the vot-
ing competence of the participants was altered. The subjects were either put all together
for a common task or separated into groups with their own responsibilities when cast-
ing their votes. And secondly, the voting sequence of the procedure was modified. The
ballots were either held all at the same time or successively, one after the other.
185 A round in my experiment did not only consist of one ballot or a single cast vote. Subjects voted for
their preferred alternative in every ballot in every round. Thus, in every round they were involved in a
dynamic coordination game with the other subjects of their group. In the next round the composition of
their group was changed. As the matching was set randomly they could not reason about the identity
of other players now being part of their group. Yet, of course each subject had experienced a variety of
reactions, successes or disappointments in previous rounds. Thus, in every new round subjects ought to
adjust their voting choices, according to their experiences made during the actual experiment.
186 For this purpose I used nonparametric statistics (cf. Sec. 5.2) instead of counterbalancing, as this would
require the assumption of a linear effect (Martin, 2008, p. 156ff).

187 By assuming that decision makers are i) policy seekers and ii) randomly assigned to each policy area I was
able to disregard the potential for bureaucratic drift (cf. Sec. 2.1) and set a more narrow focus.
188 The labels of the nine alternatives, regardless of whether A1-C3, A1-A9 or A-I, do not matter.
189 This is not a novel but validated approach when assessing the influence of organizational aspects. Already
Shepsle (1979) compared various institutional settings as the division of labor in committee systems,
specialization arrangements or the monitoring of sub-level units.
94
Altogether, the experiment used simultaneous, sequential, pooled and delegated decision-
making. Tab. 4.1 gives an overview of the specifications. The experiment comprised three
different procedures I refer to as pooling, simultaneous delegation and sequential delegation.
All procedures followed the overall design in that if no alternative was chosen by a ma-
jority, members received information on how the other players voted190 and the game
continues with a new voting. Ballots were repeated until an alternative receives the nec-
essary votes. After the group had made its decision, all players were assigned their cor-
responding number of points regardless of whether they had voted for or against this
alternative themselves.
Table 4.1: Treatment characteristics
VOTING COMPETENCE
pooled competence delegated competence
simultaneous voting pooling simultaneous delegation
VOTING SEQUENCE
sequential voting - sequential delegation
POOLING
Under pooling (henceforth POL) the group of six subjects voted for either of nine alter-
natives (A1-C3). All players cast their vote simultaneously. If neither of the alternatives
got at least the qualified majority of four out of six votes, the information on the voting
behavior of the group was shown and following a new ballot was held. This process
continued until one of the alternatives received the required majority which marked the
completion of one round.
This voting procedure is not a decision situation affected by nonseparability. All players
decide simultaneously about a specific and unique alternative to be chosen. There are no
restraints for the subjects when they cast their votes, nor any temporal sequence which
forces a separation into individual issues and conditional adaptations.
SIMULTANEOUS DELEGATION
Under simultaneous delegation (henceforth SIM) two delegations consisting of three

players each decided simultaneously over one table with nine alternatives. Each dele-
gate could only vote for a subset of the alternatives, either column or row. Subjects were
randomly assigned to each delegation. To reach a decision at least two of the three sub-
jects of a delegation had to vote for the same subset. Within each delegation votes were
cast simultaneously. If no single subset got at least the qualified majority, the informa-
tion on the voting behavior of the other delegation members was shown and the voting
continued.
Delegates were not informed about the voting behavior in the other delegation until both
delegations had reached a decision. The intersecting of both subsets determined the cho-
sen alternative. For example, one delegation decides on the column (A, B or C) and
190 Inthe lower left corner of the decision screen the other players and one’s own previous voting behavior
(in the current round) is displayed.
95
4 The Experiment
chooses set { A}. The other delegation votes on the row (1, 2 or 3) and chooses set {3};
then the resulting outcome for all six players is alternative { A3}.
SEQUENTIAL DELEGATION
Under sequential delegation (henceforth SEQ) two delegations consisting of three play-
ers each decided sequentially over one table with nine alternatives. Each delegate could
only vote for a subset of the alternatives, either column or row. Subjects were randomly
assigned to each delegation. To reach a decision at least two of the three subjects of a
delegation had to vote for the same subset. Within each delegation votes were cast si-
multaneously. If n o single subset got at least the qualified majority, the information on
the voting behavior of the other delegation members was shown and the voting contin-
ued. So far the procedure follows SIM. In contrast to the former, the voting of the two
delegations took place sequentially.191 First, one delegation decided on the column (A, B
or C). After this delegation had reached a decision, all six players were informed about
the chosen column.192 Then the other delegation voted on the row (1, 2 or 3). Again, the
intersecting set determined the chosen alternative.
The two delegation procedures SIM and SEQ differed based on the degree of information
players had available for their decision. Under SIM each delegation was not informed
about the decision of the other delegation until it had reached an agreement of its own.
Under SEQ the delegation which decided on the column “moves first” (this delegation is
further referred to as first stage). The three participants of the first stage knew that the
other delegation will be informed about their decision before they “move second” and
vote on the row (this delegation is further referred to as second stage). As the second
stage delegation already knew the results of the column vote, they faced a rather simple
decision. For example, assume in the first stage column { B} is selected. Then the sec-
ond stage decision problem over the row (1, 2 or 3) is reduced to the alternative-specific
question { B1} vs. { B2} vs. { B3}.
In a nutshell, my experiment focused on the separation of decision-making competences.
I altered the voting procedure in terms of sequence and competences of the involved
players (cf. Tab. 4.1). Under POL, my baseline, all six players decided simultaneously
from nine alternatives. Each participant had one vote and could choose to allocate it to
one specific alternative. SIM divided the group of six into two delegations of three play-
ers. Within each delegation votes were cast simultaneously, and both delegations held
their ballots at the same time. The participants were now asked to choose a subset (either
column or row) of alternatives. The intersection of the decisions of the two delegations
191 The introduction of sequence was, by no means, only a small alternation. Game theory judges the timing
of play as a fundamental characteristic of a game and classifies simultaneous and sequential games into
different categories (cf. Dixit et al., 2009). In Chap. 5 I therefore introduce appropriately adapted game
theoretic solution concepts for each design.
192 The delegation voting second was informed about the subset chosen by the first delegation. Yet, they
could only observe the collective decision but not which of the players cast their vote in favor or against
the majority decision.
96
determined the final outcome. SEQ introduced sequence to the voting process. While
within each delegation votes were still cast simultaneously, the voting of the two dele-
gations took place sequentially. Most importantly, the second stage delegation knew the
final decisions of the first stage delegation before holding its own ballot. Participants
of the first stage delegation chose subsets while participants of the second stage delega-
tion were asked to select a specific alternative out of the remaining subset (which always
contained three alternatives).
Overall, changing the voting rules generated four distinct decision situations which dif-
fered with respect to their cognitive complexity and prevailing uncertainty: POL, SEQ
first stage, SEQ second stage and SIM. Before I assign the respective manifestations of the
two characteristics to these situations (cf. Tab. 4.2), I clearly distinguish the criteria from
each other.193
COMPLEXITY
I start with complexity. This is no trivial task. Braumöller (2003, p. 212) attributed this
to the fact that “the concept is so slippery.” This holds even though highly complex and
difficult-to-analyze dynamics are common in real-world processes (Lindgren, 1991).194
Hinich (2008, p. 999) pointed out that “political and social games are so complex that
the assumption of common knowledge that all actors know all the states of nature in
the games and the conditional joint density of the states is grossly false.” Accordingly,
Koremenos (2008, p. 169) understood a complex problem “as uncertainty about behavior
or the state of world, enforcement problem, or commitment problem.” Another attempt
to define complexity was made by Jervis (1997, p.35) who referred to constellations where
“the effect of one variable or characteristic can depend on which others are present.”
Yet, these perceptions are not unambiguous as they comprise a mixture of uncertainty,
ignorance and interdependence as a cause. To structure my experimental design I strive
rather to define complexity and uncertainty independently.
I agree with Braumöller (2003, p. 212) in that the most concrete and clear solution is pro-
vided by Ragin (1987) who assessed a (causal) complex case as one that “results from
several different combinations of conditions” (Ragin, 1987, p. 20). He added further that
“multiple causes interact with one another to produce effects, and the manner in which
they interact is described by the logical operators ’and’ and ’or”’ (Ragin, 1987, p. 89-
93).195 In my experiment the decisions on column and row of a payoff table interacted to
assemble the final outcome. The level of complexity resulted from the number of multi-
ple factors which had to be taken into consideration. I did not try to measure complexity
193 Cf. Oaksford and Chater (1998) for a broader overview and compilation of contributions on the cognitive
science of human reasoning under uncertainty.
194 Lindgren (1991) conducted a simulation model analysis which encounters several kinds of evolutionary
phenomena. For example, these are periods of stasis, punctuated equilibria or large extinctions.
195 Braumöller (2003, p. 212-213) pointed out that by using this conception many operational concepts as, e.g.,
substitutability or necessary and sufficient conditions can be understood as special cases of complexity.
97
4 The Experiment
in exact figures. Rather, I restricted my classification, when following this definition, to

relative assessments.196
The splitting up of the decision-making competences implied a reduction in the number
of options which participants could select (9 alternatives vs. 3 subsets). Yet, this was true
only if supplemented by a sequence of decisions, as with simultaneous acting the other
delegation does play a role. Under SIM the mutual dynamics kept up the level of com-
plexity as nobody knew the result of the other delegation before reaching an agreement
in its own. The tasks under POL and SIM were not identical, but each demanded so-
phisticated computations in its own way. Thus, the number of the interacting causes was
reduced only under separated and sequenced decision-making, and so was the level of
cognitive complexity. This is best illustrated when considering POL in comparison to the
second stage under SEQ. Both situations allowed the participants to vote for a specific
alternative. Yet, while the first comprised five other players and nine alternatives, the
second contained two other players and three alternatives. The second stage under SEQ
therefore just resembled a relatively simple copy of POL.
UNCERTAINTY
The second varying aspect in my design was the level of uncertainty.197 Here, it is not
difficult to find a definition; rather one must select the appropriate version out of the
multitude of possibilities. Most important is the distinction between exogenous and en-
dogenous uncertainty; i.e., between known and unknown probabilities for certain out-
comes. This classification goes back to Knight (1921).198 Under exogenous uncertainty
the probabilities of occurrence for all possible events are given in advance. A lottery re-
sembles such a state where chance and associated risk are common knowledge. “More
generally, risky situations are games against nature” (Heinemann, 2005, p. 296). Contrary
to this, under endogenous uncertainty the specific probabilities of an event are unknown.
This arises, e.g., “in situations, where the outcome depends on social interaction” (Heine-
mann, 2005, p. 296).
In addition to these two clear-cut cases, Ellsberg (1961) associated situations which are
characterized neither by known risks nor by complete uncertainty but by the lack of in-
formation about relative likelihoods with the term ambiguity. Here, Weber and Johnson
(2008, p. 132) underlined clearly that “knowledge about the probability distribution of
possible outcomes of a choice can lie anywhere on a continuum, from complete igno-
rance (not even the possible outcomes are known) at one end, through various degrees
196 Cf. Stanovich and West (2000, p. 648-649) for a comprehensive discussion of statistical measures for cogni-
tive ability.
197 Cf.Weber and Johnson (2008) for a comprehensive assessment on decisions under uncertainty. The authors
covered psychological, economic, and neuroeconomic explanations of risk preferences and also included
an historical overview.
198 Knight (1921) reached this conclusion when assessing business decisions which typically involve non-
measurable risk, as they “deal with situations which are far too unique, generally speaking, for any sort
of statistical tabulation to have any value for guidance. The conception of an objectively measurable
probability or chance is simply inapplicable” (Knight, 1921, p. 231).
98
of partial ignorance (where outcomes may be known, but their probabilities not precisely
specified, denoted as uncertainty or ambiguity), to risk (where the full outcome distri-
bution is precisely specified), to certainty (where only a single, deterministic outcome
is known to result).” In their comprehensive contribution on ambiguity Eichberger and
Kelsey (2008) distinguished between the different terms in the following way: uncer-
tainty is “a generic term to describe all states of information about probabilities. The
term risk will be used when the relevant probabilities are known. Ambiguity will refer
to situations where some or all of the relevant information about probabilities is missing.
Choices are said to be ambiguous if they are influenced by events whose probabilities are
unknown or difficult to determine” (Eichberger and Kelsey, 2008, p. 4).
Within the domain of uncertainty one must differentiate, for the reason of missing infor-
mation as environmental, role or strategic uncertainty are possible. The first two can be
excluded with respect to my experimental design, though for quite different reasons. En-
vironmental uncertainty resembles a cornerstone in management decision-making (Holm
et al., 2013, p. 2). Here, uncertainty arises from the fluctuating demand of customers
(Bernstein and Federgruen, 2005), delays in project operations (Pich et al., 2002) or, more
generally, unknown environmental conditions (e.g., Aldrich, 1979; Leblebici and Salan-
cik, 1981).199 For a laboratory investigation such sources of uncertainty are quite un-
likely.200
Role uncertainty is “a commonly used experimental procedure. It consists of collecting

from the same subject responses to tasks assigned to different roles, and letting a random
mechanism determine which role’s actions will be implemented” (Iriberri and Rey-Biel,
2011, p. 160). Yet, in my experiment subject were always aware of their role. Their iden-
tity (player number), the environment (decision-making procedure) and the earning pos-
sibilities (payoff table) were always known before they had to make a decision.201 The
controlling capabilities of a laboratory allowed me not only to prevent biasing influences
but to set a specific common information level for all participants.
The term strategic uncertainty was first used by Van Huyck et al. (1991). Most impor-
tantly, the authors pointed out that “strategic uncertainty arises even in situations where
objectives, feasible strategies, and institutions are completely specified and are common
knowledge. [... Thus, it] should not be confused with uncertainty arising from incom-
plete information about other aspects of a decision maker’s environment” (Van Huyck
et al., 1991, p. 886).202 While defining the basic principles, they did “not give a proper
definition, but it seems clear that they mean the uncertainty arising from multiple equi-
199 Morris and Shin (2002, p. 2) used also the term “‘structural uncertainty” for uncertainty related to “the
underlying fundamentals.”
200 This is especially true in comparison to potential causes within the experimental design.
201 Before starting the experiment, a quiz tested the subjects’ understanding of the upcoming game and se-
cured their understanding of the displayed information.

202 Van Huyck et al. (1991) referred prominently to (Sugden, 1989, p. 881) and his “lucid critique of the view
that a rational decision maker can deduce a unique ’rational’ strategy from the information contained in
a complete information description of a game.”
99
4 The Experiment
libria” (Heinemann et al., 2009, p. 181). This left room for interpretation considering a
concrete definition.203
Holm et al. (2013, p. 3) described “competition (where the uncertainty concerns the in-
dividual’s performance relative to others) and trust (where there is a ’social’ risk that
another party does not act favorably towards the trustee) to exemplify situations involv-
ing strategic forms of uncertainty.” Thus, this type of uncertainty relates prominently
to other players’ actions (Andersson et al., 2012). Morris and Shin (2002, p. 2) extended
its context to “uncertainty concerning the actions and beliefs (and beliefs about the be-
liefs) of others.” Here, the multitude of equilibria arose out of players’ own actions and
their expectations concerning other players’ actions. Many equilibria cause computa-
tional problems as the number of possible states that need to be considered is large. In
such cases, “even if the underlying fundamentals of the problem were known for sure,
the strategic uncertainty is still all-pervasive” (Morris and Shin, 2002, p. 2). Recent exper-
imental studies have validated this influence of (strategic) uncertainty on behavior (e.g.,
Heinemann et al., 2009; Cabrales et al., 2010). More precisely, “in such situations, even the
slightest uncertainty about other players’ choices might lead a player to deviate from his
or her equilibrium strategy” (Andersson et al., 2012, p.1).204 Even worse, a heterogeneous
response to uncertainty provokes misunderstanding.205 Of course, “public information
reduces coordination failures and leads to more efficient strategies” (Heinemann et al.,
2009, p. 184). Yet, often the provision of information is not possible.
The linkage between strategic uncertainty and ambiguity becomes clear when one con-
siders their causation; both forms of uncertainty originate from the absence of relevant
information (about probabilities). Interestingly, Heinemann et al. (2009, p. 2) found that a
“subject’s certainty equivalents of coordination games are positively related to certainty
equivalents of lotteries[. ... Thus,] subjects who avoid risk or new experience also avoid
strategic uncertainty. This suggests that subjects have similar perceptions of exogenous
and strategic uncertainty if both situations are framed in a similar way.”206 By consid-
203 Due to my experimental setting I could exclude concepts as “reverberant doubt” (cf. Hofstadter, 1985,
p. 752-753), i.e., the fear of a non-realization, from my further discussion. As I always iterated ballots
until a collective agreement is reached this specific form of strategic uncertainty is not relevant.
204 Crawford and Haller (1990, p. 572) even stated that “this strategic uncertainty undermines the arguments
that players should play according to any given equilibrium and even calls into question the rationale
for playing an equilibrium strategy.” Yet, this debate seems undecided as Andersson et al. (2012, p. 1)
made the counter-argument that “in the laboratory, human subjects’ behavior in games with multiple
equilibria has also been found to be fairly stable and predictable in the aggregate”.
205 I restricted my analysis to ballots which yielded a collective result by the necessary majority of individual
votes (cf. Sec. 4.3.1). This could not answer all questions as “the behavior of people in situations of risk
and uncertainty is complex and multiply determined” (Weber and Johnson, 2008, p. 141). Further studies
may look into the complete bargaining process and include non-decisive ballots. This would enable one
to utilize the various tools provided by psychology and neuroeconomics as, e.g., one-step look-ahead
algorithm (OSLA) or reinforcement learning (Chalkiadakis and Boutilier, 2004, 2008). Those instruments
will then enable “a far more nuanced assessment and understanding of both general behavior patterns
and individual or group differences in behavior” (Weber and Johnson, 2008, p. 141).
206 In addition, Heinemann et al. (2009) investigated the “experience seeking” of their participants by us-
ing the Sensation Seeking Scale V (SSS-V, Zuckerman, 1994), a psychologist’s measure to characterize
personalities. They found that the scale resembled a good predictor for a subject’s certainty equivalents.
100
ering strategic uncertainty in my experimental design I therefore also covered the wide
range of exogenous uncertainty.
Following Brandenburger (1996) and Heinemann et al. (2009), I defined strategic uncer-
tainty as uncertainty arising from missing information about the purposeful behavior
of players in an interactive decision situation. In the words of Crawford and Haller
(1990), players in a coordination experiment “may thus bear significant uncertainty about
how other players will respond to its multiplicity of equilibria, even with complete in-
formation” (Crawford and Haller, 1990, p. 572). Therefore, the “coordination strategies
reflect their uncertainty about how their partners will respond to multiple-equilibrium
problems; this uncertainty constrains the statistical relationships between their strategy
choices players can bring about” (Crawford and Haller, 1990, p. 571).
In my experiment, most of the voting record information was missing when subjects have
to find a collective agreement under SIM. Here, each delegation was not informed about
the decision of the other delegation until they reached an agreement themselves. Un-
til the final choice was made, no player knew the specific consequences of their choice
with certainty. This distinguished SIM from SEQ which applied also separated decision-
making.207 Here, the three participants of the first stage knew that the second stage dele-
gation was informed about their choice before they had to decide themselves. Yet, while
they were able to restrict the choices of their successors they were still left in the dark
about the consequences of these actions. Thus, the first stage was associated with consid-
erable uncertainty. Contrary, the second stage finalized the decision. Because the other
delegation‘s vote was already known, the level of uncertainty was distinctly lower when
players were part of the complete or, at least, the last step of the decision process. This
holds true also for POL where no information was withheld from any player.
DECISION SITUATIONS
To summarize, POL was cognitively demanding because subjects had to consider five
other players and nine alternatives. Yet, it did not resemble a nonseparable decision
problem as the other two voting rules. Every subject was able to choose his preferred
alternative unambiguously. Neither voting competences nor voting sequences required
a separation of the decision into single issues.
Contrasting, SEQ had two stages. In the second stage subjects found themselves in a
drastically simplified (i.e., less cognitive demanding) version of POL with only two other
players and three alternatives. Subjects in the first stage had to anticipate the outcome of
the second stage for each of the three possible subsets. Yet, at least they got a chance to
set the agenda for the succeeding group; in the end, they knew which remaining subset
their successors would vote on. This reduced the complexity in comparison to that faced
by subjects under SIM, who had to consider mutual dynamics.
207 The chosen form of separated decision-making excludes all forms of signaling through collusion (commu-
nication was forbidden) or reputation building (reshuffling after each round).
101
4 The Experiment
Under SIM subjects were also exposed to a high level of uncertainty about the behavior
of others. They had to evaluate nine alternatives for six players in two delegations of
three players acting at the same time without any further information. In the first stage
under SEQ there was also no information on the behavior of its successors. Uncertainty
is related to how far to the end of the decision-making a player was involved. This can
be deduced from their choice opportunities. Could they only choose between subsets
or were they allowed to select a specific alternative? To vote for a concrete alternative
was only possible under POL or the second stage under SEQ. In relation to the other
situations the level of uncertainty was rather low as in both cases the participants took
part in the decision-making until the final and alternative-specific pick. Tab. 4.2 orders
the decision situations based on cognitive complexity and uncertainty. The (relative)
distinctions between the situations enabled me to identify the observed effects for the
characteristics separately.
Table 4.2: Characteristics of the different decision situations

EXPLANATORY NOTE
The table points out the characteristics of the four different decision situations. They can be differentiated according to
their cognitive complexity and prevailing uncertainty. The level of complexity refers to how demanding the computations
were a subject had to perform. The level of uncertainty indicates to what extent a subject remained in the dark about the
behavior of other participants.
LEVEL OF UNCERTAINTY
lower higher
lower second stage sequential delegation first stage sequential delegation
LEVEL OF COMPLEXITY
higher pooling simultaneous delegation
Complexity and uncertainty are two main characteristics which come along with non-
separability. As discussed in Sec. 2.5, the previous (theoretical) literature attributed them
to cause opportunities for strategic manipulation in form of a first-mover advantage
(ARGUMENT 2) as well as sub-optimal outcomes (ARGUMENT 3). The first aim of my
experiment was to examine these arguments empirically. In accordance with literature
on laboratory experiments, the alterations in uncertainty and complexity led to further
changes in subjects’ behavior. Chap. 5 and Chap. 6 discuss this literature on behavioral
patterns and their expected variation. Those will be formulated as specific expectations
for my experiment. To investigate if these expectations are confirmed was the second
aim of my empirical analysis; and thus, the theoretical discussion about nonseparability
is supplemented with further arguments on its impact.
4.2.7 Two procedures per session
I implemented two procedures in every experimental session. At the beginning of the

experiment, the subjects were informed that the experiment would have one specific de-
cision rule for some rounds and that this rule would be altered in later rounds. They
were not told in advance how many rounds they would play under either rule. In fact,
the number of rounds for the two procedures was not even known to the instructor. In-
stead, they were endogenously defined while the experiment was running.
102
Before conducting a session, the laboratory had to be booked for the corresponding time
slot. In this duration, the participants were brought into the lab, the instructions were
explained, several rounds of the experiment were played, the participants filled out a
questionnaire and all subjects received their payoffs. The available time slot for the voting
2
experiment was approximately 3 of the overall reserved time. The procedure changed
after half of this time had gone by. The number of rounds played so far was not decisive.
If the committees needed more time to reach a decision they played fewer rounds. If they
agreed quickly they participated in a higher number of rounds.
As I implemented two decision rules in every session, the examination of the data was
a mix of between-subject and within-subject analysis208 . In a within-subjects design, each
participant is tested under each procedure (Morton and Williams, 2010, p. 86-87). The
results of the same subject under the different procedures are compared. In a between-
subjects design, each participant is tested under one procedure only (ibid.). The results of
different subjects under each condition are compared. My experiment resembled neither
exactly. No subject is tested under all or only one procedure. This followed the idea of
randomized partial counterbalancing (cf. Mitchell, 2003, Chap. 13, p. 485).
In my analysis of the experimental data I used both methods as they complement each
other.209 The between-subject design implies fewer biases like EDE, treatment confound-
ing, etc. The within-subject design reduces individual variability and noise. This en-
abled also to obtain a within estimate of the treatment effect.210 “The accuracy of this
approach depends on whether any order biases cancel one another out across the two
orders” (Charness et al., 2012, p. 6). A requirement for combining both methods is that
subjects are tested under different modifications of the same factor (MacKenzie, 2012). In
my experiment the treatment intervention altered the decision-making procedure.211 For
the between-subject analysis I used only the data gathered under the first decision rule in
each session. For the within-subject analysis I utilized also the data gathered under the
second decision rule compared to the first procedure in each session.212
4.2.8 Payment mechanism
It is still an ongoing discussion whether and how to pay participants in experiments.

Amir et al. (2012) investigated if financial motivation matters at all. Therefore, they re-
208 Cf. Martin (2008, Chap. 8, p. 148-170) for an excellent juxtaposition of the two approaches.
209 Cf. Charness et al. (2012) for a discussion of the advantages and disadvantages of the methods.
210 Isaac and Walker (1988a,b) analyzed how individuals change their behavior in response to a change in
stakes in public good experiments.

211 MacKenzie (2012) plainly illustrated the necessary conditions for a combined analysis: “One group of
participants is tested under condition A, a separate group is tested under condition B, and so on. The
assumption here is that condition A and condition B are different levels of the same factor. For example
the factor might be device and the levels might be mouse, trackball, and touchpad. In experiments with
more than one factor, it is possible to use a within-subjects (repeated-measures) assignment for the levels
of one factor and a between-subjects assignment for the levels of another factor.”
212 The distinction between first and second decision rule also accounts for possible learning or experience
effects of having played a different procedure before.
103
4 The Experiment
peated canonical economic games online using very low stakes.213 They found that their
results were comparable to those run in laboratory settings. Camerer and Hogarth (1999)
argue that no replication of an experimental study has achieved more theory conforma-
tion or less rationality violations by purely raising incentives. The authors do not call
for more studies without financial incentives generally. The benchmark is always set
by “previous [related] research showing that financial incentives did not matter in their
task” (Camerer and Hogarth, 1999, p. 25). As I link my results to previous work on col-
lective decision (cf. Sec. 4.1.2) I followed their lead by using financial incentives.
If participants would be paid for every round using an accumulated payoff mechanism
(APM), the experiment would clearly run into wealth, hedging or portfolio effects (cf.
Cox, 2010; Ham et al., 2005; Heinemann, 2005). I would also not be able to regard every
round as a single game because of spill-over effects between decision problems. In or-
der to avoid such effects, the total earnings were determined by a random round payoff
mechanism (RRPM)214 . Using this mechanism one or a few rounds are randomly chosen
as the basis for a subjects payment (Morton and Williams, 2010, p. 279).
In addition to preventing wealth effects RRPM has the additional advantages that mone-
tary payoffs in each period can be raised. This may increase the salience subjects assign to
their choices (Morton and Williams, 2010, p. 279). So the payment mechanism influences
a subject’s perception: every round comes closest to a single one-shot game. This fol-
lows Kuziemko et al. (2011, p. 17) in abstaining from dynamic payoff accumulations over
rounds for “making the experiment a series of one-shot games”. The advantage of such a
series is that it is not necessary “to make assumptions regarding players time horizons”
(ibid.). This facilitates the analysis as it is not easy to distinguish what determines sub-
jects’ time horizon or how they try to maximize their payoffs. Benartzi and Thaler (1995)
argued that individuals just always maximize their current payoffs. This holds true even
in experiments where the defining outcome is the future final round. Gneezy and Potters
(1997) showed that individuals maximize a specific “evaluation period”. In my experi-
ment subjects neither knew the end of the experiment nor which rounds contributed to
their payoff. Each round should therefore have constituted such an evaluation period
and have been played as if it was one-shot. This is in line with Camerer et al. (1993,
p. 44) who concluded that even in a simple sequential game “subjects concentrated on
the current round when making decisions.” In summary, experimental subjects tend to
maximize current payoffs even in a setting over multiple rounds or when the received
payoff is explicitly based on the final balance.215
213 More precisely, Amir et al. (2012) implemented two payoff conditions: i) a “stakes condition”, in which a
subject’s payoff was based on the outcome of the game (max. $1), and ii) and a “no-stakes condition”, in
which a subjects payoff was unaffected by the outcome of the game. Overall, they found some but little
difference in behavior between the stakes and no-stakes conditions.
214 This way of paying subjects is also called random lottery incentive mechanism (Grether and Plott, 1979).
215 Kuziemko et al. (2011) explained this behavior with uncertainty about the experimental setting. Subjects
might value the current payoffs more because their distribution is known with certainty while future
rounds are undetermined.
104
Arzieli et al. (2012) explained that repeated experiments comprehend two relevant ele-
ments which determine the choices participants make. The first is the decision problem,
and the second the payment mechanism. These aspects cannot be separated from one
another, as the implication of a specific design is changed by different payment schemes
(Chandrasekhar and Xandri, 2012). Thus, the same experiment would yield different re-
sults if it is played using either an APM or RRPM (Lee, 2008).216 An experiment therefore
needs an “incentive compatible payment mechanism” (Arzieli et al., 2012, p. 15). A sub-
ject’s overall optimal strategy also has to be their optimal choice when looking at each
single decision problem individually. This follows the logic of subgame-perfect equilib-
ria (SPE) (Stahl and Haruvy, 2008). Looking at different possible payment mechanisms
Sherstyuk et al. (2011) conclude that only the payment for one randomly-selected choice
satisfies this condition without any further assumptions.
At least, similar approaches have been pursued in previous work on collective decision-
making (cf. Sec. 4.1.2). Here, S&K (2010) paid for 3 randomly selected decisions out of
20 rounds. In the experiment conducted by Charness and Rabin (2002) subjects made
between two and eight choices and were paid for one or two of their choices, which
were selected at random. Thus, it was reasonable for my experiment to use an RRPM
and to pay for a subset of all decisions made. This left open the question of how many
rounds the subset should consist. In the end, the total earnings were determined by
three randomly selected rounds. The points earned in these rounds were paid in cash
with an exchange rate of 5 Points to € 1. Additionally, all participants received a “show-
up fee” of € 4, which was also paid to all over-recruited subjects (following Morton and
Williams, 2010, Sec. 10.2.3, p. 282).217 The motivation to include the fee was mainly that
it ensured that all participants received at least some earnings. Also, current research by
Azar (2010, p. 24) suggested that such a fixed payment “does not affect the magnitude
of the pay-for-performance component”.218 Thus, a fixed show-up fee did not diminish
the induced incentives. I chose the exchange rate (and show-up fee) in a way that the
average compensation per subject and hour should correspond to the average hourly
wage of a scientific assistant.219 All these payoff regulations were common knowledge to
the participants and stated at the beginning of each session.
The decision to pay the earnings from three random rounds, instead of just one, was
made in response to the subjects’ reactions after the pilot session. The payoff tables of
my experiment displayed a wide range of possible payoff distributions (Sec. A.8). When
216 Lee (2008) directly compared both mechanisms and found evidence that the payment mechanism influ-
enced subject’s choices. With APM he detected wealth effects which were not observed using RRPM.
217 For every session with 18 subjects I invited originally 21 subjects. Thanks to this reserve only one sched-
uled session could not take place.

218 Azar (2010, p. 1) defined “relative thinking” as to consider relative differences although only absolute
differences are relevant. Using laboratory experiments he determined that this kind of thinking is limited
in scope (e.g., when people compare prices of goods) and does not apply when considering payments
for task performances.
219 This followed the AWI (2013) rules and conditions of participating in experiments and constitutes a widely
used standard procedure in conducting an experiment (e.g., CESS, 2012).
105
4 The Experiment
they were paid for one round, the subjects complained about the “uneven” or “unfair”
distribution of points and the “lack of opportunities” on a good performance. To them
it seemed that the draw of the one payment round was the only important factor and
the act of voting was more just an accessory. This can hardly account as an incentive-
compatible payment mechanism for a voting experiment. The payment of three rounds
was chosen to ensure a high motivation of the subjects during the ballots.
This RRPM or “random multiple decision selection mechanisms” (RMDSM, Arzieli et al.,
2012, p. 20) presupposes the additional assumption that “no complementarities exist be-
tween the different rounds” (Arzieli et al., 2012, p. 20). In my experimental context this
means that, for example, a subject who prefers alternative { A1} to alternative { B2} in
round 3 ( R3 : A13 ≽ B23 ) and alternative { A3} to alternative C3 in round 5 ( R5 : A35 ≽
C35 ), will prefer { A1} in round 3 and { A3} in round 5 ( A13 ; A35 ) to (i) { B2} in round 3
and { A3} in round 5 ( B23 ; A35 ) and (ii) { A1} in round 3 and {C3} in round 5 ( A13 ; C35 )
and (iii) to { B2} in round 3 and {C3} in round 5 ( B23 ; C3). Thus, preferences had to
be independent of the existence and results of other rounds. This had to account to all
possible combination of rounds and preferred alternatives.
4.2.9 The subjects
The subjects were recruited using the ORSEE software out of a subject pool of the lab-
oratory run by the Department of Economics at the University of Heidelberg. The pool
contained about 1,300 people. Nearly all of them were either students at the Universities
of Heidelberg or Mannheim. The subject pool is advertised by flyer distribution as well
as information sessions in first year bachelor classes. Those sessions inform about the ex-
perimental laboratory and the financial possibilities. Subjects who are interested register
themselves online in the ORSEE data base220 . Of course, this leaves the subject sample
with the same kind of (expected) selection bias most other pools have. Furthermore, the
self-registration of the subjects corresponds perfectly to the case of a “volunteer”. Rosen-
thal and Rosnow (2009, Chap. 3) showed that behavioral research draws their samples in
general from such populations of volunteers. Much more important than just their self-
motivation, those individuals might differ from those not finding their way into research.
However, current research suggests that experimental subjects are an appropriate subject
pool. They are not at all significantly different from the general population of which they
were drawn (Branas-Garza et al., 2012). Thus, my findings for the self-selected student
sample are generalizable to the student population; but can the results be generalized
beyond? Do, e.g., professionals behave differently than students? Here, Frechette (2011)
reviewed experiments that included both types of subjects and found, on a general level,
that results of both subject pools lead to similar conclusions.221
220 Registration page: http://147.142.190.243/orsee/public.
221 Frechette (2011, p. 1) defined professionals “as people working in an industry where the game under study
is thought to be relevant.”
106
Cleave et al. (2010) investigated if social and risk preferences of participants in laboratory
experiments represent the preferences of the population from which they are recruited.
Using 1,173 students they found that the preferences of the participants are not signifi-
cantly different from the preferences of the population they were recruited from. It seems
that the statement of selection bias is not always valid. Of course, Cleave et al. (2010, p. 1)
limited their conclusions as they “fail to find selection bias based on social and risk pref-
erences.” This tells us nothing about other characteristics. To exclude a selection bias the
specific patterns of the research question and the sample have to be considered.
Social preferences in general have received considerable attention among economists in

recent years. Falk et al. (2011) pointed out that this research was nearly exclusively fo-
cused on student samples. The authors conducted two studies to find out if this distorted
the results. They found “that self-selection does not significantly bias the social prefer-
ences measured in the laboratory” (Falk et al., 2011, p. 13) and “that student participants
and non-student subjects show very similar behavioral patterns” (ibid.). In fact, non-
students showed significantly more social behavior. The authors suggested therefore
that results from student samples might be seen as a lower bound for the importance
of pro-social behavior. Therefore, I do not regard the high percentage of students in the
subject pool as a problem when it comes to social preferences.
These findings are relatively new and the broad use of student samples was due to prac-
tical considerations. Druckman and Kam (2009) offered an excellent discussion of this
“convenience sampling”222 . In the lab this refers to the aspects of proximity, availability
and affordability. Henrich et al. (2010) conducted a meta-analysis on numerous psycho-
logical studies and found that 96% of all test subjects were from western industrialized
countries, which account only for 12% of the world population.223 The authors referred
to this group as “WEIRD” which stands for “Western, Educated, Industrialized, Rich
and Democratic”. Such divergence is a problem if the group of subjects behaves differ-
ently than the actual target population of the experimental research. Druckman and Kam
(2009) argued that this depends on the underlying data generating process. If “the treat-
ment effect is the same across populations, the nature of a particular sample is largely
irrelevant for establishing that effect“ (Druckman and Kam, 2009, p. 12).224
Following Levitt and List (2007), I argue that whether behavior inside the laboratory is
a good indicator of behavior outside the laboratory depends on the nature of the deci-
sion environment. The authors emphasized in particular the extent of scrutiny by others.
Therefore, laboratory experiments may “prove to be better suited for naturally-occurring
222 “Convenience sampling is a non-probability sampling technique where subjects are selected because of
their convenient accessibility and proximity to the researcher” (Castillo, 2009).
223 A second finding in Henrich et al. (2010) was that 67% of the subjects were themselves students of psy-
chology at American universities.

224 Druckman and Kam (2009, p. 12) clarified further that “if the underlying data generating process is char-
acterized by a homogeneous treatment effect [...], then any convenience sample should produce an unbi-
ased estimate of that single treatment effect, and, thus, the results from any convenience sample should
be easily generalizable to any other group of individuals.”
107
4 The Experiment
settings in which there is a high degree of scrutiny of actions (e.g., employer-employee

relationships, family interactions), or an emphasis on process (e.g., politics, judicial pro-
ceedings)” (Levitt and List, 2006, p. 4). Politics is closely monitored by the media225 and
the public considers its structure and process very thoroughly (e.g., McCubbins et al.,
1989; Lane and Ersson, 2000).226 Thus, out of the variety of real-world circumstances
the decision-making in politics may be one of the most suited processes for laboratory
scrutiny. The emphasis on process corresponds also well with my treatment, the modifi-
cation of the decision-making process for a group of individuals.227
To be honest, it would have been nearly impossible to bypass the problems of self-
selection or convenience sampling. Either way would have called for a considerable
amount of money and time. In Sec. 9.2 I discuss the possibility of further research to in-
clude online experiments or “experimental turks”228 (Rand, 2012) into the subject pool.
This would broaden the pool enormously, but it would rather not solve the potential
problem of selection bias.
4.3 Implications for the empirical analysis
Sec. 4.2 provides a comprehensive overview of my experimental design. It becomes clear

that it was no ordinary lab experiment as it followed the concept of “randomization
within constraints” (Martin, 2008, p. 30ff). More precisely, I established some constraints
(e.g., up to 18 subjects and two procedures per session in a fixed period of time) and
made random assignments within these constraints (e.g., random chosen payoff table,
group composition and number of rounds). The randomization enabled generalizability
and excluded confounding factors from the analysis (Bernauer et al., 2009, p. 91) while
the control secured an unbiased database.229
The augmenting features of my design allowed me to look into questions so far not in-
vestigated by other experimental studies. I compare the effect of several decision-making
structures and contrast separable and nonseparable decision problems directly. The com-
parison is performed on both the collective and the individual level. Also, I do not limit
225 Druckman and Parkin (2005) argued that the media does not only monitor but influence politics.
226 Sen (1997, p. 745) analyzed maximizing behavior and identified “process significance” as one of the most
relevant aspects in the act of choosing.
227 Overall, Levitt and List (2007, p. 154) examined five factors and their influence on decision-making: ethical
considerations, scrutiny of one’s actions, experimental context, self-selection of participants and stakes of
the game. Looking at their results the authors concluded that “being monitored proves to be the critical
factor influencing behavior in this study” (Levitt and List, 2007, p. 160).
228 Mason and Suri (2012) discussed the use of Amazon’s Mechanical Turk in detail. The central purpose of
their paper is to demonstrate the usefulness of this approach for behavioral research. In short, the online
labor market enables employers to post and workers to choose jobs. Thus, the platform provides a large,
stable, and diverse subject pool. They judged the markets to offer researchers at low cost a fast iteration
between developing theory and executing experiments. This lowers the barrier of entry for researchers.
Yet, there are also problems with respect to monitoring, simultaneous participation, data privacy, etc.
229 In particular the randomized assignment to the groups was essential as it enabled to control for the indi-
vidual characteristics of the subjects (cf. Morton and Williams, 2012, p. 20).
108
myself to one specific constellation, but use several problem specifications (i.e., character-
istics of the payoff tables). Yet, as mentioned already, these features also come with some
drawbacks. The specific experimental design has to be taken into account when analyz-
ing the resulting data. The following sections discuss the implications and corresponding
adjustments of the analysis.
4.3.1 Individual and collective level
While committee voting can be viewed as a standard game for a laboratory experiment,230
the split-up decision mechanism added a new pattern. There is experimental research on
collective decision-making but the overwhelming majority of laboratory work is focused
on individual behavior. My setting enabled the collection of collective and individual
data in the same experiment. Contrasting them follows the idea of Halfpenny and Taylor
(1973) and will bring new insights.
The collective level is important as the decision-making had group-wide (which means
in the realm of politics commonly society-wide) consequences for the allocation and dis-
tribution of wealth. Delegation is, in general, expected to increase efficiency and stability
(Bendor and Meirowitz, 2004). However, it might as well affect effectiveness and equity.
This possible trade-off complicates the assessment of the appropriate decision-making
procedure. The individual level allowed me to look through the noise of a collective
majority decision. Predictions of individual behavior are far more reliable than forecast
based on collective data where a minority of “renegade individuals [...] can upset these
predictions rather dramatically” (Halfpenny and Taylor, 1973, p. 444). As all procedures
used majority rule, not all participants had to agree on one single alternative. The thresh-
2
old for reaching a decision was always a qualified majority of 3 of all votes.231 Studies
which only consider the collective level would miss the votes of the outvoted subjects.
This juxtaposition of collective and individual level had an impact on the empirical anal-
ysis. I had to restrict my individual data to the final ballot in order to also have settled
collective result as counterpart for the computations. As subjects voted in every round
until a collective decision was reached, this concerns an individual which either belongs
to the majority that reached the threshold or one that gets outvoted.
As mentioned before, laboratory experiments facilitate the monitoring of every step of
decision-making (Ordeshook and Winer, 1980, p. 730). Thus, in my experiment I recorded
230 The variety of games used in experimental laboratory research is overwhelming. Frequently used experi-
mental settings include market organizations, individual decision-making, bargaining behavior, auction
markets, coordination games, committee (or group) voting, public good games, dictator games, ultima-
tum games, trust games, etc. (for an overview cf. Plott and Smith, 2008, p. xii). This compound can be
used to investigate all types of set-ups like games which are cooperative or non-cooperative, iterated or
one-shot, finitely or infinitely repeated, symmetric or asymmetric, zero-sum (i.e., constant-sum) or non-
zero-sum, simultaneous or sequential, played with perfect or imperfect information, played by a single
player or by multiple players, etc.
231 This was the lowest common denominator of all procedures: four out of six votes under POL and two out
of three votes in both delegation procedures.
109
4 The Experiment
all individual votes which formed the collective decisions as well as all ballots which did
not lead to a collective agreement and forced a new vote. To investigate the unification
process is an exciting research question in itself. If and how players use their votes as a
signal? What beliefs about other players do they have and how are those updated? Fol-
lowing this idea, Sec. 9.2.3 discusses in detail the possibilities of communication between
the participants of laboratory experiments. I particularly consider the issues of recording
and evaluation. However, this is material and subject for future research. The current
study focuses on the last ballot and its collective as well as individual decisions.
4.3.2 Statistical independence of observations
Every experimental session lasted for multiple experimental rounds. In between the
rounds the subjects were randomly reshuffled between groups which had to reach a de-
cision collectively. Even though reputation building in the course of the rounds of the
experiment was not possible (cf. Sec. 4.2.2), the single rounds of a session are not statis-
tically independent from each other. Measuring a subject repeatedly (i.e., over the sin-
gle rounds) leads to a non-independence of observations, the so called “session-effect”
(Frechette, 2012).
Frechette (2012, p. 485) nicely illustrated this effect with the analogy of participants as
“multiple members of a family [where] observations from siblings might exhibit more
correlation than those from individuals across households.” More precisely, the “session-
effect problem is defined as a within session correlation in the variable of interest (or the
residual) once the relevant factors are controlled for” (Frechette, 2012, p. 485). The de-
pendence of observations has to be taken into account, regardless of whether the analysis
uses parametric or nonparametric tests (Siegel, 2001).
Statistically independent observations can only be obtained at the session level. This
leaves two possible solutions to investigate the results at both the collective and the indi-
vidual level. Firstly, regression analysis enables the use of pooled data from all sessions
of a given treatment by considering clustering at the subject or session level. Secondly,
(non)parametric tests can be based not on single rounds but on session averages per
treatment of the variables in question;232 even though this has the disadvantage that it
considerably reduces the number of observations (i.e., from the number of rounds played
to the number of sessions).233
4.3.3 Rather qualitative than quantitative findings
It is always difficult to link experimental findings to the real world. Morton and Williams
(2010, p. 196) concluded that “the proof of external validity is always empirical”. Thus,
232 Frechette (2012) discussed in detail the sources and implications of session-effects. In particular, he also
pointed out that using session averages may not solve the dependence problem.
233 Cf. Vanberg (2008) for an application of the two approaches.
110
further experiments with variations of, e.g., the target population, subject recruiting and
experimental method are essential to verify the insights found (cf. Shadish et al., 2002,
p. 21). For example, in the case of social preferences the work of Carpenter and Seki
(2005) supported the external validity with a field experiment. However, the authors
also admitted that the exact laboratory measurements seem to have less in common with
the data gathered in “daily work lives” (Carpenter and Seki, 2005, p. 20).
Levitt and List (2006, 2007) strongly emphasized the orientation of laboratory experi-
ments on qualitative insights. Taking into account the findings of neighboring disciplines
they pointed out that “the wealth of psychological literature suggests that there is only
weak evidence of cross-situational consistency of behavior” (Levitt and List, 2007, p. 160).
This insight is strongly linked to the aim of the experiment; “if the role of experiments
shifts from testing theories to motivating the development of new theories [this] has the
danger of creating its own world” (Schram, 2005, p. 236) which prohibits the generaliz-
ability of findings.
Quite apart from reality, already changing experimental environments resulted in vary-
ing behavior and different parameter measurements. Charness et al. (2007a) conducted
“lost-wallet game” experiments in the laboratory and through the internet. The differ-
ence between the settings is that “the internet methodology increases the social distance
to a high degree” (Charness et al., 2007a, p. 101). In the laboratory the participants ob-
served who else took part in the experiment, i.e., they recognized that all co-players are
students234 . But in experiments conducted through the web the other participants could
be anybody. The authors found that differences in behavior are linked to the variation of
the social distance.235
My analysis and conclusions were therefore not interested in “quantitative magnitudes”

(Levitt and List, 2006, p. 5), i.e., reporting exact estimates for behavioral parameters.
Rather, my goal was to identify and explain “qualitative findings” (Levitt and List, 2006,
p. 5), i.e., general patterns of behavior. I focused on differences between my treatments
which were all accomplished in the laboratory. Thus, not the absolute but relative values
are decisive, as those are much more credible and generalizable (Levitt and List, 2007).
4.3.4 Responsibility for the results
SIM was a special case when it comes to the question of responsibility for the collective
outcome. Here, no specific alternative was selected by any subject; a unique feature of
this procedure. Instead, the two delegations each chose a set whose intersection then
234 For a study on the effect of performing in an experiment together with classmates (though not at the
university, but in school) cf. Belot and Van De Ven (2009).
235 Research on social distance represents a wide field. Some contributions referred simply to the impact of
known family names of co-players (Charness and Gneezy, 2008). Others looked worldwide for compar-
isons. For example, Ruffle and Sosis (2006) compared Israeli kibbutz members and city residents while
Bouckaert and Dhaene (2004) focused on inter-ethnic aspects between Belgian and Turkish businessmen.
111
4 The Experiment
determined the final result. No delegation was informed about the vote of the other
delegation until the own collective decision was reached.236
During the experiment always an equal number of column and row delegations voted
simultaneously. Their belonging to a corresponding delegation was set in advance by the
matching algorithm. I used the resulting outcome of the intersection to determine the
earned points of the subjects for that round. Yet, the correspondence of two delegations
was set by chance. For example, when a column delegation voted for set { A} and the
matching procedure assigned a row delegation that voted for set {3}, then alternative
{ A3} was the collective outcome for the whole group, i.e., both delegations. But when
the column delegation would have been matched with a row delegation that voted for
set {1} the result of the voting procedure would be alternative { A1}. Thus, the empirical
analysis had to take the matching procedure into account.
I implemented this by using for each alternative not its observed frequency of occurrence
but its conditional outcome probability. This combined the choices of each column and
each row delegation that voted on a payoff table in any of the 13 experimental sessions.
Thus, I calculated for both delegations over all rounds under SIM for each payoff table the
probabilities for choosing one of the three sets. For example, the probability of alternative
{ B2} to be the selected as final outcome is the probability of set { B} being selected by the
column delegation times the probability of set {2} being the outcome selected by the row
delegation.237 When I looked into the decision made under the different procedures I
therefore used this probabilistic result for SIM. This means, e.g., that the points obtained
through a group from a payoff table correspond to the respective selection probability
weighted average over all alternatives of a payoff table.
4.3.5 Summary of key characteristics
This section summarizes the discussed key characteristics of the experimental design and
their implications for the empirical analysis. Tab. 4.3 lists the characteristics one by one
and states the appropriate empirical response.
The empirical analysis in the next two chapters follows these requirements. First, I take a
look at the collective level and the influence of institutional rules on collective decisions in
Chap. 5. Then, I alter the focus of the analysis and look for the determinants of individual
choices in Chap. 6. Previously, Sec. 4.4 provides a descriptive overview of the collected
observations as starter.
236 Of course, subjects may have inferred what the other subjects were about to vote, but the actual result of
the other group was unknown until a delegation’s own decision was finalized. Chap. 6 looks more into
the matter of sophisticated behavior, i.e., anticipatory voting.
237 For example, assume that a payoff table was used 13 times under SIM. Of the 13 column delegations seven
choose set{ B} and of the 13 row delegations four choose set{2}. Hence, the probability of observing
7 4 28
alternative{ B2} as final outcome equals 13 × 13 = 169 = 16.6%.
112
4.4 Descriptive information
Table 4.3: Design characteristics and their impact on the empirical analysis
EXPLANATORY NOTE
The table lists specific details of my experimental design and how the empirical data analysis took them into account. All
characteristics have been discussed in earlier sections of this chapter which are indicated.
DESIGN CHARACTERISTIC IMPLICATION FOR THE EMPIRICAL DATA ANALYSIS

multiple payoff tables and their random assignment requires a normalization of the payoff tables
(Sec. 4.2.4)
multiple rounds (Sec. 4.2.5) requires a trend analysis of the results
split of voting competence (Sec. 4.2.6) requires a distinction between delegation and group level
voting sequence (Sec. 4.2.6) requires a distinction between first stage and second stage
under sequential delegation
two decision rules per session (Sec. 4.2.7) requires a within-subject and between-subject comparison
contrasting individual and collective level (Sec. 4.3.1) restricts the individual data to the last ballot in each round
no statistical independence across rounds (Sec. 4.3.2) restricts (non)parametric tests to session averages
social distance within the laboratory (Sec. 4.3.3) restricts the findings to relative differences among the
various procedures
simultaneous delegation used the intersection of sets requires the usage of conditional outcome probabilities
as collective outcome (Sec. 4.3.4) under simultaneous delegation
Overall, I conducted 13 experimental sessions between May 2011 and February 2012 with
168 participants. Subjects were 57% female and on average 22.8 years old. Nearly 88%
were 25 or younger. Of all subjects 92% were students, of which 46% studied economics.
The mean semester of all student subjects was 4.5 (SD 3.3). On average, subjects had
participated in 4.7 (SD 3.5) experiments before.238 The subjects played between 5 and 16
rounds which took between 1:10h and 2:45h.239 Average payoffs for a subject equaled
8.97€/h (SD 1.76€/h). Including the expenditures for a pilot session in December 2010
the expenses for the experiment sum up to € 3.422.
The distribution of voting rules in my experimental sessions is shown in Tab. 4.4. In every
session I applied two different rules. In 5 of the 13 sessions I used POL and SIM as well
as POL and SEQ. In the remaining 3 sessions the subjects played under SIM and SEQ.
Over all sessions I obtained data on 221 group decision, of them I collected 76 under
POL, 85 under SIM and 60 under SEQ. This represents the number of collective results,
but I have distinctly more individual data. Every group decision consists of six individ-
ual votes which leaves me with 1326 individual votes that constituted these collective
238 The demographic information was obtained as part of a post-experiment survey. I discuss its results in
detail in Chap. 7.
239 This time period extends from the entry of the first participant into the laboratory to the payment of the
last participant. It includes in addition to the actual voting experiment the assignment of cubicals, the
handing out of instructions, answering questions, etc. The variation resulted from several reasons such
as, e.g., how long it took the participants to answer the post-experiment survey or how many comments
were made by the subjects during the payout phase (cf. Sec. 7.1).
113
4 The Experiment
results.240 As discussed before (cf. Sec. 4.3.2), these are not statistically independent ob-
servations which can only be obtained at the session level. Here, the mean values were
calculated based on average on 8.50 (SD 3.20) group decisions.241
Table 4.4: Data structure at the session level

EXPLANATORY NOTE
The table depicts the voting rules used on the individual sessions of the experiment. Under each rule the session averages
of the variables in question can be calculated to secure statistically independent observations.
NUMBER OF EXPERIMENTAL SESSION
1 2 3 4 5 6 7 8 9 10 11 12 13 Σ
pooling X X X X X X X X X X 10
PROCEDURE simultaneous delegation X X X X X X X X 8
sequential delegation X X X X X X X X 8
Next, I look more thoroughly into the data structure of the observations. Tab. A.7 shows
the frequency of use of each of the payoff tables. The random assignment rule for the
payoff tables resulted in an uneven distribution of the tables across the procedures. More
interesting than the usage of single tables under each voting rule is the distribution of the
characteristics of the payoff tables, i.e., if the tables are (non-)constant-sum games and if
they contain a core alternative (cf. Sec. 4.2.4). Yet, for the discussion of the core concept
under all three procedures additional derivations are necessary. I explain the required
considerations and describe its application in detail in Sec. 5.2. For now, I restrict the
overview to the criterion of constant-sum or non-constant-sum game and, what is also
important, if a procedure was used as first or second decision rule within an experimental
session (cf. Sec. 4.2.7).
Tab. 4.5 gives an overview of the corresponding data structure for between-subject and
within-subject analysis. The between-subject analysis used the data gathered under the
first decision rule in each session; i.e., 28 POL, 60 SIM and 38 SEQ decisions.
Table 4.5: Data structure according to the order of decision rules

EXPLANATORY NOTE
The table depicts the number of observations in terms of collective decisions made in the experiment. The observations
are divided according to procedure and first or second decision rule of a session.
DECISION RULE
first second Σ
pooling 28 48 76
simultaneous delegation 60 25 85
PROCEDURE
sequential delegation 38 22 60
Σ 126 95 221
For the within-subject analysis I could also use the observations gathered under the sec-
ond decision rule of a session. Here, my data comprised 48 POL, 25 SIM and 22 SEQ
results. The juxtaposition of POL with SIM or SEQ enabled me to investigate how sub-
jects react to the divided decision-making (competencies or sequence). When conducting
POL after SIM, the divided competencies vanished. When POL followed after SEQ, both
240 Inaddition, when a ballot did not reach the necessary majority another ballot was held. Thus, I recorded
overall 4695 individual votes.
241 Looking at each voting rule separately, the session level data comprised on average 7.60 (SD 3.88) POL,
10.63 (SD 2.69) SIM and 7.50 (SD 0.87) SEQ collective results.
114
divided competencies and sequential voting were removed. Thus, of 48 POL results,
24 took place after SIM and 24 after SEQ. Reversing the sequence, I conducted 25 SIM
rounds after POL to investigate what patterns of voting behavior change when decision
competencies are split up. Furthermore, 22 SEQ rounds were run after SIM to analyze
what happens if I keep divided competencies and split up the decision sequence.
As explained in Sec. 4.2.4, the experiment contained constant-sum as well as non-constant-

sum payoff tables.242 Tab. 4.6 displays the frequency of the experimental observations
separated by this criterion (following Zizzo and Tan, 2009). The distribution of constant-
sum and non-constant-sum tables shows nearly equal frequencies for the different pro-
cedures. 27 of 76 POL (35.5%), 35 of 85 SIM (41.2%) and 22 of 60 SEQ (36.6%) results
emerged from rounds which used constant-sum tables.
Table 4.6: Data structure according to constant-sum and non-constant-sum table

EXPLANATORY NOTE
The table depicts the number of observations in terms of collective decisions made in the experiment. The observations
are divided according to procedure and constant-sum or non-constant-sum payoff table.
CONSTANT- SUM TABLE
yes no Σ
pooling 27 49 76
PROCEDURE
Σ 84 137 221
Of course, these different payoff table characteristics can be combined. Tab. 4.7 separates
the number of observations according to whether it is the first or second decision rule
of the session and whether a constant-sum and non-constant-sum table is used. Taking
both aspects together, e.g., 12 POL, 26 SIM and 12 SEQ results were obtained from a
constant-sum table used under the first decision rule. The fluctuation in the number of
observations resulted from the random assignment of payoff tables (cf. Sec. A.11).
Table 4.7: Data structure according to the order of decision rules and constant-sum or
non-constant-sum table
EXPLANATORY NOTE
The table depicts the number of observations in terms of collective decisions made in the experiment. The observations are
divided according to procedure, first or second decision rule of a session and constant-sum or non-constant-sum payoff
table.
CONSTANT- SUM TABLE
yes no
Σ
DECISION RULE DECISION RULE
first second Σ first second Σ
pooling 12 15 27 16 33 49 76
simultaneous delegation 26 9 35 34 16 50 85
PROCEDURE
sequential delegation 12 10 22 26 12 38 60
Σ 50 34 84 76 61 137 221
To secure a valid empirical analysis I list in the following, always explicitly, the respective
number of observations. This refers in Chap. 5 to the amount of collective decisions and
in Chap. 6 to the number of individual votes.
242 In
a constant-sum tables the overall sum of each alternative is the same; in non-constant-sum tables the
overall sum varies between alternatives (cf. Sec. 4.2.4).
115
4 The Experiment
116
5 The influence of institutional rules on
collective decisions
In this chapter I take a look at the decisions made in the experiment at the collective
level. This addresses the first part of my second key question, which is whether and, if
so, how nonseparability affects collective behavior. As I explore the aggregated level, my
focus is on the characteristics of the finally selected alternative. I am interested in how
an alternative must be “composed” to allow for a joint agreement. Do the groups simply
follow the theoretical baseline prediction or is there more to it? If so, what else influences
the decision-making and how do votes differ among the decision procedures? The an-
swers to these questions provide a first step towards an understanding of the influence
nonseparability exerts on (collective) behavior.
The subsequent sections are structured as follows. I proceed cautiously and first evaluate
the reliability of my experimental design. Therefore, I assess my results in comparison
to previous research. More specifically, I compare in Sec. 5.1 the most conventional part
of my design (the pooled decision-making) to the previous work of S&K (2010) on self-
interest and fairness in majority decision-making. Only then do I look at all the data of
my voting experiment. For the analysis, I establish a reliable benchmark in two steps.
First, I derive in Sec. 5.2 the theoretical baseline solution by extending the core concept to
all decision procedures. Second, I discuss in Sec. 5.3 the issue of evaluating a collective
outcome in general terms, i.e., how one can measure different criteria such as equality,
justice, inclusion, etc. This is related to the elementary question of public policy formu-
lated by Buchanan and Tullock (1962) for the trade-off between efficiency and effective-
ness in collective decision-making. I derive statistical measure for both categories and
formulate associated hypotheses. Using these benchmarks, I examine the influence of
nonseparability on the collective results in Sec. 5.4. Here, I start with an assessment of
the explanatory power of the baseline prediction.243 This allows me to explain some but
not all of the deviations observed. The comparison continues on the basis of the empiri-
cal computation of the efficiency-effectiveness metrics. Finally, I summarize the findings
of this chapter in Sec. 5.5.
243 This section illustrates in detail how I structure my empirical evaluation of the collective results. All
statistical methods, analysis and forms of presentation are explained. Later sections will always refer to
these descriptions.
117
5 The influence of institutional rules on collective decisions
5.1 Reliability of the experimental design
I start by looking into the reliability244 of the experimental design. Therefore, I compare
my findings with previous laboratory work on collective decision-making. As I adapt the
general experimental environment presented by S&K (2010), I contrast my results with
their findings.245 The identical aspects between the designs are: a group of individuals
decides on distributing points among them; majority rule; subjects interacted over mul-
tiple rounds; groups were randomly reshuffled after every round; ballots were repeated
until the majority threshold was reached and all previous votes of a round were common
knowledge.
However, there are also some aspects which differ significantly. S&K (2010) used five-
person committees which voted on eight alternatives while I used six-person committees
and offered nine alternatives. These deviations are due to my paramount research goal.
The number of six players enabled me to split-up the overall group into two delegations
of three players. Nine alternatives enabled me to disperse them into three subsets con-
taining three alternatives each. These modifications of the design affected the probability
of reaching a collective agreement. This probability P of agreeing at all in a one-shot
majority vote (if players cast their vote randomly) was calculated as
 majority treshold
1
P= ∗ n altervatives ∗ nmin winning coalitions (5.1)
n altervatives
Thus, it depends foremost on the number of alternatives n alternatives . The chance that an
1
actor chooses an alternative is just n alternatives . The majority treshold determines the nec-
essary number of actors who have to vote for the same alternative. This can be any
of the feasible alternatives and the collective agreement can be reached by any of the
possible minimum winning coalitions whose number was given by nmin winning coalitions =
   
n actors 5
. Thus, an absolute majority vote between five actors has =
majority treshold 3
 
6
10 and a ballot of six actors has = 15 minimum winning coalitions.
4
For five players voting on eight alternatives this results in a probability of 15.63% for
agreeing in a one-shot majority vote. It drops by a factor of 7.59 to 2.06% for six players
holding a ballot on nine alternatives. Of course, in both experiments ballots were held
repeatedly until an agreement was reached. But the drop in probability documents that
the task of reaching a compromise was more difficult in my experimental design.
244 Ifollow the definition of Carmines and Zeller (1979, p. 11), according to which reliability is understood as
“the extent to which an experiment, test, or any measuring procedure yields the same results on repeated
trials”.
245 This follows the idea of Guttman (1945) for test-retest reliability which emphasizes that in “dealing with
empirical data in any field, the question should be raised: if the experiment were to be repeated, how
much variation would there be in the results?” (Guttman, 1945, p. 256)
118
5.1 Reliability of the experimental design
Another difference between the experiments is that all payoff tables of S&K (2010) con-
tained a core alternative246 . In contrast, I explicitly designed some of my tables without
such equilibrium. This is related to the different research goals of the two experiments.
S&K (2010) focused on the performance of the core in majority decision-making. More
precisely, they were interested in its robustness against social preferences. I also looked
into this question but did not stop with it. In my analysis the measurement of core sta-
bility was not the main research interest. Therefore, I used different payoff tables, some
with and some without a core. I also altered other characteristic of the tables (cf. Sec. 4.2.4)
which eventually resulted in a variety of decision problems.
As already mentioned, S&K (2010) were interested in the relevance of social preferences.
They specified a two-way error components model for the selection probability of the
core; more precisely, a multilevel mixed-effects logistic regression (S&K, 2010, p.675). The
error component of their model is clustered within each session and within each round of
the experiment. In their most simple model the independent variables comprise only the
stability of the core alternative. This stability is measured by a monotonic transformation
of an Equity, Reciprocity and Competition (ERC) utility function.
The ERC model was introduced by Bolton and Ockenfels (2000). It includes social pref-
erences in an individual’s utility function. Specifically, it implies that a player’s utility
decreases the more the player’s own payoff deviates from the mean payoff in their group.
The ERC utility function sets a possible trade-off between self-interest and a concern for
equity as shown in Eqn. 5.2.
 2
1
Uij (yij , σij ) = ai yij − bi σij − (5.2)
n
The utility U of individual i as part of a group of n at alternative j is determined by two

components: their absolute payoff yij and their relative payoff. This relative payoff is
yij
given by the difference of their share σij = ∑nj=1 yij
to the average (i.e., globally equal)
1 ai
share n. The ratio of bi indicates the relative importance player i places on their pay-
ment compared to their regard for equality. The theory expects that ai and bi > 0 which
precludes all types of players who prefer inequality, both altruism and envy. Previous
applications of the ERC model to experimental data supported these expectations (e.g.,
Bolton and Ockenfels, 2000; Fehr and Schmidt, 1999).
S&K (2010) incorporated an ordinal stability ranking into the regression analysis in the
form of dummy variables, an approach first proposed by Walter et al. (1987). The dum-
mies resembled the log odds ratio between the levels of stability. A coefficient corre-
sponded to the change in the probabilities of the selection of the core if the stability de-
creased to the next lower level. The ranking indicated for every payoff table for which
246 S&K (2010, p. 671) stated that all their tables “contain a unique core alternative under the assumption of
rationally acting and egoistically motivated committee members.”
119
values of b the ERC utility function predicted a change of the core alternative. The au-
thors included seven dummies dividing the whole value range of b into eight intervals
(from b = 217 up to b = 2000). In their regression they found that, as expected, all signif-
icant coefficients of the dummies had negative signs and “infer that the probability of the
selection of the core increases monotonically with the core’s stability” (S&K, 2010, p. 677).
I replicated the analysis of S&K (2010) to ensure the general reliability of my experimental
design. As the first step, I re-estimated their regression model using all 480 observation
of their experiment.247 I modeled the two-way error components as random coefficients
for the rounds and for the sessions of the experiment. Tab. 5.1 shows in detail the original
results as well as my replication. My calculations mirrored theirs almost entirely.
Table 5.1: Two-way error components model

EXPLANATORY NOTE
The table shows the results of the replication of the S&K (2010) two-way error components model. Statistically significant
(two-tailed) at the 0.1 level * and at the 0.01 level ***, SE in parentheses.
N = 480 Dependent Variable: SELECTION OF THE CORE

Values reported by Own calculations
S&K (2010, p. 676, table 3)
FIXED PART
constant 1.89*** 1.88***
(0.38) (0.37)
Stability217 -1.52*** -1.49***
(0.43) (0.42)
Stability403 -1.62*** -1.63***
(0.41) (0.41)
Stability625 0.55 0.55
(0.48) (0.48)
Stability841 -1.23* -1.23*
(0.68) (0.68)
Stability999 0.55 0.58
(0.72) (0.71)
Stability1512 -1.55 -1.58
(1.14) (1.13)
Stability2000 1.52 1.51
(1.11) (1.11)
RANDOM PART
Residual SD between sessions 0.25 1.019
Residual SD between rounds 0.73 0.22
Log likelihood -218.61 -218.86
Next, I applied the two-way error components model for the selection probability of the
core to my own results. Due to the different research goals, not all data obtained meets
the S&K (2010) criterion for a deterministic equilibrium. Thus, not every procedure or
payoff table of my experimental sessions was suitable for a comparison and the number
of observations for the regression was significantly smaller. More specifically, I used all
collective decisions made on payoff tables with a core alternative under POL; this left me
with 58 observations. I accounted for the lower amount of observations by implement-
ing only two dummy variables, which resulted in three intervals of ERC stability: low,
medium and high.248 Payoff tables with low stability deviated from the core for b ≤ 600,
247 I am very grateful to Jan Sauermann (University of Cologne) for providing me with the data.
248 I also considered models with more than three intervals. Yet, when specifying more than two dummy
120
5.2 Theoretical benchmark predictions
medium stability included tables which deviate for 600 < b ≤ 1200 and high stability
implied an unchanged core for b > 1200. This approach was not different from S&K
(2010) but less exact. Tab. 5.2 presents the results.
Table 5.2: Selection of the core

EXPLANATORY NOTE
The table shows the estimates of the two-way error components model for the selection of the core. Statistically significant
(two-tailed) at the 0.1 level *, SE in parentheses.
N = 58 Dependent Variable: SELECTION OF THE CORE
FIXED PART
constant 1.099
(0.816)
Stability600 -0.073
(0.603)
Stability1200 -1.645*
(1.11)
RANDOM PART
Residual SD between sessions 0.342
Residual SD between rounds 0.819
Log likelihood -37.162
Due to the small number of observations the results should be considered with caution.
Nevertheless, they were in line with the findings of S&K (2010). A lower stability led to
a lower probability of selecting the core. This correspondence of my most conventional
treatment with previous research indicates the reliability of my experimental design.
In this section I establish the theoretical baseline for the subsequent empirical analysis
of the collective results. Sec. 4.2.4 argues that the appropriate baseline equilibrium is the
concept of the core as in S&K (2010). A core is characterized as a unique cooperative solu-
tion under the assumption of rationally acting individuals.249 It follows the deterministic
definition of being an alternative for which no single player or subgroup has an incentive
to leave the coalition supporting it (Peleg and Sudhoelter, 2003, Chap. 3 and 12). Hence,
the core alternative beats all other alternatives in pair-wise comparison.250
In relation to previous experiments my design introduced complex decision-making pro-
cedures. For example, consider the lower probability of agreeing in a one-shot major-
ity vote in my experiment compared to S&K (2010). As discussed above, the differences
variables, the optimization procedure (STATA’s xtmelogit) omitted the additional variable because of
collinearity. Thus, the small number of observations restricted the regression to two independent vari-
ables.
249 More specifically, the core concept “satisfies individual, coalition, and collective rationality, inasmuch as it
includes only divisions of the payoff such that the players receive at least as much as they could guarantee
for themselves by acting independently, every proper subset of the players receives at least as much as
it could guarantee for itself by acting together, and the totality of players receives at least as much as it
could guarantee for itself by acting collectively as a grand coalition” (Colman, 2003, p. 144).
250 This section investigates the core performance on the collective level. In Sec. 6.3 I introduce and analyze a
probabilistic extension of the core on the individual level.
121
arose because I chose a group of six instead of five subjects as well as nine instead of eight
alternatives. But not only have these larger numbers made it more complex. The splitting
of decision competences and the alignment to sequential voting introduced completely
new dimensions into the game, e.g., a high level of uncertainty (cf. Tab. 4.2).
RATIONALITY OF GROUP DECISIONS
This higher complexity of the decision task made the baseline analysis even more inter-
esting. Did the game-theoretical concept perform well even under these difficult condi-
tions? A distinct argument in favor is the recurrent observation that team decisions are
typically closer to standard game theoretic predictions than individual decisions (Cooper
and Kagel, 2005).251 In their election experiment Forsythe et al. (1996, p. 375) concluded
that, overall, “voters cast votes strategically and in equilibrium consistent manners.” In
accordance, Bornstein and Yaniv (1998, p. 106) found “that groups are more rational than
individuals” and Sutter (2008, p. 3) noted that this applies to “a broad variety of strategic-
and non-strategic tasks”.
An often cited reason for the differences in behavior of groups and individuals is that
interaction within the groups affects the actions taken at the intra-group level (Putnam,
1988). Also, the groups can use the knowledge and skills of several people and thus over-
come individual cognitive shortcomings (Gigerenzer and Gaissmaier, 2011).252 This ar-
gument is strongly advocated by Surowiecki (2004) who traced the inferiority of individ-
ual decisions in various fields back to the aggregation of information within groups.253
Recent contributions investigated whether behavioral differences can be explained by the
fact that motivation differs for individuals and groups. Using laboratory experiments,
Kocher and Sutter (2005) found that groups are less trusting than individuals. This is
in line with explanations of social psychology254 which argue that the discontinuity ef-
fect between individual and collective action (Schopler and Insko, 1992) is based on the
blurred accountability in collectives (cf. Sec. 6.1.4), mutual encouragement of antisocial
actions within a group (Wildschut et al., 2001, 2003), the fear of others (Insko et al., 1990)
and the reluctance to trust interacting groups (Kugler et al., 2007). All these aspects influ-
ence behavior in a way that greed is increased and cooperation diminished. Sauermann
(2012, p. 100) summarized this as “compared to individuals the behavior of groups is
approaching the model of the Homo economicus”.255
251 Cf. Levine and Moreland (1998) for a comprehensive literature overview of the differences of individual
and group behavior.
252 In my experiment subjects could not interact or decide jointly in consultation. Yet, the collective choice
procedure required nevertheless an agreement of individual choices. At least the voting process, i.e. the
repeated ballots within a round, provided the opportunity for mutual coordination.
253 Cf. Demertzis (2009) and Mannes (2009) for current contributions to the undecided debate on the “Wisdom
of the Crowds”.
254 Cf. Cooper and Kagel (2005) for a comprehensive survey of psychology literature on team versus individ-
ual play.
255 In his study Sauermann (2012, p. 100) supplemented the ERC model for collective actors and found that
social preferences might be another explanation for behavioral differences of individuals and groups.
122
5.2.1 Derivation of the credible core
Sec. 4.4 discusses the data structure of the observations according to whether a procedure
resembled the first or second decision rule within a session and if a constant-sum and
non-constant-sum table was used. Yet, so far I have omitted the characteristic if a payoff
table contained an equilibrium.
The “classical” core concept (Gillies, 1959) is only applicable to POL because standard
cooperative game theory studies mostly static one-shot situations (for an overview cf.
Peleg and Sudhoelter, 2003). This did not correspond to SIM and SEQ, which exhibited
a dynamic and successive decision-making.256 Yet, the core concept is not limited to a
specific institutional rule (for a bicameral extension cf. Bottom et al., 2000).257 For reasons
of consistency I aimed for the same theoretical baseline for all four procedures. Thus,
I adjusted the necessary definitions and assumptions of the core to a dynamic game.258
This did not change my intention to apply cooperative game theory or the fundamental
logic behind the core concept. In all procedures the core consisted of the allocation for
which no individual or subgroup within the coalition voting for this alternative can do
better by deserting the coalition (Peleg and Sudhoelter, 2003).259
Research on dynamic or repeated cooperative game theory has been sparse compared to
the work done in non-cooperative settings. Only recently has this field developed some
promising enhancements for a more general and applicable use of cooperative game the-
ory. A recent overview addressing the core and other cooperative game theory concepts
like the bargaining set or coalition formation in repeated, iterated or dynamic settings
can be found in Lehrer and Scarsini (2011).
When looking into cooperative game theory by analyzing international pollution control
strategies, Dockner and Long (1993) pointed out that a cooperative approach may lead
to the first-best solution. But this also requires a high degree of commitment of every
participants “that is not likely to be feasible in practice” (Dockner and Long, 1993, p. 16).
When such commitment cannot be ensured, an equilibrium solution has to rely on indi-
vidual strategies to promote cooperation. The crucial question is whether single players
(or subgroups) have any incentive to deviate from an arrangement. If yes, this constitutes
no stable equilibrium. Only self-enforcing agreements (respectively strategies) can lead
to cooperative equilibria.
256 Already Kramer (1972, p. 171) pointed out that “in general the sophisticated outcome [...] is sensitive to
the voting procedure, or order of voting, adopted”.
257 Bottom et al. (2000) derived a bicameral core by using median lines in two-dimensional policy space.
Testing their predictions empirically in deliberation experiments the authors recognized a clustering of
results around the core areas.
258 My dynamic game deviates from the setting discussed by Long (2010) as the “state of the system” (my
payoff table) did not change over time (within a round). The tables were altered between rounds; i.e., they
stayed the same until a collective decision was reached. So, my experiment did not include a “transition
equation” (Long, 2010, p. 4). This ruled out open-loop Nash equilibrium (OLNE), Markov-perfect Nash
equilibrium (MPNE) or feedback Nash equilibrium (FBNE) applications (cf. Long, 2010, p. 4).
259 A condensed version of this derivation is part of Fleig and Finke (2013, appendix 3).
123
A major concern was if in dynamic games without contractual agreements cooperation

between players is possible. Here, Tolwinski (2003) showed that negotiated agreements
can be enforced by suitably defined strategies. Equilibria have to satisfy the principle
of optimality along the (temporal) equilibrium trajectory. Thus, cooperative game the-
ory can constitute the theoretical baseline as long as the solution concept is implemented
based on self-enforcing individual strategies. These strategies must account for the dy-
namic and successive characteristics of the game; in my setting this referred to the divi-
sion into two delegations as well as the voting sequence.
Next, I had to consider the separated voting of delegations. When studying the core of
a finitely repeated discounted cooperative game, Oviedo (2000) re-defined the repeated
cooperative game as a repeated game where in each round the agents play a cooperative
game. Most importantly, he proved that the core of a repeated cooperative game con-
tains the core of the original cooperative game. I used this idea to account for divided
decision-making competencies under delegation. Accordingly, I treated SIM and SEQ
as a repeated game, where in each stage (decision of column and row delegation) the
participants played a cooperative game.
Kranich et al. (2005) considered the finite horizon of predetermined games and studied
three different core concepts: the classical core, the strong sequential core and the weak
sequential core. These concepts vary in the degree to which they take the temporal struc-
ture of the game into account. In a similar way Becker and Chakrabarti (1995) looked
at infinite horizon capital allocation models. They defined the recursive core depend-
ing on previous decisions up to that period. These approaches followed earlier work
of Bernheim et al. (1987a) and Bernheim and Whinston (1987b) on coalition-proof Nash
equilibria. Lehrer and Scarsini (2011) pursued this idea further and identify the credible
core as the implementation of a subgame-perfect equilibrium under cooperative game
theory. Ray (2007) reached a similar result when incorporating aspects of cooperative
and non-cooperative game theory towards a broader definition of the core. He referred
to the temporal structure of the game as determined by the farsightedness of a coalition
when evaluating the credibility of the core.260
To summarize, an equilibrium concept has to be based on self-enforcing individual strate-
gies. This does not affect or prevent my decision to apply cooperative game theory. A
repeated cooperative game can be defined as a repeated game, where in each round a
cooperative game is played. However, the sequential structure of a game must be taken
into account. As with standard theory on strategic games, all “sequential-move games
require players to consider the future consequences of their current moves before choos-
ing their actions” (Dixit et al., 2009, p. 79). Here, the equilibrium solution is obtained
through backward induction as “rollback equilibrium” (Dixit et al., 2009, p. 79).261 I used
260 Credibility represents a central leitmotiv of both cooperative and non-cooperative game theory (Gul, 1997).
261 This
corresponds to the extension of Shepsle’s (1979) structure-induced equilibria with perfect-foresight
expectations by Denzau and Mackay (1981). Such expectations allow the identification of conditionalities
between sequential played issues.
124
this insight to account for the effects of successive decision-making under delegation.
The stability conditions for the core had to apply throughout the entire game against all
credible deviations. Therefore, the sequence of coalitions was taken into account to iden-
tify a subgame-stable order (Hellman, 2008) which led to the credible core as theoretical
equilibrium prediction.262
SEQUENTIAL DELEGATION PROCEDURE
For SEQ the credible core was identified through a two-stage backwards induction. In
each stage the subjects played a cooperative game. I applied the classical core concept to
ensure that no individual had an incentive to deviate from the solution found.
First, I identified for the second stage whether a core alternative existed in any of the
three sets. If, and only if, all three sets contained a core alternative, this constituted the
hypothetical outcome of the collective decision if the first stage agreed on the correspond-
ing set. The second delegation had no incentive to deviate from the core of a set when
finally confronted with it. This led to a “reduced” payoff table for the first stage. From
the original nine alternatives only three were left (each of the three sets now contained
only the left-over core alternative). If, and only if, this “reduced” decision problem of the
first stage revealed a core alternative, it represented the solution of the dynamic setting,
i.e., the credible core.
For clarification, please imagine a decision under SEQ. First, a delegation decides on the
column and then a second delegation on the row. If the first stage delegation chooses
set { A} the second stage opts for row {3}, which gives the collective outcome of { A3}.
If the first stage delegation would choose set { B}, the second stage would vote for row
{2}, and if the first stage delegation agreed on set {C }, the second stage would settle
for row {3}. This leaves { A3, B2, C3} as possible outcomes. These three alternatives
constitute the “reduced” decision problem for the first stage. If a comparison of these
three alternatives for the first stage members reveals that a core alternative exists, this
constitutes the credible core of the complete decision-making procedure.
SIMULTANEOUS DELEGATION PROCEDURE
Under SIM the two delegations did not vote in a predefined sequence. Thus, nor can the
stages of the solution concept follow a given order. I implemented this indecisiveness
by mirroring the decision-making process. This gave me a second game tree in reversed
order. No tree or one or both trees might have provided predictions for a core. But due to
the level of uncertainty a stable equilibrium occurred if, and only if, it was the same for
both processes, i.e., both sequences predicted the same core. This alternative constituted
then the mutual best response of the delegations to each other; i.e., a subgame-perfect
equilibrium under cooperative game theory.
262 Crawford
and Haller (1990) showed that even under uncertainty the concept of subgame-perfect equilib-
rium holds also for repeated coordination games.
125
For clarification, please imagine a decision in my experiment. Analogous to SEQ, the

first stage decides on the column and the second stage on the row of a payoff table. The
equilibrium result of this process is alternative { B2}. If, and only if, a first stage now
deciding on the row and a second stage choosing the column leads to the same prediction,
then this alternative constitutes the credible core. If the second prediction differs from the
first, then this payoff table does not contain a stable equilibrium under SIM.
Since these descriptions of the equilibrium prediction under SEQ and SIM are somewhat
abstract, in the following I illustrate the core concept across the three different voting
rules with a concrete example. Fig. 5.1 shows a payoff distribution that was actually
used in my experiment in all decision-making procedures (cf. Tab. A.7). The payoff table
represents a non-constant-sum game and contains, under the assumption of rational and
self-interested actors, a unique equilibrium prediction under POL, SEQ and SIM.
Figure 5.1: Example of the determination of the core alternative

EXPLANATORY NOTE
The figure shows payoff table 13 as used in my experiment (cf. Sec. A.8). The equilibrium predictions for this non-constant-
sum table are A1 under POL and C3 under SIM as well as SEQ.
A B C
Player 1 22 9 14
Player 2 16 11 10
Player 3 30 6 14
1
Player 4 17 2 42
Player 5 3 21 9
Player 6 22 17 18
Player 1 5 20 33
Player 2 12 41 1
Player 3 33 7 8
2
Player 4 15 14 1
Player 5 14 22 38
Player 6 19 18 10
Player 1 8 7 20
Player 2 4 26 14
Player 3 19 17 13
3
Player 4 10 12 30
Player 5 42 19 1
Player 6 20 16 32
Under POL { A1} constitutes the core. This alternative beats all other eight alternatives
in a pairwise comparison by majority vote.263
Under SEQ the group of six players is divided into two delegations with players 4, 5
and 6 voting first on the column and players 1, 2 and 3 voting afterwards on the row
of the collective outcome. The credible core is identified through backwards induction.
If the first stage delegation would choose set { A}, the second stage would opt for row
{1} (by the votes of players 1 and 2). If the first stage delegation would choose set { B},
the second stage would vote for row {2} (by the votes of players 1 and 2). If the first
stage delegation would choose set {C }, the second stage would settle for row {3} (by the
263 Alternative{ A1} wins the pairwise comparisons according to the following winning coalitions: against
{ A2} by the votes of players 1, 2, 4 and 6; against { A3} by the votes of players 1, 2, 3, 4 and 6; against
{ B1} by the votes of players 1, 2, 3, 4 and 6; against { B2} by the votes of players 1, 3, 4 and 6; against
{ B3} by the votes of players 1, 3, 4 and 6; against {C1} by the votes of players 1, 2, 3 and 6; against {C2}
by the votes of players 2, 3, 4 and 6; against {C3} by the votes of players 1, 2, 3 and 5.
126
votes of players 1 and 2). This leaves { A1, B2, C3} as possible outcomes which the first
stage delegation has to consider (“reduced” decision problem). With two votes in favor
(players 4 and 6) and one vote against (player 5) then {C3} beats both { A1} and { B2}.
Thus, {C3} constitutes the credible core under SEQ.
Under SIM the reversed order of the two stages is also considered. Thus, the group of
six players is again divided into two delegations but players 1, 2 and 3 vote first on the
row and players 4, 5 and 6 vote afterwards on the column of the collective outcome. If
the first stage delegation would choose set {1}, the second stage would opt for column
{C } (by the votes of players 4 and 6). If the first stage delegation would choose set {2},
the second stage would vote for column { A} (by the votes of players 4 and 6). If the first
stage delegation would choose set {3}, the second stage would settle for column {C } (by
the votes of players 4 and 6). This leaves {C1, A2, C3} as possible outcomes which the
first stage delegation has to consider (again a “reduced” decision problem). Next, with
two to one votes {C3} beats both { A2} and {C1} (players 1 and 2 in favor). Thus, as the
equilibrium predictions for both stage orders match, {C3} constitutes the credible core
under SIM.
5.2.2 Distribution of the equilibrium solution over the payoff tables
The adaptation of the core concept for the two delegation procedures led to the distri-
bution of payoff table characteristics described in Sec. A.9. Tab. A.6 shows that for each
procedure at least four payoff tables existed which corresponded to every possible table
property combination - with and without a core alternative as well as constant-sum and
non-constant-sum game. Due to the differences in determining the (credible) core the ta-
ble assignment to the categories differed between procedures. Above all, this concerned
whether a table contained a core at all. And if so, it also affected the specific equilibrium
within a payoff table between procedures.264
More important than the general structure alone is the actual frequency of use of each ta-
ble. Here, as shown in Sec. A.10, the random assignment rule resulted in, as expected, an
uneven usage of payoff tables. The interaction of these two elements resulted in the data
structure of all collected observations of my experiment. Sec. A.11 discusses the number
of observations with respect to the distribution of the table characteristics. For now, I
limit the analysis to the equilibrium prediction. Thus, Tab. 5.3 displays the frequency of
observations according to the existence of a deterministic core alternative. During each
procedure I investigated payoff tables with and without equilibrium to observe decision-
making under a variety of configurations (cf. Sec. 4.2.4).
264 Thisis in line with Hammond and Miller (1987) who demonstrated that bicameralism can create core
solutions in settings where simple majority rule is affected by cycling among outcomes and vice versa.
In Sec. A.8 I indicate below every payoff table if it constituted a constant-sum or non-constant-sum game
and the respective core alternative for every procedure (if one exists).
127
Table 5.3: Data structure according to the existence of an equilibrium alternative

EXPLANATORY NOTE
The table depicts the number of observations in terms of collective decisions made in the experimental sessions. The
observations are divided according to procedure and whether or not the payoff table used contained a core alternative.
CORE ALTERNATIVE
Σ
does exist does not exist
pooling 58 18 76
PROCEDURE
Σ 148 73 221
Overall, I conducted more experimental rounds with tables containing a core solution
(148 to 73). Looking into the procedures, the distribution of payoff tables with and with-
out a core showed equal amounts under SIM (42 to 43). For POL and SEQ many more ob-
servations using tables with a core alternative were collected (under both rules in around
75% of the cases). Summing up, for the performance analysis of my theoretical baseline
this gave me 58 collective observations under POL as well as 42 under SIM and 48 under
SEQ.
5.3 How to judge a collective decision
My experiment contrasted the situation in which the collective decision is reached in the
group as a whole to the situation in which the group is split into two delegations which
must decide simultaneously or sequentially. The most intriguing aspect of this approach
was the opportunity to compare the behavior and outcomes of the joint and the separate
game structure within one experimental design.
It also distinguished my experiment from previous research which investigated the im-
plications of different decision rules. I discuss this literature in Sec. 5.3.1 and define sta-
tistical measures for evaluating the chosen alternative out of and according to it. Next,
Sec. 5.3.2 derives corresponding hypotheses for the collective results. The specific oper-
ationalization of the metrics used is explained in Sec. 5.3.3. These statistical measures
served, in addition to the theoretical equilibrium, as the second starting point for the as-
sessment of the selected alternatives. Only by establishing this credible classification was
the analysis able to reach conclusive findings.
5.3.1 Decisions costs and welfare effects
Laws and regulations coordinate the various activities of individuals within every soci-
ety. Such rules are of paramount importance for all individual and collective decision-
making.265 Brennan and Buchanan (2000) defined the investigation and evaluation of so-
cial rules as the subject matter of modern political economy. Since Buchanan and Tullock
(1962) various contributions have contrasted the pros and cons of different k-majority
265 For a more comprehensive overview cf. Mueller (2003, part II, p. 67-208).
128
rules266 ; most prominent are unanimity and absolute majority rule267 . This also includes
studies which introduced veto players as a key element instead of different voting thresh-
olds (e.g., Chen and Ordeshook, 1998; Kagel et al., 2010) as the implications are the same
as under unanimity.268
Current research does not provide a clear-cut answer for the search of the optimal so-
cial choice mechanism.269 For example, investigating jury decision rules Guarnaschelli
et al. (2000, p. 407) found “fewer outcomes analogous to incorrect convictions under una-
nimity rule than under majority rule”; certainly a positive effect. Consistently, many
theoretical computations led to the conclusion that “for outcomes that are both Pareto
optimal and Pareto superior, unanimity rule outperforms majority rule” (Dougherty and
Edward, 2012, p. 655). Colomer (2001) discussed different voting rules such as unanimity,
majority and plurality vote as well as various types of political regimes. He investigated
parliamentarism with either majority or proportional representation and also presiden-
tialism for its social efficiency. In the end he concluded that unanimity rule equilibria
are Pareto optimal and, maybe even more striking, that majority rule equilibria rarely
exist. The Pareto argument is further illustrated by Mueller (2003). In its very elementary
logic unanimity ensures that only Pareto improvements are possible because otherwise a
member would contradict their own interests.
Yet, these theoretical findings are weakened as they not hold up in simulations which
use “random proposals and sincere voting” (Dougherty and Edward, 2012, p. 662). Also,
Colomer (1999, p.543) found that unanimity decisions heavily “depend on the initial state
or the status quo. The closer the status quo is to the ideal points of the actors, the more
restricted, more biased and likely less socially efficient the set of decisions by unanimity
tends to be.” Such a categorization implies that general statements of superiority should
be avoided.
Contrary to the theoretic assessment that “unanimity rule is the only voting rule certain
to lead to Pareto optimality” (Johnson, 1991, p. 161), multiple experimental evidence sug-
gests “that majority [rule] can produce larger welfare effects” (Sauermann and Glasmann,
2011, p. 373). Colomer (2001, p. 71-73) even came to the exact opposite conclusion, namely
that majority rule is, indeed, better at obtaining Pareto-optimal outcomes than unanim-
ity rule. Sauermann and Glasmann (2011, p. 391) attributed this to the fact that when
266 Dougherty et al. (2009, p. 1) defined a k-majority rule to require at least k from a total of N individuals to
vote in favor while it holds that N2 < k ≤ N. Extensions to the cases of non-voters or “votes to abstain”
are discussed in Dougherty and Edward (2004).
267 Those constitute two special cases of k-majority rule: majority rule for k = N and unanimity rule for
2
k = N (Dougherty et al., 2009, p. 1).
268 The well-known contribution of Tsebelis (2002) clearly demonstrated to what extent and how veto players
shape policy.
269 In addition to unanimity and absolute majority rule this statement also holds true for alternative vot-
ing mechanisms. For example, Forsythe et al. (1996) conducted laboratory election experiments under
plurality rule, approval voting, and Borda rule. The authors also studied the effects of (non-binding)
pre-election polls. Crucial for the search for the optimal electoral system (but unfortunately also disap-
pointing) is ”that Condorcet losers occasionally win regardless of the voting rule or presence of polls”
(Forsythe et al., 1996, p. 355).
129
evaluating different democratic voting norms “the majority rule is a strong incentive to
cooperate”270 .
In the end, the assessment for the optimal social choice mechanism depends on the deci-
sion costs which are set against the welfare effects of the voting rule (Buchanan and Tul-
lock, 1962; Kaiser, 2007a).271 Miller and Vanberg (2014) directly investigated the effects of
different decision rules on the costs of decision-making in a laboratory experiment. They
observed more rejections, bullish behavior and costly delays under unanimity rule. This
implies “support of less-than-unanimity decision rules” (Miller and Vanberg, 2014, p. 20)
in multilateral bargaining situations. Interestingly, Miller and Vanberg (2014) stated that
an issue worth exploring in future research is the effect of group size. This way one could
test the classic argument of Buchanan and Tullock (1962), which states that the costs of
decision-making increase with the size of the decision-making body.272 This is feasible in
my design when looking at groups of six as well as three members.
I followed the contributions discussed by evaluating the collective majority decisions

based on decision costs and welfare effects. This represents a trade-off between (cost)
efficiency and (welfare) effectiveness. Those two categories were evaluated using five
key figures which I derived from the literature on collective decisions: decision-making
efficiency, social welfare allocation, distribution of wealth, approval rate and stability. Subse-
quently, I explain the origin and intention of each metric.
EFFICIENCY
First, I look at the efficiency aspect. As many collective decisions are typically time-
consuming, its distinction is straight forward. Efficiency depends on the ability to reach
a decision in a parsimonious way, i.e., to minimize transaction costs (Coase, 1937, 1960).
For clarification, in my experiment neither the sum nor the distribution of points of an
alternative is altered within a round across the single ballots. This distinguishes my con-
ception of efficiency from experiments where the available amount of points decreases
with the number of ballots.273 Instead, it corresponds to the literature on management
theories in which “efficiency is the achievement of the ends with the least amount of re-
sources” (Olum, 2004, p. 6). In other words, to solve the collective agreement problem in
as few steps as possible. I designate this measure decision-making efficiency.
270 The work of Dougherty et al. (2009) showed yet another possible explanation, the specific experimental
setting. The authors concluded “that unanimity rule may not be particularly adept at selecting Pareto
optimal outcomes if the starting point is not in equilibrium” (Dougherty et al., 2009, p. 23).
271 This consideration goes far back to Wicksell (1896) and his assessment of the Swedish tax system.
272 A non-agreement on a topic is generally seen as disadvantage or waste of time. But under some cir-
cumstances it may, however, be desirable. For example, in their jury experiments Guarnaschelli et al.
(2000, p. 419) found that “larger juries may convict fewer innocent defendants than smaller juries under
unanimity.”
273 For example, in the bargaining experiment of Miller and Vanberg (2014) a group takes a vote on a proposed
payoff distribution of an initial endowment. “If a simple majority accepts the proposal, the game ends
and each player receives his allocated amount. If not, the pie shrinks by a certain factor and a new round
begins. Thus, the costs of bargaining consist of the lost surplus” (Miller and Vanberg, 2014, p. 6).
130
EFFECTIVENESS
Next, I turn toward statistical measures for their effectiveness. The range of possible
criteria for assessing welfare effects is larger and more diversified than for efficiency.
Most clearly relevant and immediately apparent is the amount of allocated points which
results in the social welfare allocation of the subjects.
In addition, not only the sheer scale of welfare274 but also its distribution is relevant
(Rawls, 1971). Already Black (1948, p. 29) emphasized the “persistence of disharmony
and discord” in collective decisions. More often than not the equilibrium solution pre-
dicts a highly unequal distribution of outcomes in which the majority cares little about the
minority’s well-being (Saunders, 2010b). Thus, I also consider the distribution of wealth.
Interestingly, previous experiments have proved that distribution issues are related to the
classification of participants.275 Allocating subjects to be a “row” or “column” player, as
in my design, influences preferences (Charness et al., 2007b; Goette et al., 2006). For exam-
ple, Chen and Li (2009) measured higher earnings for ingroup matching in their alloca-
tion games. Banuri et al. (2011) investigated the reasons for nepotism276 and the influence
of anti-nepotism laws. They found that prohibitions reduce trust and that, with salient
membership, subjects engaged in socially costly nepotism. More generally, whether the
resulting effect was socially desirable or not (i.e., whether it led to more social welfare)
depended on the particular environment or type of the game. Yet, decisive was always
the salience of membership, i.e., the subjects’ identification with their assigned group
(Charness et al., 2007b, p. 1362).277
As well as the realized outcome itself, its emergence and persistence also have to be
considered. In a general context, participation and inclusion are desired and necessary
for a functioning democracy (Schäfers and Zimmermann, 2005).278 This question refers
to research on coalition-formation (e.g., Schofield, 1996) and is interesting with respect
to differences in the size of winning coalitions across procedures. S&K (2010), Miller and
Vanberg (2014) as well as Sauermann and Glasmann (2011) found that people make use of
their veto rights. It seemed reasonable to expect that subjects use their opportunities for
excluding supernumerary players when aiming for the majority threshold. Thus, groups
274 In terms of my experiment, scale refers to the amount of points accumulated by the group when reaching
a collective agreement.
275 Research on (group) identity is a wide field. Regarding the methodological framework, Sutter (2008)
extended the findings on group membership of Charness et al. (2007b) who had focused on strategic
games also to non-strategic settings. Identity is also a fairly universal pattern. In general, Fehr et al.
(2008) showed that the understanding of identity already develops between the ages of 3 to 8 years. More
specific studies investigated how personal relations affect the relation between managers and employees
(Brandts and Sola, 2006) or looked at identity when making investment decision (Güth et al., 2005).
276 Banuri et al. (2011, p. 1) defined nepotism as “discrimination in favor” of a group member relative to the
population (cf. Becker, 1971).

277 Brewer (1999) advocated a distinction between ingroup favoritism and outgroup discrimination. Yet, my
design does not allow distinguishing between them. Thus, I was not able to determine whether there is
affection towards ingroup or rejection of outgroup subjects.
278 Admittedly, in my laboratory experiment such a value judgment was less clear. That all participants go
along is no common wisdom for such an abstract context.
131
should opt for small coalitions to enforce beneficial alternatives (for a literature overview
on games with equilibria in minimal winning coalitions cf. Chen and Ordeshook, 1998).
This is supported by experimental findings. For example, Berl et al. (1976, p. 473) ob-
served minimal winning coalitions “four out of five times” in their collective voting and
van de Kragt et al. (1983) recorded predominantly “minimal contributing sets” in their
public good experiment. Taking this into account, I examined the approval rate of every
collective agreement. With respect to persistence, I considered the stability among the
different decision procedures. It is an important purpose of organizational procedures
to ensure reliable results (e.g., Shepsle, 1979) and secure durable policies (Colomer, 2001,
p. 208). That it is reasonable to avoid cycling in majority decisions is one of the few undis-
puted understandings in public choice (Shepsle and Cox, 2007).
5.3.2 Hypotheses about the treatment effect
Sec. 2.5 states common arguments on the influence of NSP on decision-making. Further-
more, the section points out that people are subject to cognitive limitations (Oaksford
and Chater, 1992; Stanovich and West, 2000) and are not able to consider thoroughly all
implications and restrictions of a decision rule. I look more thoroughly into this matter
in Chap. 6 when assessing individual voting behavior and variations in cognitive capa-
bilities. For now, it suffices to distinguish between the four decision situations unraveled
in Tab. 4.2: POL SIM as well as first and second stage SEQ.
The arguments were formulated in general terms as they focus on basic patterns. In the
following I refine them with testable hypotheses according to the concrete design of my
experiment. POL, which constitutes my most conventional treatment, serves as a genuine
link to prior contributions. Here, a two-dimensional decision problem was submitted to
a group deciding collectively. The two dimensions of the decision problem (i.e., row and
column of the payoff table) were clearly interlinked. Thus, the basic principle of nonsep-
arability implies that those parts should not be separated. If they are split nonetheless
the concept of NSP predicts a “first-mover” advantage for sequential (ARGUMENT 2) and
a sub-optimal collective outcome for simultaneous decision-making (ARGUMENT 3). In
terms of the decision situations in my experiment this relates to a contrasting juxtaposi-
tion of i) first and second stage under SEQ to determine a possible first-mover advantage
and of ii) POL and SIM to investigate potentially sub-optimal outcomes.279 I accomplish
and structure this comparison based on the just in Sec. 5.3.1 derived statistical measures.
The first two hypotheses280 aim at social welfare allocation and outcome stability.
279 ARGUMENT 3 distinguishes between decisions made jointly and separately. This corresponds to the com-
parison of POL and SIM. Yet, SIM introduced the highest level of both cognitive complexity and un-
certainty (cf. Tab. 4.2). Thus, it was more difficult for the participants to fully comprehend the decision
problem under SIM than under SEQ. When evaluating the results I therefore also discuss the differences
between the two delegation procedures.
280 The hypotheses follow those two defined in Sec. 3.3.3. Therefore, I continue the number sequence.
132
HYPOTHESIS 3 (first-mover advantage): Under the sequential delegation pro-

cedure the first stage members achieve a higher social welfare and more sta-
bility than the second stage participants.
HYPOTHESIS 4 (sub-optimality): The simultaneous delegation procedure achieves

a lower social welfare allocation and less stability than the pooling procedure.
Next, I focus on the distribution of wealth and the approval rate of the collective decision.
Here, I also have to consider the insights about group identity; under both delegation
procedures the subjects were assigned to vote either on the column or the row of the
payoff table. Taking first-mover advantage and group identity together, under SEQ the
first stage participants should decide in accordance with their mutual benefit. But then
the second stage was condemned to fight for the leftovers (especially in the case of a
constant-sum table). Thus, consensus should be less common in the second stage because
of scarce resources. Also, if the assignment proves salient enough, the participants should
have been more willing to compromise under SIM than under POL.281
HYPOTHESIS 5 (scarcity): Under the sequential delegation procedure the first

stage members achieve a more equal distribution and a higher approval rate
than the second stage participants.
HYPOTHESIS 6 (compromise): The simultaneous delegation procedure achieves

a more equal distribution and a higher approval rate than the pooling proce-
dure.
In addition to the configuration of a collective decision the classic argument of Buchanan

and Tullock (1962) focuses on the swiftness with which it is achieved. They stated that
decision costs increase with the size of the electorate. I investigated this group size effect
by contrasting decisions between groups of six and groups of three members. Of course,
the necessary negotiations for reaching a collective decision depend also on the intricacy
of the task. So a simpler problem should lead to a faster solution.
HYPOTHESIS 7 (intricacy): Under the sequential delegation procedure the first

stage members achieve a lower decision-making efficiency than the second
stage participants.
HYPOTHESIS 8 (swiftness): The simultaneous delegation procedure achieves

a higher decision-making efficiency than the pooling procedure.
To summarize, all hypotheses for the evaluation of the collective decisions are listed in
Tab. 5.4. This illustrates that the hypotheses 3, 5 and 7 address differences in the per-
formance of first and second stage participants under SEQ. Under SIM both delegations
had to perform the exact same task and there was no distinction according to succession.
Thus, no differences in performance between these delegations were expected. More
281 Thishypothesis is supported by existing literature on social preferences. Previous studies found that they
are stronger among subjects who interact in small groups compared to those in large groups (Fehr and
Schmidt, 1999). The question remains whether the reduction from six to three persons is sufficient.
133
specifically, under SIM the two delegations achieve the same decision-making efficiency,
social welfare allocation, inequality distribution, approval rate and stability. I exploited
this fact to conduct a robustness check against artificial findings due to, e.g., an unac-
counted imbalance in the payoff tables between rows and columns.
Table 5.4: Hypotheses for the treatment effect

EXPLANATORY NOTE
The table summarizes the hypotheses about the collective results. They are structured according to the respective statisti-
cal measures. For every hypothesis its designation, the involved decision situations as observation unit and the expecta-
tion of their relative performance are given.
HYPOTHESIS EXPECTATION
SOCIAL WELFARE ALLOCATION and STABILITY

3 first-mover advantage under sequential delegation the first stage performs better than the second stage
4 sub-optimality simultaneous delegation performs worse than pooling
DISTRIBUTION OF WEALTH and APPROVAL RATE

5 scarcity under sequential delegation the first stage performs better than the second stage
6 compromise simultaneous delegation performs better than pooling
DECISION - MAKING EFFICIENCY

7 intricacy under sequential delegation the first stage performs worse than the second stage
8 swiftness simultaneous delegation performs better than pooling
5.3.3 Operationalization of the statistical measures
Before I turn to the empirical analysis in the next section, I describe subsequently the
empirical properties and concrete implementation of the five statistical measures.
In my experiment efficiency refers to a delay of the collective decision. This was mea-
sured by the number of ballots necessary to finally come to a collective agreement.
SOCIAL WELFARE ALLOCATION
Welfare effects were related to the collectively gathered sum of points. Due to the random
assignment rule of the payoff tables (cf. Sec. 4.2.4) I normalized the results. This enabled
me to compare different tables and procedures. Eqn. 5.3 shows the corresponding calcu-
lation. Every (non-constant-sum) payoff table offers the subjects a continuum of alterna-
tives of which one inhibits the maximal xmax and one the minimal xmin attainable amount
of points for the group. The transformation function sets the actually achieved amount of
points xi in relation to this continuum. It transforms the continuum in a way that the new
value range extends for every payoff table from 0 to 100.282 The transformed amount of
points T ( xi ) then indicates how successful a decision was in terms of this range. Thus,
282 The normalization corresponds to a linear scale transformation instead of a z-standardization (Studen-
mund, 2006, p. 541-544). Through normalization the results from different payoff tables became compa-
rable by conversion into a homogeneous scale. Standardization would, in addition, always lead to mean
of zero and a SD of one for the standardized variable (for proof cf. Gujarati and Porter, 2008, p. 173ff and
appendix 6A, Sec. 6A.2).
134
in the analysis each decision’s relative allocation performance is reported instead of just
presenting the absolute sum of points.283
100
T ( xi ) = × ( xi − xmin ) (5.3)
xmax − xmin
DISTRIBUTION OF WEALTH
I measured distributional aspects using the standard deviation as well as the Gini coef-
ficient (Gini, 1912; Hirschman, 1964) of an alternative. Both are well-known and reliable
indicators of inequality (e.g., Bellu and Liberati, 2006; Haughton and Khandker, 2009).
As with the allocation of social welfare its distribution can be expressed in absolute or rel-
ative terms. In the following, each decision’s relative inequality performance is reported
instead of just presenting the absolute amount of standard deviation. For the standard
deviation this was calculated in accordance with Eqn. 5.3.284 The value range of the Gini
coefficient is 0; n− 1
(Wagschal, 2009, p. 130).285 In order to roughly keep this range I
 
n
normalized the coefficient to the interval [0; 1]. This was achieved by using 1 instead of
100 in Eqn. 5.3. Here, 0 means a maximal uniform distribution and 1 represents a maxi-
mal inequality.286 Both metrics show how close the committee has come to reaching an
equal split. The greater the concentration is, i.e., the imbalance in the selected alternative,
the higher the numerical values.
APPROVAL RATE
The extent of agreement for every collective decision was determined endogenously in
my design. Groups could agree unanimously on an alternative or enforce a decision on
the basis of majority rule. In either way the selected outcome determined the payoffs
for all group members. There was no reward or deduction for consensus. I counted the
approval rate by the size of the winning coalition (i.e., the number of matching votes)
which reached the collective agreement.
283 Through normalization the results are expressed as a relative success rate. Imagine a collective decision
on a payoff table of which the highest sum of points of an alternative equals 80 and of which the lowest
amount is 30. A group that manages to select the alternative which adds up to 80 points was completely
successful and ends up, according to Eqn. 5.3, with a success rate of 80100 100
−30 × (80 − 30) = 50 × 50 = 100.
On the other hand, if the group ends up with an alternative that adds up to 55 points (i.e., midway
between 80 and 30), its success rate is only 80100 100
−30 × (55 − 30) = 50 × 25 = 50.
284 Through normalization the results are expressed as relative inequality. Imagine a collective decision on a
payoff table of which the largest imbalance is 80 and of which the smallest is 30. A group that manages to
select the alternative with an inequality of 30 effectively agrees on the most equal split which, according
to Eqn. 5.3, results in a relative inequality of 80100 100
−30 × (30 − 30) = 50 × 0 = 0. On the other hand, if
the group ends up at an alternative with an inequality of 55 points (i.e., midway between 80 and 30), its
relative inequality performance is 80100 100
−30 × (55 − 30) = 50 × 25 = 50.
285 In the case of my experiment, a group of six players (n = 6) decided on the distribution of points. This
 
resulted in a possible range of 0; 0.83 .
286 The term “maximal” should be interpreted as the maximal possible distribution in this decision in the
experiment. For example, if a choice between alternatives provides the Gini coefficients of 0.2, 0.4 and
0.7 then a normalized Gini coefficient of 0 would refer to 0.2 and of 1 to 0.7 for the result.
135
STABILITY
The robustness of the collective decision was not directly tested in my experiment. Once
an agreement was reached, the respective alternative was finalized and not again put to
vote (cf. Sec. 4.2.4). I therefore approximated the persistence of outcomes by means of
the homogeneity of the collective results. This was measured by how homogeneously
the voting pattern within and between treatments was when facing the same decision
problem (i.e., the same payoff table).
Summing up, the five statistical measures for the evaluation of the collective decisions
were operationalized as follows:
• Decision-making efficiency is counted as the number of ballots conducted.
• Social welfare allocation is expressed as the relative allocation performance.
• Distribution of wealth is measured as the relative inequality performance.
• Approval rate is represented by the size of the winning coalition.
• Stability is approximated by the homogeneity of the collective results.
TREND ANALYSIS
When one investigates the influence of NSP based on these metrics a static assessment
is not sufficient. An examination of the results over time is also necessary (cf. Tab. 4.3).
With every round of the experiment the experience of the subjects with the experimen-
tal setting grew. While a session progressed, they were trained in the process and knew
how reaching an agreement worked. This should be visible, e.g., in terms of fewer bal-
lots per collective decision, a higher social welfare allocation and greater stability in later
rounds compared to those at the beginning. Yet, it is not clear after how many rounds a
change should occur. Because of the vague context I operationalized this computation by
means of a trend analysis. In this way I avoid having to specify a concrete cut-off point.
The same consideration applies to first and second decision rule of a session. Thus, I
expected a better performance in terms of decision-making efficiency, social welfare al-
location, distribution of wealth, approval rate and stability after participants had gained
experience.
5.4 Results
This section investigates the collective results of my voting experiment. As first step
I look into the performance of my theoretical baseline. In addition to its contribution
about collective behavior, this subsection serves an important additional purpose. The
discussion of the experimental design has shown that in the empirical assessment some
aspects are of particular importance (cf. Sec. 4.3.5). The evaluation had to consider trend
136
5.4 Results
analysis, statistical independence of observations, etc.287 When assessing the theoretical

benchmark predictions I discuss the necessary steps of the empirical analysis in detail.
I outline the used test statistics and explain the structure of the summary tables which
present my results (e.g., observations separated according to be first or second decision
rule of a session). Subsequent analyses follow this pattern; I use the same methods, but I
do not discuss every aspect in such detail.
In the tradition of Buchanan and Tullock (1962, cf. also Kaiser, 2007) collective decisions
resemble a trade-off between decision costs and welfare effects. In other words, one
has to choose between efficiency and effectiveness of the reached agreement. Efficiency
turns toward the ability to reach a decision in a short and parsimonious way. Here, the
consequences of the chosen alternative do not matter. Effectiveness, on the other hand,
depends solely on the specifics of the selected option. It is not important how long it took
until the decision was made; only its output properties count. This is the second step of
my empirical analysis. I look successively at the derived statistical measures and formu-
lated hypotheses (cf. Sec. 5.3.2); this provides a thorough overview on the achievements
of the different decision procedures. I also analyzed possible trends of the metrics over
time, i.e., over the progressing experimental rounds. This implied a lot of additional ta-
bles. To keep the analysis comprehensible I include them in Sec. A.12. This serves solely
for the purpose of a concise presentation. All results are discussed and clearly linked to
the corresponding table in the appendix.288
5.4.1 Selection of the core
In their simpler design S&K (2010, p. 674) found that 69% of all their committee decisions
resulted in the selection of the core alternative. As I modify their general experimental
environment I considered this value as a reference point. With respect to my data, not
all observations were suitable for a comparison. But unlike in Sec. 5.1, I did not have to
confine myself solely to POL. I used all collective decisions made on payoff tables with a
core alternative. Across all procedures these were 148 observations.
Before comparing the core performance one last modification was necessary. Under SIM
the final outcome is determined as the intersection of the two delegations’ decisions. Both
delegations chose only a subset and not a specific alternative. This is an important differ-
ence to SEQ where at least the second stage delegation picked one concrete alternative.
Thus, the performance of the core in predicting the collective voting behavior was cal-
culated at the level of the two delegations; i.e., the percentage with which a delegation
chose the subset that belonged to the credible core. As every group decision consisted of
two delegation decisions the number of observations at this level was twice as large.
287 Tab. 4.3
summarizes the key characteristics and their implications for the empirical analysis.
288 Without getting ahead of myself, I have moved the trend analysis into the appendix because it contains
only few significant results. The knowledge gained is therefore rather low.
137
The core performance under each treatment is shown in Tab. 5.5289 . As discussed before,
observations are not statistical independent across rounds (cf. Sec. 4.3.2). Therefore, the
unit of observation must be session averages per procedure of the variable in question
(cf. Frechette, 2012).
Table 5.5: Core performance

EXPLANATORY NOTE
The table shows the performance of the core in predicting the collective decisions. Each cell indicates the number of correct
predictions in relative terms. The observations are separated by procedure and for being the first or second decision rule
of a session. The unit of observation are session averages of the collective decisions per procedure.
PROCEDURE first decision rule second decision rule
N mean SD N mean SD N mean SD
pooling 9 42.6% 16.4 2 45.8% 5.9 7 41.7% 18.6
simultaneous delegation
delegation level 8 40.6% 13.9 6 37.8% 10.4 2 49.0% 25.0
sequential delegation 8 39.6% 27.6 5 30.8% 29.1 3 54.2% 7.2
It is quite apparent that the performance of the core in forecasting the collective decisions
was lower in my experiment than in the case of S&K (2010, p. 674). For every procedure,
the concept could predict about 40% of the decisions. For SIM this refers to the delega-
tion instead of the group level (as discussed in Sec. 4.3.4), where a core set selection was
independent of the other delegation. So, a non-selection of one delegation did not imply
a non-selection for both. A performance of about 40% might not be considered as a very
good predictor, but it is also not necessarily a poor one. The frequency of selection is
clearly above a 11% probability of randomly drawing one of the nine alternatives.
When examining the results across procedures, decision rules, etc., I make my assess-
ments not just by looking on the obtained average values and applying a rule of thumb.
Every single statement is based on a nonparametric test (Siegel, 1988). This is either the
Wilcoxon matched-pairs signed-ranks test (Wilcoxon, 1945)290 for within-subject compar-
isons or the Mann-Whitney U-test (Mann and Whitney, 1947) for between-subject analy-
sis (cf. Sec. 4.2.7). Both are nonparametric rank sum tests and evaluate whether the cen-
tral tendency of two samples correspond,291 they differ according to their assumption
whether the samples used are related. I stratified my observations according to proce-
dure, group and delegation level, first and second decision rule, as well as under SEQ due
to first and second stage. For reasons of clarity I do not report results for every pairwise
289 For reasons of clarity the presentation of the joint separation for both table properties has been omitted as
it contained no additional information.
290 I used the signed-ranks test (Wilcoxon, 1945) and not the sign test of matched pairs (Arbuthnott, 1710;
Snedecor and Cochran, 1989), because the signed-ranks test takes into account not only the positive or
negative sign of differences, but also the extent of the differences between the paired samples.
291 Contrary to a parametric two sample t-test the assumption of a normal distributed dependent variable is
not necessary (Gibbons, 1976). This is required as provident tests for normal distribution (Royston, 1992;
Shapiro and Wilk, 1965) on almost all obtained variables turn out negative. The violation of this assump-
tion would result in unreliable inferences and misleading interpretations of the t-test (for an assessment
when either test cannot be applied due to extremely skewed data cf. McElduff et al., 2010). Although it
is general assumed that the t-test is more powerful (because nonparametric tests convert the observed
values into ranks) this does not hold for large samples (cf. Motulsky, 2010, Chap. 37).
138
5.4 Results
comparison. Instead, I list all significant findings, including the respective (two-tailed)
significance level. All unreported values signal an insignificant test.
Looking more into the details of Tab. 5.5, neither comparison across procedures resulted
in a clear pattern. Although the performance values seemed higher for the second de-
cision rule at first glance, the differentiation for decision-rule resulted in no statistically
significant difference under POL, SIM or SEQ. This was also true when considering all
observations together.
In my experiment subjects participated in multiple rounds in every session. So, in ad-
dition to the overall performance of the core also its progression over the rounds is of
interest. After the subjects got acquainted to the game (i.e., after some training), it is rea-
sonable to assume that their voting patterns might change (cf. Sec. 4.2.5). If this is true,
the performance of the core should be better in later rounds.
I analyzed the development over the single rounds by using STATA ’s somersd package
(implemented for STATA by Newson, 2002),292 a nonparametric rank-statistic.293 This
package performs a maximum likelihood fit to obtain association measures (and CI) of a
predictor and the dependent variable. In other words, it depicts the difference between
the probability that two variables are concordant and the probability that they are not
(Newson, 2013, p. 1). The estimated value D might be interpreted as a “predictor perfor-
mance indicator” (Newson, 2006a, p. 312). In terms of my analysis, it showed the ability
of the experimental round to predict the performance of the core, i.e., if the core alterna-
tive was chosen more often in later rounds.
Tab. 5.6 shows the results of the trend analysis on the performance of the core.294 The
observations are split according to procedure and being the first or second decision rule
within a session. For none other than two estimates a significant trend was discovered. In
both cases the trend occurred within the second decision rule of an experimental session.
Thus, the trends were observed only after a procedural change. Merely repeating the
game did not alter the core performance in any of the procedures. Only in combination
with altering the decision rules did repetition lead the subjects to achieve the equilibrium
result.
Under SEQ I found a positive and under SIM I detected a negative trend over the rounds.
The finding for SEQ was more intuitive as it also corresponded to the findings in Tab. 5.5.
All SEQ rounds as second decision rule were run after playing SIM before (cf. Sec. 4.4).
Thus, the procedural change (the introduction of sequential voting) in combination with
292 The main advantage of this STATA package is that it calculates CI using jacknife variances (Newson, 2006b).
More precisely, it computes the CI of rank order statistics calculated by the Wilcoxon rank-sum test
(Wilcoxon, 1945). This considers that “nonparametric methods are in fact based on population parame-
ters, and that these parameters should be estimated, with sample statistics and confidence limits, instead
of following the traditional practice of calculating only P-values for the sample statistic” (Newson, 2006a,
p. 309).
293 Somers’ D is an ordinal measure of association and was introduced by Somers (1962).
294 The z-scores are calculated as z = estimated value and enable a comparison of the effect size of different
standard error
variables.
139
Table 5.6: Trend analysis of core performance

EXPLANATORY NOTE
The table shows the results of the trend analysis on the performance of the core in predicting the collective decisions. It
contains the number of observations, the Somers’ D coefficient, its z-score and 95% CI. The observations are separated by
procedure and for being the first or second decision rule of a session. Statistically significant (two-tailed) at the 0.05 level
** and at the 0.01 level ***. The unit of observation are collective decisions.
Independent variable: NUMBER OF ROUND Dependent variable: SELECTION OF THE CORE
N D z-score 95% CI
pooling 58 -0.04 -0.45 -0.21 0.13
first decision rule 18 0.09 0.55 -0.25 0.44
second decision rule 40 -0.11 -1.04 -0.32 0.10
delegation level 84 0.06 0.83 -0.8 0.20
second decision rule 28 -0.28** -2.43 -0.51 -0.06
PROCEDURE
sequential delegation 48 0.14 1.42 -0.05 0.33

first decision rule 30 -0.05 -0.36 -0.31 0.22
second decision rule 18 0.43*** 3.43 0.19 0.68
all 148 0.45 0.91 -0.05 0.14

repeated play led to a better performance of the core. This is not surprising as the sequen-
tial decision-making facilitated the coordination of the two delegations by reducing the
level of uncertainty. All SIM rounds as second decision rule were run after playing under
POL before (cf. Sec. 4.4). The negative trend tells us that the selection of the credible core
became less and less the longer the experiment continued. As the trend was discovered
at the delegation level it is not artificially caused by aggregation. Rather, the complexity
of SIM crowded out the amount of theory-compliant behavior.
This evaluation of the theoretical benchmark prediction made two aspects very clear:
i) the overall performance of the core was around 40% which left some space for other
explanatory factors and ii) there was no common trend over the rounds for the core per-
formance. Due to the complexity of the task the majority of participants seemed not able
or not willing to follow the theoretical benchmark. Yet, if the solution concept was not
decisive, what were the crucial criteria for selecting an alternative? The next subsections
look into this question.
5.4.2 Social welfare allocation and stability
Assessing the social welfare allocation I restricted my analysis to payoff tables which
represent non-constant-sum games.295 Tab. 5.7 displays the results. The trend analysis
(cf. Tab. A.9) shows no significant changes for any procedure. Also, combining results
across all procedures and looking only at decisions made when used as first or second
decision rule made no difference.
295 To
look into what a group has earned when each option summed up to the same amount of points would
make no sense.
140
5.4 Results
Table 5.7: Social welfare allocation

EXPLANATORY NOTE
The table shows the social welfare allocation of the collectively selected alternative for the different procedures. The obser-
vations are separated by procedure and for being the first or second decision rule of a session. In addition, I differentiate
under SIM for group and delegation level and under SEQ for first and second stage. The unit of observation are session
averages of the collective decisions per procedure.
PROCEDURE N mean SD min max
pooling 10 64.2 15.2 39.3 77.3
first decision rule 2 68.6 4.4 65.5 71.7
second decision rule 8 63.2 17.0 39.3 77.3
simultaneous delegation 8 55.0 7.6 45.9 66.4

delegation level 8 50.8 0.9 50.1 52.8
sequential delegation 8 63.0 12.8 41.5 87.2

first stage 8 43.8 16.1 22.0 74.2
second stage reduced 8 68.4 27.4 15.0 100.0
Overall, subjects were able to obtain most points under POL. In particular compared to
SIM the distinction is notable (p = 0.096). At the delegation level, SEQ significantly
outperformed SIM (p = 0.006). Of course, for SIM I used probabilistic outcomes due
to the selection of sets instead of alternatives by the participants. This results in values
around 50% for both the maximum and minimum of points under SIM.
Within POL the differences between first and second decision rule are negligible. The
same is true for SIM, while under SEQ the welfare allocation is higher as first decision
rule (p = 0.053). Also evident is that under SEQ the second stage surpassed the first
stage (p = 0.093). Please note that I restrict the evaluation to the actual remaining choice
of the second stage delegation. The term “reduced” refers to the problem of choice when
only one remaining set is left over from the first stage decision-making. Here, it becomes
obvious that the second stage was able to secure on average 68.4 of the possible points.
This represents the overall best performance.
So far I discussed the performance of the different procedures in relation to each other.
Yet, it is worthwhile to also have an absolute point of comparison. Tab. 5.8 lists the social
welfare allocation which would result if the group had in all decisions on tables with
equilibrium in fact chosen the core alternative.
The comparison to the actually obtained results shows that under POL and SIM the
groups achieved a higher welfare allocation as predicted by the core. Yet, only POL
performed significantly better (p = 0.009). Please note, that any deviation from the core
alternative which resulted in more welfare for the group cannot, by definition, have been
a Pareto-improvement.296 Under SEQ the group ended up with fewer points (p = 0.063).
296 For every core alternative it holds that under the rationality assumption no individual or subgroup within
the coalition supporting it can be better off by deserting the coalition (cf. Peleg and Sudhoelter, 2003).
141
Table 5.8: Social welfare allocation of core alternatives

EXPLANATORY NOTE
The table shows the social welfare allocation which would result if the group had always chosen the core alternative.
Obviously it includes only the observations made with tables containing such equilibrium. Please note that the different
procedures lead to different core alternatives. The data is separated according to decision rule.
PROCEDURE N mean SD
pooling
always core 9 30.9 20.9
actual result 10 62.0 33.5
sequential delegation
A second measure that was mentioned in my hypotheses is the stability of collective de-
cisions. I approximate stability by means of a dyadic comparison. This comparison an-
alyzed the collective decisions and individual votes when subjects were exposed to the
same situation, i.e., deciding on the same payoff table in the same role.297 Although this
is only an auxiliary measure, the comparison provided some evidence on how uniform
or different the voting patterns occurred. Standard errors and CI were obtained using
STATA ’s binomial CI calculator.298 More precisely, despite the programs’ default option
of binomial exact intervals (cf. Clopper and Pearson, 1934) and the “nearly universal use”
of the Wald interval (Brown et al., 2001, p. 115) I calculated “Wilson CI” (Wilson, 1927).
Following Brown et al. (2001), in comparison this method is more powerful, parsimo-
nious and less error prone. The results are shown in Tab. 5.9.
Table 5.9: Stability within a voting procedure

EXPLANATORY NOTE
The table shows the approximated stability of the decisions among the different procedures. The data is separated ac-
cording to decision rule and differentiates between collective and individual decisions. The percentage of corresponding
decisions is determined as dyadic comparison of identical outcomes in similar situations (i.e., the same payoff table)
within procedures.
PROCEDURE N % of corresponding decisions SE 95% CI
pooling
collective 66 19.7 4.9 11.9 30.8
individual 1878 24.3 1.0 22.4 26.3
collective 75 22.7 4.8 15.7 33.3
individual 2394 55.3 1.0 53.3 57.3
collective 48 37.5 7.0 25.2 51.6
individual 1140 53.5 1.0 50.6 56.4
In this table the number of observations does not refer to the vote of a subject or a collec-
tive decision. Instead it depicts the frequency of occurrence of a specific choice situation
297 As discussed in Sec. 4.3.1, I restricted the analysis of the individual votes to the last ballot in each decision-
making. Always, these final votes of individuals sum up to the collective outcome.
298 Source:
http://www.stata.com/help.cgi?cii (accessed February 25, 2013).
142
5.4 Results
with which more than one subject,  delegation or group was confronted. The number re-
nidentical situations
sults as N = given nidentical situations ≥ 2.299 The computations show
2
that stability at the collective level is not high under any procedure. In relative terms,
POL was as stable as SIM. SEQ outperformed POL (p = 0.034) and SIM (p = 0.074); this
indicates that the lower level of complexity due to the introduced sequence facilitated the
collective coordination. At the individual level the variance under POL was clearly the
highest (SIM: p = 0.000; SEQ: p = 0.000). The delegation procedures no longer differ and
reach at least a level of 50% corresponding decisions.
One interesting facet represents the comparison across procedures. Here, I looked into
the question how the votes differ between subjects voting on the same payoff table but
under various procedures. Tab. 5.10 shows that at the collective level the decisions vary
strongly between all procedures. Looking at the individual level, the highest coherence
was observed, as one might have expected, between SIM and SEQ. In both procedures
ballots were held in the group of three and every individual can only influence one di-
mension of the decision-making (either column or row). Here, at least 50% of the individ-
ual votes matched. Interestingly, individual votes under POL corresponded in the same
amount (around 40%) to SIM as well as to SEQ.300
Table 5.10: Stability across voting procedures

EXPLANATORY NOTE
The table shows the approximated stability of the decisions among the different procedures. The data is separated ac-
cording to decision rule and differentiates between collective and individual decisions. The percentage of corresponding
decisions is determined as dyadic comparison of identical outcomes in similar situations (i.e., the same payoff table) across
procedures.
C omparison of individual votes N % of corresponding decisions SE 95% CI
among DIFFERENT PROCEDURES
pooling and simultaneous delegation
collective 105 4.8 2.1 2.1 10.7
individual 607 39.2 2.0 35.4 43.2
pooling and sequential delegation

collective 88 10.2 3.2 5.5 18.3
individual 494 39.1 2.2 34.9 43.4
simultaneous and sequential delegation

collective 89 12.4 3.5 7.0 20.8
individual 520 50.8 2.2 46.5 55.1
Of my formulated hypotheses two were concerned with social welfare allocation and
stability. A better performance in both metrics was expected for the first stage com-
pared to the second stage under SEQ (HYPOTHESIS 3) and for POL compared to SIM
299 For example, three players in the same situation lead to three dyadic comparisons (A-B, A-C, B-C), four
players in the same situation lead to six dyadic comparisons (A-B, A-C, A-D, B-C, B-D, C-D), five players
in the same situation lead to ten dyadic comparisons (A-B, A-C, A-D, A-E, B-C, B-D, B-E, C-D, C-E, D-E),
n
etc. Thus, the number of observations follows just the binomial coefficient = k!∗(n!
n−k)!
(Heinrich,
k
2006, p. 13 and 223) for k = 2 and n ≥ 2. This relationship also applies to collective decisions.
300 Here, I analyzed if the choice of a specific alternative under POL corresponded to the choice of the corre-
sponding set under a delegation procedure. For example, if an individual under POL chooses alternative
{C2} a corresponding decision for a column delegation individual under SIM or SEQ is set {C }.
143
(HYPOTHESIS 4). HYPOTHESIS 3 is rejected; in fact, it is the other way around. The second
stage outperforms the first stage which could not exploit moving first. HYPOTHESIS 4 is
confirmed when looking at social welfare allocation as SIM clearly performed worse than
POL (which is also the only procedure under which the welfare allocation significantly
surpasses the theoretical baseline). However, this does not apply to stability. Here, only
SEQ surpasses SIM. At the individual level POL received the lowest stability rating. This
might result from different thresholds in reaching a collective decision. Under POL even
with two diverging votes a majority could be reached. The next section picks up this
argument when discussing the approval rate.
5.4.3 Distribution of wealth and approval rate
The distribution of welfare is nearly as important as its allocation in the first place. I
looked into the income distribution of my results by using two empirical measures: the
standard deviation as well as the Gini coefficient of an alternative. Compared to the
previous assessment of social welfare allocation I could also include constant-sum tables
into the analysis (which now considers all collected 221 collective outcomes). It holds for
both metrics that high values reflect a large imbalance and low values represent a small
amount of inequality. As equality might be considered as a generally desirable property,
low scores represent a good performance. I start with the standard deviation as criterion
for the relative inequality performance of a collective outcome. The results are shown in
Tab. 5.11.
Table 5.11: Standard deviation

EXPLANATORY NOTE
The table shows the standard deviation of the collectively selected alternative for the different procedures. The observa-
tions are separated by procedure and for being the first or second decision rule of a session. In addition, I differentiate
under SIM for group and delegation level and under SEQ for first and second stage. The unit of observation are session
averages of the collective decisions per procedure.
pooling 10 56.0 11.8 32.5 72.0


first stage 8 47.7 8.2 34.3 62.2
second stage reduced 8 48.9 13.5 30.3 67.1
Contrasting first and second decision rule across all procedures provides no findings
except under POL (p = 0.046). The trend analysis (cf. Tab. A.9) shows no results except
144
5.4 Results
for one; weak significance suggests that under SIM as first decision rule the inequality
decreased over the rounds.
The performance of SIM is good in general as the the smallest imbalances were obtained
under this procedure. Comparing the results of SIM and SEQ as first decision rule, the
inequality is significantly lower under SIM (p = 0.006). Looking at the realized minimal
and maximal results the tendency of SIM towards 50% outcomes was not as pronounced
as with respect to social welfare allocation (cf. Tab. 5.7).
As an absolute point of comparison Tab. 5.12 presents the standard deviation which
would have resulted if groups had always agreed in tables with equilibrium on selecting
the core. The juxtaposition shows no difference for POL. Under SEQ (p = 0.036) and SIM
the groups ended up with a higher inequality (p = 0.012).
Table 5.12: Standard deviation of core alternatives

EXPLANATORY NOTE
The table shows the standard deviation which would result if the group had always chosen the core alternative. Obviously
it includes only the observations made with tables containing such equilibrium. Please note that the different procedures
lead to different core alternatives. The data is separated according to decision rule.
PROCEDURE N mean SD
pooling
I also used the Gini coefficient to assess the relative inequality performance. This is
a common measure for the distribution of income within a society (cf. Haughton and
Khandker, 2009). Since I do not intend to show duplicated results in detail, I moved
them in Sec. A.13. Here, Tab. A.10 displays the results for the Gini coefficients under the
different procedures. The table also shows the Gini coefficient which would have oc-
curred if the group had always chosen the core alternative. Overall, the comparison of
the Gini coefficient with the standard deviation increased the reliability of my analysis as
the values confirmed the previous results. Firstly, the good performance under SIM.301
Secondly, as when considering the standard deviation, the comparison with its counter-
part, i.e., the actual obtained results in equilibrium tables, is not decisive for POL. Under
SEQ (p = 0.012) and under SIM (p = 0.012) the results of the collective decision-making
possessed a higher Gini coefficient than the core alternatives.
In my hypotheses the distribution of points was linked to the approval rate of a decision.
How many votes each decision received is listed in Tab. 5.13. Please note that the value
range of the approval rate is from 66.6 ( 23 majority) to 100 (unanimity).302 Already at first
301 At the group level SIM outperforms SEQ (p = 0.100).
302 Under POL two different majorities are possible, i.e., 4 or 5 out of 6 players.
145
glance it is clear that majority decisions were dominant across all procedures and mod-
ifications. Especially under POL unanimous votes were very rare. The only significant
difference is found between the same subjects that acted first under POL and subsequent
under SIM (p = 0.080). Here, unanimity was still unlikely but occurred at least to a small
extent under SIM. The trend analysis (cf. Tab. A.9) suggested more unanimity decision
when SEQ was used as second decision rule.
Table 5.13: Approval rate

EXPLANATORY NOTE
The table shows the approval rate for the different procedures. The observations are separated by procedure and for being
the first or second decision rule of a session. In addition, I differentiate under SEQ for first and second stage. As both
delegation procedures vote in delegations (of three players) I show the approval rate at this level. The unit of observation
are session averages of the collective decisions per procedure.
pooling 10 68.71 1.86 66.67 72.22


first stage 8 70.49 7.99 66.67 88.89
second stage 8 72.74 6.00 66.67 83.33
Two hypotheses were directed to the relative inequality distribution and approval rate.
A lower inequality and a more encompassing agreement were anticipated for the first
stage in relation to the second stage under SEQ (HYPOTHESIS 5) and for SIM compared
with POL (HYPOTHESIS 6). HYPOTHESIS 5 is rejected. I could not observe more unanimity
voting in the second stage and found no differences in distributional patterns across the
two stages under SEQ. HYPOTHESIS 6 is only partially confirmed. For inequality the
findings show that SIM outperformed SEQ (but not POL) at the delegation level. With
respect to the approval rate it is undeniable that majority and not unanimity decisions
dominated. I found in particular very little unanimity under POL.
5.4.4 Decision-making efficiency
A central aspect of decision-making procedures is their efficiency. How long does it take
to reach a collective agreement? The answer to this question is shown in Tab. 5.14. Of
course, with respect to the two delegation procedures the efficiency of reaching a decision
in a delegation was important. Therefore, I also list these results. While the previous
discussed effectiveness measures hardly showed a change over time this was different for
efficiency. But, contrary to expectations, the number of necessary ballots increased over
the progressing rounds (cf. trend analysis in Tab. A.9) and from first to second decision
rule (p = 0.000) when looking at all procedures together.
146
5.4 Results
Table 5.14: Decision-making efficiency

EXPLANATORY NOTE
The table shows the efficiency of collective decision-making for the different procedures. The observations are separated
by procedure and for being the first or second decision rule of a session. In addition, I differentiate under SIM for group
and delegation level and under SEQ for first and second stage. The unit of observation are session averages of the collec-
tive decisions per procedure.
pooling 10 5.25 1.16 3.25 6.9


first stage 8 1.62 0.23 1.25 2.00
second stage 8 1.10 0.09 1.00 1.25
Looking at the data, two aspects stand out under POL. Firstly, no group was able to agree
in the first ballot. This corresponds well to the previously discussed low probability of
agreeing in a one-shot ballot under this procedure. Secondly, while the minimum num-
ber of ballots necessary was 2, at one instance impressive 32 ballots were required before
a majority agreed to vote for the same alternative.303 Turning towards the statistical eval-
uation, the trend analysis found under POL a decrease of efficiency over the rounds and
when used as second decision rule (cf. Tab. A.9).
Under both delegation procedures the efficiency measures were considerably better than
under POL. No round needed more than seven ballots and most agreements were al-
ready reached after 3 ballots were hold. This fast decision-making was stable for first and
second decision rule as well for the delegation level. Comparing the observations across
procedures shows that SIM (p = 0.053) and SEQ (p = 0.046) clearly outperformed POL.
Due to the sequential process it is not possible for SEQ to reach consensus on a collective
decision of the whole group in only one ballot. Even if both delegations agree in their
respective first vote, two ballots are always necessary. In the first ballot alone the first
stage delegation participated. Only if this delegation immediately reached an agreement
could the first vote of the second stage be held next. Thus, the undistorted values could
only be obtained at the delegation level. Here, the efficiency of SEQ was as high as for
SIM. Furthermore, when used as first decision rule SIM showed a higher efficiency than
as second one (p = 0.088) and the second stage of SEQ possessed the highest overall
efficiency and was significantly faster than the first stage (p = 0.012).
Sec. 5.3.2 formulates two hypotheses which were concerned with the decision-making
efficiency. A faster collective agreement was expected for the less complex problem of
303 The case of 32 ballots may constitute a special outlier as the next highest number was 13 rounds.
147
the second stage in relation to the first stage under SEQ (HYPOTHESIS 7) and for the
smaller electorate under SIM compared with POL (HYPOTHESIS 8). Both hypotheses are
confirmed. In fact, both delegation procedures showed clear efficiency advantages over
POL. Surprisingly, the efficiency decreased in the number of rounds (cf. Sec. A.12) and
from first to second decision rule.
5.4.5 Robustness
This section serves as robustness check for the previous discussed analysis. It is intended
to correct alleged findings which resulted not from behavioral patterns but artificial ta-
ble compilations. The hypotheses 3, 5 and 7 addressed differences in the performance
of first and second stage participants under SEQ. SIM mirrored the decision-making of
SEQ except for the sequential voting. Due to the instead simultaneously hold ballots no
differences between the two delegations were expected. Thus, the juxtaposition of the
performance of these two delegations enables a determination of the robustness of the
treatment effects. Tab. 5.15 shows the results for the two delegations under SIM.
Table 5.15: Robustness of treatment effects

EXPLANATORY NOTE
The table shows the social welfare allocation, standard distribution, Gini coefficient and decision-making efficiency of
the two delegations under SIM. The designation of the groups is based on their responsibilities in reaching a collective
decision, i.e., to vote for either a column or a row (cf. Sec. 4.2.6). The unit of observation are session averages of the
collective decisions under SIM.
SIMULTANEOUS DELEGATION N mean SD min max
SOCIAL WELFARE ALLOCATION

column delegation 8 45.0 4.8 35.9 52.4
row delegation 8 56.7 3.9 53.1 64.9
STANDARD DEVIATION
row delegation 8 55.9 10.2 46.9 78.9
GINI COEFFICIENT
row delegation 8 0.41 0.04 0.36 0.46

row delegation 8 1.29 0.15 1.00 1.50
HYPOTHESIS 3 stated a first-mover advantage under SEQ in terms of social welfare allo-
cation and stability. I rejected this hypothesis as I found that the second stage performed
better. This needs to be corrected slightly as the row delegation under SIM, which mir-
rors the second stage under SEQ, also outperformed its counterpart (p = 0.012). The
difference is smaller as under SEQ but still significant.
With respect to HYPOTHESIS 5 I found no differences in distributional patterns for the

two delegations under SIM. Neither the robustness check for standard deviation, Gini
coefficient or approval rate displays significant values.
148
5.5 Chapter summary
Following HYPOTHESIS 7 I expected a higher decision-making efficiency for the second

stage under SEQ. This was found in the analysis. Looking at SIM, I obtained a somewhat
higher efficiency for the row delegation (p = 0.076) which mirrors the second stage under
SEQ. Again, the difference is smaller than and not as significant as under SEQ. However,
also these findings need to be taken into account.
5.5 Chapter summary
This chapter started the empirical analysis of my experiment. First, Sec. 5.1 proves the re-
liability of the experimental design. I analyzed my data with a method previously used
by S&K (2010). As the design of the experiments is similar but not equal, the compila-
tion showed similar but not identical findings. However, my results indicate the same
patterns as found in previous work. This fortified the reliability of my most conven-
tional treatment. Next, I set up a benchmark for the upcoming investigation of collective
results. Firstly, I explain in Sec. 5.2 the derivation of the credible core for all decision pro-
cedures. Secondly, Sec. 5.3 is concerned with further aspects which might have driven the
behavior of the experimental subjects. The section discusses the trade-off between reach-
ing a decision in an appropriate amount of time and the quality of this settlement. A
long-established assessment when it comes to collective decision-making (Buchanan and
Tullock, 1962; Kaiser, 2007a). In accordance with this literature, I formulate hypotheses
for my design. Those are investigated in Sec. 5.4 empirically.
Overall, the theoretical solution concept could explain approximately 40% of the obser-
vations. This left room for other factors which might have driven the behavior of the
experimental subjects. It also reinforces my argument for the additional assessment cri-
teria for collective decisions.
Looking at those criteria I found, unsurprisingly, that delegation increased efficiency in

terms of decision-making speed and that subjects facing less complex tasks agreed faster.
Interestingly this was not true for experience with the game. Weak evidence showed a
slower decision-making over the progress of the experimental sessions. Contrasting this
efficiency gain against the effectiveness of the procedures provided ambiguous insights.
The review of the existing theoretical literature disclosed two arguments on the effect of
nonseparability (cf. Sec. 2.5). Most importantly, ARGUMENT 2 expected a first-mover ad-
vantage when a decision characterized by nonseparable preferences is taken separately
and sequentially. For this no evidence is found. With respect to social welfare allocation
and decision-making efficiency my results show that it is rather the other way around. In
consideration to the robustness check (cf. Sec. 5.4.5) this argument must be maybe some-
what attenuated, but a confirmation of the hypothesis is far away.
The proof for ARGUMENT 3 which expected a sub-optimal outcome when a decision char-
acterized by nonseparable preferences is taken separately and simultaneously is suffi-
149
cient. The social welfare allocation was highest under the pooled decision mechanism.
Thus, for policies characterized by nonseparable issue dimensions the delegation of com-
petencies must be well coordinated to prevent sub-optimal behavior. But if the term of
effectiveness is understood in a more comprehensive way the results become less clear.
Under POL the outcomes possessed also the most uneven distribution of points and the
lowest amount of stability in collective results. Yet, in terms of stability SIM performed
worse when compared to SEQ. It holds across all procedures that different decision mech-
anisms lead to substantially different outcomes. This process dependency is an important
insight.
These findings are not related to the approval rate. This rate stays nearly unchanged
across procedures and over experimental rounds; majority outcomes clearly dominated
the ballots in my experiment. This suggests that subjects used their tactical opportunities;
they excluded supernumerary players, formed minimal winning coalitions, etc. At this
level of analysis it is not possible to determine if the outvoted players are intractable
blockers or just poor outcasts.304
In fact, the small number of decisions which was reached by unanimity is the starting
point for the next chapter which extends the analysis to the individual level. Under all
procedures and table characteristics nearly always an outvoted minority of players ex-
isted. If my analysis would stop at the collective level the votes of these subjects would
be lost. The final outcome only reflects the votes (and preferences) of the winning ma-
jority. Therefore, I look into the individual voting behavior in Chap. 6 and determine the
individual driving forces for the observed collective results.
304 Chap. 7looks in detail into this matter. Here, I discuss the results of a post-experiment survey which
asked the subjects for their criteria when allocating their vote. One question in particular focused on
which other players the participants had considered when making their decision.
150
6 The determinants of individual choices
In this chapter I look into the disaggregated results of my experiment. This enables me
to examine whether and, if so, how nonseparability affects individual behavior. In par-
ticular, I discover the individual driving forces for the identified collective behavior. As
individual intentions and collective results do not always go hand in hand, this facili-
tates a much more distinct analysis. Using individuals as observational units enables
me to fade out the noisy patterns of collective majority decisions (Halfpenny and Taylor,
1973). In my experiment some individuals are outvoted and, thus, the collective decision
does not represent their choices. As in first-past-the-post voting the collective outcome
does not indicate dissenting voices. Absolute majority ballots are held in groups of six
and three members in my laboratory experiment. This raises the question of whether the
exclusion is extensive enough to cause such trouble. Its maximum is reached with two of
six or one of three votes. Yet, Goeree and Holt (2001, p. 4) argued that models “that in-
troduce (possibly small) amounts of noise into the decision-making process can produce
predictions that are quite far from any Nash equilibrium”.305 In other words, even the
smallest deviations can exert a significant influence when incorporated in an analytical
setting.
The following sections take a look at all final individual votes the participants made in
the experiment, i.e., the single votes which made up the collective decision. The ana-
lytical process is similar to the preceding chapter. At first, in Sec. 6.1 I discuss previous
literature on behavioral patterns of individual decision-making. Sec. 6.2 links the distinc-
tive features of my design, complexity and uncertainty, to individual voting behavior and
the participant’s considerations. I discuss the implications of individual voting consid-
erations and derive the resulting expectations. Thereafter, Sec. 6.3 brings the behavioral
patterns and voting considerations together for the empirical analysis. I present in de-
tail the composition of the statistical concept and its variants. In Sec. 6.4 I contrast the
performance of various model specifications to determine the best with regard to my ob-
servations by means of a model comparison. The units of observation for the statistical
model are all final votes of each individual in the final ballot of an experimental round.
Finally, Sec. 6.5 summarizes the findings.
The main empirical results of this chapter are also discussed in “Delegation, Uncertainty
and Social Preferences in Majority Decisions?” (Fleig and Finke, 2013). This article uses
305 Cf.Crawford et al. (2010, Chap.2) for a more comprehensive overview of equilibrium concepts that include
noise in form of, e.g., some type of error or cognitive limitations.
151
the conducted laboratory experiment to investigate the variation of individual choice be-
havior in response to changing voting rules. As the goal of this chapter is similar, the
empirical content of the two contributions largely matches. Compared with the article,
the discussion of behavioral patterns of individual decision-making (Sec. 6.1) and partic-
ipants’ tactical considerations (Sec. 6.2) are expanded. Furthermore, additional models
including various co-variates are examined (Sec. 6.4.2).
6.1 The literature on behavioral patterns of individual

decision-making
The analysis in the preceding chapter clearly demonstrated that the cooperative game
theory equilibrium only partly explains the observed behavior. Its performance in pre-
dicting the outcome of the laboratory experiment left a lot of room for additional explana-
tory aspects. The core alternative is characterized as a unique cooperative solution under
the assumption of rationally acting individuals. A self-interested single player or sub-
group has no incentive to opt for another choice. However, as not all collective decisions
match the prediction, the question is what other aspects drive the voting behavior.
It is far beyond the scope of this study to provide an all-encompassing insight into be-
havioral research and theories on human behavior. A good starting point for such an
objective is Cooper et al. (2007).306 This section gives an overview of the most promi-
nent, and, in the context of my experiment, relevant behavioral patterns of individual
decision-making; i.e., I will not discuss endowment effects or status quo bias. Both are
well-established behavioral patterns, but my experimental design implements neither an
initial endowment nor an SQ.307
MORE THAN JUST PURE SELF - INTEREST
The tenet of the purely self-interested Homo economicus (cf. Persky, 1995) has provided
a fruitful basis for many insights in social science (Kirchgässner, 2008). However, it is
widely accepted that “real” people behave differently, especially if social interaction is
involved (Coleman and Ostrom, 2009). Looking around, we see people living together in
families, sharing their goods in communities, donating to the needy, etc. Even when we
look as far back as Ancient Greece, Aristotle taught us that people are by nature social be-
ings (cf. zoon politikon, Höffe, 2005). During the last two decades, behavioral economists
and an increasing number of experimental social scientists have begun to systematically
306 There are also valuable contributions which focus on more specific aspects; Sheppard (1998) and Camerer
(2003) looked into game-theoretic models to explain behavior, Provis (2000) incorporated ethics, Char-
ness and Garoupa (2000) discussed reputation, Bazerman and Lewicki (1985) as well as Mitchell (1985)
highlighted aspects of negotiations within organizations, etc.
307 Endowment effect (cf. Plott and Zeiler, 2011; Thaler, 1980) and status quo bias (cf. Knetsch and Sinden,
1984; Fernandez and Rodrik, 1991) describe a similar behavior; “once a person comes to possess a good,
she immediately values it more than before she possessed it” (Tversky and Kahneman, 1991, p. 1041).
152
6.1 The literature on behavioral patterns of individual decision-making
question the assumption that individuals only selfishly maximize their own material in-
terest (cf. Palfrey, 2009; Henrich et al., 2001). In particular, the researchers aimed to iden-
tify the forces driving individual and collective decision-making (e.g., Ehrhart et al., 2007;
Hastie and Dawes, 2010). Often, the explanatory power of their models fitted to experi-
mental data could be successfully increased by allowing for additional aspects in an in-
dividual’s utility function (e.g., Engelmann and Strobel, 2004; Löwenstein et al., 1989).308
These empirical studies experienced, as I did, that the actual results “deviate system-
atically from the game-theoretic prediction based on self-interest. These deviations are
naturally interpreted as evidence of social norms (what players expect and feel obliged
to do) and social preferences (how players feel when others earn more or less money)”
(Camerer and Fehr, 2004, p. 90).309 In contrast, not everyone and everything is focused on
the common good. Most prominently, Smith (1776) taught us that progress and prosper-
ity originate from self-interest. Not every action which has positive external effects has a
social motive. For example, a baker makes loaves of bread to sell them, not to improve
social welfare. Recent research in behavioral economics, social psychology and neuro-
sciences has begun to reconcile these two perspectives on human nature (e.g., Bräutigam,
2005; Buller, 2005). Departing from the idea of man as Homo economicus it has become
widely accepted that human interaction is partly motivated by both kinds of preferences
(cf. Henrich et al., 2001). That includes both caring (e.g., fair and trusting) as well as
misgiving (e.g., competitive and spiteful) intentions (cf. Eisenkopf and Teyssier, 2010;
Engelmann and Strobel, 2004).310
6.1.1 Other-regarding preferences
The inclusion of additional aspects, besides one’s own payments when making decisions,
implies a change of focus. Unlike self-interest, which can simply be read as the payoff a
subject receives, the considerations now take on a relative perspective. A famous quote
of Helson (1964) states that a human’s “perceptual apparatus is attuned to the evaluation
of changes or differences rather than to the evaluation of absolute magnitudes. When we
respond to attributes such as brightness, loudness, or temperature, the past and present
context of experience defines an adaptation level, or reference point, and stimuli are per-
ceived in relation to this reference point”.311 While this concerns the general way of per-
308 Ockenfels (2007, p. 3) pointed out that to be different from Homo economicus does not imply that people
behave irrationally or chaotic. Humans follow their own rationality, which may differ from pure self-
interest; but their behavior still is systematic, predictable and, thus, can be modeled.
309 In their contribution Camerer and Fehr (2004) pointed out the important role of experimental methods
in this field of research. Within an experiment it is possible to “carefully control players’ strategies,
information, and possible payoffs” (Camerer and Fehr, 2004, p. 90). This secures the necessary validity
for identifying the behavioral aspects.
310 Levitt and Dubner (2009, 2011) explain manifold and in detail that “people aren’t ’good’ or ’bad’. People
are people, and they respond to incentives. They can nearly always be manipulated - for good or ill - if
only you find the right levers” (Levitt and Dubner, 2011, p. 125).
311 Kahneman and Tversky (1979, p. 227) illustrated this vividly as “an object at a given temperature may be
experienced as hot or cold to the touch depending on the temperature to which one has adapted.”
153
ception, the “same principle applies to non-sensory attributes such as health, prestige,
and wealth. The same level of wealth, for example, may imply abject poverty for one
person and great riches for another - depending on their current assets” (Kahneman and
Tversky, 1979, p. 227). This insight is not only true for laboratory research or a specific
environment. “Overwhelming evidence shows that humans are often more sensitive to
how an outcome differs from some reference level than to the absolute level of the out-
come itself” (Rabin, 1998, p. 4).
ABSOLUTE VS . RELATIVE GAINS
The experimental laboratory resembles a rather abstract environment for participants.

They are admitted to a variety of roles and have to act accordingly. To judge and evaluate
in this setup may be a challenging task. It is thus not surprising that subjects aim to
assess their position with respect to a certain benchmark; while subjects try to evaluate
their relative position, these “individuals desire to occupy a (subjectively) better position
than their peers” (Ok and Kockesen, 2000, p. 533).312 This attempt to be relatively better
off is a continuous pattern found in many experiments (Kurzban and Houser, 2005). It
is sometimes referred to as competitiveness (e.g., Crawford, 1985; Mitchell, 1985) and
sometimes, more negatively, as envy (e.g., Kirchsteiger, 1994; Beckman et al., 2002) or
spitefulness (e.g., Levine, 1997).
Using lottery and money-transfer games Kuziemko et al. (2011) identified a “last-place
aversion”. They found that the last-place player and the second-to-last-player (in terms
of an income ranking) were willing to bear a high risk for the chance of moving up in rank
in order to defend their position. The authors linked their results to the empirical ques-
tion of why especially low-income individuals often oppose more redistributive policies.
They concluded that people do not only passively compare their relative situation, but
actively defend their interests. Ockenfels (1999, p. 2) argued that “pure altruism” is rare
and that altruistic behavior is more likely to be driven by concerns for a specific relative
position.
Looking at the importance of relative positions “one ubiquitous pattern stands out: [...]
in a wide variety of domains, people are more averse to losses than they are attracted to
same-sized gains” (Rabin, 1998, p. 5). The best known example of this is the work of Kah-
nemann and Tversky (Kahneman and Tversky, 1979; Tversky and Kahneman, 1981).313
These researchers showed that human choices depend on the outlook, or respectively the
prospect, of the reference environment. It is crucial in which context or frame a person
has to make a decision. Potential gains or potential losses are perceived differently and
people try to prevent losses even at high costs.314
312 This insight is a well-known fact which goes back at least to the “relative income hypothesis” of Veblen
(1899).
313 Cf. Heukelom (2007) for an overview on their work and the origin of behavioral economics.
314 The above discussed and recognized patterns of behavior of endowment effect and status quo bias are
perfectly in line with loss aversion.
154
In summary, humans look for reference levels to make evaluations. The more unusual
or complex a decision situation, the more this occurs (Dufwenberg et al., 2008; Schmidt,
2010). My experiment did not introduce an SQ which could have served as a focal point.
Thus, subjects with such (comparison) intentions were left with only one possible ref-
erence point, the payoffs of the other subjects. Lopomo and Ok (2001, p. 2) described
such subjects as “interdependent” because their utility depends not only on their abso-
lute level of earnings but also on the relative share of the total surplus. The question is
with which intentions a participant assessed the other players’ payoffs.
6.1.2 Social preferences
Camerer and Fehr (2004, p. 90) defined social preferences as “how players feel when oth-
ers earn more or less money” than themselves. Many experimental investigations have
found empirical evidence of such concerns (e.g., Charness and Rabin, 2002; Fischbacher
et al., 2008). Those “are demonstrated to play a complex role in explaining cooperative
behavior” (Frohlich and Oppenheimer, 1996, p. 117).315 Most importantly, they could be
identified in various settings and forms;316 so Löwenstein et al. (1989, p. 426) reasoned
that “people care about the outcomes of others. We sacrifice our own interests to help
loved ones or harm adversaries.” While many studies analyzed 2-player games (e.g.,
Bolton and Ockenfels, 2000; Fehr and Fischbacher, 2003) this also applies to collective
decisions.317 In particular, S&K (2010) explicitly designed their experiment to test for
social preferences in majority decisions. Overall, Fehr and Fischbacher (2002) pointed
out that the failure to model social preferences may misguide the researcher in answer-
ing fundamental economic questions.318 Interestingly, Diermeier and Gailmard (2006)
concluded that social preferences may exist but depend on the reservation value and
decision-making process. This is another good reason to investigate my data for social
preferences; I do not specify a reservation value and use different voting procedures.
SOCIAL WELFARE PREFERENCES
The most prominent models of social preferences implement a self-centered measure for
fairness (Cox and Sadiraj, 2012) in which a subject’s deviation from the payoff of others
or the average payoff drives its utility. A closely related but different approach was put
forward by Andreoni and Miller (2002) and Charness and Rabin (2002). These authors
315 A common finding for collective settings is that in public good games a compulsory fee which is enforced
on everyone if an initial subgroup reaches the threshold increases the chances of provision of the public
good (cf. Dawes et al., 1986).
316 Sec. A.14 gives an overview on experimental games which have frequently been used to measure social
preferences.
317 In general, the existing literature finds social preferences to be stronger in subjects who interact in small
groups than in subjects who interact in large groups (Fehr and Schmidt, 1999). Yet, this statement is based
on comparing very different sets of decision rules. On the one hand, we find highly structured 2-player
games, whereas market competition dominates the study of larger groups (Schmidt, 2010).
318 I will not deny that there are also more pessimistic voices. For a critical review on the empirical evidence
for social preferences cf. List (2009).
155
advocate a model of social welfare preferences which is characterized by individuals who

care about the size of the pie, i.e., the actors attach a positive weight toward the aggre-
gated surplus off all participants.319 Accordingly, they choose the alternative which max-
imizes their utility taking into account the sum of all actors’ payoffs. Charness and Rabin
(2002) carried out 32 experiments and found strong support for this model. However,
S&K (2010), whose design is very similar to mine, found no evidence in favor in their ex-
periment where “social welfare seems to be a poor explanation for behavior in majority
decision making” (S&K, 2010, p. 675). I therefore apply a model of self-centered fairness.
However, this model will also be able to reflect social welfare preferences if they prevail.
INEQUALITY AVERSION
In Sec. 5.1 I introduce the ERC model of fairness. Yet, in my individual level analysis I
opted for a similar but different approach to incorporate social preferences. I used the
model of inequality aversion proposed by Fehr and Schmidt (1999). It is formulated as
stated in Eqn. 6.1. An individual’s utility consists of three components: their own payoff
yi , the difference between their own and another’s payoff if the individual gets less Di−
and the difference between their own and another’s payoff if the individual gets more
Di+ . Subsequently, I refer to Di+ as advantageous inequality and to Di− as disadvanta-
geous inequality.320 In the experiment the construction of the payoff tables constrains
Di+ to the interval [0; 100] and Di− to the interval [0; 20].
Ui (yi ) = αi yi − β i Di− − δi Di+ (6.1)
, where Di− = 1
∑ j̸=i max y j − yi , 0 and Di+ = 1
∑ j̸=i max yi − y j , 0 .
   
n −1 n −1
How does this model differ from ERC? Instead of comparing actor i’s share to the glob-
ally equal distribution ( n1 ), the inequality aversion model suggests a pairwise compari-
son to all other actors j ̸= i. For example, imagine a five-player committee decision in
which player i = 1 is pivotal and able to choose between two alternatives with the payoff
vector { A} = {10, 20, 20, 0, 0} and { B} = {10, 10, 10, 10, 10}.321 The ERC model cannot
discriminate between the two alternatives. No matter how important social preferences
1
are, player 1 gets a fair share of 5 with either of the two alternatives. Thus, ERC ignores
any inequality between other players (Engelmann and Strobel, 2000, p. 8). By contrast,
in the inequality aversion model Ui depends on the exact values of δi and βi . Although
both models lead to similar predictions for two-player games, the predictions differ for
multiplayer games (cf. Engelmann and Strobel, 2004).
In their theoretical discussion Fehr and Schmidt (1999, p. 822) assumed that i) 0 < δi < αi
and ii) δi < β i . The first assumption rules out two types of players: those who like being
319 Thisapplies only in the case of settings where side payments are not possible. Otherwise such behavior
could also be led by strategic, instead of social, considerations.
320 Goeree et al. (2002, p. 267) used the designations of “guilt” and “envy” to characterize the parameters.
321 The underlined numbers indicate the payoff of player i = 1.
156
better off than others and those who place a higher importance on their relative gains
than on their absolute gains. The second assumption states that a relative loss diminishes
the utility more than an equally sized relative gain increases it, i.e., “losses resonate more
than gains” (Rabin, 1998, p. 5). From a theoretical perspective, it is important to note
that ERC and the inequality aversion concept have been developed for games in which
a subject’s preference for advantageous inequality does not alter equilibrium behavior,
in particular the ultimatum game, the dictator game and public good games (Fehr and
Schmidt, 1999, p. 851).322 Accordingly, any theoretical discussion starts with the assump-
tion that if social preferences affect individual behavior, they must reflect a concern for
fairness and reciprocity.
This assumption does not hold true for plurality voting under majority rule as applied in
my experiment. Here, the more general question is how far players are willing to forgo an
increase in absolute payoffs to improve their own or other players’ relative positions. To
answer this question I could not use the ERC model as it does not differentiate between
relative gains and losses. This was the main argument for using the inequality aversion
model of Fehr and Schmidt (1999). However, the originally provided value ranges for
the model parameters restrict the possibility to measure those aspects by focusing on
socially desirable intentions. Yet, as the model provides both parameters nonetheless, I
also used them to investigate possible envious preferences. I therefore did not implement
in advance the defined range constraints but left them arbitrary.323
6.1.3 What makes people social?
Scientific research concerned with evolutionary dynamics and human nature indicates
that mankind is far from homogeneous (Buller, 2005; Weibull, 1995).324 This has led to
an “increasing emphasis on the importance of individual differences in understanding
and modeling behavior and dynamics in experimental games and decision problems”
(Kurzban and Houser, 2005, p. 1803). In the context of social preferences this leads to
the question of if and how individuals vary in their degree of cooperativeness. Looking
at the interface between personality psychology and economics, Borghans et al. (2008)
pointed out that in psychology it is a well-known fact that certain personality traits are
more malleable than cognitive ability over the life cycle.
“Measuring the magnitude of the concern people have for others, sometimes called Social
Value Orientation (SVO), has been an interest of many social scientists for decades” (Mur-
phy et al., 2011, p. 771). This concept highlights the role of individual characteristics for
322 Not surprisingly, games which showed the clearest evidence for the existence of social preference are
dictator and ultimatum games (cf. Camerer and Thaler, 1955).
323 Even Fehr and Schmidt (1999, p. 850) acknowledged that the assumption, according to which players who
like to be better off than others do not exist, is “unsatisfactory”.

324 Questioning standard theorems of evolutionary psychology, Buller (2005) argued that human minds are
not calibrated to one prehistoric base but continually adapt during both evolutionary time and individual
lifetimes. This leads to the large number of varieties and generates a polymorphic population.
157
decision-making. Here, Bradler (2009, p. i) observed strong reactions “when the payoffs
for both subjects can be perfectly equalized.” She also found a decrease of cooperative-
ness when the other player was better off. Both aspects support the idea of inequality
aversion.325 Overall, she concludes “that a need for fairness and equality fundamentally
influences individual decision-making processes” (Bradler, 2009, p. i).
Social psychology literature (e.g., SVO) suggests that people can be classified as competi-
tors, cooperators, and individualists (Komorita and Parks, 1995). This research fits very
well with work in experimental as well as behavioral economics that classifies people
as “spiteful” (Herrmann and Orzen, 2008; Saijo and Nakamura, 1995), “selfishly payoff-
maximizing” (Eckel and Grossman, 1998) or “altruistic” (Cason et al., 2004). Applying
evolutionary simulations, Lomborg (1996) calculated a stable population of three types:
cooperators, cautious cooperators and non-cooperators. This is exemplary for the lit-
erature in this field across the different disciplines. Most contributions end up with
three different types: i) spiteful competitors trying to achieve better payoffs than others
(cf. Sec. 6.1.1), ii) cooperators caring for others or aiming for a high welfare of the group
(cf. Sec. 6.1.2), and iii) individualists who focus just on themselves.
Of course, once you classify people into these categories you are also interested in their
proportion of appearance.326 Here, Löwenstein et al. (1989) conducted an experiment in
which subjects had to review different negotiations. They varied the dispute type be-
tween personal and business as well as between relationships designated as positive and
negative. They categorize their subjects “according to their preferences toward advan-
tageous inequality” (Löwenstein et al., 1989, p. 438). They identified saints, who pre-
ferred equality over inequality, regardless of the relationship, and loyalists, who pre-
ferred equality in positive but advantageous inequality in negative relationships. Ruth-
less competitors sought advantageous inequality under all conditions. The proportions
of their sample amounted to 22% saints, 39% loyalists and 29% competitors. Andreoni
and Miller (2002) tested the consistency of altruism in a modified version of the dictator
game. They found that selfish preferences accounted for about half of the subjects, while
about 33% of subjects divided equally between both players, and another 20% gave most
tokens to the person with the highest redemption value. With public good games and an
type-classification algorithm Kurzban and Houser (2005) identified 20% of their subjects
as free-riders, 13% as cooperators and 63% as reciprocators. The authors pointed out that
their distribution of types is similar to the findings of Fischbacher et al. (2001).
Fehr et al. (2001, 2002b) showed that these different types can be modeled accordingly
by allowing for heterogeneous social preferences (HSP). In their work the explanatory
power of the statistical model rests on the interplay of strictly egoistic as well as inequity-
averse subjects. Traub et al. (2009) studied voting on redistribution. They found that in-
325 On the one hand, subjects are willing to help others “even if it is costly to them” (Bradler, 2009, p. 17). Yet,
on the other hand, envy causes a crowding out of benevolent behavior (Güth et al., 1982).
326 Kurzban and Houser (2005) offered a comprehensive overview of literature in this field.
158
corporating HSP improves the predictions significantly compared to the standard model.
Erlei (2004) and Dittrich and Ziegelmayer (2010) are further examples that build on the
assumption of HSP. Dittrich and Ziegelmayer (2010) originally conducted bilateral gift
exchange experiments to investigate the impact of loss aversion. After adjusting their
theoretical predictions for social preferences, the authors aimed to identify the social in-
tentions of individuals by endogenous types using a stochastic choice model. Their data
can be explained best when allowing for at least four types of HSP. A promising ap-
proach is Erlei (2004), who combined social preferences, social-welfare preferences and
reciprocity into one model of heterogeneous actors.327
With respect to my research it is important if these differences in people’s fairness con-

cerns matter for majority decisions. Hoechtl et al. (2012) argued that this depends on
whether fair-minded voters are pivotal, which in turn is dependent on the distribution of
voters and the design of the electoral system. Thus, the authors concluded that “fairness
concerns matter when few fair-minded voters are sufficient to tip the balance in majority
voting, but do not matter much when many are needed” (Hoechtl et al., 2012, p. 1416).
To summarize, the literature agrees that individuals vary in their degree of cooperation.
The explanatory power of multiple models has been improved by accounting for this
heterogeneity and allowing for deviations from rational choice. “However, there is little
agreement concerning the best abstraction and the relative importance of the distinct
psychological factors” (Ert et al., 2011, p. 258). In Sec. 6.3 I discuss co-variates whose
impact on social considerations was assessed when estimating the statistical model.
6.1.4 Reciprocity
An important fact, which was also discovered in experimental behavioral research, is

the dependency of behavior on accountability and context.328 In other words, the role of
personal involvement is of immense importance to an individual’s actions (Branas-Garza
et al., 2009). This is aptly illustrated by Dana et al. (2007) who have found that reducing
accountability for actions in dictator games leads to significantly less generous behavior.
The insights discovered on reciprocity “constitute a departure from neoclassical theory”
(Cox and Sadiraj, 2012, p. 927). Here, depending on the prior actions of others, the in-
tentions of subjects are altered. Deviating from “the conventional assumption that these
preferences are stable [, ...] context-dependent preferences can capture the possibility that
agents are motivated in part by reciprocity” (Sobel, 2005, p. 392).
With respect to the aforementioned social preferences it is important “to discriminate

between behavior motivated by reciprocity and behavior motivated by nonreciprocal
327 Perhaps the most unambiguous evidence for the usefulness of this approach is its ability to explain many
behavioral anomalies (Goeree and Holt, 2001).
328 Fehr and Schmidt (2000, p. 1) summarized that “in recent years experimental economists have gathered
overwhelming evidence that systematically refutes the self-interest hypothesis and suggests that many
people are strongly motivated by concerns for fairness and reciprocity.”
159
other-regarding preferences” (Cox and Deck, 2005, p. 633). Altruism is very different
from reciprocity. The first implies that an actor takes a costly action to increase the wel-
fare of others irrespective of their prior behavior (Bradler, 2009). Reciprocity, on the other
hand, means non-selfish action conditioned on the actions of others (Camerer and Fehr,
2004, p. 56). Such corresponding behavior is typically observed among principals and
agents.329 Most importantly, this might extend to vindictive actions as hostile acts pro-
voke punishment intentions (Herne et al., 2012; Rabin, 1993).
With regard to the literature on reciprocity,330 a wide variety of papers in economics (e.g.,
Andreoni et al., 2003; Bandiera et al., 2005), especially behavioral economics (e.g., Al-
Ubaydli and Lee, 2012; Charness, 2004), and psychology (e.g., Baumeister et al., 2001)
investigated the possible shape of a “reciprocity function”. Although until now not all
questions have been answered and a lot of research is still being conducted,331 two main
conclusions seem valid. Firstly, the shape of the “reciprocity function” is concave332 (Al-
Ubaydli and Lee, 2009; Bellemare and Kroger, 2007) and secondly, that negative reci-
procity is clearly stronger than positive reciprocity (Baumeister et al., 2001; Offerman,
2002). Again, this corresponds well with the previously discussed human perspective
that “losses resonate more than gains” (Rabin, 1998, p. 5).
Above all, the decisive aspect with respect to reciprocity is monitoring (Al-Ubaydli et al.,
2010). Only when one observes the actions of others can one respond appropriately.333
This holds true for individuals who are monitoring as well as being monitored (cf. Bandiera
et al., 2005). Yet, it makes reciprocity less crucial for my experiment as my design explic-
itly introduces restricted monitoring capabilities. When deciding simultaneously, neither
of the two delegations observes the vote of the other; and even when under SEQ the
second stage is informed about the first stage decision before its vote, it is not possible
to attribute the outcome to an individual but only to the complete first delegation. The
applied majority bargaining makes the accountability of results towards one individual
difficult and, thus, complicates a responsive reaction.334
6.1.5 Risk aversion
Up to now, I have determined that it is necessary to control for social preferences when
assessing the individual votes of my experiment. A second, also important finding in
329 As found, e.g., by Dittrich and Kocher (2011) when looking into a shirking game.
330 Cf. Ockenfels (1999, Chap. V) for a comprehensive overview on reciprocal behavior.
331 Cf. Cox and Sadiraj (2012) or Nicklisch and Wolff (2012) for an up-to-date overview.
332 A concave reciprocity function implies that more trust, confidence or benevolence leads to more reci-
procity, but at a diminishing rate.

333 With respect to what ’respond appropriately’ means, Rabin (1993, p. 1281) concluded that “people like to
help people who are helping them, and to hurt those who are hurting them.”
334 Many findings of behavioral patterns in laboratory experiments presume the comparison of one’s own
payoff with others. Yet, such an evaluation becomes complicated when the setting is moved away from
standard two-player games (Boyd and Richerson, 1988).
160
behavioral research is missing so far: risk aversion.335 Most prominently, Kahneman

and Tversky (1979) discovered that subjects reveal risk averse preferences over relative
gains. A large amount of literature argues that risk averse actors chose alternatives which
maximize their minimal payoff or minimize their maximal loss (e.g., Palacios-Huerta and
Volij, 2008).336 In ultimatum games the “fear of rejection” (Lopomo and Ok, 2001, p. 2) has
been found to be responsible for a large part of the results (Camerer, 2003).337 While game
theory predicts that the proposer receives (nearly) all of the pie, this is rarely observed in
empirical research (Bahry and Wilson, 2006). However, recent studies reported that the
“lemon avoidance heuristic“ (Ert and Erev, 2008) is also unusual.
Importantly, a subject’s risk aversion is inherently a relative concept. Under conditions
of uncertainty, subjects are not willing to risk being worse off than either their peers
or the SQ.338 However, one should not treat risk and inequality attitudes as equivalent.
Experimental evidence suggests that these do not necessarily correspond (Traub et al.,
2009). Thus, both should be assessed separately.
Yet, before I followed this approach I had to determine if risk aversion was relevant in
my experimental design; but this assessment could be made manifestly. The level of
uncertainty and its variation represents, besides complexity, the most prominent aspects
and central treatment in my experiment. Thus, I could not leave such a commonly found
aspect of human behavior out of my analysis.339 In addition to social preferences I also
controlled for risk aversion.
The standard approach of estimating risk aversion resonates as constant relative risk
aversion (CRRA) where the utility of a subject for a given sum of money y is represented
as in Eqn. 6.2 (Morton and Williams, 2010, p. 274).340 Using this formulation a subject’s
revealed risk aversion parameter r is assumed to be constant and relative across the pos-
sible range of payments. Following Holt and Laury (2002) I assumed that for r = 1 the
natural logarithm is employed.341
(1−r )
yi
Ui (y) = (6.2)
(1 − r )
335 The above discussed framing effects and prospect theory (cf. Sec. 6.1.1) go hand in hand with the concept
of risk aversion (Cox and Harrison, 2008; Heinemann, 2005) and the construct of concave utility functions
(Chajewska and Koller, 1999; Gorman, 1968).
336 Palacios-Huerta and Volij (2008) identified this strategy in strategic games with professional soccer play-
ers. The authors also clarified two designations, minimax and maximin; both pursue the same goal. Min-
imax is a decision rule that aims to minimize the possible maximum loss. Alternatively, maximin tries
to maximize the possible minimum gain. Maximum loss and minimum gain refer both to the worst case
scenario in which the subjects wants to be as successful as possible.
337 Other important aspects are, of course, fairness and also partially confusion (Prasnikar, 1997).
338 Schmidt (2010, p. 9) used the term “social risk aversion” to describe this tendency.
339 Cf. Harrison and Rutstroem (2008) for a review on experimental evidence on risk aversion.
340 Cf. Eeckhoudt et al. (2005) for a comprehensive overview on how to recognize, quantify, analyze and in-
corporate risk into decision-making processes. The volume focuses on economic and financial decisions
as, e.g., portfolio choices.
341 Holt and Laury (2002) also pointed out that the division by (1 − r ) is necessary for increasing the utility
when it holds that r > 1.
161
When estimating the CRRA parameter, r < 0 indicates a risk seeking individual, r = 0
a risk neutral person and r > 0 a risk averse character. Although I estimated the CRRA
of my subjects, I am aware of the difficulty of generalizing it. Berg et al. (2005, p. 4211)
pointed out “that researchers must be extremely careful in extrapolating a person’s or
group of persons’, risk preferences from one institution to another. Without appropriate
benchmarks on the preferences of individuals, researchers can mistake changes in be-
havior caused by risk preferences for change in behavior caused by other stimuli such as
information or rule changes.” The inclusion of risk aversion did not serve the purpose
of comparing its value to other studies but rather of accounting for a qualitative pattern
in subjects’ behavior (cf. Sec. 4.3.3). In other words, it corresponds to the inclusion of a
control variable which checks whether the results are robust for risk averse preferences.
6.2 Sincere and sophisticated voting
In economics, the market is the single most important mechanism for collective choice.
Here, Smith (1962, 1964) clarified that social preferences are irrelevant in perfectly trans-
parent markets, when separable preferences are assumed, because they do not affect
individual equilibrium strategies. Accordingly, Dufwenberg et al. (2008) investigated
whether agents with altruistic preferences differ from selfish agents in perfectly com-
petitive markets. They defined a separability condition on the basis of whether agents’
preferences can be represented by a weighted sum of internal utility functions. If the
condition holds, “agents who care directly about the welfare and opportunities of others
cannot be distinguished from selfish agents in market settings” (Dufwenberg et al., 2008,
p. 631). In other words, it is impossible for any subject to enforce a fair or equal outcome
if there is competition. By insisting on their fair share, the players only hurt themselves;
they cannot prevent the other market participants from trading. Thus, a market in which
agents have social preferences is observationally equivalent to a market in which each
agent only cares about themselves (Schmidt, 2010, p. 6).
This changes once we move our perspective away from the perfect “coordination through
the invisible hand of the price mechanism” (Larsson, 1993, p. 87). In the realm of politics,
majority decisions replace markets as the single most important choice mechanism (Er-
lenmaier and Gersbach, 2001, p. 2).342 Of course, most democracies do not actually use
simple majority voting in their legislative processes but rather stabilize them by a system
of checks and balances, the division of power and representative elements (Beitz, 1990,
Chap. 4, p. 60). These systems allow for a certain degree of crossover between voting and
trading, but standard plurality voting does not; here, votes are cast simultaneously. This
results in complex and uncertain environments.343
342 Majority voting constitutes the most widely spread decision rule for democratic government. Any democ-
racy is commonly associated with political equality and majority rule (Saunders, 2010a,b).
343 In economics, these aspects are summed up as nontransparent markets which cause risk aversion (Levy,
2006; Pratt, 1964).
162
With a view to general elections, the literature considers two types of voters or, respec-
tively, voter behavior (cf. Clinton and Meirowitz, 2004; Herrmann, 2012).344 At one ex-
treme are those who give up any tactical considerations right away and decide to vote
sincerely. Such a voter neither considers another subject’s choice nor the logic of the col-
lective decision problem; they are solely focused on their own payoffs. For the sincere
type, discriminating between social preferences and self-interest is easily accomplished
as long as the importance of social preferences is strong enough to influence individuals’
choices. At the other extreme are those voters who make an effort not to waste their vote
on a minority position and who therefore behave strategically (cf. Kramer, 1972).345 As-
sessing the prevalence and realism of this voting type, Riker (1982a, p. 169) claimed that
such “strategic voting is an ineradicable possibility in all voting systems [and] almost
always present in legislatures”. Costa-Gomes et al. (2001, p. 1193) defined this sophis-
tication as “the extent to which behavior in games reflects attempts to predict others’
decision, taking their incentives into account”.346
Yet, even with perfect information on their own and others’ preferences, voters will find
it difficult to judge the consequences of their vote choice for two reasons. First, even if
there is a unique equilibrium strategy, it may be difficult to identify. Second, others have
the same problem, which only reinforces the voter’s struggle with their own strategy.
This uncertainty increases with the number of alternatives put to vote, the size of the
electorate, and the possible arrangements of preferences among voters (Ordeshook and
Palfrey, 1988).347 Therefore it becomes difficult to infer preferences from given votes.
In my experiment the participants face a complex and uncertain environment. When

investigating behavior displayed under such a challenging task, the key problem is to
separate the effect of social preferences from subjects’ cognitive short comings and hu-
man error (Goeree et al., 2002). The additional explanatory value of, e.g., incorporating
fairness and extending the utility function (cf. Eqn. 6.1) depends on subjects’ ability to
calculate their resulting advantageous and disadvantageous inequality, which decreases
with rising complexity.348
One solution to this problem is an experimental design which aims at controlling for
social preferences. For example, in Johnson et al. (2002) other-regarding preferences of
subjects were turned off by introducing robots as bargaining partners. Jacquemet and
344 Cf. Merrill and Grofman (1999, part I) for an introduction to (spatial) models of voter behavior.
345 Cf. Costa-Gomes et al. (2001) for an extensive overview on literature which studied strategic behavior and
Herrmann (2012, p. 64) for contributions which have demonstrated strategic voting empirically.
346 Another incisive description was given by Kramer (1972, p. 170) who clarified that for a sophisticated
strategy “it is necessary for a voter, to make the best use of his vote, to attempt to predict the contingency
likely to arise: that is to say, how the others are likely to vote.”
347 Ordeshook and Palfrey (1988, p. 441) pointed out that the distinction for sincere and strategic voting is not
limited to elections but “represents an important contribution to our understanding of committees, of

institutions, and of the opportunities to manipulate outcomes by the manipulation of institutions.”
348 Moreover, subjects now have to form beliefs regarding the relative emphasis other players place on social
preferences as compared to their self-interest (Andersson et al., 2012). Obviously, the latter argument
only concerns sophisticated types who try to anticipate other players’ strategies and actions.
163
Zylbersztejn (2011) neutralized the effect of relative comparisons by altering the pay-
off structure towards more symmetry between their players. Another solution to this
methodological problem is offered by Diermeier and Gailmard (2006) who imposed a
bargaining protocol including a random assignment of roles and role specific reservation
values. Subjects are then aware of their chances for either role as well as their actual
role at the moment they make their particular choice. In summary, in situations which
preclude either sophisticated voting or social motivation the model parameters of the
other can be directly estimated. This does not hold true for plurality voting under major-
ity rule, as in my design, where a subject’s cognitive capacity is decisive for making the
necessary (other-regarding) computations.
BOUNDED RATIONALITY
The distinction between sincere and sophisticated voting is supported by a further argu-
ment: the cognitive capacity of every individual is limited. This insight and the desig-
nation as bounded rationality go back to Simon (1957) who raised the questions of “how
do human beings reason when the conditions for rationality postulated by the model of
neoclassical economics theory are not met?” (Gigerenzer, 2008, p. 124). The subsequent
research program argued that humans learn cognitive shortcuts in everyday exercises
(Gigerenzer, 2008). Such heuristics guide their behavior with the result of more or less ac-
ceptable outcomes. Yet, they definitely prevent lavish considerations (Kahneman, 2011)
by turning to “decision rules that reduces the computational burden to feasible levels”
(El-Gamal and Grether, 1995, p. 1137). The question is what heuristics people use and in
what environment they work or fail (Gigerenzer and Selten, 2001). Current research indi-
cates that “a) individuals and organizations often rely on simple heuristics in an adaptive
way, and b) ignoring part of the information can lead to more accurate judgments than
weighting and adding all information, for instance for low predictability and small sam-
ples” (Gigerenzer and Gaissmaier, 2011, p. 451).
In addition to the confinement of humans’ cognitive capacity in general, a laboratory

experiment may restrict it further. Acting in experimental settings can be an awkward
task depending on the game played. In my setup, subjects had to cope with four differ-
ent decision situations, which introduced different levels of complexity and uncertainty
(cf. Tab. 4.2). Taking all together, under such circumstances an actor often cannot under-
take all required computations as its rationality is bound by cognitive limitations.
What does this mean for sincere and sophisticated voting in my experiment? The higher
the level of complexity or uncertainty, the more sufficient simple approaches will be to
understand the observed behavior. Contrary to this, any sophisticated equilibrium strat-
egy will diminish because “uncertainty constrains the statistical relationship between
their strategy choices players can bring about” (Crawford and Haller, 1990, p. 571). Thus,
rational predictions were expected to explain more voting behavior under POL than un-
der SIM as well as under the second stage of SEQ in comparison to the first one.
164
Yet, if basic patterns become more important, which will that be? Many heuristics as-
sume a pre-existing social environment such as peer groups or a successful role model
whose behavior will be imitated (cf. Kahneman et al., 1982). In the controlled environ-
ment of the laboratory this is not feasible. Also widespread are actors who choose the
alternative promising the highest payoff; in other words, just “take the best” (Gigerenzer
and Goldstein, 1999; Gräfe and Armstrong, 2010). Yet, in iterated games this choice will
then be conditioned upon each alternative’s cue validity (Rosch and Lloyd, 1978), i.e., an
alternatives chance for success in a majority vote. In order to be selected an alternative
2
needed at least 3 of all participants voting for it.
WINNING COALITION
In Sec. 6.1.1 I discuss humans’ inherent compulsion for comparison. When talking about
the general principle, the aspect of group size seems unimportant, i.e., people compare
themselves with one or several others all the time (e.g., Lopomo and Ok, 2001; Azar,
2010). Yet, when looking at the exact strength of this influence, the size of the reference
group may well play a role. Levine (1997) examined results of ultimatum and centipede
experiments with respect to altruism and spitefulness. He aimed to access the disper-
sion of these characteristics in the sample by varying the number of participants. While
altruism was independent of the number of participants, he found less spitefulness in
larger groups. Levine (1997, p. 614) argued that “one explanation of spite is that it is
really competitiveness, that is, the desire to outdo opponents. In this case, it is not the
total utility of opponents that matters, but some measure of their average or maximum
utility.” Thus, even if it seems reasonable to expect subjects to evaluate their payoffs in
comparison to other players’ payoffs, it is not clear how subjects perceive and evaluate
these other payoffs.
Investigating centipede games, Tremewan et al. (2012) identified differences of experi-
mental subjects with respect to ingroups and outgroups.349 In my bargaining game under
majority rule the collective agreements were reached within a minimal winning coalition
between 80-96% of all decisions (cf. Sec. 5.4). The majority rule splits the subjects into an
ingroup, i.e., the winning coalition, and an outgroup, i.e., the outvoted subjects. To be-
long to this winning coalition is an understandable desire for every subject as it has the
advantage of being able to co-decide which alternative gets selected. Outvoted players
have no influence on the collective result. This puts the subjects into a strategic dilemma.
From the above we know that the relative position is important (Sec. 6.1.1), so subjects
avoid low payoffs as others would be relatively better off.350 On the other hand, if they
aim too high, they might be outvoted by the other subjects and would probably end up
with a very low turnout. Considering the previously discussed risk aversion (Sec. 6.1.5)
349 Focusing on level-k estimations Tremewan et al. (2012) discovered a difference in beliefs about other play-
ers and emphasized the role of uncertainty. Participants behaved as if they could completely predict the
actions of ingroup members but only partly the behavior of outgroup members.
350 In non-constant-sum tables it is possible that a low payoff still goes along with being better off than others.
But this is only possible in alternatives with a very low overall sum and, thus, low cue validity.
165
subjects were expected to try hard not to be left out and to keep someone beneath them.
In their ultimatum experiment Güth et al. (1982) found that proposers just maximized
given their fear of rejection. I anticipated that these considerations would also take place:
subjects had to take into account the voting threshold when deciding on the alternatives.
Therefore, I considered it likely that only alternatives receive votes which offer at least
four players an appealing amount of points.351 In my statistical model I controlled for this
by implementing a specific winning coalition dummy in the strategic voting component.
6.3 Statistical model
Sec. 6.1 and Sec. 6.2 discussed well-known and verified patterns of behavior discovered
in laboratory experiments. To assess their relevance for my design, I incorporated them
into different model specifications and compared their relative performance in explain-
ing the experimental results. Thus, I explored different utility functions of which each
includes specific parameters to uncover the pattern that best matches my observations.
This is similar to the approach of Ert et al. (2011) and their choice prediction competition
for models of social preferences and to Smirnov (2009) who conducted a voting experi-
ment and compared the explanatory power of random, sincere, strategic and risk averse
models. Also, Goeree et al. (2002, p. 265) compared “likelihoods” of variant models to
determine the adequacy of behavioral assumptions on altruism.
The underlying idea of the model to be estimated can be summarized as follows. When
investigating how individuals decide when confronted with a discrete choice problem,
the simplest form of model is represented by a standard conditional logit model. Here,
an individual compares their payoffs across all alternatives to reach a decision. Corre-
spondingly, when looking at individual behavior within collective decision-making, the
most basic model takes into account only each player’s own gains. Yet, subjects who also
consider the payoffs for other players and act accordingly strategically can be more suc-
cessful. Therefore, the payoffs of single alternatives have to be weighted with respect to
their probability of constituting the collective choice.352 Following my theoretical solu-
tion concept, the weight corresponds to the probability of an alternative to represent the
core.353 More specifically, the weight is determined as the probability of an alternative j
to survive against all other alternatives in a pairwise comparison across all possible win-
ning coalitions. In other words, if there exists just one other alternative k which at least
one specific player constellation with a sufficient number of players prefers over j, the
351 Under both delegation procedures two sub-committees have to decide with at least two out of three votes.
Thus, under all procedures an alternative must have received at least four votes to be selected.
352 For example, consider a collective choice of six players and an alternative like {88, 1, 1, 1, 1, 1} with five
low and just one high number of points. While this option might be tempting for player 1 it is certainly
not for the other players. Therefore, strategic considerations assign only a low probability of being the
collective choice to this alternative.
353 This approach is different from the one I applied in Chap. 5 where I used a deterministic definition. In the
following, the core alternative is determined probabilistically.
166
not-preferred alternative is not the core because the core does not lose in any pairwise
comparison. This transformation follows the idea of QRE which adjusted the basic logit
model for players’ errors (McKelvey and Palfrey, 1995, 1998). Yet, it is unrealistic to ex-
pect that all participants behave that way. Therefore, a second transformation allows for
sincere and sophisticated behavior (Sec. 6.2). This mixture stage has the advantageous
properties that one does not need to know the distribution of behavior354 and that the
model allows for observed as well as unobserved heterogeneity.
Next, Sec. 6.3.1 explains the implementation of the behavioral parameters in a utility
function. The operationalization of the random utility model is shown in Sec. 6.3.2. Here,
I derive step-by-step the probabilistic individual choice function for sincere and sophis-
ticated voting. Next, Sec. 6.3.3 introduces the mixture stage of the model and discusses
various co-variates which might have influenced subjects’ social or strategic considera-
tions. Finally, Sec. 6.3.4 summarizes the different specifications for the empirical analysis.
6.3.1 The robustness of self-interest
This section focuses on the operationalization of the previously discussed behavioral

patterns. The aspects considered are self-interest, social preferences and risk aversion.
Following Bolton and Ockenfels (2000, p. 171), the extended (expected) utility function
might be designated “motivation function because it emphasizes [...] a statement about
the objectives that motivate behavior during the experiment.” The function also contains
a parameter of random shock which depicts a subject’s errors due to a player’s mistakes,
cognitive incapacities, etc.355 This follows Glasgow et al. (2012), who criticized the fre-
quent use of conditional logit models when analyzing government choices. The authors
argued instead in favor of a mixed logit extension which includes random coefficients to
allow for unobserved heterogeneity.356
The utility level Uij of a purely self-interested actor i is shown in Eqn. 6.3. It is solely
given by their payment yij when alternative j is selected.
Uij (yij ) = yij (6.3)

354 When estimating the model I did not know whether or to which extent any player behaved strategically.
355 The interpretation of such shocks differs between disciplines in social sciences. Psychology focuses on why
a subject’s behavior becomes more and more ambiguous when stimuli get weaker (Luce, 1959). Here,
“stochastic elements represent intra-personal variations in utility levels, or perception errors, which may
cause a subject to choose differently when faced with the same stimuli. In econometrics literature [...],
stochastic choice models are usually applied to cross sectional data, containing decisions of many individ-
uals. The stochastic elements are interpreted as inter-personal variation, or heterogeneity, in preferences.
Experimental game theorists are mostly agnostic about the interpretation of the stochastic elements. In
the laboratory, noise may be due to distractions, perception biases, miscalculations, or due to heteroge-
neous preference shocks such as feelings of envy, spite, or altruism. Regardless of the interpretation of
the noise, the effect can be particularly important in an interactive context where players’ payoffs are
sensitive to others’ decisions” (Goeree et al., 2005, p. 363).
356 This argument also relaxes the independence of irrelevant alternatives (IIA) assumption.
167
Next, I looked for the robustness of this baseline assumption against random shocks and
a player’s concern for fairness. Thus, I was interested in the size of the parameters neces-
sary to change the self-interest prediction. As I had no ex-ante knowledge as to whether
a subject would vote sincerely (i.e., without consideration of other subjects’ choices) or
sophisticatedly (i.e., with a solution concept for the collective choice problem in mind),
the operationalization had to accommodate both types.
I implemented the random shock ε i as privately observed, mean zero random distur-
bances.357 Furthermore, the use of independent and identically distributed (iid) extreme
value error terms followed common logistic mixture model parameterization (Azaiez,
2010, p. 7). For sincere voting types the disturbance was operationalized by the relative
value of the best as compared to the second best alternative. With respect to sophisticated
voting types, I used the core as a theoretical solution concept. Thus, the error captured
the necessary change for altering the core prediction. The inclusion of the random shocks
leads to the utility function shown in Eqn. 6.4.
Uij (yij ) = yij + ε i (6.4)
When looking at concerns for fairness, I differentiated between sincere and sophisticated
voting in the same manner. For sincere types I measured fairness by the relative dif-
ference bi between the alternative with the highest score of the subject compared to the
1
average and, thus, (theoretical) fair share of n (cf. Eqn. 5.2).358 As before when consid-
ering random shocks, this operationalization does not take into account the payoffs or
strategies of other players. For sophisticated voting types I used again the core as bench-
mark. Here, I assessed the importance of fairness defined by inequality aversion (Fehr
and Schmidt, 1999) as shown in Eqn. 6.5. In order to overcome the two-dimensionality
of the concept, I identified the combined level of (δi + β i ) necessary to change the core
prediction.
Uij (yij ) = αi yij − β i Dij− − δi Dij+ + ε i (6.5)
In addition to random shocks and fairness I also investigated the impact of subjects’ risk
attitudes as the experimental design provided decision situations at different levels of
uncertainty. Here, I followed the common approach of CRRA and transformed the util-
ity function as discussed in Eqn. 6.2. Risk aversion extends to a subject’s complete utility
function. Thus, it influences an individual’s payment as well as their social preferences,
357 This corresponds to the original specification of the QRE framework (McKelvey and Palfrey, 1995).
358 I used the ERC concept of fairness to account for social preferences of sincere voting types. Such subjects
do not consider other subjects’ choices or payoffs; but that is not necessary as ERC compares an actor’s
share only to the globally equal distribution ( n1 ).
168
which implies that the actor should dislike taking risks that are not taken by their refer-
ence group.359 In this way the degree of risk aversion r transforms Eqn. 6.5 into Eqn. 6.6.
  (1−r )
  αi yij − β i Dij− − δi Dij+ + ε i
Uij yij = (6.6)
(1 − r )
ROBUSTNESS OF THE CORE
The payoff tables possess different degrees of robustness of the core against modifying
the assumption of pure self-interest. This enabled me to estimate the respective param-
eters based on individual votes. Yet, before turning to the model implementation I look
into the correlation between the different aspects. Fig. 6.1 reveals that the robustness
of the core against random shocks and inequality aversion are correlated. The figure
shows also fairness conceptualized along the ERC model (Bolton and Ockenfels, 2000,
cf. Sec. 5.1). It is obvious that the correlation across all three characteristics is high.
Figure 6.1: Robustness of the core alternative

EXPLANATORY NOTE
The graph illustrates the correlation between payoff tables’ robustness of the core against random shocks and fairness
considerations. I use both fairness conceptualizations: inequality aversion (Fehr and Schmidt, 1999) is indicated as △ and
ERC (Bolton and Ockenfels, 2000) is indicated as ⃝. The values are standardized and in the case of the inequality aversion
concept the combined amount of both parameters is shown.
Robustness of the core against fairness
Robustness of the core against random shocks
Due to this high correlation it was almost impossible to isolate the effect of any particular
type of disturbance by analyzing purely collective choices.360 This would bring about a
high risk of confusing the effect of subjects’ cognitive incapacities (e.g., random error) for
359 Schmidt (2010) attributed this idea to explain herding strategies, where subjects spend their money on
identical portfolios (e.g., houses) despite a known risk that prices are likely to drop.
360 This is not unique to my experiment. A similar correlation can be found, e.g., in the payoff tables con-
structed by S&K (2010, p. 681-684). Their payoffs show a clear interrelation between the robustness
against random shocks and social preferences.
169
social preferences or even strategic considerations. Fortunately, the prospects of identify-

ing social preferences are less gloomy when analyzing individual choice behavior. This
reinforced my decision to contrast collective and individual behavior. At the individual
level, the greatest challenge is the limited ex-ante knowledge about a subject’s level of
sophistication. An appropriate statistical model therefore had to incorporate both sincere
and sophisticated voting types.
6.3.2 Random utility model
Overall, my experiment represents a relatively unstructured voting game. In such a set-

ting it is difficult to separate the effect of subjects’ cognitive incapacities from behavioral
parameters. Thus, I estimated random utility models to account for random disturbances
in an individual’s utility functions as, e.g., misjudgments and errors.361 Following the
approach of Goeree et al. (2002, p. 262) this was implemented by a probabilistic choice
function which was supplemented by a theoretical equilibrium concept. This allowed for
mixing sincere and sophisticated types as well as for estimating utility (or motivation) pa-
rameters in a combined model.362 Following Harrison (2008, p. 2) this “joint estimation is
essential for inference about the model as a whole”. The optimization was accomplished
through conditional logistic regression.
Most influential for the choice of this model was the contribution of Goeree et al. (2002).
They played one-shot public goods games and looked for differences in internal returns
for the subjects themselves and external returns to their co-players. The authors specified
a logit equilibrium model in which choice was stochastic to obtain maximum likelihood
estimations. In accordance with my model, the utility function of the subjects contained
not only self-interest but also altruism parameters. Palfrey and Prisbrey (1997) used a
similar approach when investigating warm-glow altruism in voluntary contributions ex-
periments.363 To assess the actions of their subjects the authors estimated response func-
tions to treatments at the aggregated and the individual level using probit models.
SINCERE VOTING
I start by formulating the probabilistic choice of a sincere voting type. Such a subject is
focused on their own payoffs and pays no attention to another subjects’ (probable) choice.
Thus, for sincere voting types their choice model is identical to the standard conditional
361 Random utility models go back to the work of McFadden (1974, 1976, 1982). The approach has been
extended to strategic settings several times (Signorino, 1999, p. 282), among others by McKelvey and Pal-
frey (1995, 1996, 1998) for their well-known QRE concept. For an up-to-date assessment on the empirical
content and limitations of the concept cf. Haile et al. (2008).
362 Signorino (1999) employed statistical strategic discrete choice models to a simple crisis interaction setup.
Most important, he demonstrated how to directly incorporate the theorized strategic interaction into the
discrete choice models.
363 “Warm-glow preferences mean that the act of contributing, independent of how much it increases group
payoffs, increases a subject’s utility by a fixed amount” (Palfrey and Prisbrey, 1997, p. 830).
170
logistic regression.364 This is shown in Eqn. 6.7 where pij is the probability with which an
individual actor i votes for an alternative j. The probability results from the ratio of the,
by individual’s cognitive capacity λ weighted, utility Uij of alternative j to the aggregate
benefit of all nine alternatives k = 1, 2, . . . , 9 (each also weighted by λ).365 Therefore,
every alternative possesses a positive probability to be chosen by a subject (pij > 0 for
j = 1, 2, . . . , 9). Obviously, “the choice probabilities are proportional to an exponential
function of the expected payoffs” (Goeree et al., 2002, p. 263). Thus, “non-optimal choices
can occur, but the probability of this is inversely related to their cost” (Goeree et al., 2002,
p. 262). In terms of my experiment, the lower the payoff of an alternative j compared to
the highest payoff of the subject in another alternative k ̸= j, the smaller the probability
of j to be chosen by the subject.
 
exp λUij
pij = 9 (6.7)
∑k=1 exp (λUik )
In situations where all subjects vote sincerely the model is identical to the standard con-
ditional logistic regression, also at the group level. Thus, the probability of the collective
outcome is simply the likelihood by which at least four players choose the same alterna-
tive coincidentally.
The additive utility function Uij may be composed of various components. For example,
absolute payoff or inequality preference and the respective parameters α, ε, and b which
measure their influence. Importantly, these parameters and λ are not actor-specific.366 In
this model it is impossible to estimate λ and the single parameters of Uij simultaneously
because such a model would be unidentified (Signorino, 1999, p. 284ff). Here, I follow the
standard assumption according to which λ = 1; alternatively, the resulting estimates can
be interpreted as the joint effect of cognitive capacity and utility parameter (i.e., λ × β)
(Signorino, 1999, p. 284). In general, this implies that a higher cognitive capacity or a
lower level of randomness decreases the effect of all components in Uij .
SOPHISTICATED VOTING
A sophisticated voter is not solely focused on their own payoffs but takes into account
other subjects’ choices and the logic of the collective decision problem. Thus, for sophis-
ticated voting types their choice function is complemented with the core solution concept
to ensure the theoretical consistency of actions and beliefs. The individual choice prob-
ability of alternative j then coincides with the probability of j being the core. Eqn. 6.8
presents the probability that actor i prefers alternative k over j in a pairwise contest.
exp (λUik )
pikj =   (6.8)
exp λUij + exp (λUik )
364 Goeree et al. (2002, p. 262) argued that the “logit probabilistic choice function” resembles a “convenient
specification for empirical work”.
365 In this simple form the denominator ensures that the probabilities of all nine alternatives add up to 1.
366 It follows that the estimations obtained display just a “representative subject” (Palfrey and Prisbrey, 1997,
p. 835).
171
Yet, the core is defined at the collective level. Thus, the decisive question is with which
probability alternative k wins a pairwise contest with alternative j when voted on in the
group. For this, I calculate the probability by which each of the twenty-two theoretically
feasible winning coalitions c would chose k over j. Eqn. 6.9 gives an example of this
probability of a minimal winning coalition of the players 1, 2, 3 and 4 (c = {1, 2, 3, 4}).
p1,2,3,4kj = p1kj ∗ p2kj ∗ p3kj ∗ p4kj (6.9)
From Eqn. 6.9 it follows that the probability of alternative j to survive a pairwise contest
with alternative k over all twenty-two possible winning coalitions c is equal to:367
22
∏
 
p jk = 1 − pckj (6.10)
c =1
In other words, as long as one of the winning coalitions has a high probability of choosing
k over j, the latter cannot be the core alternative because the core alternative beats every
other alternative (i.e., it does not lose to any other alternative). Overall, the subjects could
choose between nine alternatives k = 1, 2, . . . , 9. Thus, in Eqn. 6.11 I multiply p jk over all
eight other alternatives k ̸= j in order to retrieve the probability with which alternative j
wins every possible pairwise contest.
9
pj = ∏ p jk (6.11)
k̸= j
ADDING - UP CONSTRAINT
So far, Eqn. 6.11 states the predicted probability with which alternative j is adopted by
a majority of subjects in the first round. The same holds true for sincere voting where
the predicted probability results from the standard conditional logistic regression at the
group level (cf. Eqn. 6.7). It is straightforward that the likelihood for reaching an agree-
ment in this first round decreases with the size of the random disturbance.368 Yet, the
ballots of each round were iterated until a collective agreement had been reached. To ac-
commodate this characteristic, I implement an adding-up constraint according to which
the sum of the probabilities p j over all nine alternatives equals one. This is shown in
Eqn. 6.12.
9
∑ pj = 1 (6.12)
j =1
367 The majority threshold of four out of six players has 15 minimal winning coalitions but is also reached by
six possible coalitions of five and one unanimous coalition of all players.
368 For λ = 0 each player acts randomly and throws an octahedral dice.
172
Sec. A.15 illustrates the predictions for the collectively chosen alternatives for different
levels of random error for all payoff tables. The graphs are separated according to sophis-
ticated and sincere voting. The resulting deviations are clearly visible; these differences
were to be expected, as the two types are based on diverging models. However, the next
section explains how they can still be integrated into a single model.
6.3.3 Mixture stage
My baseline assumption was that the subjects follow the theoretical solution concept of
the core, i.e., they choose accordingly for sophisticated types. Yet, they are restricted
by human cognitive capabilities. I expected at least some degree of sophisticated vot-
ing; in other words, I did not argue that all subjects voted in a sophisticated way all the
time. Therefore, my statistical model composes a mixture stage which enabled me to es-
timate individuals’ motivation while allowing for both sincere and sophisticated voting
types.369 In the end, the statistical model had to compute the probability with which each
group collectively chooses any of the nine alternatives.370
If we take all of this together and use a subject’s probability of sophisticated voting vil
similar to a weighting factor, a subject i’s probability to vote for alternative j in a ballot l
is then given by Eqn. 6.13. Here, psophisticated refers to Eqn. 6.11 and psincere to Eqn. 6.7.
sophisticated sincere
pijl = vil ∗ pijl + (1 − vil ) ∗ pijl (6.13)
This leads to the question of what determines this probability of voting sophisticatedly.
Sec. 6.2 argued that being part of the winning coalition might be a clue. Eqn. 6.14 defines
the possible dependency of the degree of sophistication in a more general way. It shows
a subject i’s probability of voting sophisticatedly vil in a ballot l and indicates possible
co-variates h which may explain a subjects’ degree of sophistication (cf. Tab. 6.1).
 
exp α + ∑hH=1 β h xhil + ε il
vil =   (6.14)
1 + exp α + ∑hH=1 β h xhil + ε il
CO - VARIATES
The range of possible variables which influence the extent of sophistication was restricted
by the limited diversity of my sample. Of all subjects 92% were students and 88% of
them were 25 years old or younger. Thus, it made no sense to use a “student-dummy”
as well as “salary” or “age” as explanatory variables. Instead, I used the prior laboratory
369 When discussing various theories of decision-making under uncertainty El-Gamal and Grether (1995,
p. 1137) concluded that “current literature does not support the conclusion that subjects are sufficiently
homogeneous to be described by a single theory.”
370 Cf. Redner and Walker (1984) for a methodological assessment of estimating mixture density parameters.
173
experience (i.e., the number of prior experiments the subject had participated in) as well
as the previous knowledge of the subjects with respect to game theory. This knowledge
was measured by four questions in the post-experiment survey: i) “Do you know the
research area of game theory?”, ii) “Do you know what prisoner’s dilemma means?”, iii)
“Have you attended a lecture on experimental methods?” and iv) “Have you attended a
lecture on micro-economics?”371 The answers of the subjects were than normalized into
an index reaching from 0 (i.e., four times no) to 1 (i.e., four time yes). Further, I include
a winning coalition dummy (cf. Sec. 6.2). This is different from other co-variates which
are determined exogenously. Yet, the coalition membership evolves endogenously in the
course of the experiment.
Besides the sophistication, the amount of social considerations might also hinge upon ex-
ogenous factors. Therefore, I looked also into the explanatory power of co-variates at the
utility level. More precisely, I interacted the parameters for inequality aversion (Eqn. 6.5)
with potentially influential characteristics. Here, I could draw on a wide range of pre-
vious research. Ackert et al. (2009, p. 16) aptly summarized that “social behavior can
be predicted by years of education, gender, university major, age, and primary house-
hold support. Participants who are less educated, female, non-business majors, older,
and not primarily supported by their spouse are more likely to vote against their self-
interest. Risk preferences are related to gender, but do not appear to be systematically
related to age or altruistic behavior.” As discussed earlier, the limited diversity of my
sample restricts my choice of potential variables; of the mentioned aspects only gender
and business major are feasible. Looking further into the experimental literature discloses
an “unresolved debate about whether economists are different than other professionals”
(Croson, 2005, p. 138).372 Precisely because this has so far not been completely clarified, it
makes sense not to be restricted to economics students but to enable volunteers from all
university faculties to participate as subjects in experiments.373
The argument for the influence of gender on cooperation is supported by many experi-
mental findings (for a comprehensive review cf. Croson and Gneezy, 2009). For example,
Charness and Rustichini (2011, p. 77) found that “females cooperate significantly more
than males”. Another aspect of their work is even more interesting. The authors con-
ducted prisoner’s dilemma games and divided the participants into rooms. Then, sub-
jects “played the game once with an audience of the same group (’at home’) and once
with an audience of the other group (’away’)” (Charness and Rustichini, 2011, p. 77).
371 As the experiment was conducted in Heidelberg these are translations of the original German questions.
372 Croson (2005, p. 138) offered a literature overview of contributions which investigate the question of
whether economists (or economics students) are different from other professionals (or non-economics
students). On the one hand, they are found to free-ride more in social dilemmas (Marwell and Ames,
1981; Frank et al., 1993, 1996) and to offer less in ultimatum games (Carter and Irons, 1991). On the other
hand, they are judged to be more cooperative in a lost-letter experiment (Yezer et al., 1996) and less likely
to cheat compared to sociologists and political scientists (Laband and Biel, 1999). Also, Frey and Meier
(2005, p. 170) argued that their measured “lower contribution of business economists [and not economists
in general], compared to other students, is due to self-selection rather than indoctrination.”
373 Croson (2005, p. 138) assessed this “experimental economics practice [as] a sensible and conservative one.”
174
I did not separate my participants into different rooms, but on a more general level this
finding refers to the aspect of social distance which I discuss in Sec. 4.3.3. My subjects did
not come from different pools and they did not play the game through various (technical
or design-specific) means. Yet, an aspect which differed is how many other participants
they knew (either just by sight or quite well). Here, more familiar faces will bring about
a feeling of being at home. I operationalized this aspect by controlling the number of
students within the laboratory at the same time who share the same major.
Overall, important aspects are the gender of the subject, whether or not the subject is an
economics student and the number of fellow students from the same faculty in the labo-
ratory. For the strategic component I used the number of prior participation in laboratory
experiments and subjects’ previous knowledge of game theory. In addition, I differentiate
for subjects belonging to the winning coalition. Tab. 6.1 provides a descriptive overview
of these co-variates; all associated computations are discussed in Sec. 6.4.2.
Table 6.1: Co-variates for the random utility mixture model

EXPLANATORY NOTE
The table depicts descriptive information for the co-variates used in the random utility mixture model estimation. The
variables are separated according to utility and strategic level. For every co-variate I depict its average, SD, mean, minimal
and maximal value. For gender a 0 indicates a male and a 1 a female participant. For an economics student and a winning
coalition a 0 indicates false and a 1 indicates true. The variables fellow students and prior experiments show the amount
of the respective variable. Theory knowledge displays its quantity normalized to the range of 0 (min) and 1 (max).
co-variates N mean SD median min max
UTILITY COMPONENT
gender 168 0.57 0.50 1 0 1
economics student 155 0.46 0.48 0 0 1
fellow students 155 3.75 3.12 2 0 11
STRATEGIC COMPONENT
winning coalition 1326 0.74 0.46 1 0 1
prior experiments 168 4.67 3.46 4 0 20
theory knowledge 168 0.48 0.33 0.5 0 1
STATISTICAL MODEL
Finally, by insertion of all the single elements in Eqn. 6.13 the derivation results in the
likelihood function Eqn. 6.15 which is used for estimating the models. I apply standard
maximum-likelihood techniques (following Goeree et al., 2002, p. 262) and use GAUSS’
(Version 9.0.2, build 1114) constrained optimization algorithm (cf. Sec. A.1).
 
H
  exp α + ∑ β x
h=1 h hil + ε il
L pijl | yij , xhil =   (6.15)
1 + exp α + ∑hH=1 β h xhil + ε il
 
9 22
exp (λUik )
∗ ∏∏ 1−∏  
exp λUij + exp (λUik )
k ̸ = j c =1 iϵc
   
exp α + ∑hH=1 β h xhil + ε il
 
exp λUij
+ 1 −   ∗ 9
1 + exp α + ∑hH=1 β h xhil + ε il ∑k=1 exp (λUik )
175
While the first two terms in the formula refer to sophisticated voting the following two
terms express sincere voting. In both cases, the first expression indicates the relative
weight of the respective behavior and the second expression represents the according
probabilistic choice function. For sophisticated voting the probability depends on the
prospect of an alternative j to be superior to any other alternative k over all possible
winning coalitions c. For sincere voting I use standard conditional logistic regression.
6.3.4 Model specifications
This section briefly summarizes my theoretical model and explains four concrete spec-
ifications. The baseline model only comprehends purely self-interested actors (MODEL
I ). Starting from here, I supplemented my model with the three most relevant patterns
of individual behavior which I discussed previously: sophistication, inequality aversion
and risk aversion. All specifications are estimated using conditional logistic regression.
It is in particular difficult to filter out the sophisticated component of behavior. I ac-

counted for this by including both sincere and sophisticated voting types (MODEL II ).
This followed the approach of Goeree et al. (2002) and resulted in a random utility mix-
ture model (RUMM). The voting type distinction required a mixture stage that allows
evaluating the extent to which the theoretical solution concept correctly predicts subjects’
behavior. In other words, such a model contains two levels; one at which the utility of
the subject was calculated (absolute payoff and relative payoff) and one which differen-
tiates for the degree of sophistication with which this utility was maximized (e.g., choose
highest payoff vs. choose highest payoff after weighting by its cue validity).
In addition, I incorporated further aspects into the strategic component of the model.
Most importantly, I differentiated between the participants according to their affiliation
to the winning coalition (MODEL III) as explained in Sec. 6.2.
The last model incorporated social preferences (MODEL IV) by implementing the inequal-
ity aversion concept of Fehr and Schmidt (1999). The big advantage of this approach is
its ability to distinguish between advantageous and disadvantageous inequality.
Finally, one significant pattern of individual behavior is still missing. I was also con-
cerned with the impact of subjects’ risk attitudes as the design provided four different
decision situations with different levels of uncertainty. This follows the call of Bradler
(2009, p. 20) who emphasized that “research could make a further approach to merge
theories of social preferences and theories on choice under risk together.” Risk aversion
extends to an individual’s complete utility function. Thus, each model was estimated
without (variant a) and with (variant b) risk aversion. The following section discusses
the estimates of these eight models. In addition, I also consider co-variates and their
explanatory power.
176
6.4 Results
6.4 Results
While Chap. 5 discusses the experimental results by looking at the group performances,
the following analysis investigates how those originated from individual choices. Of
course, the insights found on the collective level (e.g., an increase of decision-making
efficiency for delegation settings) remain valid. Yet, the individual data supplements the
previous results with information on motive and sophistication of the votes.
As dependent variable, I used those individual choices which caused the collective out-
come, i.e. the final ballots of every round. As I report parameter estimates for preferences
based on last choices only, this follows the assumption that an individual’s objectives
remain “stable in the short run, [i.e.] for the duration of the experiment” (Bolton and
Ockenfels, 2000, p. 171). This does not rule out that the “weights individuals give these
objectives may well change over the long term, with changes in age, education, political
or religious beliefs, and other characteristics” (Bolton and Ockenfels, 2000, p. 171).374
The computations were carried out for four different models: one which includes only
self-interest (MODEL I), one that allows for a mixture of sincere and sophisticated voting
(MODEL II), one that differentiates sophistication for members of the winning coalition
(MODEL III) and one that applies the inequality aversion concept (MODEL IV). Each model
was estimated without (variant a) and with (variant b) incorporating risk aversion.
The various models enable me, through their juxtaposition, to draw inference on subjects’
motivation for their observed votes. In other words, I test the functional specification
and perform (relative) validation tests. Thus, I am able to investigate the accuracy of
the part-sincere-part-sophisticated hypothesis, to determine the relative importance of
self-interest as compared to social preferences, whether risk aversion played a role, etc. I
base the comparison on the significance of the parameter estimates, the likelihood of the
statistical models and the Akaike information criterion (AIC, Akaike, 1973).375
6.4.1 Random utility mixture model estimates
This section discusses the conditional logistic regression model estimates for the various
procedures. As explained in Sec. 4.2.6, the voting procedures exposed the participants to
374 For a macro-analysis of data sets from various ultimatum game laboratory experiments cf. Prasnikar
(1997). Most important, she found that people’s perception of fairness is stable not only within one round
but even with repeated play. Also, Borghans et al. (2008) examined the predictive power of personality
and the stability of personality traits over the life cycle. They found that cognitive and personality traits
evolve, but to different degrees and at different stages. As the important temporal dimension refers to
the life cycle, it is at least clear that these traits are stable across situations.
375 The AIC is a popular measure of the relative accuracy of fit of statistical models. It balances the degree
of a model’s variance and the extent of contained biases against each other (Burnham and Anderson,
2004). Based on the maximized likelihood function of a model, it assigns a penalty term for the number
of included parameters; the “model with minimum AIC value is chosen as the best model to fit the data”
(Bozdogan, 2000, p. 63). For a comprehensive discussion of its principles, asymptotic properties and
inferential capabilities cf. Bozdogan (1987, 2000).
177
decision situations with different levels of uncertainty and complexity. Thus, looking at
the (relative) importance of behavioral patterns within and across procedures, I am able
to assign them to aspects of the decision-making process. For all procedures I list eight
model specifications of which each comprises two sections. The upper section displays
the parameter estimates of the utility component; i.e., self-interest, inequality aversion
and risk aversion. The lower section displays the estimates for an actor’s level of sophis-
tication; i.e., the proportion of sophisticated voting subjects. I differentiate prominently
for the proportion when accounting for the members of the winning coalition. Altogether,
this yields complicated and diversified tables. To facilitate the contemplation, the tables
display the estimated parameters and corresponding z-scores; this enables an importance
comparison between different variables.376 The score is calculated for every parameter as
estimated value 377
z= standard error . Thus, it measures by how many standard errors the estimated value
deviates from 0. I used robust standard errors for all computations.378
It is crucial to note that the indicated degree of sophistication represents a special index.
The given estimates are to be interpreted as values of the cumulative distribution func-
tion of the standard normal distribution. Specifically, an estimate ≤ −3 indicates 100%
percent sincere voting while an estimate of ≥ 3 points to 100% sophisticated behavior.
The values in between represent mixed shares; hence, at an estimate of 0 there is 50% of
either type.379 This representation has a crucial limitation. When displaying mixed vot-
ing behavior, i.e., some sophisticated as well as some sincere voting, it is not clear how
much of the explanatory power is arising from either of the sophisticated or the sincere
model. In other words, the data cannot tell which model explains the observations. Nev-
ertheless, such an estimate still rejects the hypothesis that the mixture probability is 0.
Thus, it proves the argumentation of Sec. 6.2 that the ballot is, in fact, composed partly of
both sophisticated and sincere voting behavior.
As a baseline, I use the results obtained under POL which are shown in Tab. 6.2. MODEL
I focused exclusively on self-interest which, unsurprisingly, turned out to be positive
and significant. Including risk aversion shows that subjects were highly risk averse; i.e.,
small gains were relatively more important. With MODEL II the analysis moved beyond
the standard conditional logit by implementing the mixture stage. This proves worth
the effort as the model’s explanatory power increased about as much when consider-
ing risk aversion. The estimates provided significant evidence of mainly sophisticated
voting. Including sophistication and risk aversion together weakened the evidence and
376 Assuming a standard normal distribution, the following z-scores and significance levels (two-tailed) are
associated: z = ±1.645 ←→ p = 0.1; z = ±1.96 ←→ p = 0.05; z = ±2.58 ←→ p = 0.01; etc.
377 The actual calculation rule for z-scores is z = observation value−sample mean . Yet, as the logit model mean
standard error
equals 0 I could apply the simplified version.
378 With respect to robust standard errors, cf. Greene (2003, p. 267) for a general introduction, Croux et al.
(2003) for a comprehensive discussion of examples, applications as well as scope of the approach and
King and Roberts (2012) for an assessment of risks and problems of this method.
379 Corresponding to the shape of the normal distribution this is no linear relationship. For example, a dis-
pensation of 75% sincere and 25% sophisticated voting is reached at −0.675 while a distribution of 33.3%
sincere and 66.6% sophisticated voting is reached at 0.415.
178
Table 6.2: Model estimates under pooling
EXPLANATORY NOTE
The table depicts the estimates of all eight logit models under the pooling procedure. As dependent variable, I used all individual choices of the final ballot of every corresponding experimental
round. For all independent variables the table shows the beta coefficient in addition to the corresponding z-scores based on robust standard errors in parentheses. The pseudo R² is calculated
following McFadden (1974).
N = 456 individual votes MODEL I MODEL II MODEL III MODEL IV
Variant of risk aversion self-interest self-interest and mixture mixture and risk winning winning coalition inequality inequality and
risk aversion aversion coalition and risk aversion aversion risk aversion
UTILITY
COMPONENT
self-interest 0.095 5.331 0.165 1.775 0.160 0.376 0.075 0.199
(13.069) (3.822) (33.850) (2.175) (18.171) (41.330) (6.972) (20.557)
risk aversion 1.219 0.782 0.290 0.354
(11.723) (5.154) (3.150) (4.687)
D+ 0.046 0.088
(5.848) (49.111)
D− -0.067 -0.090
(-8.753) (-47.105)
STRATEGIC
COMPONENT
constant 2.834 0.784 0.656 0.434 1.150 0.8631
(38.786) (1.994) (3.787) (24.260) (2.322) (40.613)
winning coalition 8.625 4.146 7.712 4.894
(40.721) (83.914) (9.128) (74.808)
Log likelihood -958.18 -878.16 -877.32 -858.98 -844.44 -832.59 -813.11 -805.59
Pseudo R² 0.04 0.12 0.12 0.14 0.16 0.17 0.19 0.20
AIC 1918.36 1760.32 1758.64 1732.96 1694.88 1673.18 1636.22 1623.18
179
6.4 Results
effect of both. Yet, they stayed significant and retained their estimates. Next, MODEL
III investigated the sophistication of members within and outside the winning coalition
separately. Here, the estimated probability of encountering sophisticated voting types
was around 74% for subjects outside and almost 100% for subjects within the winning
coalition. Also, the level of revealed risk aversion dropped significantly compared to
the previous models. Finally, MODEL IV looked into the parameters for inequality aver-
sion. The subjects disliked getting less than others (negative D − ), while at the same time
they revealed a highly significant preference for getting more than others (positive D + ).
In other words, the subjects exhibited a highly competitive behavior as they were moti-
vated by relative gains; this stays significant when controlling for sophistication as well
as for risk aversion. Across all model specifications, the explanatory power clearly im-
proved when adding risk aversion. The increase was highest for the simplest model; for
more multifaceted specifications the value of adding further parameters was less exten-
sive. Yet, under POL the best performing specification was MODEL IV which includes the
mixture component as well as inequality and risk aversion.
That subjects were motivated by relative gains contradicts some of the previous work
on social preferences where altruistic behavior in collective decisions was identified (e.g.,
Goerg et al., 2007; Levin, 2009). Yet, those findings were mostly obtained in other exper-
imental designs such as the ultimatum game, dictator game, trust game, gift exchange
game or public good game (cf. Sec. A.14). These are games in which a subject’s competi-
tiveness (i.e., the preference for advantageous inequality) does not alter their equilibrium
behavior (Fehr and Schmidt, 1999, p. 850). My design placed the participants into a dif-
ferent context. Goeree et al. (2002, p. 267) pointed out that the decision context affects an
individual’s preferences as in bargaining situations “the focus is more clearly on issues
of division”, which makes altruism less important and inequity aversion more salient. In
line with my results, Dittrich and Ziegelmayer (2010, p. 9) also found a lack of aversion
to advantageous inequality in their bilateral gift exchange game.
Next, I looked at the model estimates for the second stage under SEQ. This resembled a
simpler version of the decision situation under POL as the number of players and alter-
natives declines. The results are shown in Tab. 6.3. When considering only self-interest
(MODEL I) they appeared rather similar to POL. One notable difference was less risk aver-
sion and that the increase brought by it to the explanatory power was rather small. This
changed once the mixture stage of MODEL II allowed for sincere and sophisticated vot-
ing types. The estimated probability of encountering sophisticated voting was 25%, and
13% when controlling for risk aversion. The dominance of sincere voting also prevailed
in both groups (≥90%) when accounting for insiders and outsiders of the winning coali-
tion separately in MODEL III. In general, including the strategic component improved the
model significantly compared to the simplest version. MODEL IV completed the analysis
by looking at inequality aversion but the knowledge gained was very little. The only sig-
nificant parameter indicated a rejection of advantageous inequality. Yet, this disappeared
180
Table 6.3: Model estimates of the second stage under sequential delegation
EXPLANATORY NOTE
The table depicts the estimates of all eight logit models for the second stage under the sequential delegation procedure. As dependent variable, I used all individual choices of the final ballot
of every corresponding experimental round. For all independent variables the table shows the beta coefficient in addition with the corresponding z-scores based on robust standard errors in
parentheses. The pseudo R² is calculated following McFadden (1974).
UTILITY
COMPONENT
self-interest 0.080 0.347 0.467 1.141 1.287 1.757 0.842 1.605
(7.234) (2.192) (5.333) (0.623) (2.244) (1.738) (2.031) (1.350)
risk aversion 0.563 2.799 1.191 0.902
(3.697) (4.925) (1.657) (4.217)
D+ -0.743 0.065
(-2.258) (0.195)
D− -0.080 -0.497
(-0.248) (-0.456)
STRATEGIC
COMPONENT
constant -0.618 -1.120 -1.444 -1.609 -1.525 -1.431
(-2.443) (-3.528) (-3.056) (-2.947) (-3.112) (-2.991)
winning coalition -0.373 -1.091 -1.313 -0.541
(-0.683) (-1.609) (-2.559) (-0.985)
Log likelihood -161.85 -159.36 -145.77 -135.90 -133.01 -125.21 -126.99 -124.32
Pseudo R² 0.18 0.19 0.26 0.31 0.33 0.37 0.36 0.37
AIC 325.70 322.72 295.54 277.80 272.02 258.42 263.98 260.64
181
6.4 Results
Table 6.4: Model estimates of the first stage under sequential delegation
EXPLANATORY NOTE
The table depicts the estimates of six logit models for the first stage under the sequential delegation procedure. As dependent variable, I used all individual choices of the final ballot of the
first stage of every corresponding experimental round. As only for this stage is a division by beliefs necessary, the respective two variants are shown in Tab. 6.5 separately. For all independent
variables the table below shows the beta coefficient in addition to the corresponding z-scores based on robust standard errors in parentheses. The pseudo R² is calculated following McFadden
(1974).
N = 180 individual votes MODEL I MODEL II MODEL III
Variant of risk aversion self-interest self-interest and mixture mixture and risk winning winning coalition
risk aversion aversion coalition and risk aversion
UTILITY
COMPONENT
self-interest 0.057 1.150 0.104 0.552 0.110 0.857
(6.940) (1.600) (6.055) (7.229) (5.820) (2.746)
risk aversion 0.886 0.530 0.670

(3.784) (3.282) (2.021)
D+
D−
STRATEGIC
COMPONENT
constant -0.448 -0.584 -2.358 -4.124
(-1.322) (-1.675) (-2.347) (-4.857)
winning coalition 0.452 0.491
(0.418) (0.506)
Log likelihood -164.26 -155.65 -155.71 151.86 -149.01 -142.51
Pseudo R² 0.17 0.21 0.21 0.23 0.25 0.28
AIC 330.52 315.30 315.42 309.72 304.02 293.98
182
6.4 Results
when controlling for risk aversion. It seems that subjects abstained from choosing very
profitable alternatives because they feared that those allow no consensus. Across all spec-
ifications for the second stage under SEQ, the best performance was achieved by MODEL
III , which includes the mixture component and risk aversion.
Tab. 6.4 stays with SEQ but turns to the decision-making of the first stage under this pro-
cedure. Here, players have to consider the two other players of their own group as well
as the subsequent choices of the three players in the second group. Thus, they were ex-
posed to a higher level of uncertainty when making their choices. MODEL I showed that
self-interest and risk aversion played an important role. This also holds true for all other
model specifications under this procedure. The estimates of the strategic component in
MODEL II and MODEL III resembled the results of the second stage. Sincere voting was
predominant (≥67%) and differentiating for the winning coalition did not change this
but rather made the lack of sophistication even more explicit (≤8%). Looking at the two
stages of SEQ together, it becomes obvious that risk aversion and sincere voting dom-
inated both. Here, the CRRA parameter as well as the degree of sophistication (across
all subjects) was stronger in the second stage. This high risk aversion is surprising as I
characterized this decision situation with a low level of uncertainty (cf. Tab. 4.2)
So far, MODEL IV is missing. In Sec. 6.3.2 I stated that the utility parameters and λ were
not estimated to be actor-specific as such a model would be unidentified (Signorino, 1999,
p. 284ff). Thus, the results in Tab. 6.4 are based on the assumption that subjects in both
stages possess the same parameter estimates. In other words, subjects in the first stage
believed that their colleagues in the second stage would share their motivation. This
might be considered as a rather strong and unrealistic assumption (henceforth this as-
sumption is referred to as strong beliefs). Tab. 6.5 contrasts these beliefs with a second
variant that assumed that players at the first stage believe that all remaining alternatives
are equally likely to be elected in the second stage. To implement this, I set all second
stage parameters equal to zero (henceforth this assumption is referred to as flat beliefs).380
As in the second stage, MODEL IV without risk aversion indicated that subjects dislike
advantageous inequality. This effect became again insignificant when accounting for risk
aversion. The differences resulting from the two belief assumptions are not so trivial.
With flat beliefs, subjects showed distaste for relative losses and risk aversion was less
important. Also, the differentiation for belonging to the winning coalition turns out to be
significant. Yet, this modified the supremacy of sincere voting only in the second decimal.
Across all models for the first stage under SEQ the best performing model was MODEL IV
based on flat beliefs and including the mixture component, risk aversion and inequality
aversion. The better performance of this belief assumption corresponds to the obtained
parameters. As most voting was sincere, the assumption of flat beliefs corresponds best
to the underlying focusing of subjects on themselves.
380 I restrict the comparison of different beliefs to the fully specified model in order not to overload the anal-
ysis. However, the same argument could be made for each specification.
183
Table 6.5: Model estimates under sequential delegation separated for beliefs
EXPLANATORY NOTE
The table depicts the MODEL IV estimates for the first stage under the sequential delegation procedure for different beliefs.
As dependent variable, I used all individual choices of the final ballot of the first stage of every corresponding experimen-
tal round. For all independent variables the table below shows the beta coefficient in addition with the corresponding
z-scores based on robust standard errors in parentheses. The pseudo R² is calculated following McFadden (1974).
N = 180 individual votes MODEL IV
Variant of risk aversion inequality aversion inequality and risk aversion

Beliefs strong strong flat
UTILITY COMPONENT
self-interest 0.2911 0.667 0.2918
(3.127) (3.038) (3.871)
risk aversion 0.570 0.288
(1.939) (4.181)
D+ -0.1827 -0.031 0.097
(-2.327) (-0.318) (1.279)
D− 0.056 -0.052 -0.280
(0.769) (-0.553) (-3.604)
STRATEGIC COMPONENT
constant -4.581 -4.584 -4.334
(-5.421) (-5.421) (-5.919)
winning coalition 0.234 0.234 1.023
(0.370) (0.370) (5.304)
Log likelihood -141.81 -141.81 -138.05

Pseudo R² 0.28 0.28 0.29
AIC 295.02 295.62 288.10
Finally, Tab. 6.6 lists the estimates for SIM. Due to the high amount of uncertainty and
complexity under this procedure I obtained highly significant estimates of risk aversion
in all four specifications. MODEL II and MODEL III indicated the from SEQ familiar pat-
tern that sincere voting prevailed and that the affiliation to the winning mattered only
little. Looking at MODEL IV , I found no evidence for social preferences at all. The knowl-
edge gained when moving from the simplest to the most diverse model is lowest under
SIM in comparison to the other procedures. Almost the full extend was already reached
by adding risk aversion. However, on consideration of likelihood and AIC, the best per-
forming specification was MODEL III which includes the mixture component and risk
aversion.
6.4.2 The influence of co-variates
Sec. 6.3.3 discusses possible influencing factors which might explain why a subject acted
socially and sophisticatedly or not. Yet, the concrete expectations for my specific de-
sign are not clear. On the one hand, substantial literature discusses influential proper-
ties found in many studies. On the other hand, on examination of the most similarly
constructed experiments, e.g., Goeree et al. (2002) and S&K (2010), no influence of co-
variates was observed.381 Moreover, additional variables further increase the intricacy of
the objective function to be estimated. With respect to my number of observations, this
381 More precisely,
Goeree et al. (2002) found on average that altruism parameters were the same for men and
women. However, they observed that male altruism was significantly more dispersed.
184
Table 6.6: Model estimates under simultaneous delegation
EXPLANATORY NOTE
The table depicts the estimates of all eight logit models under the simultaneous delegation procedure. As dependent variable, I used all individual choices of the final ballot of every corre-
sponding experimental round. For all independent variables the table shows the beta coefficient in addition to the corresponding z-scores based on robust standard errors in parentheses. The
pseudo R² is calculated following McFadden (1974).
UTILITY
COMPONENT
self-interest 0.062 0.998 0.134 1.025 0.127 1.087 0.113 1.193
(12.216) (3.932) (6.599) (3.510) (6.956) (50.785) (2.194) (24.297)
risk aversion 0.864) 0.789 0.806 0.982
(9.757) (7.710) (61.550) (31.458)
D+ -0.053 -0.003
(-1.182) (-0.256)
D− -0.052 0.000
(-1.239) (0.027)
STRATEGIC
COMPONENT
constant -0.511 -1.811 -1.382 -4.091 -6.602 -4.389
(2.186) (3.531) (-3.011) (-10.386) (-10.654) (-88.667)
winning coalition -0.171 -0.966 -0.990 -1.122
(-0.337) (-1.407) (-1.462) (-22.669)
Log likelihood -448.86 -416.68 -431.58 -412.02 -427.49 -407.37 -410.34 -407.12
Pseudo R² 0.20 0.26 0.23 0.26 0.24 0.27 0.27 0.27
AIC 899.72 837.36 867.16 830.04 860.98 822.74 830.68 826.24
185
6.4 Results
should not be overdone. However, in this section I look into the influence of co-variates
for two reasons. Firstly, because of the possibility of inductive findings and, secondly,
because this step resembles a robustness test for the results obtained in my experiment;
and therefore it increases their validity.
In accordance with the previously discussed literature the co-variates were added into
both parts of the statistical model. For the strategic component these were the amount
of prior participation in laboratory experiments (prior experiments) and subjects’ knowl-
edge of game theory and experimental methods (theory knowledge). The utility com-
ponent accounts for the gender of the subject and either if the subject is an economics
student (econ student) or the number of fellow students of the same faculty in the lab-
oratory. Here, the co-variates were included in the form of interaction effects with the
inequality aversion coefficients. Starting from the baseline of a male, non-economics stu-
dent (or a subject without fellow students) with no prior participation in experiments
and no theory knowledge, the co-variates indicate how a change in one of these aspects
affected social behavior or strategic voting; i.e., they enable the comparison between sub-
groups of participants. The results are shown in Tab. 6.7.382
Under POL I obtained the same findings as before; self-interest was highly relevant, sub-
jects behaved competitively (they liked advantageous and loathed disadvantageous in-
equality) and sophisticated voting was significant. Neither of the co-variates influenced
the results. The differences between specifications for either economics students or fel-
low students seem negligible. In fact, the statistical measures of the models’ predictive
power rarely change between any procedure for these two co-variates. Looking at the
utility component of the delegation procedures, I obtained significant estimates for sub-
jects’ characteristics. In the second stage SEQ I found already familiar patterns (subjects
disliked relative gains) but also a significant influence of the number of fellow students.
Interestingly, a larger number of colleagues with the same field of study amplified the
rejection of disadvantageous inequality. Yet, the results in the second stage of SEQ were
obtained from a reduced data set. When using all observations, the logit model was
“completely determined” due to “hidden collinearity” within my co-variates (cf. Srib-
ney, 1999).383 Here, a specific pattern across the co-variates had only one outcome. There
are various strategies to deal with this problem as, e.g., omitting the variable responsi-
ble, centering the predictors or excluding the affected observations (cf. Kleinbaum et al.,
382 Irestrict the investigation of co-variates to subjects’ social behavior and strategic voting in order to main-
tain a comprehensible analytical framework. Manifold experimental studies have looked into the influ-
ence of personal aspects on risk aversion (cf. Cox and Harrison, 2008; Eriksson and Simpson, 2010). While
I included this behavioral pattern in my utility model (cf. Sec. 6.1.5), it is not my main research objective
to identify the impact of personal characteristics on risk aversion. Current contributions also discuss
possible systematic relationships across distinct aspects of individual preferences such as risk aversion,
social preferences, reciprocity, etc. (e.g., Ackert et al., 2009). However, these considerations are beyond
the scope of my study.
383 The symptoms are similar to those of multicollinearity as, e.g., the co-variates have insignificant coeffi-
cients when entered jointly in the regression, but each has a significant coefficient when entered individ-
ually.
186
Table 6.7: Model estimates including co-variates
EXPLANATORY NOTE
The table depicts the influence of co-variates on social preferences and sophisticated behavior in all four decision situations. As dependent variable, I used all individual choices of the final ballot
of every experimental round. The included co-variates for the strategic component are the number of prior participation in laboratory experiments (prior experiments) and subjects knowledge
of game theory and experimental methods (theory knowledge). For the utility component two variants are considered; once the gender of the subject and if the subject is an economics student
(econ student) and once gender together with the number of fellow students in the laboratory (fellow students). For all independent variables the table shows the beta coefficient in addition to
the corresponding z-scores based on robust standard errors in parentheses. The pseudo R² is calculated following McFadden (1974).
POL SEQ SEQ SIM

second stage first stage
N = 456 N = 168 N = 180 N = 480
UTILITY COMPONENT
self-interest 0.069 0.065 1.326 1.278 0.181 0.242 0.121 0.120
(4.935) (5.156) (3.334) (3.194) (1.437) (2.115) (2.318) (2.298)
D+ 0.046 0.045 -1.079 -1.036 -0.187 -0.264 -0.080 -0.076
(3.564) (3.693) (-3.391) (-3.187) (-1.663) (-2.411) (-1.723) (-1.619)
D + ×gender -0.001 -0.001 0.121 0.116 0.159 0.124 0.040 0.039
(-0.111) (-0.065) (1.534) (1.397) (2.551) (2.764) (2.267) (2.133)
D + ×econ student 0.010 -0.097 0.126 -0.003
(0.840) (-1.195) (2.284) (-0.154)
D + ×fellow students 0.002 -0.075 0.025 -0.005
(0.918) (-1.015) (2.424) (-0.265)
D− -0.067 -0.075 0.528 0.632 -0.074 -0.064 -0.031 -0.032
(-5.484) (-6.135) (1.453) (1.917) (-0.463) (-0.455) (-0.675) (-0.629)
D − ×gender 0.001 0.003 0.196 0.131 0.023 0.012 0.010 0.021
(0.061) (0.280) (0.748) (0.756) (0.188) (0.160) (0.326) (0.626)
D − ×econ student -0.015 -0.027 0.063 -0.066
(-1.118) (-0.129) (0.545) (-1.865)
D − ×fellow students 0.001 -0.069 0.024 -0.010
(0.317) (-1.864) (1.181) (-1.640)
STRATEGIC
COMPONENT
constant 4.446 4.590 -1.1543 -1.021 -0.137 -0.266 -0.908 -0.915
(4.359) (4.297) (-1.189) (-1.204) (-0.222) (-0.455) (-1.203) (-1.607)
prior experiments -0.092 -0.123 -0.094 -0.182 -0137 0.015 -0.013 -0.002
(-1.058) (-1.376) (-0.417) (-1.369) (-0.059) (0.196) (-0.113) (-0.120)
theory knowledge -0.668 -0.042 0.406 0.423 -0.342 -0.455 -1.740 -1.733
(-0.602) (0.378) (0.284) (0.394) (-0.374) (-0.487) (-1.469) (-1.573)
Log likelihood 840.27 840.03 121.20 119.96 140.06 140.48 408.55 408.79
Pseudo R² 0.16 0.16 0.39 0.39 0.29 0.29 0.27 0.27
AIC 1704.54 1704.06 266.40 263.92 304.12 304.96 841.10 841.58
187
6.4 Results
2007, p. 365ff). I chose the last option; after identifying the responsible pattern based on
the predicted logits, I excluded all observations which comprised prior experiments = 0
and theory knowledge = 0. These were 12 observations and applied to four of the 168
participants.
In the first stage of SEQ I found that subjects disliked advantageous inequality. This be-
havior was stronger for men compared to women, for non-economics students compared
to economics students and the less fellow students took part in the same experiment.
These insights correspond partially to the results under SIM; here, subjects also rejected
advantageous inequality and this effect was weaker for women. Furthermore, economics
students refused disadvantageous inequality to a greater extent.
In summary, the distinction between the economics student and fellow student co-variate
is not decisive as models’ predictive accuracy rarely changed. When focusing on co-
variates of the strategic component, I observed no differences for the degree of sophisti-
cation. This does not mean that a combined effect of the variables cannot be significant,
but clearly the differences between subgroups were not. Looking at the co-variates of the
utility component, I found some effects but no general pattern that holds true across all
procedures.
6.5 Chapter summary
This chapter analyzes the experimental data on the individual level. I start in Sec. 6.1
with an introduction to common patterns of individual behavior found in prior studies.
As many of them were discovered and validated in laboratory experiments, it was likely
that they would also play a role in my design. Sec. 6.2 addresses the limitations of hu-
mans’ cognitive capacities (Simon, 1957) and explains that an experiment such as mine
has to account for both sincere and sophisticated voting subjects (Goeree et al., 2002).
Next, Sec. 6.3 brings these insights together and derives the statistical model. It includes
various specifications to cover the different behavioral patterns, contains a mixture stage
to distinguish for sophistication and comprises error terms to filter out the noise of col-
lective decisions. Overall, I specified eight model variants which are compared in Sec. 6.4
by juxtaposition in order to determine the impetus for an individual’s behavior. Not de-
cisive were various co-variates which I tested for their influence on social considerations
as well as sophistication.
If one looks across the models it becomes clear that decision makers’ substantial motiva-
tion must be modeled with a view to their risk aversion. A model’s predictive accuracy
increased when including CRRA; this pattern was constant across all variations of group
size and decision rule. The importance of risk aversion in an environment characterized
by uncertainty is no new finding. Yet more important, also in collective decisions risk
aversion has to be modeled on the individual level.
188
6.5 Chapter summary
A second element that enhanced model performance across all specifications was the
consideration of sincere and sophisticated voting behavior at the same time. The im-
plementation of the mixture stage (cf. Sec. 6.3.3) allows the estimation process to allocate
the behavior observed to the corresponding concept. This holds true even when sincere
voting is almost exclusively present as it facilitates the identification of noise within the
collective agreement. Overall, sophisticated voting took place predominantly only un-
der POL. This is most evident when compared to SIM, which was constituted entirely of
sincere choices. Here, a high level of uncertainty in combination with a high level of com-
plexity prevented subjects from evaluating the conceptualized solution concept. As the
computations became too demanding, they stopped playing in a sophisticated manner.
POL also resembles the only procedure under which a distinction for the extent of so-
phistication based on participation in the winning coalition turned out to be significant.
Of the two stages under SEQ, the second disclosed more strategic voting even if only at
a low level. Thus, the ability for sophisticated behavior decreases according to the level
of uncertainty.
In my experiment, I found that subjects were motivated by competition rather than by

fairness-related considerations. First of all, subjects cared about their own payoff and
were risk averse with respect to it. The aversion was highest in the second stage of
SEQ where subjects seem solely concerned about avoiding the lowest payoff. Secondly,
subjects disliked relative losses. The disapproval is significant under POL and in both
stages of SEQ. Yet, it was most pronounced in the first stage where subjects were primar-
ily afraid of earning less than others. At the other extreme, under SIM neither relative
gains nor losses are significant; i.e., subjects disregarded other players and stayed purely
self-interested. Thirdly, under POL subjects’ disclosed preferences for advantageous in-
equality. My results are therefore clearly not in line with the expectations of the Fehr and
Schmidt (1999) inequality aversion model.
When the insights into strategic voting and social considerations were linked, the preva-
lence of sincere behavior and the distaste of inequality coincided. The best performing
model in both stages of SEQ as well as under SIM was MODEL III, which does not include
social preferences. Yet, their consideration very well enhanced the explanatory power of
the models under POL. It seems that social preferences may be a luxury which subjects
only take into account when they are reasonably certain about, e.g., other players’ moti-
vations, choices, and, accordingly, the resulting consequences of their own votes. Goeree
et al. (2002, p. 267) showed that the decision context affects an individual’s preferences.
Yet, for this insight to be maintained it is foremost essential that the encountered task
must be logically comprehended. Uncertainty causes subjects to focus more heavily on
their immediate self-interest. Granted, looking only at one’s own payoff is less demand-
ing while considering others requires more complex thoughts. So, complexity might
also drive the degree of self-focusing. Yet, a fair solution (e.g., a uniformly distributed
alternative) has also focal point qualities which help players to coordinate under any cir-
189
cumstances.384 In my experiment such an outcome (as well as unanimity, cf. Sec. 5.4) was
rarely achieved even at low levels of complexity.
In particular, the high degree of risk aversion observed in the second stage of SEQ is
puzzling. This task was the simplest one my subjects encountered. They were not left
uncertain about further developments and, thus, the level of uncertainty was also low.
Looking at the two characteristics which I used to distinguish the four decision situations
(cf. Tab. 4.2), I must judge my rating system as an approach which is too static. It turns out
that I did not take (sufficient) account of a specific source of uncertainty when developing
my design, i.e., the unification process. The collective decision-making in the second
stage of SEQ was simple and handled quickly; Tab. 5.14 revealed that this task hap the
highest efficiency across all procedures. This left subjects without any information about
the other players as most rounds ended with the first ballot. In contrast, under POL most
collective decisions required at least four ballots. This enabled the subjects to reach a
deeper understanding of the occurring problem.385
Overall, the findings suggest that uncertainty and the complexity of a task reduced sub-
jects’ ability to act in a sophisticated way and undermined the relevance of social prefer-
ences. Thus, nonseparability in conjunction with the respective institutional arrangement
indeed influenced individual behavior. More precisely, subjects who faced a difficult set-
ting disregarded others and focused on their own material payoff. This was aptly sum-
marized by Cabrales et al. (2010, p. 2276), who stated that “where strategic uncertainty
conflicts with social preferences in terms of their respective recommendations [...] the
former seems to be subjects’ primary concern.” Every aspect that blurs the ability to rea-
son diminishes social considerations. By contrast, subjects who faced less demanding
challenges were more likely to compare their own to other players’ payoffs. Thus, using
the means of delegation must maintain players’ ability to understand the impact of their
individual choices on the final outcome. If not, they will focus on their own self-interest
and disregard others because taking them into account would make an evaluation of the
situation even more complicated.
Furthermore, a consideration of the payoffs of co-players does not automatically imply

that subjects become socially engaged. In my experiment many behaved competitively.
Yet, aiming to be more successful is also an action requiring one to consider others. I
do not judge whether a competitive attitude is good or bad. Competition can serve as
a discovery procedure and lead to welfare levels not achievable through collusion.386
However, this conclusion should be independent from the direction of inequality or the
investigated setting. In fact, social preferences under majority rule seem unpredictable;
384 Colman (2003, p. 139) pointed out that “focal point selection in pure coordination games is inexplicable,
though it is easily achieved in practice”.
385 As discussed in Sec. 4.3.1, the course of the unification process is an exciting puzzle in itself. Further
research (cf. Sec. 9.2.3) might look thoroughly into this matter, which is beyond the limits of this study.
386 This is aptly demonstrated by Kirstein and Schmidtchen (2003) in classroom experiments on the invisible
hand argument of Adam Smith (1776).
190
6.5 Chapter summary
at some point true fairness as exhibited in two-player games turns into competitiveness.
Consequently, the findings recommend that real-world decision rules should be designed
to reduce delegates’ uncertainty about the political consequences of their choices. In
order to preserve the relevance of social preferences, single decision makers should not
only be experts in the subjects at hand, but they should also possess sufficient information
on the preference distribution and the strategies of their co-deciders.
191
192
7 Post-experiment survey
I conclude my evaluation of the laboratory experiment with this chapter. The following
sections present additional information which I collected in a mandatory post-experiment
survey after the voting experiment. In the questionnaire I asked the participants for
some personal information, such as profession, subject and semester of study (if they
were students), age, gender, etc.387 Overall, the main part of the survey focused on the
decision-making of the participants during the experiment. The questionnaire asked sub-
jects whether their decisions made in the experiment had been guided by a specific rule,
questioned them about their focus in the process of reaching a decision, and about how
they carried out the payoff comparisons, as well as about their criteria for evaluating
competing alternatives. For all these issues the questionnaire provided ’yes or no’ an-
swers as well as free-input fields.
This short survey could not solve all unanswered questions, but it provided useful in-
formation, most importantly with respect to the validity of the experimental design. The
participants were asked to describe and classify their decisions; this revealed the par-
ticipants’ assessment of whether and how nonseparability affected their behavior under
each institutional arrangement. In combination with the voting data this delivered fruit-
ful insights. Also, using an inductive approach to look into the questionnaire led to new
aspects of the analysis.
The following sections contain the empirical analysis of the subject’s responses. I start
in Sec. 7.1 with some general comments the participants made about the experiment as a
whole. Then Sec. 7.2 analyzes subjects’ entries in the binary questionnaire. Next, I turn
towards the free-input fields, in which the participants described “with their own words”
their decision-making process. I analyze the data using a qualitative content analysis
following Mayring (2002, 2010) in Sec. 7.3. Furthermore, I use in Sec. 7.4 the text scaling
software WORDFISH developed by Proksch and Slapin (2009), to perform a computational
text analysis. Finally, Sec. 7.5 summarizes the results of this chapter.
7.1 General comments
In laboratory experiments it is common practice to give participants the opportunity to

comment on the experiment; either while answering a post-experiment survey or when
387 The descriptive information is discussed in Sec. 4.4.
193
receiving their payoffs. Yet, often subjects might not want to give an elaborate commen-
tary. This may depend on the duration of, interest in, or excitement about the experimen-
tal task. And sometimes subjects might provide a lot of remarks that have nothing to do
with the experiment at all. Fortunately, I did not experience anything of this kind during
my investigations.
In my experiment all subjects were given the voluntary option to comment on any as-
pect of the experiment by typing remarks in a free-input field.388 Overall, 49% of the
participants took advantage of this opportunity. The statements received are concerned
with the overall duration of the experiment and the waiting times within the sessions. In
addition, the participants asked many questions about communication opportunities.
I conducted an experiment on collective decision-making. Such a collective decision was
only reached when enough individual players agreed on an alternative. As I reshuffled
the groups after every round, the fastest group always had to wait until all remaining
groups had agreed as well. The same principle holds true for many individual experi-
ments because re-allocating takes time. But a collective agreement results in delays larger
than in individual tasks. At the beginning of every experimental session this was ex-
plained to the participants in order to appeal to their patience. Nevertheless, the waiting
time between the experimental rounds was the main object of criticism. About 42% of all
comments (i.e., 21% of all participants) expressed such complaints. While the delays were
an issue across all procedures, the strongest rejections are related to POL. As shown in
Sec. 5.4, under this procedure it took the longest for participants to reach an agreement.
For future experimental contributions this might be a valuable key insight to prevent
participants from this sort of negative experience. Possibilities are, for example, to use
smaller groups, providing distracting screens or tasks, disbanding group reshuffling, etc.
Admittedly, with sessions lasting 1:10 to 2:45h, the experiment was at the upper limit
with respect to the overall duration. This was expressed in some of the comments which
stated that the sessions “took quite a while”. Overall, 15% of all comments were related
to this topic; however, collective tasks need time, sometimes much more than individual
ones. No experimental session exceeded the time duration announced in the recruiting
email send to all participants.
During my whole experiment, communication between the subjects was forbidden. Yet,
as subjects were engaged in a collective task they missed this opportunity. Interestingly,
the participants did not complain about this in the questionnaire. The desire for com-
munication was rather expressed verbally to the experimenter when collecting their pay-
offs.389 Participants wanted to speak their mind to others about their (from their point
of view not understandable) behavior, to coordinate collaboration, to scrutinize the in-
terest in others’ thoughts, etc. The reasons to ask for overall silence are still coherent
388 More specifically, the question asked “Is there something left you want to tell us?” (Translation of the
originally German question).
389 I did not record the payment process, so I can only state a rough estimate: to the best of my knowledge
approximately 30% of the participants asked about communication and why it was forbidden.
194
7.2 Binary questionnaire
(cf. Sec. 4.2.2), but the clarity and frequency with which the participants desired conver-
sation opportunities with others makes me recommend it for future experiments. Espe-
cially in the context of the classification between sincere and sophisticated behavior it is
interesting how participants would make use of their ability to communicate. In Sec. 9.2.3
I discuss in detail the possibility of using communication in a much more pivotal role.
Finally, experimental results can be distorted by EDE (cf. Sec. 4.2.2), i.e., a change in sub-
ject behavior according to what they think is appropriate behavior (Zizzo, 2010, p. 2). The
comments allowed to look for evidence of this. For example, this includes statements like
that the underlying principle was discovered, that the participant was not deceived, that
the expected solution technique had been applied, that the participant was able to un-
cover the intentions of the experimenter, etc. I found no such indications of EDE in the
comments of the subjects.
7.2 Binary questionnaire
The binary questionnaire restricted subjects to answering with “yes” or “no”. Of course,
subjects could also opt to no respond (respectively not “clicking” on either of the corre-
sponding check-boxes), but this did not occur at all in this part of the survey. Tab. 7.1
shows the questions and answers of the questionnaire.390
Table 7.1: Results of the binary questionnaire
N = 168 YES NO
“Did you follow a specific decision rule?” 87% 13%
“Which players did you focus on?”

myself 99% 1%
myself and others 88% 12%
others 1% 99%
“As the group was split up, on which players did you focus on?”
myself 65% 35%
myself and the members of my delegation 46% 54%
the members of my delegation 4% 96%
all six players 38% 62%
“Which criterion of an alternative did you consider?”

the amount of my payoff 73% 27%
an equal distribution of the payoffs 34% 66%
the sum of all payoffs 24% 76%
the difference between the lowest and highest payoff 22% 78%
The results reveal that 87% of the participants answered the most basic question, if they
“followed a specific decision rule”, with yes. After all, that 13% did not follow a rule is
390 The questions in Tab. 7.1 are translations of the originally German questionnaire.
195
surprising, as the question does not exclude simplest rules. When I asked for a subject’s
focus in the process of payoff comparison, almost all subjects took their own payoffs into
consideration (99%). Yet, at least 88% also paid attention to other players, and 1% focused
solely on others. Thus, as expected, one’s own payoff was most important. Under either
delegation procedure the pattern becomes less explicit. When the group was split up,
still a majority of 65% focused on themselves, but 46% took their delegation members
into account, and 38% considered all six players. Only 4% focused solely on the three
members of their delegation. These numbers suggest that altering the decision-making
mechanism resulted in subjects shifting their focus to some extent from their own to-
wards other people’s payoffs. The last question asked for the criteria by which subjects
evaluated the payoffs. A majority of 73% focused on the amount of their own payoff.
Approximately a quarter of all participants paid attention to the sum of all payoffs (24%)
or the difference between the lowest and highest payoff (22%). About 34% looked out
for an equal distribution. While a subject’s own success still dominated the individuals’
attention, this verifies social considerations. Of course, considering others and finally
also voting according to their benefit may be two different things. But to look at relative
criteria is a first step towards “fairness”.
Tab. 7.2 shows the amount of different criteria a subject considered. 55% of participants
stated just one criterion. 33% percent considered two and 10% three criteria when making
their decision. The extremes seem negligible. Only 1% considered either none or four
different criteria.
Table 7.2: Number of different criteria a subject considered
N = 168
# of criteria 0 1 2 3 4
% of participants 1% 55% 33% 10% 1%
Overall, 44% of the subjects considered more than one criterion at a time. Tab. 7.3 takes a
look at the correlation between the different criteria. The highest value of -0.29 is obtained
for a subject’s own payoff and an equal distribution of all payoffs. However, according
to a two-tailed t-test no coefficient turned out to be significant.
Table 7.3: Correlation coefficient between different criteria
N = 168 the amount of the sum of difference between the equal distribution
my payoff all payoffs lowest and highest payoff of payoffs
the amount of my payoff 1.00

the sum of all payoffs -0.09 1.00
difference between the lowest
0.04 -0.03 1.00
and highest payoff
equal distribution of payoffs -0.29 -0.14 -0.11 1.00
To summarize, most participants stated to have considered just one criterion when mak-
ing their decision. For over 50% this was just their payoff. Thus, they neither focused to
196
7.3 Qualitative content analysis
a large extent on the members of their delegation nor did they focus on all six players.
However, at least batches of social behavior were indicated; a finding that holds for all
decision-making specifications.
The questionnaire contained free-input fields for various questions. In these fields par-
ticipants could describe with their own words how they reached their decisions. I used
those answers to conduct a qualitative content analysis, which enabled me to use an in-
ductive approach (Mayring, 2010, p. 67ff). This allowed me to identify important quotes
and to generate categories out of the data itself. To discover these aspects I developed the
coding guide (following Mayring, 2002), which is shown in Sec. A.16.391 It contains dif-
ferent categories of quotes, their corresponding definition, example statements as well as
coding rules. As according to the guide, the comments of the participants were allocated
into the respective categories.
Tab. 7.4 shows the results of the respondent’ comments in the free-input field “Please de-
scribe your decision rule with your own words?” Overall, 89% of participants replied to
this question. Most subjects stated that they tried to maximize their own payoffs (30%)
or their own payoffs under consideration of the other players’ votes (43%). This corre-
sponds to the results of the binary questionnaire (Tab. 7.1). Only a few participants were
looking for a fair alternative (8%), and even less frequently they stated that their decision
followed a Mini-max rule (3%) or was guided by a high sum off all points (3%).
Table 7.4: Free-input field “Please describe your decision rule with your own words?”
N = 168
empty / no response 11%

“my own payoffs taking the probable vote of the other players into account” 43%
“my own payoffs” 30%
“fair / consensus alternative, none player worst” 8%
“minimize my risk” 3%
“highest average / sum off all points” 3%
Next, I asked which other players or payoffs the subjects focused on. Tab. 7.5 contains
the results. 18% of all participants did not answer this question. The others focused, not
surprisingly, mainly on themselves (41%). Looking at the other categories one interesting
pattern can be seen. A subject’s own delegation did not get more attention than both
delegations (each 14%). About 7% focused on the payoffs of the other delegation. About
5% took into account which other players would be likely to vote for the same alternative.
This corresponds to the decision rule mentioned above about “taking the probable vote
of the other players into account”. It is important to note that the evaluation of other
players does not necessarily correspond with social preferences. The consideration of
391 The coding guide in Sec. A.16 is only available in German.
197
other player’s payoffs may be a first step when taking their well-being into account. In
other words, it is a necessary but not a sufficient condition. Assessing the other payoffs is
also perfectly in line with evaluating the probability of an alternative to get selected from
a pure self-interested perspective.
Table 7.5: Free-input field “Which players did you focus on?”
N = 168

“myself” 41%
“my delegation” 14%
“the other delegation” 7%
“both delegations / all six players” 14%
“players (in both delegations) that are likely to vote with me” 5%
The next question investigated the criteria for making a decision. Tab. 7.6 shows that 14%
of the subjects did not reply, and that for the rest the main focus rests on a subject’s own
payoff one more time. This is expressed in the amount of points (36%), by reaching at
least the second-best result (4%) or by avoiding poor results (12%). Again, a consider-
able amount of statements mentioned the own payoff conditioned on the probability of
success of the corresponding alternative (13%). The same logic underlies an alternative
which poses at least four high payoffs (8%) as the threshold for an alternative to be se-
lected as final outcome was at four players. Altruistic motives played only a minor role,
the difference between the lowest and highest payoff is mentioned by 7% and an equal
distribution of payoffs by 6% of the subjects.
Table 7.6: Free-input field “Which criterion of an alternative did you consider?”
N = 168

“the amount of my payoff” 36%
“avoid poor payoffs for myself” 12%
“at least my second-best payoff” 4%
“my own payoff, taking the probability of each alternative into account” 13%
“at least four high payoffs” 8%
“the difference between the lowest and highest payoff” 7%
“an equal distribution of the payoffs” 6%
The last free-input field asked the participants directly what aspect of their decision-
making changed when the decision-making split up. This happened in different ways
(cf. Sec. 4.2.6): Firstly, when the experiment moved from POL to SIM the allocation of
voting competences split up. Secondly, the voting sequence was modified when altering
between SIM and SEQ. Thirdly, both aspects changed when subjects took part in POL
and SEQ.
Tab. 7.7 reveals that 38% of the participants left this text field empty and for 7% it made
no difference. Taking a look at the respondents, 23% said that their own payoffs mattered
more. The tactical aspect of the decision-making got more important for 14%. In the
statements the subjects often indicated that they “tried” to anticipate the decision of other
198
players. This suggests that they do not think of themselves as being successful in using
this strategy. Furthermore, reaching a compromise (9%) and a subject’s own delegation
(7%) became more important.
Table 7.7: Free-input field “What changed when the decision-making procedure split
up?”
N = 168

“nothing changed” 7%
“my own payoff mattered more” 23%
“it was more important to anticipated the decision of other players” 14%
“reaching a compromise was more important” 9%
“my own delegation got more important” 7%
Next, Tab. 7.8 shows the results of the same question separately, according to the different
modifications of the decision-making procedure; i.e., which aspects changed between the
first and second decision rule of an experimental session. Looking at the answers for
the single modifications three interesting divergences emerge. Firstly, when altering the
voting sequence fewer subjects stated that their own payoff had gotten more decisive
(14% vs. 24% or 25%). Secondly, at the same time the anticipating of other players’ votes
became more important (31% vs. 13% or 17%). This suggests that introducing sequence
into the decision mechanism led to more strategic voting. Thirdly, when dividing voting
competences, reaching a compromise got more significant (14% vs. 3% or 5%). Here, the
loss of voting rights seems to influence in favor of a more balanced result. Sec. 7.4 looks
deeper into these differences.
Table 7.8: Free-input field “What changed when the decision-making procedure split
up?” separated by treatment
Aspect of the decision-making voting voting voting competence

procedure which was modified competence sequence and sequence
N 72 36 60
empty 35% 42% 40%

“nothing changed” 7% 6% 8%
“my own payoff mattered more” 24% 14% 25%
“it was more important to anticipated the 13% 31% 17%
decision of other players”
“reaching a compromise was more important” 14% 3% 5%
“my own delegation got more important” 8% 6% 5%
Overall, the comments of the participants in all free-input fields confirmed that a subject’s
own payoff was the main driving force of decision-making. This insight is complemented
by tactical considerations. Approximately 41% of the subjects described their decision
rule as focused on their own payoffs, but they were “taking the probable vote of the
other players into account”. This approach adjusted each alternative by its specific cue
validity; a strategy that became more important the more complex or uncertain the task
encountered was.
199
7.4 Word scaling
In this section I use the same free-input fields as for the qualitative content analysis.
The text was processed by the word scaling algorithm WORDFISH (Proksch and Slapin,
2009). This computational content analysis focused on the identification of similarities
(respectively differences) from a subject’s perspective when the decision rules changed.
How did the participants describe their reaction to a modification of voting competences
or sequence? In contrast to the analysis of the voting data, which is explicitly set in
contrast to previous experiments (e.g., S&K, 2010), this investigation follows an inductive
research interest.
WORDFISH is written in R statistical language (Version 2.14.2., release date 2012-02-19).392

The algorithm extracts (spatial) positions from text documents. More precisely, using
word frequencies WORDFISH places different documents into a single latent dimension of
discrepancy by means of maximum likelihood estimations. The main argument for using
WORDFISH was that it is a scaling technique and does not need any anchoring document.
This distinguishes it from other computer-based content analysis approaches which use
reference texts (e.g., Laver et al., 2003).393 Instead, WORDFISH relies on a statistical model
of word counts, more precisely a Poisson distribution (Poisson, 1837) of word frequen-
cies. This proceeding fits my research questions as I did not have a specific baseline (or
reference or anchor) document.
My analysis followed to a great extent the well documented example of Slapin and
Proksch (2008). In a first preparation step the text data was “cleared up”. I performed a
spell check, removed quotation marks, hyphens, slash and backslash signs, etc. In addi-
tion, I had to replace the German specific mutated vowels as well as the letter “ß”. Thus,
the spelling of words became consistent across all documents.
Next, I applied the text mining package included in R (tm package, Feinerer et al., 2008).
This package transforms all letters into lower case and removes all numbers from the
text. In addition, it comprises a stemming algorithm and a stop-word dictionary. A
stemmer algorithm omits morphological and inflexional endings from words and returns
the stemmed root word, so similar words are captured as one. For example, a stemmer
reduces the words “fishing”, “fisher”, and “fished” to the joint root word “fish”. By using
a stop-word dictionary the text is further cleared of all meaningless conjunctions, articles,
etc. The package already contains dictionaries for many languages which can be edited
manually. Applying these functions left a term-document matrix which had significantly
fewer unique words than the original texts. This enabled a more efficient estimation.
392 Thehomepage of the r-project can be found at http://www.r-project.org.

393 Fora comprehensive overview on different text-analysis methods cf. Slapin and Proksch (2008, p. 705-708)
as well as Benoit et al. (2005) and Benoit and Laver (2007a).
200
7.4 Word scaling
7.4.1 The latent dimension of discrepancy
When applying the WORDFISH algorithm to the documents the texts were structured in
the following way. Firstly, I separated the comments into the categories decision rule
(“Please describe your decision rule with your own words?”), focus of attention (“Which
players or payoffs did you focus on?”) and decision criteria (“Which criteria of an alter-
native did you consider?”). These categories are from now on referred to as rule, payoff
and criteria. Secondly, I divided the comments accordingly to the changes of procedure.
Therefore, I was able to compare the reaction of participants when changing the alloca-
tion of competences and the voting sequence (alternate between POL and SEQ) to reac-
tions when changing the allocation of competences (alternate between POL and SIM) or
the voting sequence (alternate between SIM and SEQ) alone. Fig. 7.1 plots the position
estimates of the latent discrepancy dimension.
Figure 7.1: WORDFISH: latent discrepancy dimension

EXPLANATORY NOTE
The figure depicts the WORDFISH estimates for the latent discrepancy dimension over three categories of comments. The
chosen order of the categories is not meaningful but arbitrary. The dashed lines resemble 95% CI.
The absolute position values are not decisive; rather, the main results arise out of the
relative spatial positions. Subjects experienced a change in allocation of competences
alone as considerably different from a change in voting sequence or both modifications
together. I found this pattern in all three categories of comments: whether I asked for
decision rules, the focus of attention or the criteria the subjects took into account. The
results were also validated by the obtained CI, which enabled a distinct demarcation.394
Overall, the modification of competences or voting sequence resulted in significantly dif-
ferent reactions. When both modifications took place together, the effects of altering the
voting sequence dominated and the reactions appeared even stronger.
394 Following Slapin and Proksch (2008, p. 710), the CI were obtained through 500 times parametric bootstrap-
ping (cf. Efron and Tibshirani, 1993). As uncertainty estimates they can be seen as robustness indicators.
201
7.4.2 Word weights and fixed-effects
WORDFISH also offers the possibility to identify the words which substantiate the spa-
tial positioning. The algorithm returns two parameters for every word. The first is a
word fixed effect, which captures the fact that some words are used more frequently than
others in all documents. The second displays a word’s weight. This resembles the im-
portance of each word in discriminating between positions. An effective analysis should
result in two types of word categories: i) frequent words without spatial meaning, re-
sulting in large fixed effects and low weights, and ii) substantial words for the spatial
discrimination, containing lower fixed effects but larger word weights. A corresponding
two-dimensional scatter-plot should reflect these findings by resembling an “Eiffel tower
of words” (a slender peak and a broad basement with a blank center, Slapin and Proksch,
2008, p. 715).
The scatter-plots in Fig. 7.2 contain the estimated word fixed effects and word weights for
the three comment categories. Importantly, the labeling of positive and negative effects
or weights only facilitates the interpretation of the plot. Those are just spatial categories
calculated by the WORDFISH algorithm, representing left and right or up and down. The
terms have no pejorative or favorable meaning. Furthermore, no category is preferable to
the other. Looking at the figures the expectations are confirmed. Words with high fixed
effects scatter around a weight of zero, but words with low fixed effects show a higher
absolute weight. The patterns are the same for all three categories.
Figure 7.2: WORDFISH: word-weights vs. word fixed effects

EXPLANATORY NOTE
The figures depict the WORDFISH estimates for word fixed effects and word weights. The results are separated according
to the three free-input fields of decision rule, focus of attention and decision criteria.
Decision rule Focus of attention Decision criteria
Word fixed effects
Word fixed effects
Word fixed effects
Word weights Word weights Word weights
Tab. 7.9 lists the words with the highest impact factor in alphabetical order.395 Again, the
labels exhibit no judging or quality relation of the two directions. Looking at the deci-
sion rules, positive comments mentioned frequently terms as “balanced”, “enforceable”,
“fair”, “majorities” or “four players”. Negative comments contained more often such
terms as “individual”, “utility maximizing” or “profit”. This indicated that the spatial
dimension resembles a kind of universalism - egoism dimension.
395 As robustness test I conducted the calculations with and without applying the stemming algorithm to the
text documents. The results do not change between these two modifications.
202
7.4 Word scaling
Table 7.9: Words with the highest impact factor
Category Terms Classification
balance, compare, enforceable, fair, four players,

positive good, higher, largest, majorities, maximum, one, universalism
optimal, part, strategy
Decision rule
area, field, high, individual, little, profit, search,
negative egoism
sum, utility maximizing
agree, at least, fifty points, game, good, large, most
positive likely, oriented, others, second-best, solution, three exclusive
players, twenty points, unlikely
Focus of attention
aware, column selection, field, high, largely, low,
negative overall, possible solution, respond, row selection, inclusive
six players, worst
adjust, average, equal, few players, group, more
positive players, much, opportunity, payoffs, prefer, profit, moderate
realistic, simultaneously
Decision criteria
all players, difference, majority, never, one
strategic utility
negative alternative, points, probably, strategy, delegation
maximization
members
Taking a look at the comments on the focus of attention the terms “agree”, “others” and
“three players” were in the positive part. The negative side contained often the expres-
sions “overall”, “column selection”, “row selection” and “six players”. This indicated
an inclusive - exclusive aspect (whole group vs. own delegation). Both sides contained
references to the strategic aspect of the decision problem (“unlikely”, “most likely” or
“possible solution”) and a resulting risk avoidance (“second-best” or “worst”).
The positive comments on decision-making criteria were “equal”, “group”, “average”,
“more players” and “realistic”. On the negative side were the terms “all players”, “dif-
ference”, “majority”, “probably” and “strategy”. So the positive aspects were categorized
as moderate and consensus-orientated and the negative terms pointed towards strategic
utility-maximization.
To illustrate, Tab. 7.10 shows words with a word weight around zero. These words were
mentioned very often but did not exert a significant influence when calculating positions.
This does not mean that these terms were irrelevant when looking at the decision-making
behavior of the participants in the experiment. The low weights just indicate that they
were not used differently between the procedures. They were often used under all de-
cision rules! It may be that these phrases resembled basic and important aspects of the
decision-making process in general. They were not affected by different forms of voting
competences or decision sequences.
Referring to the results in Tab. 7.9 the left (or negative) dimension contained the aspects
egoism, inclusiveness and strategic utility maximization. The right (or positive) dimen-
sion, on the other hand, referred to the terms universalism, exclusiveness and moderate
behavior. Linking this to the findings of the spatial position estimations Fig. 7.1 I was
able to associate the decision rule modifications with corresponding behavioral patterns.
203
Table 7.10: None decisive terms
Category Terms
agreement, alternative, anticipate, arithmetic, balanced, choose, comparison, compromise,

consequence, consideration, correlation, criterion, doubt, estimate, generality, group
Decision rule members, maximizes, maximum value, members, opportunity, payoff, players, profit,
similar, simultaneously, stable, successfully, sure, delegation, uncertain, vote, winning
chance
anticipate, attempt, avoid, balanced, beginning, best, both, choice for developing,
circumstances, column, combination, compromises, concentrated, consequence, decision,
Focus of attention error, fragrance, initially, intuitive, largest, low, lowest, maximum, necessarily, odds, options,
payoffs, players, points, predict, probability, probable, profit, relatively, result, round,
satisfied, scheme, significantly, suspected, delegation
accordingly, alternately, alternative, arithmetic, at least, balanced, best, center, difficult,
eliminate, euro, experiment, failure, high, interest, loss, lowest, maximum, minimized, more,
Decision criteria
mostly, optimal, own, plus, potential, predict, prefer, purchase, safe, small, superior, highest,
variance, variation, yield
A change of the allocation of competences (between POL and SIM) was associated with
positive positions. This linked voting competences to thoughts of universalism and ex-
clusiveness. Participants considered who had to be taken into account for reaching a
consensus. The moderate behavior may also indicate a risk minimizing approach. By
giving up the overall highest payoff a subject aimed to avoid the worst possibilities.
A change in voting sequence (between SIM and SEQ) or modifying both aspects together
(between POL and SEQ) received negative position estimates. This linked the voting
sequence to egoism and strategic utility maximization behavior. Thus, subjects seem to
have captured the theoretical implication of an alternate solution concept.
7.5 Chapter summary
This chapter supplements the empirical analysis of the experiment. In addition to the vot-
ing data it contributes insights from the participants’ perspective. The post-experiment
survey allowed for general comments on the experiment but focused on the decision-
making process of the participants. It contained ’yes or no’ questions as well as free-input
fields.
The general comments proved the desire of subjects to communicate. Across all proce-
dures the collective task sparked the desire for consultation opportunities. This desire
was expressed as a wish, not as a complaint. Some comments criticized the duration of
and delays within the sessions, but delays cannot be avoided in collective tasks. Most
importantly, I found no evidence for disturbing effects such as EDE.
The binary questionnaire revealed that most participants considered just one criterion
when making their decision, and for about 50% this was, not surprisingly, just their own
payoff. In addition, at least a few social considerations were indicated, a finding that
holds for all decision-making specifications.
204
7.5 Chapter summary
The free-input fields were analyzed using a qualitative content analysis as well as a com-
putational text analysis. The qualitative content analysis confirmed the binary findings.
A subject’s own payoff was the main driving force for the decisions made in the experi-
ment. The free-input fields supplemented this behavior with a tactical aspect as subjects
took “the probable vote of the other players into account”. Thus, the periphrastic ex-
pression of sophisticated voting receives a more vivid description. Such a procedure was
used with increasing frequency, the more complex or uncertain the decision-making pro-
cedure got.
The computational text analysis indicated that the alternation of the decision-making
procedure influenced the motives of the participants. A change of decision-making com-
petences immediately triggered the participants‘ thinking patterns for who had to be
taken into account. This led to a higher exclusiveness in voting. A change in voting se-
quence was processed through adjusted strategic utility maximization. Both patterns are
in line with the qualitative content analysis as well as the binary questions.
Overall, the most important insight of the post-experiment survey was its validation of
the design. Subjects did not misunderstand the experimental instructions. This would
have led to dissimilarities between the perceptions of what a subject thought they were
doing and their actual behavior. Fortunately, this was not observed.
205
206
8 Complex reality and limited models
The previous parts of my study plainly show the relevance of NSP; I demonstrate the
vulnerability of inference when falsely assuming separable preferences (cf. Chap. 3) and
clarify the impact of nonseparability on collective (cf. Chap. 5) as well as on individual
(cf. Chap. 6) decision-making. Yet, I focus only on the knowledge gained if NSP are in-
cluded into analytical research. The corresponding requirements for the consideration of
nonseparability have so far only been mentioned in passing. In this chapter I rectify this
deficiency and add the missing assessment of the necessary expenses for including NSP.
In Sec. 8.1 I discuss possible empirical implementations and offer a rough estimation of
related cost in terms of necessary tasks and methods. This enables an assessment of the
knowledge gained with respect to the required effort of modeling. The subsequent sec-
tions situate the argument in a broader context. All research must decide between includ-
ing more details and keeping models feasible. Sec. 8.2 discusses the concept of abstrac-
tion, its application in analytical models and its legitimacy. Hereby, I set the identified
threat to inference (ARGUMENT 1) against the benefits of simplifying assumptions. Subse-
quently, I trace in Sec. 8.3 the latest development in social science research paradigms in
which a ’new thinking’ suggests a shift away from artificial assumptions towards a more
realistic modeling of behavior. This affects analytical research in general and is most cer-
tainly relevant for the assumption of separability. My experiment clearly demonstrated
the behavioral effects of NSP (ARGUMENT 2 and ARGUMENT 3), which therefore should
not be ignored. Finally, Sec. 8.4 summarizes whether and how nonseparability should be
included into analytical research. It also places my contribution into the current state of
political science.
8.1 The necessary effort
The previous chapters investigate the possible gains of including NSP into scientific re-
search. I use both field and laboratory data for this assessment. What stands against
their observance are the necessary requirements for including nonseparability, which are
twofold. In either case, the corresponding data has to be i) collected and ii) incorporated
into the analysis. Both tasks make claims that exceed those of standard approaches and
which might challenge researchers.
In the first step the necessary data must be gathered. Collecting data can turn into an
exhausting experience, and more data most certainly implies more cost. It is obvious
207
that collecting data on NSP will lead to higher expenditures. First of all, this concerns
purely monetary resources for access rights, licenses, etc. Then, of course, also supple-
mental work hours and additional staff might be necessary. The coding of NSP in the
DEU data set (Thomson et al., 2006), reported in Sec. 3.4, may serve as an example. At
least three graduate student coders were necessary to obtain sufficient reliability. All of
them had to receive a thorough introduction to the concept of nonseparability and spent
quite a few hours of work on the coding of the 66 law proposals. Yet, such additional
effort is inevitable for an appropriate data provision. The following paragraphs provide
an overview on such tasks by summarizing the pros and cons of different methods. Un-
fortunately, none of them comes without problems.396
SURVEYS AND INTERVIEWS
Like other data on individual decision-making (e.g., ideal points, salience and believes),
the information on NSP can be collected through interviews and surveys. While the
measurement of separable preferences can be limited to an actors’ first preference, NSP
require an evaluation of actors’ utility function at several values. Thus, the collection
turns into a disproportionately more complex task (cf. Sec. 2.4).
An accurate example for measuring NSP can be found in Lacy (2001b). The author mea-
sured NSP in survey respondents’ preferences for the U.S.397 In his survey he used the
following arrangement of questions to identify NSP (Lacy, 2001b, p. 253ff). First, respon-
dents were asked for their unconditional first preference on two issues. Second, in order
to identify NSP, the issues were alternately fixed at specific levels in follow-up ques-
tions which asked for the resulting modification of the other issue. This questioning
was repeated for all combinations and orders of issues. The study clearly showed that
common survey questions about ideal position are insufficient when dealing with NSP
(cf. Sec. 2.4.6). The same proceeding holds true for expert interviews. The criteria for
selecting experts stay the same as when only concerned with separable data. But in the
proceeding of the interview the questions have to be structured in the same manner as
used by Lacy (2001b).
It holds for both data-collecting techniques that the hypothetical character of the follow-
up questions is demanding. It represents a challenge to the imagination as well as the
attention of the respondent and stresses its patience. The additional hypothetical evalu-
ations of actors’ utility functions at several values leads to longer interviews, much more
questions and a more complex task.
In addition, questions repeated iteratively may develop distorting self-dynamics. If one
asks people many times about hypothetical scenarios this will affect the answers given.
396 Cf. Schnapp et al. (2009) for a comprehensive overview on the collection and types of data commonly used
in political science research.

397 Lacy(2001a,b) found that preferences on income taxation depended on crime prevention policies, prefer-
ences on environmental pollution depended on environmental regulation, preferences on defense spend-
ing depended on social spending, preferences on immigration policy depended on the constitutional
status of English being the only official language in the U.S., etc.
208
8.1 The necessary effort
For example, in such interviews politicians do not want to look like a pushover. So they
announce to stick to their current policy (e.g., to liberalize certain industries), although
they would prefer an adjustment under the scenarios discussed (e.g., altered regulatory
safeguards).
OBSERVATIONAL DATA
In addition to directly asking individuals for their preferences, they can also be ob-
tained through behavioral observations. Such observations are not restricted to a spe-
cific setting; instead, they can take place in a subject’s everyday environment. This non-
interference enables surveyors to observe a subject’s natural behavior without distortion
(cf. Ortmann, 2005). Observational data generally offers larger amounts of data, espe-
cially in the form of longitudinal data which is otherwise difficult to obtain. However,
establishing causal effects can be statistically challenging (e.g., Hill et al., 2005).
As discussed in Sec. 6.2, two types of behavior are commonly observed in electoral pro-
cesses; these are sincere as well as strategic voting (Herrmann, 2012). When aiming to as-
sess their respective spreading, the problem is that “to identify strategic voting requires
that we know both the voter’s true values and the voter’s actual expression of the values
in a vote. From direct observation we can know only the latter. We must infer the former
from other and softer evidence” (Riker, 1982a, p. 167). This makes it difficult to distin-
guish between them. “If we rely only on the observable actions (e.g., votes) of legislators
to test theories of strategic behavior we cannot determine whether the observed behavior
represents legislators’ true preferences or whether the legislators are acting strategically”
(Clinton and Meirowitz, 2004, p. 676).398
To uncover an individual’s true preferences is also of paramount importance for identi-

fying nonseparability. Yet, in particular Chap. 3 reveals in detail how much effort it takes
to obtain individual data. NSP are subject to the same limitations and it seems difficult
to uncover the relevant relations using solely observational data.399 Observed behav-
ior nearly always offers “multiple causal paths” (Braumöller, 2003, p. 209). This implies
a high degree of “substitutability” (Braumöller, 2003, p. 215) in explanation of the data
which can resemble nothing more than an educated guess for the truth.400
Observational data may often provide in depth knowledge on a specific effect but not
on universal patterns. The argument made by Levitt and List (2007, p. 160) concerning
low “cross-situational consistency of behavior” describes this very well. As long as NSP
are not well studied and the relevant relationships and policies are not known, this ap-
398 While empirical investigations of legislative strategic voting are relatively scarce, cf. Clinton and
Meirowitz (2004) for an overview of such approaches.
399 I am not arguing that investigating nonseparability with observational data is impossible. In fact, the
problems encountered are oftentimes just similar compared to conducting a survey (e.g., costs and op-
portunities of data collection). Yet, the previous permanent neglect of NSP does not provide the necessary
empirical framework.
400 Braumöller (2003, p. 215, table 1) offered a selection of examples for international relations, comparative
and American politics.
209
proach is indeterminate but might be worthwhile in future research. Therefore, I discuss

in Sec. 9.2.2 the possibility of field experiments, i.e., a first step into this direction.
EXPERIMENTS
With respect to validity, laboratory experiments are the opposite of observational re-
search. The pros and cons of this method are assessed in Sec. 4.1 in detail. In short,
experiments possess a high degree of internal validity at the expense of external validity
(Schram, 2005). Inducing preferences according to the research question is only possible
in the laboratory (McDermott, 2002, p. 326). In particular, the monitoring capabilities as
well as the likelihood of eliminating interfering effects are excellent.
Of course, this tells us nothing about where NSP exist in reality. Such “proof of exter-
nal validity is always empirical” (Morton and Williams, 2010, p. 196). That is why fur-
ther experiments and other complementary studies are inevitable. Yet, an experiment
enables the investigation of the effects which NSP exert on decision-making processes;
thus, for “theory development, testing and refinement” (McDermott, 2002, p. 341). In
other words, one can learn about nonseparability without a long and onerous search for
suitable situations. Of course, such investigations can only be the first step in a series
of further analyses; but that is the exact purpose of an experiment and it fulfills it pretty
well (cf. Sec. 1.3.2).
SIMULATIONS
The previous methods aim to actually measure NSP. In Chap. 3 I apply a different ap-
proach; the determination of NSP is accomplished by means of simulation techniques.
This was the only possible way as the original data collection (through expert interviews)
was not concerned with nonseparability. This leaves the technique with the limitation
that is it only possible to theorize on and not prove the existence of NSP.
The simulation was based on three steps with each focusing on a single aspect of NSP
(cf. Sec. 3.4): existence, direction and reciprocity. Firstly, it was identified whether or not
actors’ preferences over two issues may be nonseparable at all. Secondly, the direction
of the nonseparability was added, i.e., whether the issues in question are supplements
or complements. Third, it was determined whether the potential nonseparability is re-
ciprocal or not (cf. Sec. 2.4.5). This coding scheme was then used for a computational
comparison of legislative models of decision-making.
Simulations should only be used in combination with a model or theory already well
proven (Clarke and Primo, 2007). The unconstrained bargaining model, constrained bar-
gaining model and agenda-setting model used in Chap. 3 fulfill this claim as they can
rely on a vast literature on legislative of decision-making (cf. Sec. 3.1).
To summarize, Tab. 8.1 lists the pros and cons of the different methods. As neither of
them comes without flaws, the most promising way forward might be to combine the
210
8.2 The concept of abstraction
various approaches. Together they can balance each other’s respective problems, as their
strengths are complementary. For example, Hamenstädt (2012) argued in favor off bring-
ing the lab into the field in order to combine the explanatory powers of both methods.
Sec. 9.2.2 will illustrate this principle further and demonstrate its possible significance
for future research. Of course, this is no easy task, but the numerous opportunities for
scientific analysis allow a closer examination of the nonseparability phenomenon.
Table 8.1: Methods of data collection for investigating nonseparability

EXPLANATORY NOTE
The table lists possible methods of data collection and their pros and cons with respect to the assembling of the necessary
information to determine NSP. To a large extent these points agree with general characterizations of the methods (e.g.,
Moses and Knutsen, 2007).
METHOD Pro Contra
Surveys and determine NSP exactly hypothetical questions and long

(expert) interviews sessions necessary
Observational data obtain “true” NSP from a multiple possible explanations make
natural setting a determination difficult
Laboratory induce preferences according reaction to induced NSP must not be the
experiments to research question same outside the laboratory
Simulations feasible even ex-post applicable only with well proven models;
can only theorize on and not prove
the existence of NSP
EMPIRICAL EVALUATION
The second step of incorporating the eventually collected data comprises the operational-
ization of NSP. Here, Sec. 2.4 discusses the theoretical aspects in detail and Chap. 3 sup-
plements an empirical implementation into different legislative models. In general, mod-
els become more complex and demanding as they incorporate additional information.
However, the fact that a problem is somewhat more complicated alone is no reason to
refrain from scientific investigation. Improvements in computational technology enable
the calculation of most sophisticated theories (cf. Humphreys, 2004).
Overall, the claims represented to the researcher when setting up the model including
NSP are not unreasonably high. Accuracy is an inevitable requirement to which scien-
tific research must comply. This holds also for the assessment of nonseparability. The
(admittedly) high demands on data quality and computational resources are no valid
counter-argument.
Abstraction resembles a fundamental aspect of analytical research, especially when spec-

ifying a theoretical model. Reality is too complex to comprehend without simplification
(Kesten and Pnueli, 1998). As the only completely accurate way to map a city is to set up a
211
1:1 template, even a vast map with a gigantic resolution would omit data. Weber argued
that it is not important if a model does not mirror reality perfectly and in every detail
(cf. Weber, 1968, p. 190ff).401 In fact, he stated that “ideal types” (e.g., a perfectly and
rigorously rational voter) only at all exist in abstract concepts (cf. Shils and Finch, 1949,
p. 89-95). Yet, without a model it would not be possible to gauge real behavior. As early
as 1836, John Stuart Mill’s famous essay “On the Definition of Political Economy; and
on the Method of Investigation Proper to It” took advantage of “a hypothetical subject,
whose narrow and well-defined motives made him a useful abstraction in economic anal-
ysis” (Persky, 1995, p. 222-223). Miljkovic (2012) aptly summarized in her preamble that
the “complexity of nature and consequently the complexity of everyday life processes
often make the mathematical models deterministically unsolvable. Moreover, if such so-
lutions do exist, usually a lot of resources are required to find them. Therefore, the idea
of approximation has developed as an irreplaceable tool for handling many problems.”
A model resembles a conceptual representation of the real data generating process (Mor-
ton and Williams, 2010, p. 194). Yet, by its very definition a model is restricted to some
aspects of a real phenomenon and different models can focus on completely different as-
pects of the same event (Lave and March, 1975, p. 3ff).402 To correctly comprehend the
model, its implications and predictions requires knowing its original purpose and the as-
sumptions made when setting it up (cf. Grosslight et al., 1991). The performance assess-
ment of every model depends on multiple characteristics, such as validity, reliability and
consistency. Downs and Wildenmann (1968) advocated that theoretical models should
first of all be evaluated with respect to their predictive accuracy and not the reference to
reality of their assumptions. However, a model’s ’cost of use’, especially in comparison
to alternative models, is essential.
Each systematic approach must thoroughly consider the assumptions made during the
research process. Bloomfield and Anderson (2010) stated that it is necessary to think
more clearly about the nature of assumptions in their discipline of economics. Bloom-
field et al. (2009) discussed typical categories of assumptions used by experimentalists.
They identified structural assumptions which describe the institutions in which agents
interact (e.g., information distribution, possible set of actions and incentives), behavioral
assumptions which characterize agents’ preferences and decision-making (e.g., the form
of the utility function) and equilibrium assumptions that describe the solution concepts
applied (e.g., Bayesian Nash equilibrium or backward induction). These categories must
all be considered when setting up an analytical model.
401 Klein (1985) examined the mode (proximity or structure) and dimensions (physical, temporal, attribu-
tional or construal) in which abstraction takes place. He identified a limit for the principle of abstraction
“since people cannot handle substantial amounts of construal abstraction” (Klein, 1985, p. 677) (which
represents a participant’s subjective perception of the content of a model). Here, “anything more than
trivial abstraction in this dimension alters the meaning of models to such an extent that human players
will not be able properly to understand them” (Klein, 1985, p. 671).
402 Long (2006, p. 5) clarified that in the very Aristotelian tradition abstraction is to be understood “as a matter
of attending to some aspects of a thing and ignoring others.”
212
In general, each additional assumption complicates the creation of a model. So why not
stay with a few assumptions and slim models? Looking into the field of behavioral eco-
nomics, and particularly into research on social preferences, Mertins (2008, p. 33) pointed
out that a “large literature has shown that we can go a surprisingly long way with very
simple models of fairness in some important classes of games.” Thus, why should more
and more complex and intricate models be necessary at all? This depends on the knowl-
edge gained (whether it is worth the effort) and the reliability of the findings. Cameron
and Morton (2002, p. 793) explained that “even inadequate formal models” possess virtue
as they “have the advantage of clarity.”
Any concepts with substantial impact should be included into prospective scientific re-
search.403 Otherwise a simplifying assumption (in case of my research the presumption
of separable preferences) has its legitimacy. This follows Occam’s razor404 which calls,
all else equal, for the simplest model. Occam’s razor constitutes a widely used principle
in economics.405 Yet while this norm has provided valuable assistance for scientific re-
search, “its continued use [...] risks significant opportunities to be missed” (Domingos,
1999, p. 409). Reid (1987, p. 551) warned that a “too ready application of Occam’s ra-
zor’ (broadly defined to champion elegance and simplicity) slows rather than speeds the
growth of economic knowledge.” There exists even stronger criticism406 which is part of
an intense controversy about its application in science in general.407 Thus, a thorough
investigation of possible insights is inevitable before rejecting alternative specifications.
This discussion, so far, seems to refer mostly to the discipline of economics. Yet, all ana-
lytical research has to face an assessment of more details included and manageable limits
of modeling. This also holds true for political science, even though a famous quote of
Otto von Bismarck states that “Politics is not an exact science” (Speech to the Herren-
haus, Otto von Bismarck, 1863). Aptly enough, Noel (2010) replied that “politics is not a
science, but it can be studied systematically” and thus be subject to analytical research.
From the beginning of this study it was laid out that researchers have to face a trade-off
between excluding NSP and keeping the analysis simple, or including NSP and dealing
403 In addition to adding new findings, confirming or refuting of existing knowledge is also a valuable con-
tribution.
404 The term goes back to William of Occam who stated in the late middle ages that “nunquam ponenda
est pluralitas sin necesitate” (Domingos, 1999, p. 409). This was translated by Tornay (1938) as “entities
should not be multiplied beyond necessity”.
405 In his marvelous description of the history of economic ideas, Heilbronner (2000, p. 103) showed that it
was David Ricardo who “gave the powerful tool of abstraction to economics.”
406 An aptly example of such criticism is the contribution “Razoring Ockham’s razor” (http://
rationallyspeaking.blogspot.de/2011/05/razoring-ockhams-razor.html, posted at May 6th 2011)

by Pigliucci (2013) where the author emphasized the necessity for a discussion on the principle’s proper
application. He also described a contradiction within Ockham’s razor; namely that “philosophers often
refer to this as the principle of economy, while scientists tend to call it parsimony. Skeptics invoke it every
time they wish to dismiss out of hand claims of unusual phenomena (after all, to invoke the “unusual”
is by definition unparsimonious, so there).”
407 Riesch (2010) interviewed 40 scientists on their views of Occam’s razor and simplicity. She received vari-
ous interpretations of the principle; the responses ranged from complete rejection to the assessment that
“Occam’s razor indeed forms an integral part of scientific method” (Riesch, 2010, p. 86).
213
with more complex requirements. A complex reality (including NSP) and abstracting
models (frequently ignoring NSP) are difficult to reconcile. So far nonseparability has
often been neglected. The justifications for not taking account of NSP varied. For “the
purpose of simplifying” (e.g., Morgan, 1990, p. 321) as well as because it “is common”
(e.g., Le Breton and Sen, 1999, p. 606) were the most frequent arguments.
I do not question the general principle of abstraction, only its overuse. The threat ne-
glected nonseparability poses to the inference of analytical models clearly substantiates
the danger of overexploitation. Again, this is not a question which applies only to the
topic of NSP. Reflecting on the adequacy of abstraction in economics, the sociologist
Emil Durkheim made the noteworthy remark as early as in 1887 that “without doubt,
in the field of economics the application of abstraction is legitimate” (Emil Durkheim, as
quoted by Swedberg and Maurer, 2009, p. 53). But he also pointed out that “just not all
abstractions are equally correct. Abstraction requires isolating a part of reality, not to let
it disappear” (ibid.).
By ignoring NSP, fundamental aspects of decision-making are faded out of analytical

consideration in favor of clearer mathematical models. Whether this is an appropriate
way of investigating social phenomena is highly questionable. My study can only be the
first step in considering this question thoroughly.
8.3 A more realistic view of human behavior
In his noteworthy essay “The Unreasonable Effectiveness of Mathematics in the Natural

Sciences“ the physicist Eugene Wigner (1960) pointed out that the mathematical struc-
ture of physics itself has again and again led to further discoveries. His main argument
was “that mathematical concepts turn up in entirely unexpected connections. Moreover,
they often permit an unexpectedly close and accurate description of the phenomena in
these connections”. This could not be a mere coincidence, even if we thus far “do not
understand the reasons of their usefulness” (Wigner, 1960).
This work has inspired much research on the phenomenon that “the enormous usefulness
of the same piece of mathematics in widely different situations has no rational explana-
tion” (Hamming, 1980, p. 82). Maybe there is a deeper reason why (it seems) “that the
laws of nature are written in the language of mathematics” (Galileo Galilei, as quoted by
Hamming, 1980, p. 82). Over the last fifty years, many contributions across disciplines
followed up by assessing the usefulness of mathematical applications for their own field
of research (e.g., Lesk (2000) for molecular biology and Tegmark (2007) for physics).408
The idea was not only appreciated but also criticized, e.g., Hamming (1980) and Gray
(2011) spoke against Wigner (1960) for giving only partial explanations or staying vague
408 For an up to date review of contributions cf. Russ (2011).
214
in his argumentation. Nevertheless, both also stated that “these problems are in fact deep
questions and worthy of further investigation” (Russ, 2011, p. 211).
Winger’s contribution was written over 50 years ago; this raises the question if it is still a
contemporary issue. With regard to its future relevance Djorgovski (2005, p. 131) argued
that “applied computer science is now playing the role which mathematics did from the
17th through the 20th centuries: providing an orderly, formal framework and exploratory
apparatus for other sciences.” If so, there is a shift into a more technical dimension, but
the basic phenomena that natural science is able to express even the most complicated
relations with simple formulas remains.
THE DIFFERENCE BETWEEN SOCIAL AND NATURAL SCIENCES
The ability of natural science to describe their object of study with simple equations has
triggered a desire in social science to work in the same way. However, we reach limits
when aiming to describe humans and their behavior in simple formulas. This is most
evident when “economists suffer from physics envy over their inability to neatly model
human behavior” (Halevy et al., 2009, p. 8). Velupillai (2005, p. 849) even claimed “that
mathematical economics is unreasonably ineffective.”
The difference between the disciplines has been summarized aptly by Taleb (2007). He
concluded that “if you know all possible conditions of a physical system you can [...]
project its behaviour into the future. But this only concerns inanimate objects. We hit a
stumbling block when social matters are involved. It is another matter to project a future
when humans are involved, if you consider them living beings and endowed with free
will” (Taleb, 2007, p. 183). It is inevitable to accept that social dynamics do not follow
incontrovertible laws of nature. North (2008, p. 25) emphasized that we exist in “a non-
ergotic world. An ergotic world would be one in which the fundamental underlying
structure is uniform and exists everywhere. In such a world, if you understand that
fundamental underlying structure and you want to solve a new problem, you go back
to fundamentals and then build your theory based on the structure. Now that is what is
done in the physical sciences and the natural sciences. The social sciences, however, have
no such tools; and, what is much more difficult - the world just keeps changing.”
This leaves the question what opportunities social science has left, if the focus on math-
ematical applications provides no answer. All in all, there are two possibilities. On one
hand, research can aim for an alternative tool which serves as simplification device. On
the other hand, research can engage and incorporate complexity as good as possible.
Halevy et al. (2009) advocated “The Unreasonable Effectiveness of Data”. The authors
focused mainly on their discipline of data mining and computer technology. Yet, their
argument is valid for all scientific research409 when applying optimization algorithms,
409 Already
Achen (1983) discussed the amount of data collected for political science research. While he
advocated a development “towards theories of data”, he also pointed to risks in this undertaking; namely
measurement error and aggregation bias.
215
scrutinizing large amounts of data or formulating extensive systems of equations. The

improvements in computer technology and the enormous growth of available data410
enable the investigation of previously inaccessible research questions (cf. Humphreys,
2004). Here, self-learning algorithms that search through exabytes of data open many
doors for science and industry.411 Conducting laboratory experiments may represent an
important building block for this prospective approach, as it enables researchers to obtain
accurate information in a controlled environment.
As an alternative, Russ (2011, p. 211) promoted the necessity to accept complexity. He

argued that focusing on rigid and idealized models or theories neither represents the
present nor the future of social research. This calls for a less idealized but more realistic
representation of behavior (Ariely, 2009) which takes into account the inherent complex-
ity of social systems. For example, the intricate dependencies within human preferences.
THE INHERENT COMPLEXITY OF SOCIAL SYSTEMS
Political science was and is heavily influenced by neighboring scientific fields. Looking at
its methodological framework Beck (2000) defined it as “welcoming discipline”. The au-
thor argued that political scientists “use a variety of methods to attack questions related
to political institutions and behavior. Although the methodological issues are defined by
our political questions, we freely use whatever methodological solutions are available.
Thus political methodology has freely drawn on insights from econometrics, psychomet-
rics, sociology, and statistic” (Beck, 2000, p. 651).412 Accordingly, Druckman and Lupia
(2006, p. 18) emphasized that “context, not methodology, is what unites our discipline”
when assessing political science.
Not surprisingly, a large part of quantitative political research followed neoclassical eco-
nomics to an assumption-centered research approach which is characterized by a “rigor-
ous corset of formal and propositional logic” (Ruckriegel, 2010, p. 3).413 Bradley (2006,
p. 17) summarized the neoclassical paradigm, admittedly somewhat exaggerated, as fol-
lowing Pythagoras’ doctrine that “all is number”.414 Of course, the formalization was not
simply a goal in itself. The introduction of the Homo economicus followed the purpose
to pursue social science in analogy to natural science tradition as an exact science (Matis,
2007). The intent of the highly formalized approach was to offer the “availability of new
logical and mathematical tools” (Walliser, 2008, p. 2).
410 The study of Manhart (2011) documented this growth in detail.
411 The study of Velten and Janata (2012) investigated the “explosion of data availability” and how this
changed not only the IT but all business sectors.
412 This collecting of methods from other fields has also been criticized. Achen (1983, p. 70) argued that
“techniques invented by statisticians, psychologist, and economics [were] often meant for very different
tasks”. A specific version of this general criticism is the disapproval of the common and uncritical use
of Euclidean utility functions to model political decision-making (cf. Sec. 2.4.2) by Milyo (2000b,a) and
Benoit and Laver (2007b).
413 This was not without controversy and in some fields of political science also a counter-movement was
visible (cf. Monroe, 2005).

414 Cf. Gill (2006) for a brilliant introduction into mathematical concepts in social science.
216
However, some assumptions made when introducing the Homo economicus are highly
controversial and have often been called into question (e.g., Opp, 1999; Taylor, 2006).
The “optimization mathematics forced economists to make very ambitious assumptions
about the intellectual capacity of its agents - the controversial assumption of perfect ra-
tionality” (Geisendorf, 2009, p. 163) is one of the most prominent examples. This doubt
about the rationality assumption is also reflected in my experimental results. The com-
plexity of decision situations affected by NSP clearly exemplified human cognitive limi-
tations.415 Also, a “false assumption is that almost all people, almost all of the time, make
choices that are in their best interest or at the very least are better than the choices that
would be made by someone else” (Thaler and Sunstein, 2009, p. 9). In particular, Ariely
(2008, p. xii) provided “a wide range of scientific experiments, findings, and anecdotes
that [exemplify] how systematic certain mistakes are.” Thus, “orthodox conceptions of
rationality are evidently internally deficient and inadequate for explaining human inter-
action” (Colman, 2003, p. 139). McFadden (2006, p. 10-11) concluded that “Homo eco-
nomicus, sovereign in tastes, steely-eyed and point-on in perception of risk, and relent-
less in maximization of happiness, is a rare species.”
The simplifying assumptions implied an artificially high level of (only seemingly) preci-
sion and scientific reliance (cf. Ortlieb, 2010). The method was bound to a serious internal
contradiction. On the one hand, complex decision situations and complicated action port-
folios were traced back to basic patterns based on simple models to analyze the dynamics
and structures of (political) behavior (Ostrom, 2005). On the other hand, obtaining the de-
sired parsimony of the model required a large amount of complexity reduction through
omission of less important aspects defined beforehand (Braun, 2013, p. 182).
Today, this is no longer state of the art.416 Mankiw and Taylor (2010, p. 864) pointed out
that “economics may borrow some methodology from the hard sciences but as a science
of human behavior some of these methods are built on ever shifting sand.” Instead of
developing even more sophisticated models, economics is following a development that
might be described as “rediscovering the human side in economics - from neoclassic
back(!) to behavioral economics” (Ruckriegel, 2010, p. 1) or “the return of the lost human -
ways to Homo sapiens economicus” (Dopfner, 2002). It follows the insight that “modern
economics must be based on a realistic description of human behavior - not, as previously,
on the assumption that we all act rationally” (Krugman, 2010). This leads to “an economic
theory based on the actual behavior of people, not [...] one that is based on how people
should behave” (Ariely, 2008, p. 265).417
415 This is evidenced by the prevailing position of sincere behavior, especially under SIM.
416 For an overview on the development of the rational choice paradigm cf. Gilboa (2010).
417 Fehr and Schwarz (2002) summarized this change incisive. For outsiders economics often seemed like
’rocket science’; like a highly technical juggling with data and formulas looking for (a relatively mecha-
nistic understood) causality. This ignored that economics is ultimately a human science; i.e., a science of
human behavior. Yet, this view has in the last few years become increasingly important again. The au-
thors attributed an important role in this development to “to the advent of micro-economic experiments,
which brought about almost a methodical revolution” (Fehr and Schwarz, 2002, p. 5).
217
The shortcomings of current theories and models are diverse.418 Akerlof (2007) assessed
macroeconomic theory based on the traditional neoclassic assumptions of the Homo eco-
nomicus as leading to inaccurate conclusions. In his remarkable speech at MIT, Friedman
(2007) introduced the term “gross-individual-product”. He claimed that it constitutes a
necessary adjustment of outdated economic theories which had not “fully been able to
capture what is happening far below the firm level, on the individual level” (Friedman,
2007). In his view this resembles a mismatch between the individual and the collective.
The standard research framework is too far away from what is observed in the economy
today. A similar plea was made by Kaufmann (2012), who argued in favor of breaking
out of standard assumptions and to pay attention to the ethical questions of competition,
instead of following the mantra of the free-market indiscriminately.
This call for a more accurate economic theory was strengthened by the recent global fi-
nancial crisis (Akerlof and Shiller, 2009). A purely rational framework was no longer
sufficient (Ariely, 2009). The speculative bubbles turned the focus on irrational behavior
and “animal spirits” in the markets (Keynes, 1936; Dow and Dow, 2011). The financial
crisis was not accountable within the standard economic models because it would have
been too complicated to calculate (Johnson, 2012). However, Dalio (2011)419 provided a
framework for economic forecast which differed from the traditional perspective on sup-
ply and demand. The framework received a lot of attention, as it could explain aspects
of the current crisis (as well as the problems in solving it) that were otherwise not under-
stood. A central part of the theory focuses on the process of portfolio decision-making.
In addition to this inner solution within the formalized approach a lot of research in
cognitive psychology, anthropology, evolutionary biology, neurology and sociology tries
to determine how people actually think (McFadden, 2013). Starting from simplifying
assumptions and ideal types has provided invaluable scientific insights (e.g., in game-
theory, McCarty and Meirowitz, 2007) but by their innermost definition those theories
are very far from “real” thinking. Here, neuroeconomic research has made huge progress
in opening up the “black box” of the brain (Camerer et al., 2004).
DellaVigna (2007) provided an overview on the current state of behavioral economics.420
He focused on deviations from the standard model in terms of preferences, beliefs and
decision-making aspects.421 Following the same principle Walliser (2008) discussed ex-
tending the framework of game theory with respect to findings of various laboratory ex-
periments on individual choices and collective interactions.422 Most attention was paid
418 Rabin (1998) reviewed various psychological findings on human judgment and behavior. He also pointed
out how those can be used to improve assumptions about individual behavior in economic theory.
419 Ray Dalio is an American businessman and founder of the investment firm Bridgewater Associates (http:
//www.bwater.com), currently the world’s largest hedge fund.

420 Cf. also Häring and Storbeck (2007) for an entertaining and very informative overview of findings.
421 Among the aspects considered by DellaVigna (2007) were time preferences (self-control problems), risk
preferences (reference dependence), social preferences, overconfidence, projection bias, framing and
menu effects, limited attention, persuasion, social pressure and emotions.
422 Walliser (2008) devoted much space to the fundamental behavioral aspect of bounded rationality which
leads to, e.g., financial bubbles, herd behavior or mass hysteria.
218
to beliefs and cognitive operations of actors in dynamic and strategic environments. The
foundations of this “cognitive science” (Walliser, 2008, p. 103) arose from previous work
on collective learning processes.
At the moment, a lot of unknown aspects and open questions still exist.423 A promising
approach was put forward by Kahneman (2011), who argued for a combination of two
systems of thought: one fast and intuitive, the other rather slow but analytical. The first
frequently leads to bad decisions, as people (unconsciously) ignore the second mode be-
cause the fast approach is more convenient.424 While Gladwell (2007) agreed that human
behavior is largely marked by mental processes which work rapidly (and automatically),
he claimed that people are able to unconsciously filter out the relevant information. This
also allows to reach good decisions quickly, even if they are based on relatively little
information or experience.425 Yet, concerning this matter Taleb (2007) prominently dis-
cussed the extreme impact of rare events on people’s perception and humans’ tendency
to find simplistic explanations for them in retrospect.
A BROADER RESEARCH FRAMEWORK
One development of this trend was the foundation of the “Institute for new economic
thinking” (INET). Its main purpose is to broaden and accelerate the development for re-
placing current economic theories which were revealed during the recent global financial
crisis to be inadequate. Therefore INET promotes research funding, community building,
hosting of conferences, etc. The “new thinking” concerns mostly economic topics but is
nevertheless important for political science. Looking at its history so far “one could argue
that the effect of economics has been felt more strongly in political science than any other
social science” (Miller, 1997, p. 1173). It is important for both disciplines to understand
how people think and what consequences a procedural change has.
The work of Thaler and Sunstein (2009) constituted a prominent example for the benefit
of combining insights from both disciplines. The authors argued in favor of libertar-
ian paternalism, i.e., using behavioral effects to influence choices in benefit of the public
good. They based their argumentation on the two systems of thought (Kahneman, 2011)
as well as on multiple other fallacies and biases in human decision-making discovered in
behavioral economics research. Interpreting these behaviors, they derived policy recom-
mendations for public areas such as healthcare, retirement saving, etc.
Another case of a more inclusive approach to economic research represents the nascent
“modern theory of economic order” (von Weizsäcker, 2012, p. 6). This framework incor-
porates insights of political economics on interactions between political and economic in-
stitutions (e.g., North et al., 2009). Taking these and future political equilibria into consid-
423 Kenning and Plassmann (2005) offered an overview of the current state of neuroeconomic research. They
also provided a basic introduction into common concepts and methods.
424 A good assessment of the theory and suggestions for future research can be found in Evans (2003).
425 Gigerenzer and Gaissmaier (2011, p. 451) consistently declared that “ignoring part of the information can
lead to more accurate judgments than weighting and adding all information”.
219
eration leads to diverging policy recommendations from traditional economic expertise

that solely aimed for the removal of market failures, inefficiencies and externalities (cf.
Acemoglu and Robinson, 2012). The reason for the disagreement is a diverging view on
the causes and consequences of economic development. Following the “iron law of con-
vergence” (this term goes back to Larry Summers, Barro, 1996) and the “modernization
hypothesis” (Lipset, 1959), “economic development spurs the introduction and mainte-
nance of higher quality institutions, including well-functioning representative democ-
racy” (Barro, 2012, p. 3). In contrast, new approaches put the “causal effect of income
and education on democracy” (Acemoglu et al., 2007, p. 27) into question and point out
the critical role of political equilibria. Here, “the political system defines the kind of eco-
nomic rules of the game and the judicial system you have” (North, 2008, p.27). Even well-
intentioned market interventions which increase the economic performance can “change
the political equilibrium in a direction involving greater efficiency losses” (Acemoglu and
Robinson, 2013, p. 2) society-wide and over time. Admittedly, the focus on political in-
stitutions and the integration of institutional and growth economics is in turn subject to
criticism (e.g., Sachs, 2012).426 This debate is not yet decided but nevertheless emphasizes
the potential insights of broadening the social research framework.427
Overall, the “new thinking” in economics (and social sciences in general) puts more
attention to institutional, historical, and psychological factors (Ruckriegel, 2010). The
knowledge of related disciplines is considered (e.g., Colman, 2003), cognitive aspects of
behavior are incorporated (e.g., Simon et al., 1992) and inter-dependencies are taken into
account (e.g., Hodge and Schwallier, 2006). Most importantly, the constitutive assump-
tions become more realistic and less idealized by including details which were previously
omitted (e.g., Ariely, 2009). This is perfectly in line with my investigation of NSP. Those
are an important part of the explanation of human behavior. As political science inves-
tigates collective decisions, the tools and methods used should, thus, be based rather on
reality than simplicity.
8.4 Chapter summary
This chapter evaluates the question if the required effort for including NSP is justified
on the basis of the newly found insights. For this purpose, I focus in Sec. 8.1 on meth-
ods and data necessary to implement nonseparability. The assessment is placed in the
broader context of using abstraction when setting up analytical models in Sec. 8.2. The
concept of abstraction enables the analysis of intricate questions by reducing realities’
426 Sachs (2012) criticized not the general idea of extending the traditional framework. Rather, he contra-
dicted with Acemoglu and Robinson (2012) sole focus on political institutions and, in particular, the as-
pects which they did not consider as, e.g., “geopolitics, technological discoveries, and natural resources”
(Sachs, 2012, p. 2).
427 It is important to note that this is not just an abstract debate within scientific circles. The ongoing dispute
is also closely covered in daily media (e.g., Braunberger, 2013).
220
8.4 Chapter summary
complexity. I do not question the general principle, but criticize its (careless) overuse.
This becomes particularly clear when it is contrasted with new trends in social science
research paradigms. Here, I describe the so called “new thinking” in behavioral research
in Sec. 8.3. This development calls for a more realistic implementation and less restricted
or stationary view of the social research environment.
As mentioned earlier in this study, I do not argue that every research project should op-
erationalize nonseparability. Although I criticize the excessive application of the simpli-
fying separability assumption, I will not call for the general (and unaudited) application
of nonseparability. It is important to note that, “although this may seem a paradox, all
exact science is dominated by the idea of approximation” (Bertrand Russell, as quoted by
Auden and Kronenberger, 1966). “Life is complex and so we must simplify our analysis
to obtain useful insights. The art of research involves creation of simplifications that pro-
vide insights based on evidence and observations” (Hinich, 2008, p. 1000). I will not deny
that “if we make our models too complex, we may lose our ability to derive useful pre-
dictions for empirical evaluation” (Morton, 1999, p. 280). In addition, Rabin (1998, p. 13)
pointed out that tractability and parsimony should be guiding principles when aiming to
make research more realistic. Also, I concur with Clarke and Primo (2007) in their effort
to modernize the use of models in political science. Models should not only be judged by
the accuracy of their deductive predictions, but also for their usefulness “in producing
empirical generalizations that may serve as a spur to further modeling efforts” (Clarke
and Primo, 2007, p. 741).
My ultimate goal is to lay the foundation for further research on and empirical modeling
of NSP. The debate in social science on the relevance of nonseparability is just starting.
My contribution is dedicated to further research in two aspects. Firstly, it draws atten-
tion to the phenomenon of nonseparability and strengthens the awareness for it. My
work proves that research in the fields of survey and referendum design, organizational
theory and institutional analysis should consider carefully whether or not NSP are rele-
vant. Both answers (yes or no) can be adequately justified using the arguments discussed
in the previous chapters. Yet, it is important to consider both options! I am committed
to advocating that the nearly unambiguous standard exclusion of nonseparability in an-
alytical research should end. NSP need not be included into every research project but
evaluated with respect to their scientific insights. If they are left out, the reasons for this
decision should be stated and consequences for the obtained results discussed.
Secondly, if a research project decides to incorporate NSP, this study comprises a detailed
overview of previously conducted theoretical as well as empirical research. I provide
empirical tools and guidelines for the operationalization of nonseparability. In addition,
I highlight concrete starting points for its assessment; e.g., the reliability of model pre-
dictions and changes in actor behavior. My study summarizes various contributions and
combines knowledge from different fields. This assists in building a better general insight
and specific understanding of NSP.
221
222
9 Conclusion and outlook
This chapter concludes my study by summarizing the previous chapters. I outline in

Sec. 9.1 the individual steps of my research and highlight my findings. Sec. 9.2 addresses
possible future extensions. Those include specific substitutions within my experimental
design as well as an alternative focus for research on NSP.
9.1 Final summary
My research addresses the question of whether NSP are of significance for analytical po-
litical science. Chap. 1 introduces this research question, demonstrates its relevance and
explains the applied research design and method. It also discusses previous contributions
to clarify the existing research gap. Whereas the theoretical concept of nonseparability is
relatively old, empirical tests are rare. So far, almost all research has used the simplify-
ing assumption of separable preferences as virtually standard. This is most evident when
thoroughly considering the amount of previous research on NSP (cf.Sec. 1.2) compared to
remaining contributions. I list the scientific fields which gain from research on NSP and
describe the further outline of the study. In particular, I point out the expected benefits
from the implementation of a laboratory experiment.
My study is devoted to clarifying the relevance of NSP. In Chap. 2 I provide a multitude

of examples; all of them strengthen my claim that NSP are commonplace. I use simple
hypothetical patterns, genuine empirical examples as well as short case studies to sup-
port my argumentation. Next, I particularly evaluate the interplay of institutions and
preferences and clarify the importance of nonseparability in this context. Common orga-
nizational proceedings as, e.g., delegation, decentralization and specialization are part of
the unanswered puzzle of the influence of NSP; this further underlines the relevance of a
thorough investigation. Building upon these explanations I describe the theoretical mod-
eling of NSP. With respect to reciprocity I add an additional theoretical aspect to its basic
definition. Discussing existing contributions, I put forward arguments about the con-
sequences when neglecting nonseparability in analytical research. Firstly, ignoring NSP
leads to a misspecification of actors’ utility functions. This implies biased and inaccurate
results of models relying on these functions. Also, when decisions affected by NSP are
separated, institutional aspects influence decision-making. Thus, secondly, if decisions
are taken sequentially, nonseparability favors the actor deciding first. Thirdly, if deci-
223
sions are taken simultaneously, nonseparability causes sub-optimal outcomes, because

strategic voting becomes increasingly difficult.
Next, Chap. 3 is concerned with the misspecification of actors’ utility functions. For this
purpose I use the field of legislative decision-making in the context of EU politics. More
precisely, I undertake empirical analyses to quantify the distortions by examining the
impact of NSP on the performance of several legislative decision-making models. These
models rely on utility functions which have to be specified according to the policies in
question. When a proposal consists of multiple dimensions, the (non)separability of sin-
gle issues has to be taken into account as otherwise the utility functions are misspecified.
The empirical calculations were based on the collected data from the DEU project (Thom-
son et al., 2006). The coding of this prominent data set for nonseparability demonstrated
that the majority of (multi-issue) EU politics are indeed affected by NSP. Moreover, in
the majority the respective decisions the effect is non-reciprocal. Thus, the extension of
the NSP concept with respect to reciprocity (cf. Sec. 2.4) is strongly encouraged. Next,
I investigate if the comparison of various models’ predictive accuracy depends on the
existence of NSP. Applying simulation techniques, I demonstrate that overlooking NSP
may have caused a substantial bias in the empirical evaluation of competing models of
EU legislative politics.428 This confirms my first key question: namely, the importance
of a correct model specification with respect to nonseparability to avoid invalid conclu-
sions. In other words, neglecting NSP poses a threat to the inference of the corresponding
research. Overall, by comparing my results to previous research I prove the impact and
evidence of NSP with real-world data.
The second part of the study (which comprises Chap. 4 to Chap. 7) investigates the impli-
cations of NSP for individual and collective decision-making. Here, I rely on a laboratory
experiment, the design of which is explained in detail in Chap. 4. I argue that collective
decision-making in committees is well-suited for laboratory scrutiny, as the level of moni-
toring and the focus on processes in decision-making correspond well to both laboratory
and real-world politics (Levitt and List, 2006). Also, as I perform a validity test on an
existing, theoretically well-known but empirically neglected concept, the experimental
method fits my purpose perfectly (Schram, 2005).
Following Levitt and List (2006, 2007) I emphasize the qualitative patterns of my find-
ings. Because of the specific laboratory environment I refrain from singling out the spe-
cific parameter estimates (e.g., for social welfare) as my main findings. Chap. 5 looks into
the results of my laboratory experiment on the aggregated and Chap. 6 on the individ-
ual level. The main findings of the experiment are fourfold. Firstly, the performance of
the deterministic equilibrium concept is rather low. The probabilistic extension of the
core at the individual level derived in Sec. 6.3.2 is far more appropriate. This indicates
clearly that also in collective decisions behavioral aspects such as risk aversion and so-
428 Thebias arises from the different degree of restrictions incorporated into the models. This is particularly
unsatisfactory for those researchers interested in the effect of procedural aspects.
224
9.1 Final summary
phistication must be modeled at the individual level. Such a comparison of collective and
individual level analysis may allow new insights even into already familiar concepts. An
argument in favor of this juxtaposition was the frequent occurrence of minimal-winning
coalitions. The information provided by the outvoted subjects is only accessible at the
individual level. On the other hand, the collective level might be blurred by the noise of
single seditious individuals (cf. Goeree et al., 2002).
Secondly, the long-established trade-off between efficiency (decision costs) and effective-
ness (allocation and distribution of wealth) in collective decision-making (cf. Buchanan
and Tullock, 1962) is the focus of Chap. 5. I found that, as expected, delegation increased
efficiency in terms of decision-making speed. The insights on effectiveness are more am-
biguous as delegation provided more stability and fewer imbalances. However, as pre-
dicted by theory, with respect to social welfare allocation the pooled decision-making
procedure delivered the best performance, whereas unaccounted NSP led to sub-optimal
outcomes and Pareto-inferior decisions. This highlights the necessity of well-coordinated
delegation. In other words, the organizational structure must enable actors to assess their
behavior’s impact on the final outcome to prevent less than optimal behavior.
Thirdly, institutions are not neutral, because different decision mechanisms led to sub-
stantially different outcomes. This confirms my second key question that nonseparability
in conjunction with the respective institutional arrangement indeed influences collective
as well as individual behavior. In the specific case of my investigation, unaccounted non-
separability increased sincere behavior; at the same time sophisticated social considera-
tions are replaced by simplistic self-interest. This indicates an interaction among institu-
tions and preferences which calls the neo-classical view in economics (Stigler, 1950) that
preferences are exogenously fixed (i.e., that behavior is driven by hedonic utility) into
question. This is in line with current cognitive science and psychologists’ findings that
“people’s decisions can be highly sensitive to situational factors, even when such factors
are unrelated to the actual utility of that course of action” (Ariely and Norton, 2008, p. 13).
The task for the subjects in my experiment was marked with a high level of complexity
and uncertainty; these two aspects accompany every decision affected by NSP. Here, my
design differs from standard laboratory experiments which typically use simple games
and setups.429 If one does not account for the possible influence of an institutional setting
when measuring individual preferences, this bears the risk of mistaking the respondent’s
conditional response for their genuine preference. Disregarding conditional preferences
at the individual stage jeopardizes the conclusions drawn at the collective level.
Fourthly, the first-mover advantage in delegation settings based on standard game the-
ory expectations was not found in my experiment. On the contrary, being part of the
429 The term “simple” is in fact quite often used by experimenters to describe their setup; e.g., Berl et al.
(1976) tested the core concept in a “simple n-person cooperative nonsidepayment game”, Charness and
Rabin (2002) looked into social preferences “with simple tests” while Ert et al. (2011) used “simple exten-
sive form games” for the same purpose and Chou et al. (2009) emphasized the control over game form
recognition when playing “a simple two person guessing game”.
225
first stage when deciding on a nonseparable, but nonetheless split, problem resulted in
a below average performance. The high level of complexity and uncertainty prevented
the subjects from performing the necessary computations. As both aspects varied across
decision procedures, the experimental results clearly showed the crowding out of sophis-
tication. Subjects were not able to strategically comprehend the effects of their votes. The
simplifying assumptions that individuals can at any rate or in any situation order their
preferences amongst a variety of alternatives and choose the optimum is therefore par-
ticularly problematic with regard to NSP. This would imply an artificially high level of
computational skills (Sen, 1997).
The voting experiment was accompanied by a post-experiment survey, which is analyzed

in Chap. 7; it contained ’yes or no’ questions as well as free-input fields. Most impor-
tantly, the answers obtained confirm the validity of my design. Describing their own
voting behavior, subjects stated that they focused on a limited number of criteria when
making their decision. This was mainly, not surprisingly, their own payoff. Looking at so-
phisticated voting the participants took the probable vote of co-players into account and
maximized their conditional utility. This speaks strongly for the introduced probabilistic
core concept. The alternation of the decision-making procedure influenced the motives
of the participants. Also, the comments prove the desire of subjects to communicate. Col-
lective decision-making is a social exercise, and subjects wondered why communication
was excluded for such a task.
Finally, Chap. 8 takes the expense for the implementation of nonseparability into consid-
eration. In particular, I discuss the trade-off between feasibility and accuracy in analytical
research. Here, I emphasize the indispensability of abstraction, as it enables the analysis
of intricate questions by reducing the complexity of reality; but I also criticize its overuse.
With respect to NSP I evaluate the necessary effort for incorporation as well as the knowl-
edge gained. I place my argument to consider nonseparability in the context of a current
development in social science research paradigms. This trend calls for a more realistic
and less idealized approach when investigating human behavior. This also applies to the
nonseparability of preferences, which constitutes an undeniable part of human decision-
making.
To summarize, this study comprises a multitude of references, various examples of imple-

mentation and a comprehensive theoretical discussion on the concept of nonseparability.
I demonstrate that including NSP in analytical research prevents biased and inaccurate
results. In particular, studies with respect to institutional structures should be careful
because the results obtained are not necessarily due to the organizational aspects, but
the denial of nonseparability. I do not conclude that NSP has to be considered in all re-
search, because tractability and parsimony are also guiding principles for a more realistic
research paradigm. Yet, if the simplifying assumption of separability is made, the reasons
should be given and possible consequences discussed. My work provides the following
contributions with the required toolbox.
226
9.2 Future research
9.2 Future research
In this study I discussed many important aspects when designing and conducting labo-
ratory experiments. The suitability of this approach to my specific research questions is
verified in Sec. 1.3.2 and its general abilities are laid out in Sec. 4.1. Also, Sec. 4.2 explains
in detail my design choices and their consequences. The appropriate research method de-
pends on the specific research question (Draper, 2004). However, an absolutely perfect,
flawless experiment does not exist. I chose a laboratory experiment due to its high level
of internal validity when looking at implications of NSP for individual and collective
decision-making. The control capabilities of the laboratory facilitated the operationaliza-
tion. It also made it possible to identify changes in subjects’ behavior when confronted
with (neutral) decision problems affected by NSP.
As “the proof of external validity is always empirical” (Morton and Williams, 2010, p.196)
it is necessary to complement results of laboratory research with further studies. This
calls for further experiments with variations in design, target population and subject
recruiting. Yet, it also emphasizes necessary variations in terms of the experimental
method. As always, the knowledge gained from scientific research is confined to a certain
extent by the limitations of the applied method. Many missing clues can be attributed to
specific approaches. Thus, additional research can provide fruitful new insights.430
The following sections discuss three possible aspects of future research. Sec. 9.2.1 picks
up the discussion on common problems of laboratory subject pools and evaluates the po-
tential of crowdsourcing to overcome these criticisms. Sec. 9.2.2 focuses on the problem of
external validity and argues in favor of supplemental field experiments. Finally, Sec. 9.2.3
offers an alternative research focus on coordination in collective decisions. This list does
not mean that many more opportunities for future research do not exist; however, with
regard to my own study, these are the most noteworthy and supplemental.
9.2.1 Crowdsourcing
In Sec. 4.2.9 I discuss in detail the shortcomings of laboratory experiments with respect
to subject recruiting. These are that
• all participants volunteer for the task.
• nearly all subject pools consist to an overwhelming extent of students.
• the participants might know each other.

430 Anexcellent example of how to “prove” external validity can be found in Bosch-Domenech et al. (2002).
The authors performed a meta-study of a previous beauty-contest conducted in laboratory as well as in
newspaper experiments. In addition, they themselves collected more data in classrooms, conferences,
by e-mail, or through newsgroups. This provided a rich variety of different subject pools, sample sizes,
payoffs and environmental settings.
227
Thus, “a random American undergraduate is about 4,000 times more likely than an
average human being to be the subject of such a [laboratory] study” (Experimental-
Psychology, 2012, p. 69). This really does not account for the principle of random sam-
pling.431 The problems of self-selection and convenience sampling have been discussed
often and to a great extent (e.g., Cleave et al., 2010). Although current research suggests
that students are an appropriate subject pool (Branas-Garza et al., 2012), a relatively new
approach circumvents these problems and offers interesting new opportunities. Here,
the advances in computer technology (e.g., lower costs, faster networks and wider distri-
bution) can also be used for experimental work. A special development is “crowdsourc-
ing”432 , the exploitation of online labor markets. Such a market is an online forum where
employers (or researchers) post jobs and employees (or participants) choose which tasks
to do for a payoff.433 It is important to understand that these markets are different from
just combining multiple laboratories or subject pools.434
Horton et al. (2011) judged online labor markets to be the logical next step of techno-
logical improvements in experimental research. Researchers in the early 1990s for the
first time had well-developed tools for conducting experiments over local computer net-
works. The capabilities of the World Wide Web could now be used to free research from
the logistical limitations of physical laboratories (cf. Suri and Watts, 2011).
Several firms currently offer their online services for tasks like answering surveys, ana-
lyzing texts, ranking websites, etc.435 The “most popular for scientific purposes is Me-
chanical Turk, which is run by Amazon” (Experimental-Psychology, 2012, p. 69). Al-
though the markets are still dominated by Americans, they are growing fast world-wide.
This implies a much larger and more diversified subject pool which is a clear improve-
ment compared to the “standard” laboratory participant who fits the WEIRD436 charac-
teristics (Henrich et al., 2010). However, the participants of these platforms “are still a
pretty skewed sample of humanity. In particular, they are younger and more liberal than
people at large” (Experimental-Psychology, 2012, p. 69). Buhrmester et al. (2011, p. 3)
concluded that Amazon Mechanical Turk (AMT) “participants are slightly more demo-
graphically diverse than are standard Internet samples and are significantly more diverse
431 Itis also not at all a new phenomenon. Already Reips (2000, p. 92) pointed out that “80% of all psycholog-
ical studies are conducted with students, while only about 3% of the population are students.”
432 The notion crowdsourcing is used as “an umbrella term for a variety of approaches that harness the po-
tential of large crowds of people by issuing open calls for contribution to particular tasks” (Geiger et al.,
2012, p. 2). It is clearly related to the traditional concept of outsourcing, the handover of business func-
tions and structures to third party companies. Yet, for crowdsourcing the voluntary principle is also
important (cf. Hammon and Hippner, 2012).
433 The comparison between laboratory and internet is not new to psychological research (for a comprehen-
sive overview cf. Birnbaum, 2000). Yet, the interest of economics was sparked rather recently in accor-
dance to the rise of online labor markets.
434 Cf. Frei (2009) for a detailed overview.
435 These are, e.g., Amazon Mechanical Turk (https://www.mturk.com), CrowdFlower (http://
crowdflower.com), Elance (http://de.elance.com), Freelance (https://www.freelance.de), Guru

(http://www.guru.com), oDesk (https://www.odesk.com), etc.
436 The term WEIRD stands for “Western, Educated, Industrialized, Rich and Democratic” and characterizes
the land of origin for most laboratory subjects. This is clearly not globally representative.
228
9.2 Future research
than typical American college samples.” Thus, despite being more diverse than student
samples, the markets do not solve all recruiting problems.437
All experimental laboratories are equipped with a subject-recruiting mechanism.438 Most
of the time, this is just a database containing a huge number of registered subjects in-
cluding demographic and contact information. The success of a laboratory (and of the
experiments conducted in it) almost always depends on a well-maintained, up to date
and, in particular, accurate database. The same is true for online experiments. But with-
out personal contact it is more difficult to ensure reliability. Online labor markets rely
on reputation systems (Resnick et al., 2000). Market operators also conduct bank account
checks (Horton et al., 2011, p. 6). Overall, the problems of online laboratories are the same
as those recruiting databases have. It is difficult and demanding to keep them accurate.
Rand (2012) argued strongly in favor of mechanical turks. By reviewing replication stud-
ies on AMT of previous laboratory contributions, he verified that the method is valid:
the results revealed the same effects under both conditions. Horton et al. (2011) reached
a similar conclusion with respect to internal and external validity. The author also con-
ducted truth checks of self-reported demographics. Depending on the variable the level
of reliability was between 81% and 98%.439
An argument often made in favor of the online markets is that they are relatively cheap.
Mason and Suri (2012) offered practical advice for conducting experiments in this way.
Such contributions are much-needed as current reports express concerns that participants
are exploited due to small salaries of less than $2 per hour (Experimental-Psychology,
2012, p. 70). Paolacci (2012) operates a blog that has similar goals. While promoting the
approach for experimental research in general, his objective is to establish basic guide-
lines for freelance work.
Overall, conducting experiments in online labor markets is still an unusual approach.
However, it offers easy access to a large and more diverse subject pool. As it is also a
low cost approach, it increases the possible number of observations. This provides ex-
perimental research in political science with a promising opportunity because real-world
politics usually affects, or is created by, many people (e.g., Miller, 1997; Mueller, 2003).
In general, well-designed laboratory experiments may be limited to a small number of
437 Reips (2000, p. 96) expected the “Demographics of Internet users [...] to rapidly approach similarity with
demographics of the general population.” Although this argument was made some time ago and has not
yet come completely true it might for future research change the diversity argument significantly.
438 The Laboratoire d’Economie Experimentale de Montpellier offers a list of experimental laboratories
around the world. It can be obtained at http://leem.lameta.univ-montp1.fr/index.php?page=

liste_labos&lang=eng.
439 Typical findings of laboratory experiments which have been replicated using online labor markets were,
e.g., priming and framing effects (Horton et al., 2011). In general, many current studies look for behav-
ioral differences between experiments conducted in the laboratory and via the internet. Bosch-Domenech
et al. (2002, p. 1687) referred to this as “to test the critical assumption of ’parallelism’ between the lab
and the field.” Those contributions included e.g., trust games (Fiedler and Haruvy, 2009), ultimatum
games (Goerg et al., 2007), beauty-contests (Selten and Nagel, 1998) and auctions (Lucking-Reiley, 1999).
Sec. A.17 contains a more extensive list of references on studies with experiments conducted online and
in the laboratory collected by Israel Waichman (Heidelberg University).
229
observations due to the high internal validity of the research method (cf. Sec. 1.3.2). But
political science asks specific questions about collective processes and structures. A good
example is the realm of public choice, where “the use of laboratory experiments [...] has
also increased rapidly in the last thirty years” (Schram, 2002, p. 1). Many of these exper-
iments have examined various variables and their influence on the provision of public
goods (for a detailed overview cf. Ledyard, 1995). While in reality public goods are gen-
erally provided within a large population (e.g., pension system and national defense),
few experiments investigated the effect of group size. While Isaac et al. (1994) found that
public good contributions increase with group size, Offerman et al. (1996) concluded that
voluntary contributions decrease with more group members.440 More important than
these controversial insights is the fact that, the usual number of participants in public
good experiments was only four to five, and very few studies looked at groups of more
than ten (Schram, 2002, p. 2). This shortcoming can be attributed to a large extent to the
spatial and organizational limitations of standard laboratories. In addition to the obvi-
ous problem of equipping a lab with 200 or more seats,441 a large number of participants
require exponentially greater financial resources. These are exactly the types of logistical
problems of standard laboratory experiments in which crowdsourcing might help.
It is also important to note that subjects do not necessarily realize that they are part of
a research project. They are not aware that their “employer” might be a researcher. As
the interaction takes place in their usual (online) working environment this can prevent
a Hawthorne effect442 , where subjects change their behavior because they are aware of
being monitored. This is even truer if the use of online labor markets follows the example
of the conversion of people’s social life into online networks (e.g., Facebook).
The potential of linking theoretical work to a broad empirical analysis is not in question.
However, online experiments have limitations and pitfalls of their own (Eckel and Wil-
son, 2006). It is much harder to conduct synchronized experiments and to secure high
quality work. When you only receive coded input online, a subject’s identity remains
anonymous (raising the question who actually sat in front of the screen). In particu-
lar, communication between subjects can become a critical issue. The laboratory offers a
much higher degree of control (Morton and Williams, 2010, p. 31ff). In addition, typical
online data problems have to be taken into account, such as how to keep data private,
how to maintain code security, etc. Mason and Suri (2012, p. 11ff) discussed these prob-
lems in detail and suggested possible improvements. However, it looks as though the
solution simply exchanges the shortcomings of one approach for the shortcomings of an-
440 More specifically, Isaac et al. (1994, p. 32) pointed out that not group size per se matters, but the interaction
between group size and the marginal per capita return from the public good (i.e., the marginal rate of
substitution).
441 To give an exemplary insight on the possible number of participants, the study of Diederich and Goeschl
(2011) investigated the determinants for charitable giving and is based on a large-scale field experiment
with 2,440 subjects. To include this number of persons successively, let alone simultaneously, into a
laboratory experiment is infeasible.
442 Gillespie (1991) provided an excellent review of the Hawthorne experiments.
230
9.2 Future research
other. Tab. 9.1 summarizes the arguments discussed so far, both in favor of and against
internet experiments.
Table 9.1: Pros and cons of internet experiments

EXPLANATORY NOTE
The table lists the pros and cons of conducting experiments online instead of in the laboratory. It also cites exemplary
sources as evidence for the arguments made. The illustration is based on Charness et al. (2007a, p. 91, table 1).
Arguments against internet experiments Arguments in favor of internet experiments
• Not everyone has internet access (Krantz et al., • More diverse populations (generalizability)
1997) (Buhrmester et al., 2011)
• More noise and higher variance (Shavit et al., • Demographics approach those of the general
2001) population (Kehoe and Pitkow, 1996)
• Subjects appear less attentive in Internet • Reduced experiment(er) effects as the web
experiments (Anderhub et al., 2001) might rather reflect a “natural environment”
(Harrison and List, 2004)
• Subjects fear deception (e.g., subjects do not
believe they are matched with a real person) • Lower costs and fewer physical limitations
(Eckel and Wilson, 2006) (Suri and Watts, 2011)
• Loss of control over the physical environment • No discernible differences in levels of

(Morton and Williams, 2010) rationality (Nagel et al., 2002)
The main reason to discuss crowdsourcing in the context of NSP is its ability to facili-
tate cross-cultural research (Paolacci et al., 2010).443 Various contributions (experimental
as well as non-experimental) suggested that preferences and attitudes are different across
cultural backgrounds (e.g., Fiorino and Ricciuti, 2007; Bohnet et al., 2008). Using a survey
designed accordingly, Lacy (2001b) identified in the U.S. numerous NSP between political
issues: the preferences on income taxation depend on crime prevention policies, prefer-
ences on environmental pollution depend on environmental regulation, preferences on
defense spending depend on social spending, preferences on immigration policy depend
on the constitutional status of English being the only official language, etc. But do the
same dependencies exist in other countries as well? More likely than not the political
agenda will consist of other topics (Ehmke et al., 2005; Goerg et al., 2007). Those issues
might be separable, but as this study has shown, this should not be taken for granted.
To provide a list with nonseparable policy fields across countries would be a good way
to increase the awareness of NSP. It would also offer valuable insights for research on
elections and campaigns. Of course, for such statements a representative sample is in-
evitable; whether crowdsourcing can provide this to a sufficient degree, however, stays
questionable but seems worth investigating.
9.2.2 Going to the field
The idea of bringing research on NSP “to the field” (Harrison and List, 2004) follows
similar argumentation. Compared to the laboratory setting, a field experiment inhibits
443 Cf. Eriksson and Simpson (2010) for an example of such a study. The authors conducted an online-survey
testing attitudes of risk preferences in the U.S. and India.
231
a lower degree of environmental control. Many more possibilities and distorting influ-
ences on subjects have to be considered. Nevertheless, Levitt and List (2007) pointed
out that lab-generated as well as observational data suffer from shortcomings. Here, a
“well-designed field experiment [...] can serve as a bridge connecting these two empiri-
cal approaches” (Levitt and List, 2007, p. 171). If this is done correctly, a researcher may
be able to observe a subject’s natural behavior.
Studying a subject outside the “sterile environment” (Harrison and List, 2004, p. 1009) of
a laboratory has proven to be valuable many times.444 Such contributions complemented
existing laboratory investigations in various fields very well. Bahry and Wilson (2006)
evaluated fairness in ultimatum games, Carpenter and Seki (2005) searched for social
preferences, Holm and Nystedt (2010) investigated collective trust behavior, Olken (2008)
analyzed the provision of public goods, Cohen and Nisbett (1997) looked for reasons
for the perpetuation of honor-related violence, etc.445 Going more into specific topics
of politics, Wantchekon (2003) assessed clientelism, Kuklinski et al. (1997) examined the
impact of racial prejudice, Gerber et al. (2009) scrutinized media influence on voting, etc.
I list these contributions in such detail for two reasons. Firstly, they reflect a variety of
research questions which profited from field investigations. There is no argument why
this should not also be true for research on NSP. Secondly, some of them (e.g., racial
prejudice) may very well be affected by NSP themselves.
It is important to note that the term “field experiment” does not just refer to a physical
location, and by no means are the methodological requirements lower or less stringent.
With the focus on “lab in the field” experiments Hamenstädt (2012) highlighted the po-
tential problems of self-selection and non-compliance and emphasized the importance of
randomization. Harrison and List (2004) broadened the definition of field experiments
considerably. They explained that it involves the nature of the subject pool, the informa-
tion subjects bring to the task and the type of commodity and stakes as well. In their
taxonomy a natural field experiment takes “place in the subjects’ natural environment
and, importantly, where the subjects do not know they are in an experiment” (Morton
and Williams, 2010, p. 223). Depending how far one views online labor markets as being
a natural environment, crowdsourcing (cf. Sec. 9.2.1) might belong into this category.
9.2.3 Communication and interaction
The laboratory empowers the researcher with a high degree of environmental control.
Nearly every aspect of subjects’ interaction and perception can be modified (cf. Morton
and Williams, 2010, Chap. 4). Communication is a central aspect of human interplay in
444 In political science quasi-experiments are also common (Bernauer et al., 2009, p. 91). In this section, it
is not my intention to distinguish between experiments and quasi-experiments; rather I will focus on
the differences for external validity when leaving the laboratory. For a comprehensive overview on
experimental and quasi-experimental designs cf. (Shadish et al., 2002).
445 Cf. Ortmann (2005) for an overview on field experiments in Economics and some methodological notes.
232
9.2 Future research
general and many studies have proven its influence on behavior and beliefs (e.g., An-
dreoni and Rao, 2011). It is also a fundamental aspect of group coordination. Current
research in public administration theory, organization theory and planning theory turns
more and more towards its dynamic aspects (cf. Pedersen et al., 2011). Diverging from
traditional theories, these new contributions are “pointing to the relational, interpretive,
interdependent, and interactive aspects of all coordination processes” (Pedersen et al.,
2011, p. 375).
In previous laboratory experiments two ways of dealing with communication could be

observed. One approach was to control communication “in order to prevent possible
cross-effects that occur with communication. [..] In almost all game theory experiments,
communication between subjects is not allowed except under particularly controlled cir-
cumstances” (Morton and Williams, 2010, p. 122). This cautious handling was due to the
less predictable confounding of communication with other variables of the experiment.
A second stream of research used communication as its central treatment variable. For
example, Cason and Mui (2007) found out that different forms of (non-binding) com-
munication facilitated collective coordination. Isaac and Walker (1988b) looked for sim-
ilar effects when analyzing group size effects in public good experiments. Miller and
Oppenheimer (1982) found that communication enforced the relevance of universalism.
Changes in behavior were also observed by Muren and Pyddoke (1999) after their sub-
jects were allowed to interact and build up trust in each other. In summary, these studies
agree that social interaction enhances social consideration.446 The literature is strongly
linked with work on framing effects (Tversky and Kahneman, 1981). Morton and Williams
(2010, p. 28) stated that “framing effects work when a communication causes an individ-
ual to alter the weight he or she places on a consideration in evaluating an issue or an
event.”447
Of course, interactions did not exclusively lead to cooperative aspects. Eckel and Petrie
(2011) found that information is used to discriminate between people. They revealed that
a specific “face, it appears, has strategic value” (Eckel and Petrie, 2011, p. 1497). In his
experiment Vincent (2012) informed his subjects about the name of their experimental
partners, hereby disclosing gender and race. His results showed no general trend but
suggested that “racial group behavior varies depending on the racial composition and
attitudes of the population being observed” (Vincent, 2012, p. 1).
While these examples disagreed on intensity and shape, all studies agreed that interac-
tion influences behavior. Sunitiyoso et al. (2011) investigated the net and gross effects of
social interaction as well as the direction of change of individual behavior. They iden-
tified behavior which conforms to models of social learning. Mickan and Rodger (2000,
446 However, as discussed in Sec. 4.2.2 and Sec. 6.1.2 social motivations may be enhanced by communication,
but they do not dependent on it (Fehr et al., 2002a).
447 Druckman (2004) identified various criteria for the success of framing. He concluded that a multitude of
contextual forces as well as individual attributes are important.
233
p. 205) argued that communication is one of the most important factors for group effi-
ciency. Allen and Fusfeld (1974) found that communication networks as well as seating
arrangements influence the frequency and patterns of interaction. They concluded that
“communication is influenced by the physical, architectural arrangement of the labora-
tory. Communication between individuals is very sensitive to both the horizontal and
vertical distances separating them” (Allen and Fusfeld, 1974, p. 39). But what happens
if subjects are given a choice? Lai and Lim (2012) carried out a test to uncover if sub-
jects communicate when given the (costly) possibility to do so. They let subjects chose
between delegating competence or communicating with other subjects. The authors ob-
served “significantly more choices of delegation than of communication” (Lai and Lim,
2012, p. 541). However, subjects still under-delegated as they hoped to profit from infor-
mation of their co-players.
I implemented my laboratory experiment with the z-Tree software. This program also
includes a chat feature which enables subjects to send text messages to each other (cf.
Fischbacher, 2007, p. 172). Most important, the experimenter would still control and ob-
serve the content of the messages, the degree and frequency of contact, etc. So, in general,
monitoring communication between subjects would be easy to achieve. A much more
demanding task is to develop a corresponding theory and to translate it into a specific
protocol. Communication in collective decision-making means more than merely agree-
ing on one alternative. Committee meetings possess multiple levels of mutual influence.
Communication involves an observable interchange of information and subtle interac-
tions of power, attitudes and values (Loxley, 1997). Therefore, it is difficult to decide
which kind of communication should be allowed.
One aspect refers to the technical implementation, e.g., should the discussion be based
on text alone or should it also be visual; should the contact take place in person or via
a screen? In their collusion games Cooper and Kühn (2009) observed clear differences
when allowing either unlimited pre-game communication or limited message-space dur-
ing the experiment. Dutcher (2012) investigated the effects of telecommuting on produc-
tivity and found that it has positive implications with creative and negative implications
with dull tasks. Thus, the kind of experimental task may also interact with communica-
tion and the different technical implementations. Another aspect concerns the structure
of communication. Should everybody be able to talk to everybody else or should one
host structure the discussion?448 Previous public good experiments investigated the in-
fluence of a group leader on contributions. They found that one-sided communication
increased the contributions although a lot of cheap-talk was generated (e.g., Güth et al.,
2004; Koukoumelis et al., 2012). It would be interesting to observe how this kind of com-
munication would affect my experimental setting and its distributional patterns. Yet, an
unanswered matter in public good experiments is the effect of how the group leader is
448 Kriss
and Weber (2012) provided a survey of laboratory insights on the formation of groups with(out)
leaders and the shaping of economic organization.
234
9.2 Future research
determined. The leadership role can be assigned exogenously by the experimenter or

endogenously by the participants449 (cf. Arbak and Villeval, 2007).
Overall, thinking about communication is more than just recording words; in the labo-
ratory it is possible to do more. When I mention communication as a possible part of
future research, I do not simply refer to a degree of information exchange but focus on if
and how the subjects use their interpersonal contact to influence the collective decision.
The pending task is to scrutinize how separating the decision over nonseparable issues
affects communication (and behavior).450
Weihe et al. (2008) looked at the principles of committee deliberation processes.451 De-
liberation means to scrutinize the complete process of negotiations and to focus on the
establishment of liabilities in committees (Gutmann and Thompson, 2004). This is con-
trary to the traditional rational choice approach (starting with the contributions of Black,
1958, 1991) which does not look into the concrete course of (internal) committee meetings.
Thus, the focus on liability is in fact a theoretical principal and follows the perception of
politics, which are seen as emerging from binding decisions (cf. Easton, 1957; Scharpf,
2000).
To evaluate the deliberative process it is necessary to provide an extensive documentation
of the communication process and the various interaction steps (Weihe et al., 2008, p.344).
Such an inquiry will be difficult and laborious. Weihe et al. (2008) presented a three-piece
theoretical scheme for covering the process. They dispersed the unification process into
the steps of proposal, acceptance and confirmation. The course of the discussion had
then to be sorted into these categories enabling the analysis. Of course, this did not yet
deal with the concrete operationalization of the measurement.
Again, here the high degree of environmental control of the laboratory can help. Hennig-
Schmidt (1997) used a video analysis to investigate the break-offs in bargaining nego-
tiations. In a similar fashion Nullmeier and Pritzlaff (2009) analyzed the dynamics and
power relations of committee decision-making. The difficult part is to set up an appropri-
ate coding scheme to transcribe the video material into a specific score notation452 . This
notation must then be divided into a small section and thoroughly scrutinized to enable
a process analysis of the deliberative action (Weihe et al., 2008, p. 344ff).
Looking into group decision-making at this level of detail enables researchers to discover
insights into the behavior of individuals. How do they react when the settings change?
Which attitude do they adopt when talking? If distinguishing between sincere and so-
phisticated behavior, which attitude do they chose? We know that the terms we use, and
how we use them, influences our perception and our actions (Pritzlaff, 2006). But how
449 This can be done in various ways: through elections, based on the performance in solving a puzzle or
answering knowledge questions, etc.
450 Cf. Bardsley et al. (2010) for a more general assessment of the current state and future development of
experimental (economic) research.

451 For a comprehensive overview on deliberative democracy cf. Bohman and Rehg (1997) and Elster (1998).
452 Cf. Weihe et al. (2008, p. 347) for an example of such a score notation.
235
do individuals form and shape collective decisions? Furthermore, in the context of NSP,
how does separating the decision over nonseparable issues alter the findings.
Using video analysis is only one possible way to monitor communication and interaction
in an experiment. There are many more, and often chat protocols or pictures might be
perfectly adequate. Nevertheless, group decision-making blurs the ability to infer from
the observed behavior. To cover as many different levels of human interaction as possible
might be a laborious but appropriate reaction to obtain validity.
236
9.2 Future research
237
Bibliography
Abbink, K. and H. Hennig-Schmidt. 2006. “Neutral versus loaded instructions in a

bribery experiment.” Experimental Economics 9:103–121.
Acemoglu, Daron, Simon Johnson, James A. Robinson, and Pierre Yared. 2007. “Reeval-
uating the Modernization Hypothesis.” NBER Working Paper 13334.
Acemoglu, Daron and James A Robinson. 2012. Why Nations Fail: The Origins of Power,
Prosperity, and Poverty. Crown Business.
Acemoglu, Daron and James A Robinson. 2013. “Economics Versus Politics: Pitfalls of
Policy Advice.” NBER Working Paper, SSRN w18921.
Achen, C. 1983. “Towards Theories of Data.” In Political Science: The State of the Discipline,
edited by A. Finifter, chapter 3, pp. 69–93. Washington, DC: American Political Science
Association.
Achen, Christopher H. 2006. “Evaluating Political Decision Making Models.” In The Eu-
ropean Union Decides, edited by R. Thomson, F. N. Stokman, C. H. Achen, and T. König,
pp. 264–298. Cambridge: Cambridge University Press.
Ackert, Lucy F., Ann B. Gillette, Jorge Martinez-Vazquez, and Mark Rider.
2009. “Risk Tolerance, Self-Interest, and Social Preferences.” mimeo. Ex-
perimental Economics Center Working Paper Series, Georgia State University.
http://excen.gsu.edu/workingpapers/GSU_EXCEN_WP_2009-04.pdf (accessed
March 2, 2012).
Akaike, H. 1973. “Information theory and an extension of the maximum likelihood prin-
ciple.” In Second International Symposium on Information Theory, edited by B. N. Petrov
and B. F. Csaki, pp. 267–281. Academiai Kiado: Budapest.
Akerlof, G. A. 2007. “The missing Motivation in Macroeconomics.” In Presidential Adress

at the 199th Conference of the American Economics Association.
Akerlof, George A. and Robert J. Shiller. 2009. Animal Spirits: How Human Psychology
Drives the Economy, and Why It Matters for Global Capitalism. Princeton University Press.
Al-Ubaydli, Omar, Uri Gneezy, Min Sok Lee, and John A. List. 2010. “Towards an under-
standing of the relative strengths of positive and negative reciprocity.” Judgment and
Decision Making, Society for Judgment and Decision Making 5:524–539.
239
Al-Ubaydli, Omar and Min Sok Lee. 2009. “An experimental study of asymmetric reci-
procity.” Journal of Economic Behavior and Organization 72:738–749.
Al-Ubaydli, Omar and Min Sok Lee. 2012. “Do you reward and punish in the way you
think others expect you to?” The Journal of Socio-Economics 41:336–343.
Aldrich, Howard E. 1979. Organizations and Environments. Prentice-Hall, New York.
Allen, Thomas J. and Alan R. Fusfeld. 1974. “Research Laboratory Architecture and the
Structuring of Ccmnunications.” Working papers, Massachusetts Institute of Technology
(MIT). 692-74.
Ambrose, M. L. and M. Schminke. 2003. “Organization structure as a moderator of the

relationship between procedural justice, interactional justice, perceived organizational
support, and supervisory trust.” Journal of Applied Psychology 88:295–305.
Amir, O., D. G. Rand, and Y. K. Gal. 2012. “Economic Games on the Internet. The Effect
of 1-Dollar Stakes.” PLOS one 7.
Amsterdam. 1997. Vertrag zur Gruendung der Europaeischen Gemeinschaft. Europaeischer

Rat. signed on 2nd October.
Anderhub, V., R. Mueller, and C. Schmidt. 2001. “Design and evaluation of an economic
experiment via the Internet.” Journal of Economic Behavior and Organization 46:227–247.
Anderson, Lisa R. and Charles A. Holt. 2002. “Experimental economics

and public choice.” mimeo. Prepared for the Encyclopedia of Public Choice.
http://web.cenet.org.cn/upfile/44224.pdf (accessed July 24, 2012).
Andersson, Ola, Cedric Argenton, and Jörgen W. Weibull. 2012. “Robustness to Strategic
Uncertainty.” Working Paper Series , Research Institute of Industrial Economics 910.
Andreoni, James, William Harbaugh, and Lise Vesterlund. 2003. “The Carrot or the
Stick: Rewards, Punishments, and Cooperation.” American Economic Review 93:893–
902. American Economic Association.
Andreoni, J. and J. Miller. 2002. “Giving According to GARP: An Experimental Test of

the Consistency of Preferences for Altruism.” Econometrica 70:737–753.
Andreoni, James and Ragan Petrie. 2004. “Public goods experiments without confiden-
tiality: a glimpse into fund-raising.” Journal of Public Economics 88:1605–1623.
Andreoni, James and Justin M. Rao. 2011. “The power of asking: How communication
affects selfishness, empathy, and altruism.” Journal of Public Economics 95:513–520.
Anonymous. 2010. “Nonseparable Preferences in Public Surveys.” mimeo. Political Stud-

ies.
Arbak, E. and M.-C. Villeval. 2007. “Endogenous Leadership: Selection and Influence.”
IZA Discussion Paper 2732.
240
Arbuthnott, J. 1710. “An argument for divine providence, taken from the constant regu-
larity observed in the births of both sexes.” Philosophical Transaction of the Royal Society
of London 27:186–190.
Ariely, D. 2008. Predictably Irrational: The Hidden Forces That Shape Our Decisions. Harper
Collins.
Ariely, Dan. 2009. “The End of Rational Economics.” Harvard Business Review .
Ariely, Dan and Michael I. Norton. 2008. “How actions create - not just reveal - prefer-
ences.” TRENDS in Cognitive Sciences 12:13–16.
Arregui, J., F. Stokman, and R. Thomson. 2006. “Compromise, Exchange and Challenge
in the European Union.” In The European Union Decides, edited by Stokman F. Achen C.
Thomson, R. and T. König, chapter 5. Cambridge University Press.
Arrow, Kenneth J. 1950. “A Difficulty in the Concept of Social Welfare.” Journal of Political
Economy, University of Chicago Press 58:328–346.
Arzieli, Yaron, Christopher P. Chambers, and Paul J. Healy. 2012. “Eliciting mul-
tiple choices: A theoretical analysis of incentives in experiments.” mimeo. Ohio
State University. http://healy.econ.ohio-state.edu/papers/Azrieli_Cambers_Healy-
CombinedDecisions.pdf (accessed October 5, 2012).
Auden, Wystan Hugh and Louis Kronenberger. 1966. The Viking Book of Aphorisms: A
Personal Selection. Dorset Press.
AWI. 2013. “Teilnahme-Regeln (AGBs).” Alfred-Weber-Institut für Wirtschaftswis-
senschaften der Ruprecht-Karls-Universität Heidelberg, http://www.uni-
heidelberg.de/md/awi/professuren/beheco/teilnahmeregeln_awi_lab.pdf (accessed
November 20, 2013).
Azaiez, Ines. 2010. “Sampling of Alternatives for Logit Mixture Mod-
els.” Master Thesis, Massachusetts Institute of Technology, http://transp-
or.epfl.ch/documents/masterTheses/AZAIEZ10.pdf (accessed November 28, 2013).
Azar, Ofer H. 2010. “Does relative thinking exist in mixed compensation
schemes?” mimeo. Ben-Gurion University of the Negev. http://thema.u-
cergy.fr/IMG/pdf/AEE_24_06_10.pdf (accessed February 17, 2012).
Bahry, D. L. and R. K. Wilson. 2006. “Confusion or fairness in the field? Rejections in the
ultimatum game under the strategy method.” Journal of Economic Behavior and Organi-
zation 60:37–54.
Bailer, Stefanie. 2006. “The European Commission and Its Legislative Activity - Not as
Integrationist and Autonomous as Believed.” Working paper of the Center for Comparative
and International Studies (ETH Zurich and University of Zurich) 24.
Bandiera, O., I. Barankay, and I. Rasul. 2005. “Social preferences and the response to
incentives: Evidence from personnel data.” Quarterly Journal of Economics 120:917–962.
241
Banuri, Sheheryar, Catherine Eckel, and Rick Wilson. 2011. “Decon-
structing Nepotism.” mimeo. Working paper, University of Texas, Dallas.
http://cbees.utdallas.edu/papers/DeconstructingNepotism.pdf (accessed September
2, 2012).
Barbera, S., J. Masso, and A. Neme. 2005. “Voting by committees under constraints.”
Journal of Economic Theory 122:185–205.
Bardsley, N., R. Cubitt, G. Loomes, P. Moffatt, C. Starmer, and R. Sugden. 2010. Experi-
mental Economics: Rethinking the Rules. Princeton: Princeton University Press.
Barmettler, Franziska, Ernst Fehr, and Christian Zehnder. 2011. “Big Experimenter Is
Watching You! Anonymity and Prosocial Behavior in the Laboratory.” Institute for the
Study of Labor (IZA) Discussion Papers 5925.
Baron, D. P. and J. A. Ferejohn. 1989. “Bargaining in Legislatures.” American Political

Science Review 83:1181–1206.
Barrett, Richard and Maurice Salles. 2006. “Social Choice With Fuzzy Preferences.” Eco-
nomics Working Paper Archive, Center for Research in Economics and Management (CREM),
University of Rennes 200615.
Barro, R. 1996. Getting It Right: Markets and Choices in a Free Society. Cambridge, MA, MIT
Press.
Barro, Robert J. 2012. “Convergence and Modernization Revisited.” NBER Working Paper
18295.
Basedow, Juergen, Joern Aldag, Justus Haucap, Peter-Michael Preusker, and Katha-
rina M. Trebitsch. 2006. “Die Privatisierung der Deutschen Bahn AG.” Sondergutachten
der Monopolkommission gemaess § 44 Abs. 1 Satz 4 GWB .
Baumeister, R., E. Bratslavsky, C. Finkenauer, and K. Vohs. 2001. “Bad is stronger than
good.” Review of General Psychology 5:323–370.
Bazerman, Max H. and Roy J. Lewicki. 1985. “Contemporary Research Directions in the
Study of Negotiations in Organizations: A Selective Overview.” Journal of Occupational
Behaviour 6:1–17.
Beck, Kurt. 2011. “Weniger Soldaten bedeutet auch weniger Kasernen - Kurt Beck ueber
die Schliessung von Kasernen und die Zukunft der Bundeswehr.” Deutschalndfunk
26.10.
Beck, Nathaniel L. 2000. “Political Methodology: A Welcoming Discipline.” Journal of the

American Statistical Association 95:651–654.
Becker, G. 1971. The Economics of Discrimination. Chicago, IL: University of Chicago Press.
Becker, Gary S. 1976. The Economic Approach to Human Behavior. Chicago University Press.
Becker, R. and S. Chakrabarti. 1995. “The Recursive Core.” Econometrica 63:401–423.
242
Beckman, Steven R., John P. Formby, James W. Smith, and Buhong Zheng. 2002. “Envy,
malice and Pareto efficiency: An experimental examination.” Social Choice and Welfare
19:349–367.
Beedham, Brian. 1993. “A Better Way to Vote.” 150 Economist Years pp. 5–8.
Beitz, Charles R. 1990. Political Equality: An Essay in Democratic Theory. Princeton Univer-
sity Press.
Bellemare, Charles and Sabine Kroger. 2007. “On representative social capital.” European
Economic Review, Elsevier 51:183–202.
Bellu, Lorenzo Giovanni and Paolo Liberati. 2006. “Inequality Analysis - The Gini In-
dex.” In United Nations - Inequality Analysis, chapter Module 040. Food and Agriculture
Organization of the United Nations.
Belot, Michele and Jeroen Van De Ven. 2009. “Friendships and Fa-
voritism on the Schoolground.” mimeo. Nuffield College, Oxford University.
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.182.3724&rep=rep1&typ
e=pdf (accessed July 4, 2012).
Benartzi, S. and R. H. Thaler. 1995. “Myopic loss aversion and the equity premium puz-
zle.” The Quarterly Journal of Economics 110:73–92.
Bendor, Jonathan, Ami Glazer, and Thomas H. Hammond. 2000. “Theories of Delega-
tion.” Annual Review of Political Science 4:235–269.
Bendor, J. and A. Meirowitz. 2004. “Spatial models of delegation.” American Political

Science Review 98:293–310.
Benoit, Jean-Pierre and Lewis A. Kornhauser. 1991. “Voting Simply in the Election of As-
semblies.” Working Papers, C.V. Starr Center for Applied Economics, New York University.
91-32.
Benoit, Jean-Pierre and Lewis A. Kornhauser. 2007. Only a Dictatorship is Efficient or Neu-
tral, volume 07-01. New York: NYU Law School.
Benoit, K. and M. Laver. 2007a. “Benchmarks for text analysis: A response to Budge and
Pennings.” Electoral Studies 26:130–135.
Benoit, K. and M. Laver. 2007b. “Estimating party policy positions: Comparing expert
surveys and hand-coded content analysis.” Electoral Studies 26:90–107.
Benoit, Kenneth and Michael Laver. 2007c. Party policy in modern democracies. Milton Park,
Abingdon, Oxon ; New York: Routledge.
Benoit, K., M. Laver, C. Arnold, P. Pennings, and M. O. Hosli. 2005. “Measuring national
delegate positions at the convention on the future of Europe using computerized word
scoring.” European Union Politics 6:291–313.
243
Benz, Arthur. 2005. “Kein Ausweg aus der Politikverflechtung? Warum die Bun-
desstaatskommission scheiterte, aber nicht scheitern musste.” Politische Vierteljahress-
chrift (PVS) 46:204–214.
Berg, Joyce, John Dickhaut, and Kevin McCabe. 2005. “Risk Preference Instability Across
Institutions: A Dilemma.” Proceedings of the National Accademy of Sciences 102:4209–
4214.
Bergstrom, Ted. 2011. “Lecture Notes on Separable Preferences.” mimeo. USCB Econ 210A.
http://www.econ.ucsb.edu/~tedb/Courses/GraduateTheoryUCSB/separability
notes.pdf (accessed September 22, 2012).
Bergstrom, T.C. and J.H. Miller. 2000. Experiments with Economic Principles: Microeco-
nomics. New York: McGraw-Hill, 2nd edition.
Berl, Janet, Richard D. McKelvey, Peter C. Ordeshook, and Mark Winer. 1976. “An Exper-
imental Test of the Core in a Simple N-Person Cooperative Nonsidepayment Game.”
Journal of Conflict Resolution 20:453–479.
Bernauer, Thomas, Detlef Jahn, Patrick Kuhn, and Stefanie Walter. 2009. Einführung in die
Politikwissenschaft. Baden-Baden: Nomos, 2nd edition.
Bernheim, B. D., B. Peleg, and M. D. Whinston. 1987a. “Coalition-proof Nash equilibria.

I. Concepts.” Journal of Economic Theory 42:1–12.
Bernheim, B. D. and M. D. Whinston. 1987b. “Coalition-proof Nash equilibria. II. Appli-

cations.” Journal of Economic Theory 42:13–29.
Bernstein, Fernando and Awi Federgruen. 2005. “Decentralized Supply Chains with
Competing Retailers under Demand Uncertainty.” Management Science 51:18–29.
Bianco, William T., Ivan Jeliazkov, and Itai Sened. 2004. “The Uncovered Set and the
Limits of Legislative Action.” Political Analysis 12:256–276.
Bianco, William T., Michael S. Lynch, Gary J. Miller, and Itai Sened. 2008. “The Con-
strained Instability of Majority Rule: Experiments on the Robustness of the Uncovered
Set.” Political Analysis 16:115–137.
Birnbaum, Michael H. 2000. Psychological Experiments on the Internet. Academic Press.
Black, Duncan. 1948. “On the Rationale of Group Decision-Making.” The Journal of Polit-
ical Economy, University of Chicago Press 56(1):23–34.
Black, Duncan. 1958. The theory of committees and elections. Cambridge Eng.: University
Press.
Black, Duncan. 1991. “Arrows Work and the Normative Theory of Committees.” Journalof
Theoretical Politics 3:259–276.
Bloomfield, Robert and Alyssa Anderson. 2010. “Experimental finance.” Johnson School
Research Paper Series 23.
244
Bloomfield, R.J., W.B. Tayler, and F. Zhou. 2009. “Momentum, Reversal, and Uninformed
Traders in Laboratory Markets.” The Journal of Finance 64:2535–2558.
Blundell, R. 1999. “Consumer Demand and Intertemporal Allocations: Engel, Slutsky

and Frisch.” In Econometrics and Economic Theory in the 20th Century: The Ragnar Frisch
Centennial Symposium, edited by S. Strom, chapter 5, pp. 147–166. Cambridge Univer-
sity Press.
BMVG, Bundesministerium der Verteidigung. 2012. “Infrastrukturmassnahmen /-

investitionen der Bundeswehr zur Umsetzung des neuen Stationierungskonzepts.”
Stellungnahme zur schriftliche Frage beim Bundeskanzleramt am 6. Maerz 3/50.
Bogart, Dan. 2009. “Nationalizations and the Development of Transport Systems: Cross-
Country Evidence from Railroad Networks: 1860-1912.” Journal of Economic History
69:202–237.
Bogart, Dan. 2010. “A Global Perspective on Railway Inefficiency and the Rise of State
Ownership, 1880-1912.” Explorations in Economic History 47:158–178.
Bohman, James and William Rehg. 1997. Deliberative Democracy. Essays on Reason and
Politics. Cambridge/London: MIT Press.
Bohnet, Iris and Bruno S. Frey. 1999. “The sound of silence in prisoner’s dilemma and
dictator games.” Journal of Economic Behavior and Organization 38:43–57.
Bohnet, I., F. Greig, B. Herrmann, and R. Zeckhauser. 2008. “Betrayal aversion: evidence
from Brazil, China, Oman, Switzerland, Turkey, and the United States.” American Eco-
nomic Review 98:294–310.
Bolton, G. E., J. Brandts, and A. Ockenfels. 2005. “Fair procedures: Evidence from games
involving lotteries.” Economic Journal 115:1054–1076.
Bolton, G. E. and A. Ockenfels. 2000. “ERC: A theory of equity, reciprocity, and competi-
tion.” American Economic Review 90:166–193.
Bolton, Gary E. and Axel Ockenfels. 2005. “The Influence of Infor-

mation Externalities on the Value of Reputation Building - An Ex-
periment.” mimeo. Working Paper Series in Economics, University of
Cologne, no. 17. http://ockenfels.uni-koeln.de/fileadmin/wiso_fak/stawi-
ockenfels/pdf/wp_series_download/wp0017.pdf (accessed October 8, 2012).
Bonica, Adam. 2012. “Operationalizing the Spatial Model with an Interactive Sur-
vey on Public Spending Preferences.” mimeo. Nex York University public files.
https://files.nyu.edu/ajb454/public/files/Bonica_Fisc_Prefs_and_Ideology_3_3.pdf
(accessed October 17, 2012).
Borghans, Lex, Angela Lee Duckworth, James J. Heckman, and Bas ter Weel. 2008. “The
Economics and Psychology of Personality Traits.” Working Papers, Geary Institute, Uni-
versity College Dublin. 200827.
245
Bornstein, Gary and Ilan Yaniv. 1998. “Individual and Group Behavior in the Ultimatum
Game: Are Groups More ’Rational’ Players?” Experimental Economics 1:101–108.
Bosch-Domenech, Antoni, Jose Garcia-Montalvo, Rosemarie Nagel, and Albert Satorra.

2002. “One, Two, (Three), Infinity...: Newspaper and Lab Beauty-Contest Experi-
ments.” American Economic Review 92:1687–1701.
Bottom, W. P., C. L. Eavey, G. J. Miller, and J. N. Victor. 2000. “The institutional effect on
majority rule instability: Bicameralism in spatial policy decisions.” American Journal of
Political Science 44:523–540.
Bottom, W. P. and P. W. Paese. 1997. “False consensus, stereotypic cues, and the percep-
tion of integrative potential in negotiation.” Journal of Applied Social Psychology 27:1919–
1940.
Bouckaert, J. and G. Dhaene. 2004. “Inter-ethnic trust and reciprocity: results of an ex-
periment with small businessmen.” European Journal of Political Economy 20:869–886.
Boyd, R. and P. J. Richerson. 1988. “The Evolution of Reciprocity in Sizable Groups.”

Journal of Theoretical Biology 132:337–356.
Bozdogan, H. 1987. “Model Selection and Akaike’s Information Criterion (AIC): The
General Theory and Its Analytical Extensions.” Psychometrika 52:345–370.
Bozdogan, H. 2000. “Akaike’s Information Criterion and Recent Developments in Infor-

mation Complexity.” Journal of Mathematical Psychology 44:62–91.
Bradler, Christiane. 2009. “Social Preferences under Risk - An experimental analysis.”

ZEW Discussion Papers, Center for European Economic Research 09-077.
Bradley, Michael J. 2006. The Birth of Mathematics: Ancient Times to 1300. Sonlight Chris-
tian.
Bradley, W. J., J. K. Hodge, and D. M. Kilgour. 2005. “Separable discrete preferences.”

Mathematical Social Sciences 49:335–353.
Brady, Henry E. 2008. “Causation and Explanation in Social Science.” In The Oxford
Handbook of Political Methodology, edited by David Collier, Henry E. Brady, and Janet M.
Box-Steffensmeier, chapter 10. Oxford: Oxford University Press.
Brambor, Thomas, Roberts Clark William, and Matt Golder. 2006. “Understanding Inter-
action Models: Improving Empirical Analysis.” Political Analysis 14:63–82.
Brams, S. J., D. M. Kilgour, and W. S. Zwicker. 1997. “Voting on referenda: The separabil-
ity problem and possible solutions.” Electoral Studies 16:359–377.
Branas-Garza, Pablo, Miguel A. Duran, and Maria Paz Espinosa. 2009. “The Role of Per-
sonal Involvement and Responsibility in Unfair Outcomes : A Classroom Investigation
2009 21: 225.” Rationality and Society 21:225–248.
246
Branas-Garza, Pablo, Antonio M. Espin, and Filippos Exadaktylos. 2012. “Students, Vol-
unteers and Subjects: Experiments on Social Preferenes.” Working Papers, Chapman
University, Economic Science Institute 12-11.
Brandenburger, Adam. 1996. “Strategic and Structural Uncertainty in Games.” In

Wise Choices: Games, Decisions, and Negotiations, edited by Richard Zeckhauser, Ralph
Keeney, and James Sibenius, pp. 221–232. Harvard Business School Press, Boston.
Brandts, Jordi and Carles Sola. 2006. “Personal Relations and their Effect on Behavior
in an Organizational Setting: An Experimental Study.” mimeo. Working paper, Univer-
sitat Autonoma de Barcelona. http://digital.csic.es/bitstream/10261/1698/1/69207.pdf
(accessed July 29, 2012).
Braumöller, B. F. 2003. “Causal complexity and the study of politics.” Political Analysis
11:209–233.
Braumöller, B. F. and G. Görtz. 2000. “The methodology of necessary conditions.” Amer-

ican Journal of Political Science 44:844–858.
Braun, Dietmar. 2013. “Theorien rationaler Wahl.” In Studienbuch Politikwissenschaft,

edited by M. G. Schmidt, F. Wolf, and S. Wurster. VS Verlag für Sozialwissenschaften,
Wiesbaden.
Braunberger, Gerald. 2013. “Wenn Ökonomen streiten: Wie entsteht Wohlstand? Ace-
moglu/Robinson gegen Jeffrey Sachs.” Frankfurter Allgemeine, Fazit - das Wirtschaftsblog
http://blogs.faz.net/fazit/2013/01/13/wenn-oekonomen-streiten-woher-kommt-
der-wohlstand-acemoglu-robinson-versus-jeffrey-sachs-889 (accessed March 17,
2013).
Brennan, Geoffrey and James M. Buchanan. 2000. The Reason of Rules: Constitutional Po-
litical Economy. Indianapolis, IN: Liberty Fund, Inc.
Brewer, M. 1999. “The Psychology of Prejudice: Ingroup Love or Outgroup Hate?” Jour-
nal of Social Issues 55:429–444.
Brown, L. D., T. T. Cai, and DasGupta A. 2001. “Interval estimation for a binomial pro-
portion.” Statistical Science 16:101–133.
Börsch-Supan, Axel and Martin Gasche. 2013. “Sozialsysteme und Demographie - Sägen
an den Säulen.” Frankfurter allgemeine Zeitung 28.01.2013:7.
Bräuninger, Thomas, Andre Bächtiger, and Susumu Shikano. 2012. “Experiment und
Simulation.” Jahrbuch für Handlungs- und Entscheidungstheorie 7.
Brunnhuber, Georg. 2006. “DB-Aufsichtsrat fuer Boersengang.” Eisenbahn-Revue Interna-

tional 8 – 9:378.
Bräutigam, S. 2005. “Neuroeconomics - From neural systems to economic behaviour.”

Brain Research Bulletin 67:355–360.
247
Buchanan, James M. 1986. The Constitution of Economic Policy. Nobel Prize Lecture.
Buchanan, James M. and Gordon Tullock. 1962. The calculus of consent: logical foundations
of constitutional democracy. Ann Arbor: University of Michigan Press.
Bueno de Mesquita, Bruce. 2004. “The methodical study of politics.” In Problems and
Methods in the Study of Politics, edited by Ian Shapiro, Rogers M. Smith, and Tarek E.
Masoud, pp. 227–248. Cambridge University Press.
Bueno de Mesquita, Bruce. 2011. “A New Model for Predicting Policy Choices Prelimi-
nary Tests.” Conflict Management and Peace Science 28:65–85.
Buhrmester, M., T. Kwang, and S. D. Gosling. 2011. “Amazon’s Mechanical Turk: A New
Source of Inexpensive, Yet High-Quality, Data?” Perspectives on Psychological Science
6:3–5.
Buller, David J. 2005. Adapting minds : evolutionary psychology and the persistent quest for
human nature. Cambridge, Mass.: MIT Press.
Burgess, Philip M. and James A. Robinson. 1969. “Alliances and the Theory of Collec-
tive Action: A Simulation of Coalition Processes.” Midwest Journal of Political Science
13(2):194–218.
Burkhart, Simone. 2009. “Reforming Federalism in Germany: Incremental Changes in-
stead of the Big Deal.” The Journal of Federalism 39.
Burnham, Kenneth P. and David R. Anderson. 2004. “Multimodel Inference: Understand-
ing AIC and BIC in Model Selection.” In Sociological Methods Research, volume 33, pp.
261–304. Sociological Methods and Research.
Butler, D. J. and C. F. Camerer. 2005. “Behavioral game theory: Experiments in strategic
interaction.” Journal of Economic Surveys 19:137–147.
Cabrales, Antonio, Raffaele Miniaci, Marco Piovesan, and Giovanni Ponti. 2010. “So-
cial Preferences and Strategic Uncertainty: An Experiment on Markets and Contracts.”
American Economic Review, American Economic Association 100:2261–2278.
Camerer, C. F. 2003. “Behavioral game theory: Plausible formal models that predict ac-
curately.” Behavioral and Brain Sciences 26:157ff.
Camerer, Colin F and Ernst Fehr. 2004. “Measuring Social Norms and Preferences Using
Experimental Games: A Guide for Social Scientists.” Institute for Empirical Research in
Economics - University of Zurich, Working Papers 97:55–95.
Camerer, C. F. and R. M. Hogarth. 1999. “The effects of financial incentives in experi-
ments: A review and capital-labor-production framework.” Journal of Risk and Uncer-
tainty 19:7–42.
Camerer, C. F., E. J. Johnson, T. Rymon, and S. Sen. 1993. “Cognition and framing in se-
quential bargaining for gains and losses.” In Frontiers of game theory, edited by K. Bin-
more, A. Kirman, and P. Tani, pp. 27–47. Cambridge, MA: MIT Press.
248
Camerer, C. F., G. Loewenstein, and D. Prelec. 2004. “Neuroeconomics: Why economics
needs brains.” Scandinavian Journal of Economics 106:555–579.
Camerer, C. F. and R. H. Thaler. 1955. “Anomalies: Ultimatums, dictators, and manners.”

Journal of Economic Perspectives 9:209–219.
Cameron, Charles M. and Rebecca Morton. 2002. “Formal Theory Meets Data.” In Polit-
ical Science: The state of the discipline, edited by Ira Katznelson and Helen V. Milner, pp.
784–804. London and New York: W.W. Norton and American Political Science Associ-
ation.
Campbell, Donald T. 1969. “Reforms as experiments.” American Psychologist 24.
Carmines, Edward G. and Richard A. Zeller. 1979. Reliability and Validity Assessment.
Quantitative Applications in the Social Science. Sage Publications, London.
Carpenter, Jeffrey and Erika Seki. 2005. “Do Social Preferences Increase Productivity?
Field Experimental Evidence from Fishermen in Toyama Bay.” Discussion paper series,
Institute for the Study of Labor 1697.
Carrubba, C. J. and A. Singh. 2004. “A decision theoretic model of public opinion: Guns,
butter, and European common defense.” American Journal of Political Science 48:218–231.
Carter, John and Michael Irons. 1991. “Are Economists Different, and If So, Why?” Journal
of Economic Perspectives 5:171–177.
Cason, T. N. and V. L. Mui. 2007. “Communication and coordination in the laboratory

collective resistance game.” Experimental Economics 10:251–267.
Cason, T. N., T. Saijo, T. Yamato, and K. Yokotani. 2004. “Non-excludable public good
experiments.” Games and Economic Behavior 49:81–102.
Castillo, Joan Joseph. 2009. “Convenience Sampling.” Experiment Resources

http://www.experiment-resources.com/convenience-sampling.html (accessed Octo-
ber 1, 2012).
CESS. 2012. “Conducting an experiment at CESS - Rules and procedures associated with
conducting an experiment.” Nuffield center for Experimental social science .
Chajewska, Urszula and Daphne Koller. 1999. Learning the Structure of Utility Functions.
Stanford: Stanford University.
Chalkiadakis, G. and C. Boutilier. 2004. “Bayesian Reinforcement Learning for Coalition

Formation Under Uncertainty.” Autonomous Agents and Multiagent Systems 04:1090–
1097.
Chalkiadakis, G. and C. Boutilier. 2008. “Sequential Decision Making in Repeated Coali-

tion Formation under Uncertainty.” Proceedings of the 7th international joint conference on
Autonomous agents and multiagent systems 1:347–354.
249
Chandrasekhar, Arun G. and Juan Pablo Xandri. 2012. “A note on payments in ex-
periments of infinitely repeated games with discounting.” mimeo. MIT Economics.
http://economics.mit.edu/files/7310 (accessed November 7, 2012).
Charness, G. 2004. “Attribution and reciprocity in an experimental labor market.” Journal
of Labor Economics 22:665–688.
Charness, Gary and Nuno Garoupa. 2000. “Reputation, Honesty, and Efficiency with
Insider Information: an Experiment.” Journal of Economics and Management Strategy
9:425–451.
Charness, Gary and Uri Gneezy. 2008. “What’s in a name? Anonymity and social distance
in dictator and ultimatum games.” mimeo. Departmental Working Papers, University of
California. http://econ.econ.ucsb.edu/papers/wp11-01.pdf (accessed October 2, 2010).
Charness, Gary, Uri Gneezy, and Michael A. Kuhn. 2012. “Experimental methods:
Between-subject and within-subject design.” Journal of Economic Behavior and Orga-
nization 81:1–8.
Charness, G., E. Haruvy, and D. Sonsino. 2007a. “Social distance and reciprocity: An
internet experiment.” Journal of Economic Behavior and Organization 63:88–103.
Charness, G. and M. Rabin. 2002. “Understanding social preferences with simple tests.”
Quarterly Journal of Economics 117:817–869.
Charness, Gary, Luca Rigotti, and Aldo Rustichini. 2007b. “Individual Behavior and
Group Membership.” American Economic Association 97:1340–1352.
Charness, Gary and Aldo Rustichini. 2011. “Gender differences in cooperation with
group membership.” Games and Economic Behavior 72:77–85.
Chen, Y. and S. X. Li. 2009. “Group Identity and Social Preferences.” American Economic
Review 99:431–457.
Chen, Y. and P. C. Ordeshook. 1998. “Veto games: Spatial committees under unanimity
rule.” Public Choice 97:617–643.
Chongwoo, Choe and Park In-Uck. 2011. “Information, Authority, and Corporate Hier-
archies.” The B.E. Journal of Theoretical Economics 11.
Chou, E., M. McConnell, R. Nagel, and C. R. Plott. 2009. “The control of game form
recognition in experiments: understanding dominant strategy failures in a simple two
person "guessing" game.” Experimental Economics 12:159–179.
Christians, Malte. 2012. “Mehrheit akzeptiert niedrigeres Rentenniveau.” Focus 41.
Clarke, Kevin A. and David M. Primo. 2007. “Modernizing Political Science: A Model-
Based Approach.” Perspectives on Politics 5:741–753.
Cleave, Blair L., Nikos Nikiforakis, and Robert Slonim. 2010. “Is There Selection Bias
in Laboratory Experiments?” Working Papers University of Sydney, School of Economics
2123/6957.
250
Clinton, J. D. and A. Meirowitz. 2004. “Testing explanations of strategic voting in legisla-
tures: A reexamination of the compromise of 1790.” American Journal of Political Science
48:675–689.
Clopper, C. J. and E. S. Pearson. 1934. “The use of confidence or fiducial limits illustrated
in the case of the binomial.” Biometrika 26:404–413.
Coase, Ronald. 1937. “The Nature of the Firm.” Economica 4:386–405.
Coase, Ronald. 1960. “The Problem of Social Cost.” Journal of Law and Economics 3:1–44.
Cohen, Dov and Richard Nisbett. 1997. “Field Experiments Examining the Culture of
Honor: The Role of Institutions in Perpetuating Norms about Violence.” Personality
and Social Psychology Bulletin 11:1188–1199.
Coleman, Eric and Elinor Ostrom. 2009. “Experimental Contributions to Collective Ac-
tion Theory.” In Cambridge Handbook of Experimental Political Science, edited by Jamie
Druckman, Donald Green, Jim Kluklinski, and Arthur Lupia, chapter 24. Cambridge
University Press.
Coleman, James. 1990. Foundations Of Social Theory. Cambridge: Belkhap Press.
Colman, Andrew M. 2003. “Cooperation, psychological game theory, and limitations of

rationality in social interaction.” Behavioral and Brain Sciences 26:139–198.
Colomer, Josep H. 2001. Political Institutions: Democracy and Social Choice. Oxford Univer-
sity Press, New York.
Colomer, Josep M. 1999. “On the Geometry of Unanimity Rule.” Journal of Theoretical
Politics 11:543–553.
Consilium, Council of the European union. 2000. “Council Regulation amending Regu-
lation (EEC) No 404/93 on the common organisation of the market in bananas.” In-
terinstitutional file - legislative acts and other instruments 99/0235 (CNS).
Consilium, Rat der europäischen Union. 1999. “Beratungsergebnisse des Vorschlag für
eine Verordnung (EG) des Rates zur Stärkung des Dialogs mit den Unternehmen
und sonstigen Beteiligten über die Gemeinsame Fischereipolitik.” Interinstitutionelles
Dossier 99/0163 (CNS).
Cooper, D. and K. Kühn. 2009. “Communication, Renegotiation, and

the Scope for Collusion.” mimeo. CEPR Discussion Paper, no. 7563.
http://www.cepr.org/pubs/dps/DP7563.asp (accessed July 23, 2012).
Cooper, David J. and John H. Kagel. 2005. “Are Two Heads Better Than One? Team
Versus Individual Play in Signaling Games.” American Economic Review 95:477–509.
Cooper, John O., Timothy E. Heron, and William L. Heward. 2007. Applied Behavior Anal-
ysis. Pearson, 2nd edition.
251
Costa-Gomes, Miguel, Vincent P. Crawford, and Bruno Broseta. 2001. “Cognition and
Behavior in Normal-Form Games:An Experimental Study.” Econometrica 69:1193–1235.
Coughlin, Peter J. 1990a. “Candidate Uncertainty and Electoral Equilibria.” In Advances

in the Spatial Theory of Voting, edited by James Enelow and Melvin Hinich. NewYork:
Cambridge University Press.
Coughlin, Peter J. 1990b. “Majority Rule and Election Models.” Journal of Economic Sur-
veys 3:157 – 188.
Cox, Gary W. 1984. “Electoral Equilibrium in Double Member Districts.” Public Choice
44:443–451.
Cox, Gary W. 1997. “Duverger’s Law and Strategic Voting.” In Making Votes Count:
Strategic Coordination in the World’s Electoral Systems, edited by Gary W. Cox, p. 315.
Cox, James C. 2010. “Some issues of methods, theories, and experimental designs.” Jour-
nal of Economic Behavior and Organization 73:24–28.
Cox, James C. and Cary A. Deck. 2005. “On the Nature of Reciprocal Motives.” Economic
Inquiry, Western Economic Association International 43:623–635.
Cox, James C. and Glenn W. Harrison. 2008. Risk Aversion in Experiments, volume 12 of
Research in Experimental Economics Series. Emerald Group Publishing.
Cox, James C., Klarita Sadiraj, and Vjollca Vjollca. 2008. “Implications of Trust, Fear, and
Reciprocity for Modeling Economic Behavior.” Experimental Economics 11:1–24.
Cox, James C. and Vjollca Sadiraj. 2012. “Direct Tests Of Individual Preferences For Effi-
ciency And Equity.” Economic Inquiry, Western Economic Association International 50:920–
931.
Craig, Stephen, David Hill, and Scott Richards. 2006. “Citizen Attitudes about Di-
vided Government in the United States: Does the Public Really Care?” mimeo.
Paper presented at the annual meeting of the Southern Political Science Association.
http://www.allacademic.com/meta/p68426_index.html (accessed July 16, 2012).
Crawford, Vincent P. 1985. “Dynamic Games and Dynamic Contract Theory.” The Journal
of Conflict Resolution 29:195–224.
Crawford, Vincent P., Miguel A. Costa-Gomes, and Nagore Iriberri.

2010. “Strategic thinking.” mimeo. University of Colifornia.
http://dss.ucsd.edu/~vcrawfor/CGCI27Dec10.pdf (accessed April 8, 2011).
Crawford, Vincent P. and Hans Haller. 1990. “Learning How to Cooperate: Optimal Play
in Repeated Coordination Games.” Econometrica 58:571–595.
Croson, R. 1999. “Handbook of experimental economics.” Journal of Economic Behavior

and Organization 40:115–118.
252
Croson, Rachel. 2005. “The Method of Experimental Economics.” International Negotiation
10:131–148.
Croson, R. and U. Gneezy. 2009. “Gender Differences in Preferences.” Journal of Economic
Literature 47:448–474.
Cross, J. P. 2012. “Interventions and negotiation in the Council of Ministers of the Euro-
pean Union.” European Union Politics 13:47–69.
Croux, C., G. Dhaene, and D. Hoorelbeke. 2003. “Robust Standard Errors for Robust
Estimators. Research.” Discussions Paper Series, Department of Applied Economics, Leuven,
Belgium. 03.16.
Dal Bo, Pedro, Andrew Foster, and Louis Putterman. 2010. “Institutions and Behav-
ior: Experimental Evidence on the Effects of Democracy.” American Economic Review
100:2205–2229.
Dalio, Ray. 2011. “A Template for Understanding - How the Economic Ma-
chine Works and How it is Reflected Now.” mimeo. Bridgewater Associates.
http://www.zerohedge.com/sites/default/files/images/user5/imageroot/2011/12/
Ray%20Dalio%27s%20Template%20for%20Understanding%20%28Oct-2011%29.pdf
(accessed December 2, 2012).
Dalton, Russell J., Wilhelm Bürklin, and Andrew Drummond. 2001. “Public opinion and
direct democracy.” Journal of Democracy 12.
Dana, J., R. A. Weber, and J. X. Kuang. 2007. “Exploiting moral wiggle room: experiments
demonstrating an illusory preference for fairness.” Economic Theory 33:67–80.
Davis, Douglas D. and Charles A. Holt. 1993. Experimental economics. Princeton, N.J.:
Princeton University Press.
Davis, O. and M. Hinich. 1966. “A mathematical model of policy formation in a demo-
cratic society.” In Mathematical applications in political science, edited by J. Bernd, vol-
ume II, pp. 175–208. Southern Methodist University Press, Dallas.
Davis, O. and M. Hinich. 1967. “Some results related to a mathematical model of policy
formation in a democratic society.” In Mathematical applications in political science, edited
by J. Bernd, volume III, pp. 14–38. University Press of Virginia, Charlottesville.
Davis, O. and M. Hinich. 1968. “On the power and importance of the mean preference in
a mathematical model of democratic choice.” Public Choice 5:59–72.
Davis, Otto A., Melvin J. Hinich, and Peter C. Ordeshook. 1970. “An Expository Develop-
ment of a Mathematical Model of the Electoral Process.” The American Political Science
Review 64:426 – 448.
Dawes, Robyn M., John M. Orbell, Randy T. Simmons, and Alphons J. C. Van De Kragt.
1986. “Organizing Groups for Collective Action.” The American Political Science Review
80(4):1171–1185.
253
DBT, Deutscher Bundestag. 2011. “Gesetzentwurf Bundesregierung: Feststellung des
Bundeshaushaltsplans fuer das Haushaltsjahr 2011 (Haushaltsgesetz 2011).” Bun-
destagsdrucksachen 17/2500.
DBT, Deutscher Bundestag. 2012. “Antwort der Bundesregierung auf die Kleine An-
frage der Abgeordneten Dorothea Steiner, Nicole Maisch, Baerbel Hoehn, weiterer Ab-
geordneter und der Fraktion BUENDNIS 90/DIE GRUENEN: Weiterentwicklung der
REACH-Verordnung - Position der Bundesregierung in der Europaeischen Debatte.”
Bundestags Drucksache 17/9524.
de Groot-van Leeuwen, Leny E. and Wannes Rombouts. 2010. Separation of Powers in

Theory and Practice - An International Perspective. Wolf Legal Publishers.
DellaVigna, Stefano. 2007. “Psychology and Economics: Evidence from the Field.” NBER
Working Paper Series 13420.
Demertzis, Maria. 2009. “The ’Wisdom of the Crowds’ and Public Policy.” DNB Working
Papers, Netherlands Central Bank 203.
Denzau, Arthur T. and Robert J. Mackay. 1981. “Structure-Induced Equilibria and Perfect-
Foresight Expectations.” American Journal of Political Science 25:762–779.
Dewan, Torun and Rafael Hortala-Vallve. 2009. “The Three A’s of Government Forma-
tion: Appointment, Allocation, and Assignment.” PSPE working papers p. 34.
Diederich, Johannes and Timo Goeschl. 2011. “Giving in a Large Economy: Price vs. Non-
Price Effects in a Field Experiment.” Discussion Paper Series, Department of Economics,
University of Heidelberg 514.
Diermeier, Daniel and Sean Gailmard. 2006. “Self-Interest, Inequality, and Entitlement in
Majoritarian Decision-Making.” Quarterly Journal of Political Science 1:327–350.
Diermeier, Daniel and Rebecca Morton. 2005. “Experiments in Majoritarian Bargaining.”

In Social Choice and Strategic Decisions, edited by David Austen-Smith and John Duggan,
pp. 201–236. Berlin and New York: Springer: Springer.
Dittrich, Dennis A.V. and Martin G. Kocher. 2011. “Monitoring and

Pay: An Experiment on Employee Performance under Endogenous Su-
pervision.” mimeo. Papers on Strategic Interaction, Max Planck Institute
of Economics, Strategic Interaction Group. 2006-23. http://epub.ub.uni-
muenchen.de/12222/1/Dittrich_Kocher_2011_mimeo_Monitoring_and_Pay.pdf
(accessed August 11, 2012).
Dittrich, Dennis A. and Anthony Ziegelmayer. 2010. “Laboratory bilateral gift

exchange: Heterogeneous social preferences.” mimeo. Jacobs University Bremen.
http://economicscience.net/content/publications (accessed August 11, 2012).
Dixit, Avinash K., Susan Skeath, and David Reiley. 2009. Games of strategy. New York: W.
W. Norton & Co.
254
Djorgovski, George. 2005. “Virtual Astronomy, Information Technology, and the New
Scientific Methodology.” Seventh International Workshop on Computer Architectures for
Machine Perception (CAMP 2005), 4-6 July 2005, Palermo, Italy pp. 125–132.
Dockner, E. J. and N. V. Long. 1993. “International Pollution-Control - Cooperative Versus

Noncooperative Strategies.” Journal of Environmental Economics and Management 25:13–
29.
Doina, Bratu Renate and Petria Nicolae. 2011. “The Structure of Financial Regulation.”
Ovidius University Annals, Economic Sciences Series, Ovidius University of Constantza, Fac-
ulty of Economic Sciences 0:128–132.
Domingos, Pedro. 1999. “The role of Occam’s Razor in knowledge discovery.” Data
Mining and Knowledge Discovery 3:409–425.
Dopfner, Kurt. 2002. “Die Rückkehr des verlorenen Menschen - Wege zum Homo sapiens
oeconomicus.” In Psychologische Grundlagen der Ökonomie, edited by Fehr, pp. 99–106.
Neue Zürcher Zeitung.
Dougherty, Keith, Brian Pitts, Justin Moeller, and Robi Ragan. 2009. “An Experimental
Study of the Efficiency of Unanimity Rule and Majority Rule.” mimeo. Indiana University
Bloomington. http://www.indiana.edu/~wow4/papers/dougherty_pitts_wow4.pdf
(accessed January 27, 2012).
Dougherty, Keith L. and Julian Edward. 2004. “The Pareto Efficiency and Expected Costs
of k-Majority Rules.” Politics Philosophy and Economics 3:161–89.
Dougherty, Keith L. and Julian Edward. 2012. “Voting for Pareto optimality: a multidi-
mensional analysis.” Public Choice 151:655–678.
Dow, Alexander and Sheila C. Dow. 2011. “Animal Spirits Revisited.” Capitalism and
Society 6:1. Berkeley Electronic Press.
Dowding, Keith and Desmond King. 1995. Introduction to Preferences, Institutions, and
Rational Choice. Oxford: Clarendon Press.
Downs, Anthony. 1957. An economic theory of democracy. New York,: Harper.
Downs, Anthony and Rudolf Wildenmann. 1968. Oekonomische Theorie der Demokratie.
Tuebingen,: Mohr.
DPA. 2011a. “Bahn: Boersengangs-Diskussion momentan ueberfluessig.” Verkehrs Rund-

schau 02.03.
DPA. 2011b. “Finanzkrise stoppt Boersengang der Bahn.” Die Zeit 14.11.
Draper, Janet. 2004. “The relationship between research question and research design.”
In Research into Practice: Essential Skills for Reading and Applying Reasearch in Nursing
and Health Care, edited by Patrick A. Crookes and Sue Davies, chapter 6, pp. 69–84.
Edinburgh: Bailliere Tindall.
255
Druckman, James and Arthur Lupia. 2006. “Mind, Will, and Choice.” In The Oxford
Handbook of Contextual Political Analysis, edited by Joseph L. Tilly, Charles and Robert E
Goodin, chapter 5. Oxford University Press.
Druckman, J. N. 2004. “Political preference formation: Competition, deliberation, and

the (Ir)relevance of framing effects.” American Political Science Review 98:671–686.
Druckman, James N. 2011. Cambridge Handbook of Experimental Political Science. Cam-

bridge ; New York: Cambridge University Press.
Druckman, James N. and Cindy D. Kam. 2009. “Students as Experimental Participants:

A Defense of the narrow Data Base.” In Cambridge Handbook of Experimental Political
Science, edited by J. N. Druckman, D. P. Green, J. H. Kuklinski, and A. Lupia. Boston:
Druckman, J. N. and M. Parkin. 2005. “The impact of media bias: How editorial slant
affects voters.” Journal of Politics 67:1030–1049.
Dufwenberg, Martin, Paul Heidhues, Georg Kirchsteiger, Frank Riedel, and Joel Sobel.
2008. “Other-Regarding Preferences in General Equilibrium.” CEPR Discussion Papers
6815.
Dutcher, E. Glenn. 2012. “The effects of telecommuting: An experimental examination.

The role of dull and creative tasks.” Journal of Economic Behavior and Organization
84:355–363.
Duverger, Maurice. 1954. Political Parties: Their Organization and Activity in the Modem
State. New York: Wiley.
Dyer, J. S., P. C. Fishburn, R. E. Steuer, J. Wallenius, and S. Zionts. 1992. “Multiple Crite-
ria Decision-Making, Multiattribute Utility-Theory - the Next 10 Years.” Management
Science 38:645–654.
Easton, David. 1957. “An Approach to the analysis of Political systems.” World Politics
9:383–400.
Eavey, Cheryl L. and Gary J. Miller. 1984. “Fairness in Majority Rule Games with a Core.”
American Journal of Political Science 28:570–586.
Eavey, Cheryl L. and Gary J. Miller. 1995. “Subcommittee Agenda Control.” Journal of
EC, European Commission. 2000. “Commission proposes solution to end banana dis-
pute.” mimeo. Communication of the European Commission to the European Council,
IP/00/1110. http://europa.eu/rapid/press-release_IP-00-1110_en.htm?locale=en (ac-
cessed February 9, 2012).
Eckel, Catherine and Rick Wilson. 2006. “Internet Cautions: Experimental Games with
Internet Partners.” Experimental Economics 9:53–66.
256
Eckel, Catherine C. and Philip J. Grossman. 1996. “Altruism in Anonymous Dictator
Games.” Games and Economic Behavior 16:181–191.
Eckel, Catherine C. and Philip J. Grossman. 1998. “Are Women Less Selfish Than Men?
Evidence from Dictator Experiments.” Economic Journal 108:726–735.
Eckel, Catherine C. and Ragan Petrie. 2011. “Face Value.” American Economic Review
101:1497–1513.
Edwards, W. and J.R Newman. 1982. Multiattribute evaluation. Beverly Hills, CA: Sage.
Eeckhoudt, Louis, Christian Gollier, and Harris Schlesinger. 2005. Economic and Financial
Decisions under Risk. Princeton University Press.
Efron, Bradley and Robert Tibshirani. 1993. An introduction to the bootstrap. New York:
Chapman & Hall.
Ehmke, Mariah D., Jayson L. Lusk, and Wallace E. Tyner. 2005. “A Multidimensional
Homo Economicus: Cultural Dimensions of Economic Preferences in Four Countries.”
Presented at the Annual meeting of the Agricultural and Applied Economics Association, July
24-27 .
Ehrhart, Karl-Martin, Roy Gardner, Juergen von Hagen, and Claudia Keser. 2007. “Bud-
get processes: Theory and experimental evidence.” Games and Economic Behavior
59:279–295.
Eichberger, Jürgen and David Kelsey. 2008. “Ambiguity.” In The Oxford Handbook of
Rational and Social Choice, edited by P. Anand, P. Pattanaik, and C. Puppe, chapter V,
pp. 113–139. Oxford: Oxford University Press.
Eisenkopf, Gerald and Sabrina Teyssier. 2010. “Envy and Loss Aversion in Tournaments.”
Research Paper Series, Thurgau Institute of Economics and Department of Economics (Univer-
sity of Konstanz) 52.
El-Gamal, Mahmoud A. and David Grether. 1995. “Are People Bayesian? Uncovering
Behavioral Strategies.” Journal of the American Statistical Association: Applications and
Case Studies 90:1137–1145.
Ellsberg, D. 1961. “Risk, ambiguity and the Savage axioms.” Quarterly Journal of Economics
75:643–669.
Elster, Jon. 1998. Deliberative Democracy. Cambridge: Cambridge University Press.
Enelow, James M. and Melvin J. Hinich. 1984. The spatial theory of voting : an introduction.
Cambridge Cambridgeshire ; New York: Cambridge University Press.
Engelmann, D. and M. Strobel. 2000. “An Experimental Comparison of the Fairness Mod-
els by Bolton and Ockenfels and by Fehr and Schmidt.” Schriftenreihe Sonderforschungs-
bereich 373, Humboldt Universitaet Berlin. 28.
257
Engelmann, D. and M. Strobel. 2004. “Inequality aversion, efficiency, and maximin pref-
erences in simple distribution experiments.” American Economic Review 94:857–869.
EP, European Parliament. 2005. “REACH - MEPs debate chem-

icals legislation and new chemicals agency environment.”
http://www.europarl.europa.eu/sides/getDoc.do?pubRef=-//EP//TEXT+IM-
PRESS+20051116IPR02381+0+DOC+XML+V0//EN (accessed October 11, 2011).
Epanechnikov, V. A. 1969. “Nonparametric estimation of a multidimensional probability

density.” Theory of Probability and Its Applications 14:153–158.
Epstein, David and Sharyn O’Halloran. 1996. “Divided Government and the Design of
Administrative Procedures: A Formal Model and Empirical Test.” The Journal of Politics,
Cambridge University Press 58:373–397.
Epstein, David and Sharyn O’Halloran. 1999. Delegating powers: a transaction cost politics
approach to policy making under separate powers. Cambridge, U.K. ; New York: Cambridge
University Press.
Eriksson, K. and B. Simpson. 2010. “Emotional reactions to losing explain gender differ-
ences in entering a risky lottery.” Judgment and Decision Making 5:159–163.
Erlei, Mathias. 2004. “Heterogeneous Social Preferences.” Working Papers in Economics,

Technische Universitaet Clausthal 0001.
Erlenmaier, Ulrich and Hans Gersbach. 2001. “Flexible Majority Rules.” CESifo Working
Paper Series 464.
Ert, Eyal and Ido Erev. 2008. “The rejection of attractive gambles, loss aversion, and the
lemon avoidance heuristic.” Journal of Economic Psychology 29:715–723.
Ert, Eyal, Ido Erev, and Alvin E. Roth. 2011. “A Choice Prediction Competition for Social
Preferences in Simple Extensive Form Games: An Introduction.” Games, MDPI. 2:257–
276.
Evans, J. 2003. “In two minds: dual-process accounts of reasoning.” TRENDS in Cognitive
Sciences 7.
Experimental-Psychology. 2012. “The roar of the crowd - Crowdsourcing is transforming

the science of psychology.” The Economist print edition May 26th:69–70.
Faas, Thorsten and Sascha Huber. 2010. “Experimente in der Politikwissenschaft: Vom
Mauerbluemchen zum Mainstream.” Politische Vierteljahresschrift (PVS) 51:721–749.
Falk, Armin. 2009. “Lab Experiments Are a Major Source of Knowledge in the Social.”
Science 326:535–538.
Falk, Armin, Stephan Meier, and Christian Zehnder. 2011. “Do Lab Experiments Mis-
represent Social Preferences? The case of self-selected student samples.” Journal of the
European Economic Association 11.
258
FDP. 2007. “Echter Wettbewerb fuer mehr Verkehr auf der Schiene.” Beschluss des FDP-
Praesidiums Pressemitteilung vom 24.09.2007, 13:06.
Feddersen, T., S. Gailmard, and A. Sandroni. 2009. “Moral Bias in Large Elections: Theory
and Experimental Evidence.” American Political Science Review 103:175 192.
Feddersen, Timothy J. 1992. “A Voting Model Implying Duverger’s Law and Positive
Turnout.” American Journal of Political Science 36:938–962.
Fehr, Ernst, Helen Bernhard, and Bettina Rockenbach. 2008. “Egalitarianism in Young
Children.” Nature 454:1079–1084.
Fehr, E. and U. Fischbacher. 2002. “Why social preferences matter - The impact of non-
selfish motives on competition, cooperation and incentives.” Economic Journal 112:C1–
C33.
Fehr, Ernst and Urs Fischbacher. 2003. “The nature of human altruism.” Nature 425:785–
791.
Fehr, E., U. Fischbacher, and S. Gachter. 2002a. “Strong reciprocity, human cooperation,
and the enforcement of social norms.” Human Nature-an Interdisciplinary Biosocial Per-
spective 13:1–25.
Fehr, Ernst, Alexander Klein, and Klaus M. Schmidt. 2001. “Fairness, Incentives and
Contractual Incompleteness.” CEPR Discussion Papers 2790.
Fehr, E., S. Kremhelmer, and K.M. Schmidt. 2002b. “Fairness and the Optimal Allocation
of Ownership Rights.” Discussion Paper Series, Governance and the Efficiency of Economic
Systems 11. Free University of Berlin, Humboldt University of Berlin, University of
Bonn, University of Mannheim, University of Munich.
Fehr, E. and K. M. Schmidt. 1999. “A theory of fairness, competition, and cooperation.”

Fehr, Ernst and Klaus M. Schmidt. 2000. “Theories of Fairness and Reciprocity - Evi-
dence and Economic Applications.” Working Papers, Institute for Empirical Research in
Economics, University of Zurich. iewwp075.
Fehr, Ernst and Gerhard Schwarz. 2002. Psychologische Grundlagen der Ökonomie. Neue
Zürcher Zeitung.
Feinerer, Ingo, Kurt Hornik, and David Meyer. 2008. “Text Mining Infrastructure in R.”
Journal of Statistical Software 25:1–54.
Felsenthal, Dan, Amnon Rapoport, and Zeev Maoz. 1988. “Tacit cooperation in three al-
ternative noncoopera tive voting games: a new model of sophisticated behavior under
the plurality procedure.” Electoral Studies 7:143–61.
Ferejohn, John A. and Morris P. Fiorina. 1975. “Purposive Models of Legislative Behav-
ior.” American Economic Review 65:407–414.
259
Fernandez, Raquel and Dani Rodrik. 1991. “Resistance to Reform: Status Quo Bias in the
Presence of Individual-Specific Uncertainty.” American Economic Review 81:1146–1155.
Ferris, Gerald R. and Wayne A. Hochwarter. 2011. “Organizational politics.” In Main-

taining, expanding, and contracting the organization, edited by Sheldon Zedeck, volume 3
of APA Handbooks in Psychology; APA handbook of industrial and organizational psychology,
pp. 435–459. Washington, DC, US: American Psychological Association.
Fey, Mark. 1997. “Stability and Coordination in Duverger’s Law: A Formal Model of
Preelection Polls and Strategic Voting.” The American Political Science Review 91:135–
147.
Fey, M. 2004. “May’s Theorem with an infinite population.” Social Choice and Welfare
23:275–293.
Fiedler, Marina and Ernan Haruvy. 2009. “The Lab versus the Virtual Lab and Virtual
Field - An Experimental Investigation of Trust Games with Communication.” Journal
of Economic Behavior and Organization 72:716–724.
Finke, D. 2009. “Estimating the Effect of Nonseparable Preferences in Eu Treaty Negotia-

tions.” Journal of Theoretical Politics 21:543–569.
Finke, Daniel and Andreas Fleig. 2013. “The Merits of Adding Complexity: Non-
Separable Preferences in Spatial Models of EU Politics.” Journal of Theoretical Politics
25:547 – 576.
Fiorina, Morris P. 1992a. Divided Government. Allyn and Bacon, Boston.
Fiorina, Morris P. 1992b. “An Era of Divided Government.” Political Science Quarterly
107:387–410.
Fiorina, Morris P. and Charles R. Plott. 1978. “Committee Decisions under Majority Rule:
An Experimental Study.” The American Political Science Review 72:575–598.
Fiorino, Nadia and Roberto Ricciuti. 2007. “Determinants of Direct Democracy.” Interna-
tional Centre for Economic Research (ICER) Working Papers 23 - 2007.
Fischbacher, Urs. 2007. “z-Tree: Zurich toolbox for ready-made economic experiments.”
Experimental Economics 10:171–178.
Fischbacher, U., A. Falk, and E. Fehr. 2008. “Testing theories of fairness - Intentions mat-
ter.” Games and Economic Behavior 62:287–303.
Fischbacher, U., S. Gachter, and E. Fehr. 2001. “Are people conditionally cooperative?
Evidence from a public goods experiment.” Economics Letters 71:397–404.
Fishburn, P. 1977. “Condorcet Social Choice Functions.” SIAM Journal of Applied Mathe-
matics 33:469–489.
Fishburn, Peter C. 1973. The theory of social choice. Princeton, N.J.: Princeton University
Press.
260
Fleig, Andreas and Daniel Finke. 2013. “Delegation, Uncertainty and Social Preferences
in Majority Decisions?” mimeo.
Forsa. 2011. “Würden Sie für die Griechen zahlen?” Statista

http://de.statista.com/statistik/daten/studie/201617/umfrage/zahlungsbereitschaft-
der-deutschen-fuer-griechenland (accessed August 3, 2012).
Forsythe, R., T. Rietz, R. Myerson, and Weber R. 1996. “An experimental study of vot-
ing rules and polls in three-candidate elections.” International Journal of Game Theory
25:355–383.
Frank, Bjoern. 1998. “Good news for experimenters: subjects do not care about your
welfare.” Economics Letters 61:171–174.
Frank, Robert, Thomas Gilovich, and Dennis Regan. 1993. “Does studying economics
inhibit cooperation?” Journal of Economic Perspectives 7:159–171.
Frank, Robert, Thomas Gilovich, and Dennis Regan. 1996. “Do economists make bad
citizens?” Journal of Economic Perspectives 10:187–192.
Franz, Stephan. 2004. “Grundlagen des oekonomischen Ansatzes: das Erk-

laerungskonzept des Homo Oeconomicus.” International Economics, Working Paper Uni
Potsdam 02.
Frechette, Guillaume R. 2011. “Laboratory Experiments: Professionals Versus Students.”

mimeo. New York University, working paper. http://ssrn.com/abstract=1939219 (accessed
March 5, 2012).
Frechette, Guillaume R. 2012. “Session-effects in the laboratory.” Experimental Economics

15:485–498.
Frei, Brent. 2009. “Paid Crowdsourcing: Current State & Progress

toward Main-stream Business Use.” mimeo. Smartsheet.com.
http://www.smartsheet.com/files/haymaker/Paid%20Crowdsourcing%20Sept%20
2009%20-%20Release%20Version%20-%20Smartsheet.pdf (accessed July 19, 2012).
Frevert, Ute. 2001. Die kasernierte Nation. Militaerdienst und Zivilgesellschaft in Deutschland.
Verlag C. H. Beck, Muenchen.
Frey, Bruno S. and Stephan Meier. 2005. “Selfish and Indoctrinated Economists?” Euro-
pean Journal of Law and Economics 19:165–171.
Friedman, Thomas. 2007. “The World is Flat 3.0.” Speech at MIT, December 2nd.
http://www.academicearth.org/lectures/the-world-is-flat (accessed July 17, 2012).
Frohlich, Norman and Joe A. Oppenheimer. 1996. “Experiencing Impartiality to Invoke

Fairness in the n-PD: Some Experimental Results.” Public Choice 86:117–135.
Frohlich, Norman and Joe A. Oppenheimer. 2007. “Justice preferences and the Arrow
problem.” Journal of Theoretical Politics 19:363–390.
261
Gailmard, S. 2009. “Discretion Rather than Rules: Choice of Instruments to Control Bu-
reaucratic Policy Making.” Political Analysis 17:25–44.
Ganghof, Steffen and Philip Manow. 2005. Mechanismen der Politik: Strategische Interak-
tion im deutschen Regierungssystem, volume 54 of Schriften des Max-Planck-Instituts fuer
Gesellschaftsforschung. Frankfurt a.M.: Campus.
Gautier, Axel and Dimitri Paolini. 2007. “Delegation and information revelation.” Journal
of Institutional and Theoretical Economics 163:574–597.
Geiger, David, Michael Rosemann, Erwin Fielt, and Martin Schader. 2012.
“Crowdsourcing Information Systems - Definition Typology, and De-
sign.” mimeo. Proceedings of the International Conference on Information Sys-
tems. https://ub-madoc.bib.uni-mannheim.de/32631/1/Geiger_et_al._-_2012_-
_Crowdsourcing_Information_Systems_-_Definition,_Typology,_and_Design.pdf
(accessed December 1, 2012).
Geisendorf, Sylvie. 2009. “The irrational foundations of neoclassical economics. How a
simplified Newtonism shaped the perception of the economy.” In Der neoliberale Markt-
Diskurs: Ursprünge, Geschichte, Wirkungen, edited by Walter Otto Ötsch and Claus
Thomasberger, pp. 151–166. Metropolis.
Gerber, Alan S., Dean Karlan, and Daniel Bergan. 2009. “Does the Media Matter? A
Field Experiment Measuring the Effect of Newspapers on Voting Behavior and Political
Opinions.” American Economic Journal 1:2:35–52.
Gerring, John. 2005. “Causation - a unified framework for the social sciences.” Journal of
Gibbons, J.D. 1976. Nonparametric Methods for Quantitative Analysis. Holt, Rhinehart, and
Winston.
Gigerenzer, G. 2008. Rationality for mortals: How people cope with uncertainty. Oxford.
Gigerenzer, G. and W. Gaissmaier. 2011. “Heuristic decision making.” Annual Review of
Psychology 62:451–482.
Gigerenzer, Gerd and Daniel G. Goldstein. 1999. “Betting on One Good Reason: The Take
The Best Heuristic.” In Simple heuristics that make us smart, edited by Research Group
ABC, chapter 4. Oxford University Press, New York.
Gigerenzer, G. and R. Selten. 2001. Bounded rationality: The adaptive toolbox. Cambridge,
Mass.
Gilboa, Itzhak. 2010. Rational Choice, volume 1. The MIT Press, 1 edition.
Gill, Jeff. 2006. Essential mathematics for political and social research. Cambridge; New York:
Gillespie, Richard. 1991. Manufacturing knowledge : a history of the Hawthorne experiments.
Cambridge England ; New York: Cambridge University Press.
262
Gillies, D. B. 1959. “Solutions to general non-zero-sum games.” In Contributions to the
Theory of Games, edited by A. W. Tucker and R. D. Luce, volume IV, pp. 47–85. Princeton:
Gilligan, Thomas W. and Keith. Krehbiel. 1987. “Collective Decision-Making and Stand-
ing Committees: An Informational Rational for Restrictive Amendment Procedures.”
Journal of Law, Economics, and Organization 3:287–334.
Gilligan, T. W. and K. Krehbiel. 1990. “Organization of Informative Committees by a

Rational Legislature.” American Journal of Political Science 34:531–564.
Gini, C. 1912. Variabilità e mutabilità. Bologna, Italy.
Gladwell, Malcolm. 2007. Blink: The Power of Thinking Without Thinking. Back Bay Books.
Glasgow, Garrett, Matt Golder, and Sona N. Golder. 2012. “New Empirical Strategies for
the Study of Parliamentary Government Formation.” Political Analysis 20:248–270.
Gneezy, U. and J. Potters. 1997. “An experiment on risk taking and evaluation periods.”
The Quarterly Journal of Economics 112:631–645.
Goeree, J. K. and C. A. Holt. 2001. “Ten little treasures of game theory and ten intuitive
contradictions.” American Economic Review 91:1402–1422.
Goeree, Jacob K., Charles A. Holt, and Susan K. Laury. 2002. “Private costs and pub-
lic benefits: unraveling the effects of altruism and noisy behavior.” Journal of Public
Economics 83:255–276.
Goeree, Jacob K., Charles A. Holt, and Thomas R. Palfrey. 2005. “Regular Quantal Re-
sponse Equilibrium.” Experimental Economics 8:347–367.
Goerg, Sebastian J., Werner Güth, Gari Walkowitz, and Torsten Weiland. 2007. “Interre-
gional diversity of fairness concerns - An online ultimatum experiment.” Jena Economic
Research Papers, Max-Planck-Institute of Economics 016.
Goette, Lorenz, David Huffman, and Stephan Meier. 2006. “The Impact of Group Mem-
bership on Cooperation and Norm Enforcement: Evidence Using Random Assignment
to Real Social Groups.” American Economic Review 96:212–216.
Goode, W. 1971. “Force and Violence in the Family.” Journal of Marriage and the Family
33:624–636.
Gorman, W. M. 1968. “The Structure of Utility Functions.” The Review of Economic Studies
35:367–390.
Grant, robert M. 1996. “Toward a knowledge-based theory of the firm.” Strategic Man-
agement Journal 17:109–122.
Gray, Jeremy. 2011. “The vagueness of Wigner’s analysis.” Interdisciplinary Science Reviews
36:214–228.
263
Green, Donald P. and Ian Shapiro. 1994. Pathologies of Rational Choice Theory. A Critique of
Applications in Political Science. New Haven-London: Yale University Press.
Greene, W.H. 2003. Econometric analysis. New Jersey: Prentice Hall.
Greiner, Ben. 2004. “An Online Recruitment System for Economic Experiments.” In
Forschung und wissenschaftliches Rechnen, edited by Kurt Kremer and Volker Macho,
volume 63 of GWDG Bericht, pp. 79–93. Goettingen: Gesellschaft fuer wissenschaftliche
Datenverarbeitung.
Grendstad, Gunnar and Per Selle. 1995. “Cultural Theory and the New Institutionalism.”
Journal of Theoretical Politics 7:5–27.
Grether, D. M. and C. R. Plott. 1979. “Economic Theory of Choice and the Preference
Reversal Phenomenon.” American Economic Review 69:623–638.
Gräfe, A. and J. S. Armstrong. 2010. “Predicting Elections from the Most Important Issue:
A Test of the Take-the-best Heuristic.” Journal of Behavioral Decision Making 25:41–48.
Grosslight, Lorraine, Christopher Unger, Eileen Jay, and Carol L. Smith. 1991. “Under-
standing models and their use in science: Conceptions of middle and high school stu-
dents and experts.” Journal of Research in Science Teaching, Special Issue: Students’ Models
and Epistemologies 28:799–822.
Güth, Werner, M. Vittoria Levati, Matthias Sutter, and Eline van der Heijden. 2004. “Lead-
ership and cooperation in public goods experiments.” Papers on Strategic Interaction,
Max Planck Institute of Economics 29.
Güth, Werner, Matteo Ploner, and Vittoria Levati. 2005. “The Effect of Group Identity in
an Investment Game.” Papers on Strategic Interaction. Max Planck Institute of Economics
06.
Güth, W., R. Schmittberger, and B. Schwarze. 1982. “An Experimental Analysis of Ulti-
matum Bargaining.” Journalof Economic Behavior and Organization 3:367–388.
Guala, F. 2005. The Methodology of Experimental Economics. Cambridge: Cambridge Uni-

versity Press.
Guarnaschelli, Serena, Richard D. McKelvey, and Thomas R. Palfrey. 2000. “An Experi-
mental Study of Jury Decision Rules.” The American Political Science Review 94:407–423.
Gujarati, Damodar and Dawn Porter. 2008. Basic Econometrics. Mcgraw-Hill Higher Edu-
cation, 4th edition.
Gul, Faruk. 1997. “A Nobel Prize for Game Theorists: The Contributions of Harsanyi,
Nash and Selten.” Journal of Economic Perspectives, American Economic Association
11:159–174.
Gutmann, Amy and Dennis Thompson. 2004. Why Deliberative Democracy? Oxford:
264
Guttman, L. 1945. “A basis for analyzing test-retest reliability.” Psychometrika 10:255–282.
Haege, Frank M. 2007. “Committee Decision-making in the Council of the European

Union.” European Union Politics 8:299–328.
Hagemann, S. 2007. “Applying ideal point estimation methods to the council of minis-
ters.” European Union Politics 8:279–296.
Hagemann, Sara and Julia De Clerck-Sachsse. 2007. “Decision-Making in the Enlarged

Council of Ministers: Evaluating the Facts.” CEPS - Policy brief 119.
Haile, Philip A., Ali Hortacsu, and Grigory Kosenok. 2008. “On the Empirical Content of
Quantal Response Equilibrium.” The American Economic Review 98:180–200.
Halevy, A., P. Norvig, and F. Pereira. 2009. “The Unreasonable Effectiveness of Data.”
IEEE Intelligent Systems 24:8–12.
Halfpenny, Peter and Michael Taylor. 1973. “An Experimental Study of Individual and
Collective Decision-Making.” British Journal of Political Science 3:425–444.
Hall, Peter A. and Rosemary C. R. Taylor. 1996. “Political Science and the Three New
Institutionalisms.” Political Studies XLIV:936–957.
Ham, J. C., J. H. Kagel, and S. F. Lehrer. 2005. “Randomization, endogeneity and lab-
oratory experiments: the role of cash balances in private value auctions.” Journal of
Econometrics 125:175–205.
Hamenstädt, Ulrich. 2012. “Das "Lab in the field" Experiment. Kontrolle und die Inte-
gration finanzieller Anreize in feldexperimentellen Forschungsdesigns anhand eines
Beispiels.” In Jahrbuch der Handlungs- und Entscheidungstheorie, edited by Thomas
Bräuninger, Andre Bächtiger, and Susumu Shikano, number Band 7, chapter 5, pp.
105–124. VS Verlag für Sozialwissenschaften, Wiesbaden.
Hamming, R. W. 1980. “The Unreasonable Effectiveness of Mathematics.” The American

Mathematical Monthly 87:81–90.
Hammon, L. and H. Hippner. 2012. “Crowdsourcing.” Business and Information Systems

Engineering 4.
Hammond, Thomas H. and Gary J. Miller. 1987. “The Core of the Constitution.” The
American Political Science Review 81:1155–1174.
Hansen, John Mark. 1998. “Individuals, Institutions, and Pub-lic Preferences over Public
Finance.” American Political Sci-ence Review 92:513–532.
Harrison, Glenn and John List. 2004. “Field Experiments.” Journal of Economic Literature
42:1009–1055.
Harrison, Glenn W. 2008. “Maximum Likelihood Estimation of Utility Functions Using

Stata.” Working Paper, Department of Economics, University of Central Florida 06-12.
265
Harrison, Glenn W. and E. Elisabet Rutstroem. 2008. “Risk Aversion in the Laboratory.” In
Risk Aversion in Experiments, edited by Glenn W. Harrison and James C. Cox, volume 12
of Research in Experimental Economics, chapter 3, pp. 41–196. Emerald Group Publishing.
Hastie, Reid and Robyn M. Dawes. 2010. Rational choice in an uncertain world : the psychol-
ogy of judgment and decision making, volume 2. Los Angeles: SAGE.
Haughton, Jonathan and Shahidur R. Khandker. 2009. Handbook on Poverty and Inequality,
by. Washington D.C.: The world Bank.
Hayes-Renshaw, Fiona, Wim van Aken, and Helen Wallace. 2006. “When and Why the
EU Council of Ministers Votes Explicitly.” Journal of Common Market Studies 44:1468–
5965.
Hayes-Renshaw, Fiona and Hellen Wallace. 2006. The Council of Ministers. London:
Macmillan, 2nd edition.
Heilbronner, Robert. 2000. The Worldly Philosophers: The Lives, Times and Ideas of the Great
Economic Thinkers. Penguin Books, 7 edition.
Heinemann, Frank. 2005. “Measuring Risk Aversion and the Wealth Effect.” In Risk
Aversion in Experiments - Research in Experimental Economics, edited by James C. Cox
and Glenn W. Harrison, number 12, pp. 293–313. Bingley, UK: Emerald.
Heinemann, Frank, Rosemarie Nagel, and Peter Ockenfels. 2009. “Measuring Strategic
Uncertainty in Coordination Games.” Review of Economic Studies 76:181–221.
Heinrich, Gert. 2006. Grundlagen der Mathematik, der Statistik und des Operations Research
für Wirtschaftswissenschaftler [Taschenbuch]. Oldenbourg Wissenschaftsverlag, 3. auflage
edition.
Hellman, Ziv. 2008. “Bargaining Set Solution Concepts in Dynamic Cooperative Games.”
MPRA Paper, University Library of Munich 8798.
Helson, H. 1964. Adaptation level theory: An experimental and systematic approach to behavior.
Harper and Row.
Hennig-Schmidt, Heike. 1997. “Break-offs in Bargaining, Evidence from a Video Experi-

ment.” Discussion Paper Serie, University of Bonn B - 402.
Henrich, Joseph, R. Boyd, S. Bowles, C. Camerer, E. Fehr, H. Gintis, and R. McElreath.

2001. “In Search of Homo Economicus: Behavioral Experiments in 15 Small-Scale So-
cieties.” American Economic Review 91:73–78.
Henrich, J., S. J. Heine, and A. Norenzayan. 2010. “The weirdest people in the world?”
Behavioral and Brain Sciences 33:61–135.
Herne, Kaisa, Olli Lappalainen, and Elina Kestilae-Kekkonen. 2012. “Ex-

perimental comparison of direct, general, and indirect reciprocity.”
266
mimeo. University of Turku. http://www.soc.utu.fi/laitokset/valtio-
oppi/oppiaine/henkilokunta/herne_lappalainen_kestila.pdf (accessed January
22, 2013).
Herrmann, Benedikt and Henrik Orzen. 2008. “The appearance of homo rivalis: Social
preferences and the nature of rent seeking.” Discussion Papers, The Centre for Decision
Research and Experimental Economics, University of Nottingham. 10.
Herrmann, Michael. 2012. “Voter uncertainty and failure of Duverger’s law: an empirical
analysis.” Public Choice 151:63–90.
Herzberg, Roberta Q. and Rick K. Wilson. 1988. “Results on Sophisticated Voting in an

Experimental Setting.” The Journal of Politics 50(2):471–486.
Heukelom, Floris. 2007. “Kahneman and Tversky and the Origin of Behavioral Eco-
nomics.” Tinbergen Institute Discussion Papers 07-003/1:25.
Höffe, Otfried. 2005. “Zoon politikon.” In Aristoteles-Lexikon. Kroener-Verlag, Stuttgart.
Hill, J. L., J. P. Reiter, and E. L. Zanutto. 2005. “A Comparison of Experimental and

Observational Data Analyses.” In Applied Bayesian Modeling and Causal Inference from
Incomplete-Data Perspectives: An Essential Journey with Donald Rubin’s Statistical Family,
edited by A Gelman and X.-L. Meng, chapter 5. John Wiley & Sons, Ltd, Chichester,
UK.
Hinich, Melvin J. 2008. “The Future of Analytical Politics.” In The Oxford Handbook of
Political Economy, edited by Barry R. Weingast and Donald Wittman, chapter 56. Oxford
Handbooks Online.
Hinich, Melvin J. and Michael C. Munger. 1997. Analytical politics. Cambridge England ;
New York, NY, USA: Cambridge University Press.
Hirschman, Albert O. 1964. “The Paternity of an Index.” The American Economic Review
54:761–762.
Hodge, J. K. 2006. “Permutations of separable preference orders.” Discrete Applied Math-

ematics 154:1478–1499.
Hodge, J. K. and P. Schwallier. 2006. “How does separability affect the desirability of
referendum election outcomes?” Theory and Decision 61:251–276.
Hoechtl, Wolfgang, Rupert Sausgruber, and Jean-Robert Tyran. 2012. “Inequality aver-
sion and voting on redistribution.” European Economic Review 56:1406–1421.
Hofstadter, Douglas. 1985. Metamagical Themas: Questing for the Essence of Mind and Pat-
tern. New York, Basic Books.
Holm, H. J. and P. Nystedt. 2010. “Collective Trust Behavior.” Scandinavian Journal of

267
Holm, Hakan J., Sonja Opper, and Victor Nee. 2013. “Entrepreneurs Under Uncertainty:
An Economic Experiment in China.” Management Science 01/2013 published online
before print January 8, 2013.
Holt, C.A. and L. R. Anderson. 1999. “Agendas and strategic voting.” Southern Economic
Journal 65:622–629.
Holt, Charles and Susan K. Laury. 2002. “Risk Aversion and Incentive Effects.” American
Economic Review 92:1644–1655.
Holyoak, K.J. and D. Simon. 1999. “Bidirectional reasoning in decision making by con-
straint satisfaction.” Journal of Experimental Psychology 128:3–31.
Holzinger, Katharina. 2009. “Vom ungeliebten Störenfried zum akzeptierten Paradigma?

Zum Stand der (Neuen) Politischen Ökonomie in Deutschland.” Politische Viertel-
jahresschrift (PVS) 50:539–576.
Horton, J. J., D. G. Rand, and R. J. Zeckhauser. 2011. “The online laboratory: conducting
experiments in a real labor market.” Experimental Economics 14:399–425.
Hotelling, Harold. 1929. “Stability in Competition.” The Economic Journal 39:41–57.
Häring, Norbert and Olaf Storbeck. 2007. Ökonomie 2.0: 99 überraschende Erkenntnisse.
Schäffer-Poeschel.
Hörl, Björn and Andreas Warntjen. 2005. “Built on quicksand? a decade of procedural
spatial models on eu legislative decision-making.” Journal of European Public Policy
12:592–606.
Hug, Simon. 2009. “Models of Multilateral Negotiations and Ratification.” mimeo.

Paper prepared for presentation at the ISA conference March 1-5, 2005, Hawaii.
http://www.unige.ch/ses/spo/static/simonhug/momn.html (accessed February 26,
2012).
Huitt, Ralph K. 1954. “The Congressional Committee: A Case Study.” The American
Political Science Review 48:340–365.
Humphreys, Marcatan and Michael Laver. 2010. “Spatial Models, Cognitive Metrics, and
Majority Rule Equilibria.” British Journal of Political Science 40:11–30.
Humphreys, Paul. 2004. Extending Ourselves: Computational Science, Empiricism, and Sci-
entific Method. Oxford University Press.
Insko, Chester A., John Schopler, and Rick H. Hoyle. 1990. “Individual-Group Discon-
tinuity as a Function of Fear and Greed.” Journal of Personality and Social Psychology
58:68–79.
Iriberri, Nagore and Pedro Rey-Biel. 2011. “The role of role uncertainty in modified dic-
tator games.” Experimental Economics 14:160–180.
268
Isaac, R.M. and J.M. Walker. 1988a. “Communication and free-riding behavior: the vol-
untary contribution mechanism.” Economic Inquiry 26:585–608.
Isaac, R.M. and J.M. Walker. 1988b. “Group size effects in public goods provision: the
voluntary contributions mechanism.” Quarterly Journal of Economics 103:179–199.
Isaac, R. Mark, James M. Walker, and Arlington W. Williams. 1994. “Group size and the
voluntary provision of public goods : Experimental evidence utilizing large groups.”
Journal of Public Economics 54:1–36.
Jacquemet, Nicolas and Adam Zylbersztejn. 2011. “What drives failure to maximize pay-
offs in the lab ? A test of the inequality aversion hypothesis.” Technical report, Doc-
uments de travail du Centre d’Economie de la Sorbonne 11036, Université Panthéon-
Sorbonne (Paris 1), Centre d’Economie de la Sorbonne.
Jeffrey, Milyo, David M. Primo, and Matthew L. Jacobsmeier. 2006. “Estimating the Im-
pact of State Policies and Institutions with Mixed-Level Data.” Working Papers, Depart-
ment of Economics, University of Missouri 0603.
Jervis, R. 1997. System Effects: Complexity in Political and Social Life. Princeton: Princeton
University Press.
Johnson, D. B. 1991. Public choice: an introduction to the new political economy. . Mountain
View: Bristlecone Books.
Johnson, E. J., C. Camerer, S. Sen, and T. Rymon. 2002. “Detecting failures of backward in-
duction: Monitoring information search in sequential bargaining.” Journal of Economic
Theory 104:16–47.
Johnson, George. 2008. The ten most beautiful experiments. New York: Alfred A. Knopf.
Johnson, Robert. 2012. “Hedgefonds-Manager: "Ich konnte es nicht ertragen".” Die ZEIT,
Interview 16.
Jones, Mark, Wonjae Hwang, and Juan Micozzi. 2009. “Government and Opposition in
the Argentine Congress, 1989-2007: Understanding Inter-Party Dynamics through Roll
Call Vote Analysis.” Journal of Politics in Latin America, German Institute of Global and
Area Studies, Hamburg 1:67–96.
Juettner, Julia, Benjamin Schulz, and Oliver Trenkamp. 2011. “Bundeswehr-Streichliste -

Angst vor der Geisterstadt.” Spiegel - Magazin 26.10.
Junge, Dirk. 2010. “Game Theoretic Models and the Empirical Analysis of EU Policy
Making: Strategic Interaction, Collective Decisions, and Statistical Inference.” In Re-
form Processes and Policy Change: Veto Players and Decision-Making in Modern Democracies,
edited by Thomas König, George Tsebelis, and Marc Debus, pp. 165–186. New York.
Kagel, J.H., H. Sung, and E. Winter. 2010. “Veto power in committees: an experimental
study.” Experimental Economics 13:167–188.
269
Kagel, John H. and Alvin E. Roth. 1995. The handbook of experimental economics. Princeton,
NJ: Princeton Univ. Press.
Kahneman, Daniel. 2011. Thinking, fast and slow. New York: Farrar, Straus and Giroux.
Kahneman, Daniel, Paul Slovic, and Amos Tversky. 1982. Judgment under uncertainty :
heuristics and biases. Cambridge: Cambridge University Press.
Kahneman, Daniel and Amos Tversky. 1979. “Prospect Theory: An Analysis of Decision
under Risk.” Econometrica 47:263–292.
Kaiser, Andre. 2007a. “James M. Buchanan, J. Gordon Tullock, The Calculus of Consent.
Logical Foundations of Constitutional Democracy, Ann Arbor 1962.” In Schlüsselwerke
der Politikwissenschaft, edited by Steffen Kailitz, pp. 56–60. VS Verlag für Sozialwis-
senschaften, Wiesbaden.
Kaiser, A. 2007b. “Neuere Beitraege zur empirischen Institutionenanalyse: Von der

Gegenstands-zur Problemorientierung.” Politische Vierteljahresschrift (PVS) 48:119–135.
Kaiser, Andre, Jan Biela, and Annika Hennl. 2013. Policy Making in Multilevel Systems.
ECPR Press.
Kaufmann, Stephan. 2012. “Appell an die Wirtschaftslehre: Lehre ohne akademische

Prostitution.” Frankfurter Rundschau printed edition April 4th.
Keeney, R.L. and H. Raiffa. 1976. Decisions with multiple objectives: Preferences and value
tradeoffs. New York: Cambridge University Press.
Kehoe, C. and J. Pitkow. 1996. “Surveying the territory.” The World Wide Web Journal
1:77–84.
Keiding, H. and B. Peleg. 2001. “Stable voting procedures for committees in economic
environments.” Journal of Mathematical Economics 36:117–140.
Kenning, P. and H. Plassmann. 2005. “NeuroEconomics: An overview from an economic

perspective.” Brain Research Bulletin 67:343–354.
Kesten, Yonit and Amir Pnueli. 1998. “Modularization and Abstraction: The Keys to
Practical Formal Verification.” Mathematical Foundations of Computer Science pp. 54–71.
Keynes, John Maynard. 1936. The General Theory of Employmen. Palgrave Macmillan.
Kinder, Donald R. and Thomas R. Palfrey. 1991. “An experimental political science? Yes,
an experimental political science.” The Political Methodologist 4:2–8.
King, Gary and Daniel Hopkins. 2007. Extracting Systematic Social Science Meaning from
Text. Weatherhead Center for International Affairs, Harvard University.
King, Gary, Robert O. Keohane, and Sidney Verba. 1994. Designing Social Inquiry: Scientific
Inference in Qualitative Research. Princeton University Press.
270
King, Gary and Margaret Roberts. 2012. “How Robust Standard Errors Expose
Methodological Problems They Do Not Fix.” mimeo. Harvard University. gk-
ing.harvard.edu/files/robust.pdf (accessed March 3, 2013).
Kirchgässner, Gebhard. 2008. Homo oeconomicus: Das okonomische Modell individuellen Ver-
haltens und seine Anwendung in den Wirtschafts- und Sozialwissenschaften, volume 3. Tue-
bingen: Mohr Siebeck Gmbh & Co. K.
Kirchsteiger, Georg. 1994. “The role of envy in ultimatum games.” Journal of Economic
Behavior and Organization 25:373–389.
Kirsch, Guy. 2004. Neue Politische Oekonomie, volume 5. Stuttgart: Lucius und Lucius.
Kirstein, Roland and Dieter Schmidtchen. 2003. “Self-interest, Social Wealth, and Com-
petition as a Discovery Procedure - A classroom experiment that makes the "invisible
hand" visible.” Discussion Paper, Center for the Study of Law and Economics 08.
Klein, Jonathan H. 1985. “The Abstraction of Reality for Games and Simulations.” The
Journal of the Operational Research Society 36:671–678.
Kleinbaum, David G., Lawrence L. Kupper, Azhar Nizam, and Keith E. Muller. 2007.
Applied Regression Analysis and Other Multivariable Methods. Duxbury Press, 4th edition.
Knetsch, Jack L. and J. A. Sinden. 1984. “Willingness to Pay and Compensation De-
manded: Experimental Evidence of an Unexpected Disparity in Measures of Value.”
König, Thomas. 2013. “Negative Preference Interdependence, Continuous Con-
cessions and the Division of Reform Gains. Making the Lisbon Treaty Work.”
mimeo. 6th Annual Conference on The Political Economy of International Organizations.
http://147.142.190.246/joomla/peio/files2013/papers/König%2030.09.2012.pdf (ac-
cessed March 23, 2013).
König, Thomas and Dirk Junge. 2009. “Why Don’t Veto Players Use Their Power?” Eu-
ropean Union Politics 10:507–534.
König, Thomas, Bernd Luig, and Stephan Marc Solomon. 2010. “Sachverstaendige und
der Einfluss von Expertise auf Reformen: Eine raeumliche Analyse der Foederalismus-
reform II.” Perspektiven der Wirtschaftspolitik 11:307–323.
Knight, Frank H. 1921. Risk, Uncertainty and Profit. Houghton Mifflin.
Kocher, Martin G. and Matthias Sutter. 2005. “The Decision Maker Matters: Individual
Versus Group Behaviour in Experimental Beauty-Contest Games.” Economic Journal,
Royal Economic Society 115:200–223.
Komorita, S. S. and C. D. Parks. 1995. “Interpersonal-Relations - Mixed-Motive Interac-
tion.” Annual Review of Psychology 46:183–207.
Koremenos, Barbara. 2008. “When, What, and Why do States Choose to Delegate?” Law
and Contemporary Problems 71:151–192.
271
Kormendi, Roger C. and Charles R. Plott. 1982. “Committee Decisions under Alternative
Procedural Rules - An Experimental Study Applying a New Nonmonetary Method of
Preference Inducement.” Journal of Economic Behaviour and Organisation 3:20.
Koukoumelis, A., M.V. Levati, and J. Weisser. 2012. “Leading by words: A voluntary
contribution experiment with one-way communication.” Journal of Economic Behavior
and Organization 81:379–390.
Kramer, Gerald H. 1972. “Sophisticated voting over multidimensional choice spaces.”

The Journal of Mathematical Sociology 2:165–180.
Kranich, L., A. Perea, and H. Peters. 2005. “Core Concepts for Dynamic TU Games.”
International Game Theory Review 7:43–61.
Krantz, J., J. Ballard, and J. Scher. 1997. “Comparing the results of laboratory andworld-
wide web samples on the determinants of female attractiveness.” Behavior Research
Methods, Instruments, and Computers 29:264–269.
Krehbiel, Keith. 1986. “Sophisticated and Myopic Behavior in Legislative Committees:

An Experimental Study.” American Journal of Political Science 30(3):542–561.
Krehbiel, Keith. 1988. “Spatial Models of Legislative Choice.” Legislative Studies Quarterly
13:259–319.
Krehbiel, Keith, Adam Meirowitz, and Jonathan Woon. 2004. “Testing Theories of Law-
making.” Research Papers, Stanford University 1860.
Kriss, Peter H. and Roberto A. Weber. 2012. “Economic Laboratory Experiments on Orga-
nizational Formation and Change.” In Handbook of Economic Organization: Intergrating
Economic and Organization Theory, edited by Anna Grandori. Edward Elgar.
Kröning, Daniel and Ofer Strichman. 2008. Decision procedures : an algorithmic point of
view. Berlin: Springer.
Krugman, Paul. 2010. “Nobelpreisträger Krugman will alte VWL-Weisheiten "beerdi-

gen".” In Interview mit dem Handelsblatt, Ausgabe vom 11.01.
Kugler, Tamar, Gary Bornstein, Matthias Sutter, and Martin G. Kocher. 2007. “Trust be-
tween individuals and groups: Groups are less rusting than individuals but just as
trustworthy.” Journal of Economic Psychology 28:646–657.
Kuklinski, James H., Paul M. Sniderman, Kathleen Knight, Thomas Piazza, Philip E. Tet-
lock, Gordon R. Lawrence, and Barbara Mellers. 1997. “Racial Prejudice and Attitudes
Toward Affirmative Action.” American Journal of Political Science 41:402–419.
Kurzban, R. and D. Houser. 2005. “Experiments investigating cooperative types in hu-

mans: a complement to evolutionary theory and simulations.” Proceedings of the Na-
tional Academy of Sciences of the United States of America 102:1803–1807.
272
Kuziemko, Ilyana, Ryan W. Buell, Taly Reich, and Michael I. Norton. 2011. “"Last-place
Aversion": Evidence and Redistributive Implications.” NBER Working Paper Series
17234. National Bureau of Economic Research, Inc.
Laband, David and Richard Biel. 1999. “Are Economists More Selfish than Other ’Social’
Scientists?” Public Choice 100:1–2, 85–101.
Lacy, Dean. 2001a. “Nonseparable Preferences, Measurement Error, and Unstable Survey
Responses.” Political Analysis 9:1–21.
Lacy, Dean. 2001b. “A theory of nonseparable preferences in survey responses.” American

Journal of Political Science 45:239–258.
Lacy, Dean. 2003. “Dependent preferences and over-time instability in survey responses.”
Presented at the annual meeting of the American political science association .
Lacy, D. and Niou. 1998a. “Nonseparable Preferences, Issue Linkage,

and Economic Sanctions.” mimeo. Presented at the Annual Meeting of
the American Political Science Association, Boston, MA, September 3-6. cite-
seerx.ist.psu.edu/viewdoc/download?doi=10.1.1.8.8872&rep=rep1&type=pdf (ac-
cessed December 18, 2012).
Lacy, Dean and Emerson M. S. Niou. 1994. “Preferences and a Problem with Referen-
dums.” mimeo. Presented at the Annual Meeting of the Public Choice Society and Economic
Science Association, April 8-10, Austin, Texas.
Lacy, D. and E. M. S. Niou. 1998b. “Elections in double-member districts with nonsepa-

rable voter preferences.” Journal of Theoretical Politics 10:89–110.
Lacy, Dean and Emerson M. S. Niou. 2000. “A problem with referendums.” Journal of
Lai, Ernest K. and Wooyoung Lim. 2012. “Authority and communication in the labora-
tory.” Games and Economic Behavior 74:541–560.
Lalive, Rafael and Armin Schmutzler. 2008. “Exploring the Effects of Competition for
Railway Markets.” International Journal of Industrial Organisation 26:443–458.
Lane, Jan-Erik. 2000. “The New Institutionalism.” In The Public Sector: Concepts, Models
and Approaches, chapter 10. SAGE Publications, London.
Lane, Jan-Erik and Svante Ersson. 2000. The New Institutional Politics: Performance and
Outcomes. London: Routledge.
Lang, Jerome and Lirong Xia. 2009. “Sequential composition of voting rules in multi-issue
domains.” Mathematical Social Sciences 57:304–324.
Larsson, Rikard. 1993. “The Handshake between Invisible and Visible Hands: Toward a
Tripolar Institutional Framework.” International Studies of Management and Organization
23:87–106.
273
Lask, Reinhard. 2010. “Anbau-Gegner sehen die "Stadthalle im Chaos" versinken.” Rgein-
Neckar-Zeitung 15.06.
Lave, Charles and James G. March. 1975. An Introduction to Models in the Social. New
York: Harper and Row.
Laver, M., K. Benoit, and J. Garry. 2003. “Extracting policy positions from political texts
using words as data.” American Political Science Review 97:311–331.
Laver, M. and M. Schilperoord. 2007. “Spatial models of political competition with en-
dogenous political parties.” Philosophical Transactions of the Royal Society B-Biological
Sciences 362:1711–1721.
Laver, Michael and Norman Schofield. 1998. Multiparty government : the politics of coalition
in Europe. Ann Arbor: University of Michigan Press.
Le Breton, M. and A. Sen. 1999. “Separable preferences, strategyproofness, and decom-

posability.” Econometrica 67:605–628.
Leblebici, Huseyin and Gerald R. Salancik. 1981. “Effects of Environmental Uncertainty

on Information and Decision Processes in Banks.” Administrative Science Quarterly
26:578–596.
Ledyard, J. 1995. “Public goods: A survey of experimental research.” In The Handbook of

Experimental Economics, edited by J. H. Kagel and A.E. Roth, pp. 111–194. NJ: Princeton
University Press.
Lee, Jinkwon. 2008. “The Effect of the Background Risk in a Simple Chance Improving
Decision Model.” The Journal of Risk and Uncertainty 36:19–41.
Lehrer, Ehud and Marco Scarsini. 2011. “On the Core of Dynamic Cooperative Games.”
Cornell University Library arXiv:1203.2832.
Lerner, A. P. and H. W. Singer. 1937. “Some Notes on Duopoly and Spatial Competition.”
The Journal of Political Economy, University of Chicago Press 45:145–186.
Lesk, A. M. 2000. “The unreasonable effectiveness of mathematics in molecular biology.”

The Mathematical Intelligencer 22:28–37.
Leuffen, Dirk, Susumu Shikano, and Stefanie Walter. 2012. “Measurement and
Data Aggregation in Small-N Social Scientific Research.” European Political Science
2012/03/30/online.
Levin, Simon A. 2009. Games, groups, and the global good. Berlin ; New York: Springer.
Levine, David K. 1997. “Modeling Altruism and Spitefulness in Experiments.” Journal of

Economic Literature .
Levine, John M. and Richard L. Moreland. 1998. “Small Groups.” In The Handbook of Social
Psychology, edited by Daniel T. Gilbert, Susan T. Fiske, and Gardner Lindzey, volume II,
chapter Part IV, 26, pp. 415–469. Boston: McGraw-Hill, 4th edition.
274
Levitt, Steven D. and Stephen J. Dubner. 2009. Freakonomics. Harper.
Levitt, Steven D. and Stephen J. Dubner. 2011. SuperFreakonomics: Global Cooling, Patriotic
Prostitutes, and Why Suicide Bombers Should Buy Life Insurance. William Morrow Paper-
backs.
Levitt, Steven D. and John A. List. 2006. “What Do Laboratory Experiments
Tell Us About the Real World?” mimeo. University of Chicago and NBER.
http://pricetheory.uchicago.edu/levitt/Papers/jep%20revision%20Levitt%20&%20
List.pdf (accessed November 6, 2010).
Levitt, Steven D. and John A. List. 2007. “What do laboratory experiments measuring
social preferences reveal about the real world?” Journal of Economic Perspectives 21:153–
174.
Levy, Haim. 2006. Stochastic Dominance: Investment Decision Making under Uncertainty,
Springer. Springer US.
Lewin, K. 1936. Principles of topological psychology. New York: McGraw-Hill.
Lichbach, Mark I. 2001. Is Rational Choice Theory All of Social Science? . Ann Arbor:
University of Michigan Press.
Lijphart, Arend. 1971. “Comparative politics and the comparative method.” American
Political Science Review 65:682–693.
Lijphart, A. 1999. Patterns of Democracy. New Haven: Yale University Press.
Lind, Edgar Allen. 1988. The social psychology of procedural justice. New York: Plenum
Press.
Lindgren, Kristian. 1991. “Evolutionary Phenomenon in Simple Dynamics.” In Artificial
Life II, edited by Christopher G. Langton, Charles Taylor, J. Doyne Farmer, and Steen
Rasmussen. Addison-Wesley, Reading, MA.
Lipset, Seymour M. 1959. “Some Social Requisites of Democracy: Economic Develop-
ment and Political Legitimacy.” American Political Science Review 53:69–105.
List, John A. 2009. “Social Preferences: Some Thoughts from the Field.” Annual Review of
Economics pp. 563–579.
Lockhead, Gregory and James Pomerantz. 1991. The Perception of Structure [Paperback].
American Psychological Association.
Lomborg, B. 1996. “Nucleus and shield: The evolution of social structure in the iterated
Prisoners dilemma.” American Sociological Review 61:278–307.
Long, Ngo Van. 2010. A survey of dynamic games in economics. Singapore ; London: World
Scientific.
Long, Roderick T. 2006. “Realism and abstraction in economics: Aristotle and Mises
versus Friedman.” The Quarterly Journal of Austrian Economics 9:3–23.
275
Lopomo, Giuseppe and Efe Ok. 2001. “Bargening, Interdependence and the Rationality
of fair division.” Levine’s Bibliography, UCLA Department of Economics .
Lowell, A. Lawrence. 1910. “The Physiology of Politics.” American Political Science Review
4:1–15.
Loxley, A. 1997. Collaboration in Health and Welfare. Jessica Kingsley Publishers, London.
Luce, D. 1959. Individual Choice Behavior. New York: Wiley.
Lucking-Reiley, David H. 1999. “Using Field Experiments to Test Equivalence Between

Auction Formats: Magic on the Internet.” American Economic Review 89:1063–1080.
Löwenstein, G., L. Thompson, and M. Bazerman. 1989. “Social utility and decision mak-
ing in interpersonal contexts.” Journal of Personality and Social Psychology 53:426–441.
MacKenzie, Scott. 2012. “Within-subjects vs. Between- subjects Designs: Which to Use?”
mimeo. Dept. of Computer Science, York University. http://www.yorku.ca/mack/RN-
Counterbalancing.html (accessed November 28, 2012).
Maier-Rigaud, Frank P. and Jose Apesteguia. 2003. “The Role of Choice in Social
Dilemma Experiments.” mimeo. Economic Discussion Papers, University of Bonn. bgse22.
http://www.wiwi.uni-bonn.de/bgsepapers/bonedp/bgse22_2003.pdf (accessed De-
cember 11, 2011).
Malecki, Michael and Matthew Gabel. 2007. “A Survey Experiment on Sat-

isfaction with Democracy: How question order shapes respondents’ eval-
uations of satisfaction with democracy in the EU.” mimeo. Paper presented
at the annual meeting of the American Political Science Association, Chicago.
http://www.allacademic.com/meta/p210366_index.html (accessed March 13, 2012).
Manhart, Klaus. 2011. “IDC-Studie zum Datenwachstum - Doppeltes Datenvolumen alle

zwei Jahre.” CIO July.
Mankiw, N. G. and Mark Taylor. 2010. Economics: Special Edition With Global Economic
Watch. Southwestern Cengage Learning.
Mann, H. B. and D. R. Whitney. 1947. “On a test whether one of two random variables is
stochastically larger than the other.” Annals of Mathematical Statistics 18:50–60.
Mann, Thomas. 2005. “REACH-Kompromiss nachbessern!” Pressemeldung 17th Novem-

ber.
Mannes, Albert E. 2009. “Are We Wise About the Wisdom of Crowds? The Use of Group
Judgments in Belief Revision.” Management Science 55:1267–1279.
March, James G. and Johan P. Olsen. 2006. “The Logic of Appropriateness.” In The Oxford
Handbook of Public Policy, edited by Michael Moran, Martin Rein, and Robert E. Goodin,
pp. 689–708. Oxford: Oxford University Press.
276
Martin, Andrew D. and Kevin M. Quinn. 2002. “Dynamic Ideal Point Estimation via
Markov Chain Monte Carlo for the U.S. Supreme Court, 1953 to 1999.” Political Analysis
10:134–153.
Martin, D. W. 2008. Doing psychology experiments, volume 7. Pacific Grove, CA:
Brooks/Cole.
Marwell, Gerald and Ruth E. Ames. 1981. “Economists Free Ride, Does Anyone Else?
Experiments on the Provision of Public Goods, IV.” Journal of Public Economics 15:295–
310.
Mason, W. and S. Suri. 2012. “Conducting behavioral research on Amazon’s Mechanical
Turk.” Behavior Research Methods 44:1–23.
Matis, Herbert. 2007. “Das Menschenbild der klassischen Nationalökonomie.” In Das
Menschenbild in der Ökonomie: eine verschwiegene Voraussetzung, edited by Reinhard
Pirker, pp. 105–140. Wien: Löcker.
Mattila, Mikko. 2004. “Contested decisions: Empirical analysis of voting in the European
Union Council of Ministers.” European Journal of Political Research 43:29–50.
May, Kenneth O. 1952. “A set of independent necessary and sufficient conditions for
simple majority decisions.” Econometrica 20:680–684.
Mayring, Philipp. 2002. Einfuehrung in die qualitative Sozialforschung : eine Anleitung zu
qualitativem Denken. Weinheim ; Basel: Beltz.
Mayring, Philipp. 2010. Qualitative Inhaltsanalyse : Grundlagen und Techniken. Weinheim ;
Basel: Beltz.
McCarty, Nolan M. and Adam Meirowitz. 2007. Political game theory : an introduction.
Cambridge [u.a.]: Cambridge University Press.
McConnachie, Alistair. 2000. “Direct democracy - Government of the People, by the
People, for the People.” Sovereignty March issue.
McCubbins, M., R. G. Noll, and B. R. Weingast. 1987. “Administrative Procedures as
Instruments of Political Control.” Journal of Law, Economics and Organization .
McCubbins, Mathew, Roger G. Noll, and Barry R. Weingast. 1989. “Structure and Pro-
cess; Politics and Policy: Administrative Arrangements and the Political Control of
Agencies.” Virginia Law Review .
McCubbins, Mathew and Thomas Schwartz. 1984. “Congressional Oversight Over-
looked: Police Patrols vs. Fire Alarms.” American Journal of Political Science .
McDermott, Rose. 2002. “Experimental Methology in Political Science.” Political Analysis
10:4:325–342.
McElduff, Fiona, Mario Cortina-Borja, Shun-Kai Chan, and Angie Wade. 2010. “When
t-tests or Wilcoxon-Mann-Whitney tests won’t do.” Advances in Physiology Education
34:128–133.
277
McFadden, Daniel. 2006. “Free Markets and Fettered Consumers.” American Economic
Review, American Economic Association 96:5–29.
McFadden, Daniel. 2013. “The New Science of Pleasure.” NBER Working Paper 18687.
McFadden, Daniel L. 1974. “Conditional logit analysis of qualitative choice behavior.”

In Frontiers in econometrics, edited by P. Zarembka, pp. 105–142. New York: Academic
Press.
McFadden, Daniel L. 1976. “Quantal Choice Analysis: A Survey.” Annals of Economic and
Social Measurement 5:363–390.
McFadden, Daniel L. 1982. “Econometric Models of Probabilistic Choice.” In Structural

analysis of discrete data with econometric applications, edited by C. Manski and D. L. Mc-
Fadden, pp. 198–272. Cambridge, MIT Press.
McKelvey, Richard D. 1976. “Intransitivities in multidimensional voting models and

some implications for agenda control.” Journal of Economic Theory 12:472–482.
McKelvey, Richard D. and Peter C. Ordeshook. 1981. “Experiments on the Core: Some
Disconcerting Results for Majority Rule Voting Games.” Journal of Conflict Resolution
25:473–482.
McKelvey, Richard D. and Peter C. Ordeshook. 1982. “Two-Candidate Elections Without

Majority Rule Equilibria: An Experimental Study.” Simulation and Games 13:311–335.
McKelvey, Richard D. and Thomas R. Palfrey. 1995. “Quantal Response Equilibria for
Normal Form Games.” Games and Economic Behavior 10:6–38.
McKelvey, Richard D. and Thomas R. Palfrey. 1996. “A Statistical Theory of Equilibrium

in Games.” Japanese Economic Review 47:186–209.
McKelvey, Richard D. and Thomas R. Palfrey. 1998. “Quantal Response Equilibria for
Extensive Form Games.” Experimental Economics 1:9–41.
McNulty, Roy. 2011. “Realising the Potential of GB Rail.” Report of the Rail Value for Money
Study .
Merkel, Wolfgang. 2010. “Volksentscheide - wer ist das Volk.” Neue Gesellschaft Frank-
furter Hefte 12.
Merrill, Samuel and Bernard Grofman. 1999. A Unified Theory of Voting: Directional and
Proximity Spatial Models. Cambridge University Press.
Mertins, Vanessa. 2008. “The Effects of Procedures on Social Interaction: A Literature

Review.” Discussion Paper Series, Institute for Labour Law and Industrial Relations in the
European Community, University Trier 06.
Mickan, Sharon and Sylvia Rodger. 2000. “Characteristics of effective teams: a literature
review.” Australian Health Review 23.
278
Miljkovic, Sladjana Lj. 2012. Iterative methods for computing generalized inverses of matrices.
Ph.D. thesis, University of Nis, Faculty of Natural Sciences and Mathematics.
Miller, Gary J. 1997. “The Impact of Economics on Contemporary Political Science.” Jour-
nal of Economic Literature XXXV:1173–1204.
Miller, G. J. and T. H. Hammond. 1990. “Committees and the Core of the Constitution.”
Public Choice 66:201–227.
Miller, G. J., T. H. Hammond, and C. Kile. 1996. “Bicameralism and the core: An experi-
mental test.” Legislative Studies Quarterly 21:83–103.
Miller, Gary J. and Joe A. Oppenheimer. 1982. “Universalism in Experimental Commit-
tees.” The American Political Science Review 76(3):561–574.
Miller, Luis and Christoph Vanberg. 2014. “Decision costs in legislative bargaining: An
experimental analysis.” Public choice (accepted). alt: 2011. Discussion Papers, Univer-
sity of Oxford, Nuffield College. 2011002.
Milyo, Jeffrey. 2000a. “Logical deficiencies in spatial models: A constructive critique.”
Public Choice 105:273–289.
Milyo, Jeffrey. 2000b. “A problem with Euclidean preferences in spatial models of poli-
tics.” Economics Letters 66:179–182.
Mitchell, Mark L. 2003. Research Design Explained. Thomson Learning.
Mitchell, Terence R. 1985. “Review and Evaluation of Research Studies on Negotiation
Behaviour.” Journal of Occupational Behaviour 6:85–91.
Mitchell, William C. 1999. “Political science and public choice: 1950-70.” Public Choice
98:237–249.
Moe, Terry. 1990. “Political Institutions - The Neglected Side of the Story.” Journal of Law,
Economics and Organizatio 6:39–77.
Monroe, Kristen R. 2005. Perestroika! The Raucous Rebellion in Political Science. New Haven-
London: Yale University Press.
Morgan, T. Clifton. 1984. “A Spatial Model of Crisis Bargaining.” International Studies
Quarterly 28:407–426.
Morgan, T. C. 1990. “Issue linkage in international crisis bargaining.” American Journal of
Morgan, T. C. 1994. Untying the Knot of War. Ann Arbor, MI: University of Michigan
Press.
Morris, Dick. 1999. The New Prince. Los Angeles: Renaissance Books.
Morris, Stephen and Hyun Song Shin. 2002. “Measuring
Strategic Uncertainty.” mimeo. Working paper, Yale University.
http://www.princeton.edu/˜hsshin/www/barcelona.pdf (accessed March 3, 2012).
279
Morton, Rebecca and Kenneth C. Williams. 2006. “Experimentation in Political Sci-
ence.” In Cambridge Handbook of Experimental Political Science, edited by Janet Box-
Steffensmeier, David Collier, and Henry Brady, pp. 1–27. Oxford: Oxford University
Press.
Morton, Rebecca B. 1999. Methods and models : a guide to the empirical analysis of formal
models in political science. Cambridge: Cambridge University Press.
Morton, Rebecca B. and Kenneth C. Williams. 1999. “Information Asymmetries and Si-
multaneous Versus Sequential Voting.” American political science Review 93:51–67.
Morton, Rebecca B. and Kenneth C. Williams. 2010. Experimental political science and the
study of causality : from nature to the lab. New York: Cambridge University Press.
Morton, Rebecca B. and Kenneth C Williams. 2011. “Electoral Systems and strategic vot-
ing (laboratory election experiments).” In Cambridge Handbook of Experimental Politi-
cal Science, edited by James N. Druckman, Donald P. Green, James H. Kuklinski, and
Arthur Lupia, pp. 369–383. New York: Cambridge Unviersity Press.
Morton, Rebecca B. and Kenneth C. Williams. 2012. “Experimente in der Politischen

Ökonomie.” In Jahrbuch für Handlungs- und Entscheidungstheorie, edited by Thomas
Bräuninger, Bächtiger, and Susumu Shikano, number Band 7: Experiment und Simu-
lation, chapter 1, pp. 81–104. VS Verlag für Sozialwissenschaften, Wiesbaden.
Moser, P. 1997. “A theory of the conditional influence of the European Parliament in the
cooperation procedure.” Public Choice 91:333–350.
Moses, Jonathon Wayne and Torbjoern L. Knutsen. 2007. Ways of knowing : competing
methodologies and methods in social and political research. New York: Palgrave Macmillan.
Motulsky, Harvey J. 2010. Intuitive Biostatistics. Oxford University Press, 2nd edition.
Möslinger, Astrid. 2010. “Heidelberg: Klotzen für die Ewigkeit.” Meier - Magazin 07 -
10:32–33.
Mueller, D. 2003. Public Choice III. Cambridge University Press, New York.
Muren, Astri and Roger Pyddoke. 1999. “Does Collusion without Communication Ex-
ist?” Research Papers in Economics, Stockholm University, Department of Economics. 11.
Murphy, Ryan O., Kurt A. Ackermann, and Michel J. J. Handgraaf. 2011. “Measuring
Social Value Orientation.” Judgment and Decision Making 6:771–781.
Nagel, R., A. Bosch, J. Montalvo, and A. Satorra. 2002. “One, two, [three], infinity: news-
paper and lab beauty-contest experiments.” American Economic Review 92:1687–1701.
Nash, John F. jr. 1950. “The Bargaining Problem.” Econometrica 18:155–162.
Neale, M. A. and M. Bazerman. 1991. Cognition and rationality in negotiation. New York:
Free Press.
280
Newson, Roger. 2002. “Parameters behind "nonparametric" statistics: Kendall’s tau,
Somers’ D and median differences.” Stata Journal 2:45–64.
Newson, Roger. 2006a. “Confidence intervals for rank statistics: Somers’ D and exten-
sions.” The Stata Journal 6:309–334.
Newson, Roger. 2006b. “Efficient Calculation of Jackknife Confidence Intervals for Rank
Statistics.” Journal of Statistical Software 15.
Newson, Roger. 2013. “Interpretation of Somers’ D under four simple models.” mimeo.
King’s College London. www.imperial.ac.uk/nhli/r.newson/miscdocs/intsomd1.pdf
(accessed April 22, 2013).
Nice. 2001. “Treaty of Nice - Amending the treaty on European union, the treaties es-
tablishing the Eurpean communities and certain related acts.” Official Journal of the
European Communities 2001/C80/01. signed on 26th February.
Nicklisch, Andreas and Irenaeus Wolff. 2012. “On the Nature of Reciprocity: Evi-
dence from the Ultimatum Reciprocity Measure.” Research Paper Series, Thurgauer
Wirtschaftsinstitut, Universitaet Konstanz. 65.
Niemi, Richard G. and Arthur Q. Frank. 1985. “Sophisticated voting under the plurality
procedure: A test of a new definition.” Journal of Theory and Decision 19:151–162.
Niskanen, William A. 1971. Bureaucracy and Representative Government. Chicago: Aldine.
Noel, Hans. 2010. “Ten Things Political Scientists Know that You Don’t.”
Political science and practical Politics Forum, Georgetown University 8.
http://www9.georgetown.edu/faculty/hcn4/Downloads/Noel_Forum.pdf (ac-
cessed July 7, 2011).
North, Douglass. 2008. “Violence.” Keynote Address, Upton Forum
www.beloit.edu/upton/assets/North.pgs.vol.I.pdf (accessed October 22, 2011).
North, Douglass Cecil. 1990. Institutions, institutional change, and economic performance.
Cambridge ; New York: Cambridge University Press.
North, Douglass C. 1997. “Nobel Prize Lecture: Economic Performance through Time.”
In From Nobel Lectures, Economics 1991-1995, edited by Torsten Persson. Singapore:
World Scientific Publishing Co.
North, Douglass C., John Joseph Wallis, and Barry R. Weingast. 2009. Violence and Social
Orders: A Conceptual Framework for Interpreting Recorded Human History. Cambridge
University Press.
Nosofsky, Robert. 1985. “Overall similarity and the identification of separable-dimension
stimuli: A choice model analysis.” Perception and Psychophysics 38:415–432.
Nullmeier, Frank and Tanja Pritzlaff. 2009. “The implicit normativity of political prac-
tices. Analyzing the dynamics and power relations of committee decision-making.”
Critical Policy Studies 3:357–374.
281
Nutzinger, Hans G. 2010. “Nobelpreis in Wirtschaftswissenschaften für Elinor Ostrom:
Ein Überblick über ihr ökonomisches Hauptwerk.” MAGKS Discussion Paper Series in
Economics 24.
Oaksford, M. and N. Chater. 1992. “Bounded rationality in taking risks and drawing
inferences.” Theory and Psychology 2:225–230.
Oaksford, M. and N. Chater. 1998. Rationality in an uncertain world: Essays on the cognitive
science of human reasoning. Psychology Press.
Oates, Wallace E. 2005. “Toward A Second-Generation Theory of Fiscal Federalism.”

International Tax and Public Finance 12:349–373.
Ockenfels, Axel. 1999. Fairness, Reziprozität und Eigennutz : Oekonomische Theorie und
experimentelle Evidenz, volume 108. Tuebingen: Mohr Siebeck.
Ockenfels, Axel. 2007. “Daten statt Dogmen.” In Ökonomie 2.0: 99 überraschende Erkennt-
nisse. Schäffer-Poeschel.
Offerman, T. 2002. “Hurting hurts more than helping helps.” European Economic Review
46:1423 1437.
Offerman, T., J. Sonnemans, and A. Schram. 1996. “Value orientations, expectations, and
voluntary contributions in public goods.” Economic Journal 106:817–845.
Ok, Efe A. and Levent Kockesen. 2000. “Negatively interdependent preferences.” Social
choice and welfare 17:533–558.
Olken, Benjamin A. 2008. “Direct democracy and local public goods evidence from a field
experiment in Indonesia.” NBER Working Paper Series 14123.
Olsen, Johan P. 2007. “Organization theory, public administration, democratic gover-

nance.” ARENA Working Papers 1.
Olson, Mancur. 1965. The logic of collective action; public goods and the theory of groups,
volume 124. Cambridge, Mass.: Harvard University Press.
Olum, Yasim. 2004. “Modern management theories and prac-

tices: an critical overview.” presented at the 15th East African
Central Banking Course at Kenya School of Monetary Studies.
http://unpan1.un.org/intradoc/groups/public/documents/AAPAM/UNPAN025765.pdf
(accessed November 25, 2013).
Opp, Karl-Dieter. 1999. “Contending Conceptions of the Theory of Rational Action.”

Journal of Theoretical Politics 11:171–202.
Opp, Karl-Dieter. 2004. “Die Theorie rationalen Handelns im Vergleich mit alternativen
Theorien.” In Paradigmen der akteurszentrierten Soziologie, edited by Manfred Gabriel,
pp. 43–68. VS Verlag für Sozialwissenschaften, Wiesbaden.
282
Ordeshook, Peter C. 1986. Game theory and political theory : an introduction. Cambridge
Cambridgeshire ; New York: Cambridge University Press.
Ordeshook, Peter C. and Thomas R. Palfrey. 1988. “Agendas, Strategic Voting, and Sig-
naling with Incomplete Information.” American Journal of Political Science 32:441–466.
Ordeshook, Peter C. and Mark Winer. 1980. “Coalitions and Spatial Policy Outcomes
in Parliamentary Systems: Some Experimental Results.” American Journal of Political
Science 24(4):730–752.
Ortlieb, Claus Peter. 2010. “Interview: Ökonomie ist eigentlich keine Wissenschaft.” FAZ
- Frankfurter Allgemeine Zeitung Ausgabe vom 08.05.
Ortman, Andreas and Gerd Gigerenzer. 2000. “Reasoning in Economics and Psychology:
Why Social Context Matters.” In Cognition, Rationality, and Institutions, edited by M.E.
Streit, U. Mummert, and D. Kiwit, pp. 131–145. Springer.
Ortmann, A. 2003. “Charles R. Plott’s collected papers on the experimental foundations

of economic and political science.” Journal of Economic Psychology 24:555–575.
Ortmann, Andreas. 2005. “Field experiments in Economics: some methodological

caveats.” In Field Experiments in Economics, edited by Glenn W. Harrison, Jeffrey Car-
penter, and John A. List, volume 10 of Research in Experimental Economics, pp. 51–70.
Emerald Group Publishing Limited.
Ostrom, Elinor. 1986. “An agenda for the study of institutions.” Public Choice 48:3–25.
Ostrom, Elinor. 1990. Governing the Commons: The Evolution of Institutions for Collective
Action. Cambridge University Press.
Ostrom, Elinor. 2005. Understanding Institutional Diversity. Princeton: Princeton Univer-

sity Press.
Ostrom, Elinor. 2006. “Converting Threats into Opportunities.” Political Science and Poli-
tics 39:3–12.
Ostrom, Elinor. 2008. “Polycentric Systems as One Approach for Solving

Collective-Action Problems.” Social Science Research Network available at SSRN:
http://ssrn.com/abstract=1304697 (accessed February 1, 2013).
Ostrom, Vincent. 1999. “Polycentricity.” In Polycentricity and Local Public Economies: Read-
ings from the Workshop in Political Theory and Policy Analysis, edited by Michael McGin-
nis, pp. 52–74. Ann Arbor, MI: University of Michigan Press.
Oviedo, J. 2000. “The core of a repeated n-person cooperative game.” European Journal of
Operational Research 127:519–524.
Palacios-Huerta, Ignacio and Oscar Volij. 2008. “Experientia docet: professionals play
minimax in laboratory experiments.” Econometrica 76:71–115.
283
Palfrey, T. and J. E. Prisbrey. 1997. “Anomalous Behavior in Public Goods Experiments:
How Much and Why?” American Economic Review 87:829–846.
Palfrey, Thomas R. 1989. “A Mathematical Proof of Duverger’s Law.” In Models of Strate-

gic Choice in Politics, edited by Peter C. Ordeshook, pp. 69–92. Ann Arbor: University
of Michigan Press.
Palfrey, Thomas R. 2009. “Laboratory Experiments in Political Economy.” Annual Review

of Political Science 12:379–388.
Paolacci, Gabriele. 2012. “Experimental Turk.” A blog on social science experiments using
Amazon Mechanical Turk experimentalturk.wordpress.com.
Paolacci, G., J. Chandler, and P. G. Ipeirotis. 2010. “Running experiments on Amazon

Mechanical Turk.” Judgment and Decision Making 5:411–419.
Pearl, J. 2009. “Causal inference in statistics: An overview.” Statistics Surveys 3:96–146.
Pedersen, Anne Reff, Karina Sehested, and Eva Sörensen. 2011. “Emerging Theoretical
Understanding of Pluricentric Coordination in Public Governance.” The American Re-
view of Public Administration 41:375–394.
Peleg, B. and P. Sudhoelter. 2003. Introduction to the Theory of Cooperative Games. Boston:
Kluwer Academic.
Persky, Joseph. 1995. “The Ethology of Homo Economicus.” Journal of Economic Perspec-
tives, American Economic Association 9:221–231.
Persson, Torsten and Guido Enrico Tabellini. 1992. Federal fiscal constitutions. Pt.1, Risk
sharing and moral hazard. London: Centre for Economic Policy Research.
Perttunen, C.D., D.R. Jones, and B.E. Stuckman. 1993. “Lipschitzian optimization without
the Lipschitz constant.” Journal of Optimization Theory and Applications 79:175–181.
Peters, B. G. 2000. Institutional theory in political science. London: Continuum. repr. ed.
Pich, Michael T., Christoph H. Lock, and Arnoud de Meyer. 2002. “On Uncertainty, Am-
biguity, and Complexity in Project Management.” Management Science 48:10008–1023.
Pigliucci, Massimo. 2013. “Rationally Speaking.” A blog on the Philosophy of Science

http://rationallyspeaking.blogspot.de (accessed October 23, 2012).
Plott, C.R. 1979. “The application of laboratory experimental methods to public choice.”
In Collective decision making: Applications from public choice theory, edited by C.S. Russell,
pp. 137–160. Baltimore: Johns Hopkins University Press.
Plott, Charles R. 1967. “A Notion of Equilibrium and Its Possiblity Under Majority Rule.”
American Economic Review 57:787–806.
Plott, C. R. 1991. “Will Economics Become an Experimental Science.” Southern Economic

Journal 57:901–919.
284
Plott, Charles R. and Vernon L. Smith. 2008. Handbook of experimental economics results.
[Volume 1]. Amsterdam ; New York: North Holland.
Plott, C. R. and K. Zeiler. 2011. “The Willingness to Pay-Willingness to Accept Gap, the
"Endowment Effect," Subject Misconceptions, and Experimental Procedures for Elicit-
ing Valuations: Reply.” American Economic Review 101:1012–1028.
Poisson, S. D. 1837. Recherches sur la probabilité des jugements en matière criminelle et matière
civile. Paris: Bachelier.
Prasnikar, Vesna. 1997. “Learning the Decision Rules in Ultimatum Games.” Working
paper .
Pratt, J. W. 1964. “Risk Aversion in the Small and in the Large.” Econometrica XXXII:122–
36.
Pritzlaff, Tanja. 2006. Entscheiden als Handeln. Campus Verlag, Frankfurt.
Proksch, Sven-Oliver and Jonathan B. Slapin. 2009. “WORDFISH: Scaling Software for
Estimating Political Positions from Texts.” http://www.wordfish.org. Version 1.3 (22 Jan-
uary 2009).
Provis, Chris. 2000. “Ethics, Deception and Labor Negotiation.” Journal of Business Ethics
28:145–158.
Putnam, Robert D. 1988. “Diplomacy and Domestic Politics: The Logic of Two-Level
Games.” International Organization 42:427–460.
Quattrone, George A. and Amos Tversky. 1988. “Contrasting Rational and Psychological
Analyses of Political Choice.” The American Political Science Review 82(3):719–736.
Rabin, M. 1993. “Incorporating Fairness into Game-Theory and Economics.” American
Economic Review 83:1281–1302.
Rabin, M. 1998. “Psychology and economics.” Journal of Economic Literature 36:11 46.
Radke, Sabine. 2011. Verkehr in Zahlen. DVV Media Group GmbH, Hamburg. 40.
Jahrgang.
Ragin, C. C. 1987. The Comparative Method: Moving Beyond Qualitative and Quantitative
Strategies. Berkeley: University of California Press.
Ragin, Charles C. 2006. “Set Relations in Social Research: Evaluationg Their Consistency
and Coverage.” Political Analysis 14:291–310.
Ragusa, Jordan, Joshua Huder Huder, and Nate Birkhead. 2012. “Rule 22.” In Dedicated
to Unlimited Debate about American Politics.
Rand, D. G. 2012. “The promise of Mechanical Turk: How online labor markets can help
theorists run behavioral experiments.” Journal of Theoretical Biology 299:172–179.
Rawls, John. 1971. A Theory of Justice. Belknap. Original 1971, überarbeitete dt. Fassung
1975.
285
Ray, Debraj. 2007. A game-theoretic perspective on coalition formation. Oxford ; New York:
Oxford University Press.
Redner, R. and H. Walker. 1984. “Mixture Densities, Maximum Likelihood and the EM
Algorithm.” Siam Review 26:195–239.
Reid, Joseph D. Jr. 1987. “The Theory of Sharecropping: Occams Razor and Economic
Analysis Winter 1987 19(4):.” History of Political Economy 19:551–569.
Reips, U. 2000. “The web experiment method: advantages, disadvantages, and solu-
tions.” In Psychological Experiments on the Internet, edited by M. Birnbaum, pp. 89–120.
Academic Press, San Diego.
Resnick, P., K. Kuwabara, R. Zeckhauser, and E. Friedman. 2000. “Reputation Systems.”

Communications of the ACM 43:45–48.
Richards, D. 1994. “Intransitivities in Multidimensional Spatial Voting - Period three

Implies Chaos.” Social Choice and Welfare 11:109–119.
Riesch, Hauke. 2010. “Simple or simplistic? Scientists’ views on Occam’s razor.” Theoria
25:75–90.
Riker, William H. 1980. “Implications from the Disequilibrium of Majority Rule for the
Study of Institutions.” The American Political Science Review 74:432–446.
Riker, William H. 1982a. Liberalism Against Populism. San Francisco: W.H. Freeman.
Riker, William H. 1982b. “The Two-Party System and Duverger’s Law: An Essay on the
History of Political Science.” American Political Science Review 76:753–766.
Risse, Mathias. 2004. “Arguing for Majority Rule.” The Journal of Political Philosophy
12:41–64.
Rosch, Eleanor and Barbara B. Lloyd. 1978. Cognition and categorization. Hillsdale, NJ:
John Wiley & Sons Inc.
Rosenthal, Robert and Ralph L. Rosnow. 2009. Artifacts in behavioral research: Robert Rosen-
thal and Ralph L. Rosnow’s classic books : a re-issue of Artifact in behavioral research, Experi-
menter effects in behavioral research and the volunteer subject. New York: Oxford University
Press.
Roth, A. E. and I. Erev. 1995. “Learning in Extensive-Form Games - Experimental-Data

and Simple Dynamic-Models in the Intermediate-Term.” Games and Economic Behavior
8:164–212.
Royston, P.J. 1992. “Approximating the Shapiro-Wilk W-Test for Non-normality.” Statis-
tics and Computing 2:117–119.
Rubinstein, Ariel. 1982. “Perfect Equilibrium in an Bargaining Model.” Econometrica

50:97–110.
286
Ruckriegel, Karlheinz. 2010. “Die Wiederentdeckung des Menschen in der Ökonomie.”
In Vortrag im Rahmen der Tagung "Ökonomische Rationalitätenkaskade" an der Universität
Koblenz-Landau, überarbeitete Fassung vom 30.12.
Rudolf, Stefan. 2011. Militaerorganisatorische und gesellschaftliche Herausforderungen bei der
"Transformation" der Bundeswehr in eine Einsatzarmee. GRIN Verlag.
Ruffle, Bradley J. and Richard H. Sosis. 2006. “Cooperation and the In-Group-Out-Group
Bias: A Field Test on Israeli Kibbutz Members and City Residents.” Journal of economic
behavior and organization 60:147–163.
Russ, Steve. 2011. “The unreasonable effectiveness of mathematics in the natural sci-
ences.” Interdisciplinary Science Reviews 36:209–213.
Rutstroem, E. Elisabet and Melonie B. Williams. 2000. “Entitlements and fairness: an
experimental study of distributive preferences.” Journal of Economic Behavior and Orga-
nization 43:75–89.
Sachs, Jeffrey D. 2012. “Government, Geography, and Growth - The True Drivers of
Economic Development.” Foreign Affairs 91.
Saijo, T. and H. Nakamura. 1995. “The Spite Dilemma in Voluntary Contribution Mecha-
nism Experiments.” Journal of Conflict Resolution 39:535–560.
Salant, Stephen W. and Eban Goodstein. 1990. “Predicting Committee Behavior in Ma-
jority Rule Voting Experiments.” The RAND Journal of Economics 21:293–313.
Samuelson, W. and W. Zeckhauser. 1988. “Status Quo Bias in Decision Making.” Journal
of Risk and Uncertainty 1:7–59.
Sauermann, Jan. 2012. “Ticken Gruppen anders? Ein Laborexperiment zur unter-
schiedlichen Motivation von Individuen und Gruppen in Verhandlungen.” In Jahrbuch
für Handlungs- und Entscheidungstheorie, edited by Thomas Bräuninger, Bächtiger, and
Susumu Shikano, number Band 7: Experiment und Simulation, chapter 5, pp. 81–104.
VS Verlag für Sozialwissenschaften, Wiesbaden.
Sauermann, Jan and Ulrich Glasmann. 2011. “Entscheidungskosten und Gemeinwohlef-
fekte demokratischer Abstimmungsregeln - eine experimentelle Untersuchung.” Poli-
tische Vierteljahresschrift (PVS) 52:373–398.
Sauermann, J. and A. Kaiser. 2010. “Taking Others into Account: Self-Interest and Fair-
ness in Majority Decision Making.” American Journal of Political Science 54:667–685.
Saunders, Ben. 2010a. “Democracy, Political Equality, and Majority Rule.” Ethics, The
University of Chicago Press 121:148–177.
Saunders, Ben. 2010b. “Why Majority Rule Cannot Be Based Only on Procedural Equal-
ity.” Ratio Juris 23:113–122.
Scharpf, Fritz W. 2000. Interaktionsformen. Akteurzentrierter Institutionalismus in der Politik-
forschung. Opladen: Leske and Budrich.
287
Schäfers, Michael and Joachim Zimmermann. 2005. Partizipation: Von Solidarität, Eigen-
verantwortung und bürgerschaftlichem Engagement [Taschenbuch]. LIT Verlag Münster.
Schmidt, Klaus M. 2010. “Social Preferences and Competition.” Discussion Papers in
Economics, University of Munich 6.
Schmidt, M. G. 2012. Der deutsche Sozialstaat. Geschichte und Gegenwart. München:
C.H.Beck.
Schnapp, Kai-Uwe, Nathalie Behnke, and Joachim Behnke. 2009. Datenwelten - Datener-
hebung und Datenbestaende in der Politikwissenschaft. Nomos.
Schneider, G., D. Finke, and S. Bailer. 2010. “Bargaining Power in the European Union:
An Evaluation of Competing Game-Theoretic Models.” Political Studies 58:85–103.
Schneider, Gerald, Bernard Steunenberg, and Mika Widgren. 2006. “Evidence with insigt:
What models contribute to EU research.” In The European Union Decides. The Empirical
Relevance of Policy Making Models, edited by R. Thomson, F. N. Stokman, C. H. Achen,
and T. König, chapter 11, pp. 299–316. Cambridge University Press.
Schoenberg, Ronald. 2000. “Constrained Optimization.” GAUSS documentation
http://econ.la.psu.edu/computing/pdf/co.pdf (accessed March 2, 2011).
Schofield, Norman. 1996. Collective Decision-Making: Social Choice and Political Economy.
Springer, 1 edition.
Schofield, Norman. 2008. The spatial model of politics. Routledge, London and New York.
Schopler, John and Chester A. Insko. 1992. “The Discontinuity Effect in Interpersonal
and Intergroup Relations: Generality and Mediation.” In European Review of Social
Psychology, edited by Wolfgang Stroebe and Miles Hewstone, pp. 121–151. Chichester:
Wiley.
Schotter, Andrew. 2006. “Strong and Wrong: The Use of Ratio-nal Choice Theory in
Experimental Economics.” Journal of Theoretical Politics 18:498–511.
Schram, A. 2005. “Artificiality: The tension between internal and external validity in
economic experiments.” Journal of Economic Methodology 12:225–237.
Schram, Arthur J.H.C. 2002. “Experimental Public Choice.” Tinbergen Institute Discussion
Papers 106.
Schuman, Howard and Stanley Presser. 1996. Questions and answers in attitude surveys:
Experiments on question form, wording, and context. Sage Publications.
Schwartz, Thomas. 1977. “Collective Choice, Separation of Issues and Vote Trading.” The
American Political Science Review 71:999–1010.
Selten, Reinhard and Rosemarie Nagel. 1998. “Das Zahlenwahlspiel-Hintergruende und
Ergebnisse.” Spektrum der Wissenschaft February:16–22.
Sen, A. 1995. “Rationality and Social Choice.” American Economic Review 85:1–24.
288
Sen, Amartya. 1997. “Maximization and the Act of Choice.” Econometrica 65 (4):745–780.
Sen, A. K. 1977. “Rational Fools a critique of the behavioural foundations of economic

theory.” Philosophy and Public Affairs 6:317–344.
Sengul, Metin, Javier Gimeno, and Jay Dial. 2012. “Strategic Delegation: A Review, The-
oretical Integration, and Research Agenda.” Journal of Management 38:375–414.
Severinov, Sergei. 2008. “The value of information and optimal organization.” RAND
Journal of Economics 39:238–265.
Shadish, William R., Thomas D. Cook, and Donald Thomas Campbell. 2002. Experimental
and Quasi-Experimental Designs for Generalized Causal Inference. Houghton Mifflin.
Shapiro, S. S. and M. B. Wilk. 1965. “An Analysis of Variance Test for Normality (Com-
plete Samples).” Biometrika 52:591–611.
Shavit, T., D. Sonsino, and U. Benzion. 2001. “A comparative study of lotteries-evaluation

in class and on the web.” Journal of Economic Psychology 22:483–491.
Shepard, Roger N. 1991. “Integrality versus separability of stimulus dimension: From

an early convergence of evidence to a proposed theoretical basis.” In The perception
of structure: Essays in honor of Wendell R. Garner., edited by Gregory R. Lockhead and
James R. Pomerantz, pp. 53–71. Washington, DC: American Psychological Association.
Sheppard, Simon G. 1998. Game-Theoretic Models of Human Behaviour. Heretical Press.
Shepsle, Kenneth A. 1979. “Institutional Arrangements and Equilibrium in Multidimen-

sional Voting Models.” American Journal of Political Science 23:27–59.
Shepsle, Kenneth A. and Mark S. Bonchek. 1997. Analyzing politics : rationality, behavior,
and institutions. New York: W.W. Norton.
Shepsle, Kenneth A. and Gary W. Cox. 2007. “Majority Cycling and Agenda Manipula-
tion: Richard McKelvey’s Contributions and Legacy.” In Positive Change in Political Sci-
ence: The Legacy of Richard McKelvey’s Most Influential Writings, edited by John Aldrich,
James E. Alt, and Arthur Lupia. Ann Arbor: University of Michigan Press.
Shepsle, Kenneth A. and Barry R. Weingast. 1984. “Uncovered Sets and Sophisticated
Voting Outcomes with Implications for Agenda Institutions.” American Journal of Polit-
ical Science 28:49–74.
Shepsle, Kenneth A. and Barry R. Weingast. 1995. Positive theories of congressional institu-
tions. Ann Arbor: University of Michigan Press.
Shepsle, Kenneth A. and Barry R. Weingast. 2004. “Structure-induced equilibrium and

legislative choice.” Public Choice 37 (3):503–519.
Sherstyuk, Katerina, Tarui Nori, and Saijo Tatsuyoshi. 2011. “Payment schemes in
infinite-horizon experimental games.” Working Papers Series, University of Hawai’i .
289
Shikano, Susumu. 2009. “Simulating party competition and vote decision under mixed
member electoral systems.” AUCO Czech Economic Review 3:270–291.
Shils, Edward A. and Henry A. Finch. 1949. Max Weber on the Methodology of the Social
Sciences. New York: Free Press.
Shu, Min. 2003. “Nonseparable Preferences, Public Opinion and Referendums on Euro-
pean Integration.” mimeo. Paper presented at the ECPR General Conference, Marburg, Ger-
many. http://www.aoni.waseda.jp/min.shu/paper.html (accessed December 8, 2012).
Shuttleworth, Martyn. 2008. “Double Blind Experiment.” Experiment Resources
http://www.experiment-resources.com/double-blind-experiment.html (accessed Jan-
uary 13, 2012).
Shy, Oz. 1996. Industrial Organization: Theory and Applications, volume 1. The MIT Press,
1st edition.
Siegel, Sidney. 1988. Nonparametric statistics for the behavioral sciences. New York: McGraw-
Hill.
Siegel, Sidney. 2001. Nichtparametrische statistische Methoden. Eschborn bei Frankfurt a.M.:
Klotz, 5., unveränd. aufl. edition.
Signorino, Curtis S. 1999. “Strategic interaction and the statistical analysis of international
conflict.” American Political Science Review 93:279–297.
Signorino, Curtis S. 2003. “Structure and Uncertainty in Discrete Choice Models.” Political
Analysis 11:316–344.
Signorino, Curtis S. and Kuzey Yilmaz. 2003. “Strategic Misspecification in Regression
Models (with . . 47(3):.” American Journal of Political Science 47:551–566.
Siminski, Peter. 2008. “Order Effects in Batteries of Questions.” Quality and Quantity:
International Journal of Methodology 42:477–490.
Simon, D., D. C. Krawczyk, and K. J. Holyoak. 2004. “Construction of preferences by
constraint satisfaction.” Psychological Science 15:331–336.
Simon, Herbert. 1957. “A Behavioral Model of Rational Choice.” The Quarterly Journal of
Simon, Herbert A. 1962. “The Architecture of Complexity.” Proceedings of the American
Philosophical Society 106.
Simon, Herbert A. 1982. Models of bounded rationality. Cambridge, Mass. ; London: MIT
Press.
Simon, Herbert A. 1993. Homo rationalis. Die Vernunft im menschlichen Leben. Frankfurt
a.M.: Campus.
Simon, Herbert A., Massimo Egidi, and Robin Lapthorn Marris. 1992. Economics, bounded
rationality and the cognitive revolution. Aldershot: Elgar.
290
Slapin, Jonathan and Sven Oliver Procksch. 2010. “Look who’s talking: Parliamentary
debate in the European Union.” European Union Politics September 11:333–357. erste
Quelle war 2008, nicht 2010.
Slapin, J. B. and S. O. Proksch. 2008. “A scaling model for estimating time-series party
positions from texts.” American Journal of Political Science 52:705–722.
Smirnov, O. 2009. “Endogenous choice of amendment agendas: types of voters and ex-
perimental evidence.” Public Choice 141:277–290.
Smith, A. 1759. The Theory of Moral Sentiment. London: A. Millar. P. 1: How selfish soever
man may be supposed, there are evidently some principles in his nature, which interest
him in the fortunes of others, and render their happiness necessary to him, though he
derives nothing from it, except the pleasure of seeing it.
Smith, Adam. 1776. An Inquiry into the Nature and Causes of the Wealth of Nations. W.
Strahan and T. Cadell, London.
Smith, Charles E., Robert D. Brown, John M. Bruce, and L. Marvin Overby. 1999. “Party
Balancing and Voting for Congress in the 1996 National Election.” American Journal of
Smith, Vernon L. 1962. “An Experimental Study of Competitive Market Behavior.” Jour-
nal of Political Economy, University of Chicago Press 70:111–137.
Smith, Vernon L. 1964. “Effect of Market Organization on Competitive Equilibrium.”

Smith, Vernon L. 1973. “Notes on Some Literature in Experimental Economics.” Social

Science working paper California Institut of Technology 21.
Smith, Vernon L. 1976. “Experimental economics: Induced value theory.” The American
Economic Review 65(2):274–279.
Smith, Vernon L. 1991. Papers in Experimental Economics. Cambridge University Press.
Smith, Vernon L. and James M. Walker. 1993. “Money Rewards and Decision Costs in
Experimental Economics.” Economic Inquiry 15:245–261.
Smithies, Arthur. 1941. “Optimum Location in Spatial Competition.” Journal of Political

Economy, University of Chicago Press 49:423–439.
Snedecor, G. W. and W. G. Cochran. 1989. Statistical Methods. Ames, IA: Iowa State
University Press, 8th edition.
Sobel, J. 2005. “Interdependent preferences and reciprocity.” Journal of Economic Literature

43:392–436.
Soberg, M. S. and M. P. Wattenberg. 2001. Mixed-member electoral systems: the best of both
worlds? Oxford, Oxford University Press.
291
Somers, Robert H. 1962. “A New Asymmetric Measure of Association for Ordinal Vari-
ables.” American Sociological Review 27:799–811.
Soto, Fabian A. and Edward A. Wasserman. 2010. “Integrality and Separability of Stim-
ulus Dimensions and Multidimensional Generalization in Pigeons.” Journal of Experi-
mental Psychology: Animal Behavior Processes 36:194–205.
Sprinthall, Richard C. 2011. Basic Statistical Analysis. Prentice Hall, 9 edition.
Sribney, Willim. 1999. “What does "completely determined" mean in
my logistic regression output?” Statalist, Stata 10 FAQ, StataCorp LP.
http://www.stata.com/support/faqs/statistics/completely-determined-in-logistic-
regression (accessed April 4, 2013).
Stahl, D. O. and E. Haruvy. 2008. “Subgame perfection in ultimatum bargaining trees.”
Games and Economic Behavior 63:292–307.
Stanovich, K. E. and R. F. West. 2000. “Individual differences in reasoning: Implications
for the rationality debate?” Behavioral and Brain Sciences 23:645.
Steunenberg, B. 1994. “Decision-Making under Different Institutional Arrangements
- Legislation by the European-Community.” Journal of Institutional and Theoretical
Economics-Zeitschrift Fur Die Gesamte Staatswissenschaft 150:642–669.
Steunenberg, Bernard and Torsten J. Selck. 2006. “Testing Procedural Models of Euro-
pean Union Legislative Decision-Making.” In The European Union Decides, edited by
Christopher H. Achen Robert Thomson, Frans N. Stokman and Thomas König, pp.
54–84. Cambridge University Press.
Steunenberg, Bernard and Frans van Vught. 1997. Political institutions and public policy :
perspectives on European decision making. Dordrecht ; Boston: Kluwer Academic Pub-
lishers.
Stigler, G. G. 1950. “The development of utility theory, I,II.” Journal of Political Economy I,
II:43, 307–327 373–396.
Stiglitz, Joseph E. 1972. “Some Aspects of the Pure Theory of Corporate Finance:
Bankruptcies and Take-Overs.” Bell Journal of Economics, The RAND Corporation 3:458–
482.
Stokman, Frans N. and Bruce Bueno De Mesquita. 1994. European Community Decision
Making: Models, Applications, And Comparisons. New Haven: Yale University Press.
Stokman, Frans N. and Reinier Van Oosten. 1994. “The Exchange of Voting Positions: An
Object-Oriented Model of Policy Networks.” In European Community Decision Making:
Models, Applications and Comparisons, edited by Bruce Bueno de Mesquita and Frans
N. Stokman, pp. 105–127. New Haven: Yale University Press.
Stoll, Heather. 2013. Changing Societies, Changing Party Systems. Cambridge University
Press.
292
Stoll, Heather M. 2004. Social cleavages, political institutions and party systems: putting pref-
erences back into the fundamental equation of politics. Dissertation, Stanford University,
Deptartment of Political Science.
Strom, Gerald Steven. 1990. The logic of lawmaking: a spatial theory approach. Baltimore:
Johns Hopkins University Press.
Strom, Kaare, Wolfgang C. Mueller, and Torbjoern Bergman. 2003. Delegation and Ac-
countability in Parliamentary Democracies. Oxford , New York.
Studenmund, A. H. 2006. Using Econometrics: A Practical Guide. Pearson: Addison-

Wesley.
Sugden, Robert. 1989. “Spontaneous Order.” Journal of Economic Perspectives III:85–98.
Sunitiyoso, Yos, Erel Avineri, and Kiron Chatterjee. 2011. “The effect of social interactions
on travel behaviour: An exploratory study using a laboratory experiment.” Transporta-
tion Research Part A: Policy and Practice 45:332–344.
Suri, S. and D. J. Watts. 2011. “Cooperation and Contagion in Web-Based, Networked

Public Goods Experiments.” PLOS one 6.
Surowiecki, James. 2004. The Wisdom of Crowds: Why the Many Are Smarter Than the Few
and How Collective Wisdom Shapes Business, Economies, Societies and Nations. Doubleday.
Sutter, Matthais. 2002. “Public Bad Prevention by Majority Voting on Redistribution -

Experimental Evidence.” Group Decision and Negotiation 11:415–428.
Sutter, Matthias. 2008. “Individual behavior and group membership: Comment.” Jena
Economic Research Papers, Max-Planck-Institute of Economics 075.
Swedberg, Richard and Andreas Maurer. 2009. Grundlagen der Wirtschaftssoziologie. Verlag
für Sozialwissenschaften.
Taleb, Nassim Nicholas. 2007. The black swan - The impact of the highly improbable. Random
House.
Taylor, Michael. 2006. Rationality and the Ideology of Disconnection. New York: Cambridge
University Press.
Tegmark, Max. 2007. “The Mathematical Universe.” Foundations of Physics 38:101–150.

Dept. of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139.
Thaler, Richard H. 1980. “Toward a positive theory of consumer choice.” Journal of Eco-
nomic Behavior and Organization 1:39–60.
Thaler, Richard H. and Cass R. Sunstein. 2009. Nudge: Improving Decisions About Health,
Wealth and Happiness. Penguin.
Thiem, Janina. 2009. Nationale Parteien im Europaeischen Parlament: Delegation, Kontrolle

und politischer Einflus. VS Verlag für Sozialwissenschaften, Wiesbaden.
293
Thomson, Robert. 2011. Resolving controversy in the European Union. Cambridge Univer-
sity Press.
Thomson, R., J. Arregui, D. Leuffen, R. Costello, J. Cross, R. Hertz, and T. Jensen. 2012.
“A new dataset on decision-making in the European Union before and after the 2004
and 2007 enlargements (DEUII).” Journal of European Public Policy 19(4):604–622.
Thomson, R., F. N. Stokman, C. H. Achen, and T. König. 2002. “Report on Commission
proposal COM(1999)582.” mimeo. DEU field report.
Thomson, R., F. N. Stokman, C. H. Achen, and T. König. 2006. The European Union decides.
Cambridge, UK ; New York: Cambridge University Press.
Thye, Shane R. 2007. “Logical and Philosophical Foundations of Experiments in the Social
Sciences.” In Laboratory Experiments in the Social Sciences, edited by Murray Webster and
Jane Sell, chapter 3, pp. 57–86. Elsevier: London.
Tirole, Jean. 1988. The theory of industrial organization. Cambridge, Mass.: MIT Press.
Tolwinski, B. 2003. A concept of cooperative equilibrium for dynamic games. Warsaw, Poland:
Polish Academy of Sciences.
Torgerson, Warren S. 1952. “Multidimensional scaling: Theory and Method.” Psychomet-
rica 17:401–419.
Torgerson, Warren S. 1967. Theory and methods of scaling. New York [u.a.]: Wiley.
Tornay, S.C. 1938. Ockham: Studies and Selections. La Salle, IL: Open Court.
Traub, Stefan, Christian Seidl, and Ulrich Schmidt. 2009. “An experimental study on
individual choice, social welfare, and social preferences.” European Economic Review
53:385–400.
Tremewan, James, Chloe Le Coq, and Alexander K. Wagner. 2012. “Social Preferences
and Bounded Rationality in the Centipede Game.” mimeo. Department of Economics,
University of Vienna. http://vimeo.com/37037476 (accessed December 12, 2012).
Tsebelis, G. 1994. “The Power of the European Parliament as a Conditional Agenda Set-
ter.” American Political Science Review 88:128–142.
Tsebelis, George. 2002. Veto players : how political institutions work. Princeton, N.J.: Prince-
ton University Press.
Tsebelis, G. and G. Garrett. 2001. “The institutional foundations of intergovernmentalism
and supranationalism in the European Union.” International Organization 55:357.
Tullock, Gordon. 1981. “Why so much stability.” Public Choice 37:189–204.
Tversky, Amos and Daniel Kahneman. 1981. “The Framing of Decisions and the Psychol-
ogy of Choice.” Science 211:453–458.
Tversky, Amos and Daniel Kahneman. 1986. “Rational Choice and the Framing of Deci-
sions.” The Journal of Business 59:251–278.
294
Tversky, Amos and Daniel Kahneman. 1991. “Loss Aversion in Riskless Choice: A
Reference-Dependent Model.” The Quarterly Journal of Economics 106:1039–1061.
van de Kragt, Alphons J. C., John M. Orbell, and Robyn M. Dawes. 1983. “The Minimal
Contributing Set as a Solution to Public Goods Problems.” The American Political Science
Review 77(1).
van Deemen, Adrian. 2009. “Structure-induced (Dis)Equilibrium.”
mimeo. Institute for Management Research, University of Nijmegen.
http://www.gate.cnrs.fr/IMG/pdf/van_Deemen.pdf (accessed October 29, 2011).
Van den Bos, Jan M. M. 1991. Dutch EC policy making: a model-guided approach to coordina-
tion and negotiation. Amsterdam: Thesis Publishers.
Van Huyck, John B., Raymond C. Battalio, and Richard O. Beil. 1991. “Strategic Uncer-
tainty, Equilibrium Selection, and Coordination Failure.” In Average Opinion Games,
volume 106, pp. 885–910. Quarterly Journal of Economics.
Vanberg, Christoph. 2008. “Why Do People Keep Their Promises? An Experimental Test
of Two Explanations.” Econometrica 76:1467–1480.
Vanberg, Viktor. 2002. “Rational Choice vs. Program-Based Behavior: Alternative Theo-
retical Approaches and Their Relevance for the Study of Institutions.” Rationality and
Society 14:7–54.
Veblen, Thorstein. 1899. The theory of the leisure class; an economic study in the evolution of
institutions. London: Macmillan & co., ltd.
Velten, Carlo and Steve Janata. 2012. Datenexplosion in der Unternehmens-IT: Wie Big Data
das Business und die IT verändert. Eine Studie der Experton Group AG im Auftrag der
BT (Germany) GmbH & Co. oHG.
Velupillai, K. Vela. 2005. “The unreasonable ineffectiveness of mathematics in eco-
nomics.” Cambridge Journal of Economics 29:849–872.
Vincent, Leah-Martin. 2012. “The Role of Gender and Race in Co-
operation and Trust: Experimental Examination of Identity Based
Strategic Interaction.” mimeo. University of California, San Diego.
http://www.econ.ucsd.edu/~rstarr/191AFall2012/191%20Final%20Paper%20-
%20Leah-Martin%20%282012%29.pdf (accessed February 2, 2013).
von Bismarck, Otto. 1863. “Speech to the Herrenhaus at December 18.” In Oxford Dictio-
nary of Quotations. 3d, p. 84 edition.
von der Leyen, Ursula. 2012. “Rede von Ursula von der Leyen, Bundesministerin
fuer Arbeit und Soziales, anlaesslich der Senkung des Rentenbeitragssatzes 2013 im
Deutschen Bundestag.” Haushaltsrede 27.09. 13:50.
von Neumann, John and Oskar Morgenstern. 1944. Theory of Games and Economic Behavior.
Princeton, NJ. Princeton University Press, 1st edition. 1944, sec.ed. 1947, th.ed. 1953.
295
von Prittwitz, Volker. 2007. Vergleichende Politikanalyse. Utb, Stuttgart.
von Weizsäcker, Carl Christian. 2012. “Eine Makro- Perspektive der Evolutionsökonomie:
Das Problem der Vorsorge.” Vortrag Evolutionsökonomischer Ausschuss Kloster Beuron
13. Juli 2012 http://www.denkwerkzukunft.de/downloads/CCvW-Beuron-Vortrag-
Jul%202012.pdf (accessed August 22, 2012).
Vredenburgh, Donald J. and John G. Maurer. 1984. “A Process Framework of Organiza-

tional Politics.” Human relations 37:47–65.
Wachendorfer-Schmidt, Ute. 2005. Politikverflechtung im vereinigten Deutschland. VS Verlag

für Sozialwissenschaften, Wiesbaden.
Wagschal, Uwe. 2009. Statistik fuer Politikwissenschaftler. Muenchen: Oldenbourg, R.
Wallenius, J., J. S. Dyer, P. C. Fishburn, R. E. Steuer, S. Zionts, and K. Deb. 2008. “Multiple
criteria decision making, multiattribute utility theory: Recent accomplishments and
what lies ahead.” Management Science 54:1336–1349.
Walliser, Bernard. 2008. Cognitive economics. Berlin: Springer.
Walter, Stephen D., Alvan R. Feinstein, and Carolyn K. Wells. 1987. “Coding Ordinal
Independent Variables in Multiple Regression Analyses.” American Journal of Epidemi-
ology 125:319–23.
Wandling, Richard. 2012. “Rationality and Rational Choice.” In 21st Century Political
Science, edited by John Ishiyama and Marijke Breuning, chapter 5, pp. 34–42. Thousand
Oaks: Sage.
Wantchekon, L. 2003. “Clientelism and voting behavior - Evidence from a field experi-
ment in Benin.” World Politics 55:399–422.
Weber, E. U. and E. J. Johnson. 2008. “Decisions under uncertainty: Psychological, eco-

nomic, and neuroeconomic explanations of risk preference.” In Neuroeconomics: Deci-
sion making and the brain, edited by P. Glimcher, C. Camerer, E. Fehr, and R. Poldrack,
chapter 10, pp. 127–144. New York: Elsevier.
Weber, Max. 1968. Gesammelte Aufsätze zur Wissenschaftslehre. Tübingen.
Webster, Murray and Jane Sell. 2007. Laboratory experiments in the social sciences. Amster-
dam ; Boston: Academic Press/Elsevier: London.
Weibull, Joergen W. 1995. Evolutionary game theory. Cambridge, Mass ; London: MIT
Press.
Weihe, Anne Cordelia, Tanja Pritzlaff, Frank Nullmeier, Tilo Felgenhauer, and Britta
Baumgarten. 2008. “Wie wird in politischen Gremien entschieden? Konzeptionelle
und methodische Grundlagen der Gremienanalyse.” Politische Vierteljahresschrift (PVS)
49:339–359.
296
Weingast, Barry R. 2009. “Second generation fiscal federalism: The implications of fiscal
incentives.” Journal of Urban Economics 64:279–293.
Weise, Frank-Juergen, Hans Heinrich Driftman, Hans-Ulrich Klose, Juergen Kluge, Karl-
Hein Lather, and Hedda von Wedel. 2010. “Vom Einsatz her Denken - Konzentra-
tion, Flexibilitaet, Effizienz.” Technical report, Bericht der Strukturkomission der Bun-
deswehr.
Wicksell, Knut. 1896. Finanztheoretische Untersuchungen: nebst Darstellung und Kritik des
Steuerwesens Schwedens. Jena: Fischer Verlag.
Wigner, Eugene. 1960. The Unreasonable Effectiveness of Mathematics in the Natural Sciences,
volume 13 of Communications in Pure and Applied Mathematics. New York: John Wiley
& Sons.
Wilcoxon, Frank. 1945. “Individual comparisons by ranking methods.” Biometrics Bulletin
1:80–83.
Wildschut, Tim, Hein F. M. Lodewijkx, and Chester A. Insko. 2001. “Toward a Reconcili-
ation of Diverging Perspectives on Interindividual-Intergroup Discontinuity: The Role
of Procedural Interdependence.” Journal of Experimental Social Psychology 37:273–285.
Wildschut, Tim, Brad Pinter, Jack L. Vevea, Chester A. Insko, and John Schopler. 2003.
“Beyond the Group Mind: A Quantitative Review of the Interindividual-Intergroup
Discontinuity Effect.” Psychological Bulletin 129:698–722.
Williamson, Oliver E. 1975. Markets and hierarchies, analysis and antitrust implications : a
study in the economics of internal organization. New York: Free Press.
Willits, Fern K. and Bin Ke. 1995. “Part-Whole Question Order Effects: Views of Rurality.”
The Public Opinion Quarterly 59:392–403.
Wilson, E. B. 1927. “Probable inference, the law of succes-sion, andstatistical inference.”
Journal of the American Statistical Association 22:209–212.
Wilson, Rick and Catherine Eckel. 2011. “Trust and Social Exchange.” In The Cambridge
Handbook of Experimental Political Science, edited by James N. Druckman, Donald P.
Green, James H. Kuklinski, and Arthur Lupia, chapter 17. New York: Cambridge Un-
viersity Press.
Wilson, R. K. 2007. “Voting and Agenda Setting in Political Science and Economics.” In
Laboratory Experiments in the Social Sciences, edited by M. Webster and J. Sell, pp. 433–
458. San Diego, CA: Elsevier.
Witt, L. A., Martha C. Andrews, and K. Michele Kacmar. 2000. “The Role of Participa-
tion in Decision-Making in the Organizational Politics-Job Satisfaction Relationship.”
Human relations 53:341–358.
Wolf, Frieder. 2010. “Enlightened Eclecticism or Hazardous Hodgepodge? Mixed Meth-
ods and Triangulation Strategies in Comparative Public Policy Research.” Journal of
Mixed Methods Research 4:144–167.
297
Würzner, Eckart. 2010. “Das Signal ist angekommen - Interview mit Oberbürgermeister
Dr. Eckart Würzner zur Stadthallen-Erweiterung.” Rhein-Neckar-Zeitung 15.06.
Xia, Lirong, Vincent Conitzera, and Jerome Lang. 2011. “Strategic Sequential Voting in
Multi-Issue Domains and Multiple-Election Paradoxes.” Preprint submitted to Journal of
Economic Theory .
Yezer, Tony, Robert Goldfarb, and Paul Poppen. 1996. “Does Studying Economics Dis-
courage Cooperation?: Watch What We Do, Not What We Say or How We Play.” Journal
of Economic Perspectives 10:171–186.
Zaller, J. and S. Feldman. 1992. “A Simple Theory of the Survey Response - Answering
Questions Versus Revealing Preferences.” American Journal of Political Science 36:579–
616.
Zizzo, Daniel. 2010. “Experimenter demand effects in economic experiments.” Experi-

mental Economics 13:75–98.
Zizzo, Daniel John and Jonathan H. W. Tan. 2009. “Game Harmony: A Behavioral Ap-
proach to Predicting Cooperation in Games.” ICBBR Working Papers, International Cen-
tre for Behavioural Business Research 4.
Zohlnhöfer, Reimut. 2008. “Foederalismusreform und die Entwicklung der Zustim-

mungsbeduerftigkeit von Bundesgesetzen: Versuch einer Klaerung.” Zeitschrift fuer
Parlamentsfragen 39:415–419.
Zohlnhöfer, Reimut. 2009. “Der Politikverflechtungsfalle entwischt? Die Effekte der

Foederalismusreform I auf die Gesetzgebung.” Zeitschrift fuer Politikwissenschaft 19.
Jahrgang:39–76.
Zuckerman, Marvin. 1994. Behavioural expressions and biosocial bases of Sensation Seeking.
Cambridge: Cambridge University Press.
298
299
300
Appendix
A.1 Software used
GAUSS
In Sec. 6.4 I use GAUSS’ (Version 9.0.2, build 1114) constrained optimization algorithm453
(Schoenberg, 2000) to estimate a random-utility mixture model for an individual’s choice
determinants. This follows the approach of Goeree et al. (2002) but adds sincere and so-
phisticated voting types. In situations where all subjects vote sincerely this task could
be accomplished by using conditional logistic regression which implies a “logit equilib-
rium” (Goeree et al., 2002, p. 262). When including sophisticated voting the correspond-
ing choice probabilities are adjusted according to the equilibrium solution concept.
STATA
STATA (Version 12.0) serves as main computational tool for the multiple calculations in
this study. Most of the commands used can be categorized as belonging to the standard
tool box. Not quite ordinary are the application of a local polynomial smoother which
used STATA ’s kdensity454 estimation in Sec. 3.5 and the implementation of a two-way er-
ror components model in Sec. 5.1 by using STATA’s xtmelogit455 (multilevel mixed-effects)
logistic regression. Also, I apply STATA ’s somersd package456 in Sec. 5.2 to assess trends
over single rounds of my experiment.
MATLAB
I use MATLAB ’s (Version 7.12.0.635) “Global Optimization Toolbox”457 (Version 3.2 - Re-
lease 2011b) and the related extension “Genetic Algorithm and Direct Search Toolbox”458
in Sec. 3.5 for simulating the effect of different NSP values. Based on the idea of Pert-
tunen et al. (1993) this derivative-free global optimization algorithm enables me to find
the global optimum of the objective function. Due to the discontinuity of the function
453 Source: http://www.gaussian.com/g_tech/g_ur/k_opt.htm.
454 Source: http://www.stata.com/help.cgi?kdensity.
455 Source: http://www.stata.com/help.cgi?xtmelogit.
456 The somersd package can be obtained from the Statistical Software Components (SSC) archive, which is
accessible under http://www.repec.org. SSC is one of the most important download sites for user-
written STATA packages.
457 Source: http://www.mathworks.de/products/global-optimization.
458 Source: http://www.mathworks.com/help/releases/R13sp2/pdf_doc/gads/gads_tb.pdf.
301
this is impossible while using standard optimization methods for constrained nonlinear
optimization problems such as Sequential Quadratic Programming.
R / WORDFISH
In Sec. 7.4 I use the word scaling algorithm WORDFISH 459 (Proksch and Slapin, 2009),
written in R statistical language (Version 2.14.2., release date 2012-02-19). The algorithm
extracts (spatial) positions from text documents using word frequencies and places differ-
ent documents into a single dimension of discrepancy by means of maximum likelihood
estimations. The scaling technique does not need any anchoring documents to perform
the analysis. Instead, it relies on a statistical model of word counts, more precisely a
Poisson distribution (Poisson, 1837) of word frequencies.
My analysis also takes advantage of the text mining package already included in R (tm
package, Feinerer et al., 2008). This package possesses a stemming algorithm and a stop-
word dictionary. A well-documented example of how to apply WORDFISH can be found
in Slapin and Proksch (2008).
z-Tree
I implemented my experiment using the z-Tree software (Version 3.3.11) developed by

Fischbacher (2007). The acronym stands for Zurich Toolbox for Readymade Economic
Experiments. The software facilitates to develop and carry out laboratory experiments
as many standards are predefined. The program can also be adjusted to fit a wide range
of experimental designs. It requires no previous programming knowledge and can be
licensed free of charge. Under http://www.iew.uzh.ch/ztree/index.php the current
version (3.3.12) and much more information (z-Tree Wiki, FAQ, etc.) can be obtained.
459 Source: http://www.wordfish.org.
302
A.2 Weighted Euclidean distance including nonseparable preferences

The standard weighted Euclidean distance (cf. Hinich and Munger, 1997) is represented by WED (θ, x ) = (θ − x )T A (θ − x ), where in
a d-dimensional
 space
 θ = (θ1 ,θ2 ,θ3 ,...,θd ) describes an actor’s unconditional first preference, x = ( x1 ,x2 ,x3 ,...,xd ) describes the policy and
a11 ... a1d
 . . 
A= . . . ...  describes each dimension’s salience in its main diagonal (a11 , ..., add ) and the conditionality between two dimensions
 . 
ad1 ... add
in the secondary diagonals (ad1 , ..., a1d ).
Applying the standard equation to a simple two-dimensional example (dimensions i and j) results in
   
 
 aii aij θ i − xi
WED (θ, x ) =  θ i − xi θ j − xj (A.1)
a ji a jj θ j − xj
Eqn. A.1 applies to the case of reciprocal nonseparability. In the case of non-reciprocal nonseparability (cf. Sec. 2.4.5) an actor’s conditional
allocation dimension j is indiscriminant with respect to the direction towards which the outcome deviates from the unconditional ideal point
in dimension i. An increase in the absolute distance between their unconditional ideal policy and the outcome causes a decrease in their
conditional allocation preferences. Accordingly, I modify the standard equation by using absolute distance |θi − xi | on the policy dimension:
   
 
 aii aij |θ i − xi |
WED (θ, x ) =  |θ i − xi | θ j − xj
a ji a jj θ j − xj
  

      |θ i − xi |
=  aii |θ i − xi | + aij θ j − x j a ji |θ i − xi | + a jj θ j − x j
θ j − xj
       
303
= aii |θ i − xi ||θ i − xi | + aij θ j − x j |θ i − xi | + a ji |θ i − xi | θ j − x j + a jj θ j − x j θ j − x j
   
As it always holds that |θ i − xi ||θ i − xi | = (θ i − xi )2 and that θ j − x j |θ i − xi | = |θ i − xi | θ j − x j the equation can be simplified to
304
      2
WED (θ, x ) = aii (|θ i − xi )2 + aij θ j − x j |θ i − xi | + a ji |θ i − xi | θ j − x j + a jj θ j − x j
and finally to
     2
WED (θ, x ) = aii (|θ i − xi )2 + aij + a ji θ j − x j |θ i − xi | + a jj θ j − x j (A.2)
Eqn. A.2 corresponds to Eqn. 2.11 in Sec. 2.4.5.

A.3 Structure of the DEU field reports and expert interviews
The field reports of the DEU project (Thomson et al., 2006) collected information on 66
contested law-making proposals. The reports include the following data, which has been
gathered by expert interviews:
• Descriptive information on each proposal as identifying number, legislative proce-

dure, date of introduction and revision,
• Descriptive information on the data collection as date and duration of the inter-
view, name, nationality and profession of the expert, evaluation of the informant’s
knowledge on the topic, etc.
• The number of single issues within each proposal,
• Position and salience of national governments, the European Commission and the
European Parliament on each issue.
Most important is the issue specification of the proposal. This was obtained through the
interviews and accompanied with background information obtained, e.g., from the Leg-
islative Observatory460 . The interviews enabled the determination of ideal position and
salience of actors. Tab. A.2 provides an overview on the content of the expert interviews.
Although the reports did not ask for NSP in particular they proved to be a very helpful
starting point for the investigation into potential nonseparability (Tab. 3.1).
Table A.2: Structure of the DEU expert interviews
INFORMATION OF INTEREST SAMPLE QUESTIONS AND TASKS
Issue specification List the single issues contained in proposal X.

Specify each single proposal.
Position specification Visualize each issue continua as a scale from 0 to 100.

State the substantive meaning, in terms of a policy position, of as many points
on the scale as possible.
What are the preferences of the stakeholders regarding each dimension?
Why is this issue of more importance to some stakeholders than to others?
What arguments do the stakeholders use for their preferences?
Did stakeholders shift in preferences and positions?
Contestation Purpose of Commission proposal?

Why are these issues contested?
Did stakeholders make threats or promises?
What relationships exist between the issues?
To what extent are they independent from each other?
Further comments Features of the decision-making situation not contained in these issues?
Additional information which is important?
460 The Legislative Observatory constitutes the European Parliament’s database for monitoring EU decision-
making processes. It can be accessed online under http://www.europarl.europa.eu/oeil/home/home.
do.
305
A.4 Three-dimensional contour plots
The three graphs in Fig. A.1 and in Fig. A.2 depict the three-dimensional spatial models
for different levels of nonseparability. The graphs correspond to the two-dimensional
representations of the illustrative case studies in Sec. 3.3. As their counterparts, the plots
depict member states’ unconditional ideal positions, the SQ ante, the winset for QMV, the
actual policy outcome and the model predictions. The “utility hills” indicate the Nash
product within the winset to the SQ. Accordingly, the thick lines demarcate the border of
the winset. Please note that due to the undulating contour not all information is visible.
Figure A.1: Three-dimensional contour plots for Council regulation COM1999/163

EXPLANATORY NOTE
The three graphs illustrate the Council regulation COM1999/163. The regulation was concerned with the representation
of fishery organizations at the EU level. Its two issues were the extent of national representation and the corresponding
funding. These two issues form the dimensions of the base area. The utility level is represented as height. Since the utility
product does not possess a meaningful unit, I leave the height dimension without index. Instead, the highest peak (or
plateau) of the utility hill indicates the maximum relative gain within the Winset. For calculation of the utility gains I
assumed a positive but non-reciprocal nonseparability and therefore applied Eqn. 2.11. The SQ was (35/50) and the final
outcome (100/50) for this proposal.
LABELS
Counterpart to Fig. 3.1: separable preferences

A (72.4 / 59.7), B (64.9 / 69.3) and C (59.5 / 71.8).
306
Counterpart to Fig. 3.2: 50% nonseparable preferences
A (99.9 / 59.7); B (86.7 / 73.1); C (92.0 / 70.5).

A (85.2 / 44.5); B (63.9 / 57.5); C (75.7 / 71.4).
307
Figure A.2: Three-dimensional contour plots for Council regulation COM1999/582
EXPLANATORY NOTE
The three graphs show Council regulation COM1999/582. This regulation amended the organization of the common
market in bananas. Its two issues were the type of import regime that would be adopted and the transitional period
during which a tariff quota would apply. These two issues form the dimensions of the base area. The utility level is
represented as height. Since the utility product does not possess a meaningful unit, I leave the height dimension without
index. Instead, the highest peak (or plateau) of the utility hill indicates the maximum relative gain within the Winset. For
calculation of the utility gains I assumed mutually positive and reciprocal nonseparability and applied Eqn. 2.9. The SQ
was (0/100) and the final outcome (50/40)for this proposal.
LABELS
of America.
Counterpart to Fig. 3.3: separable preferences

A (24.1 / 68.3), B (11.8 / 62.4) and C (31.3 / 55.8).
308
A (28.4 / 57.1); B (18.8 / 57.7); C (32.0 / 55.7).

A (50.0 / 40.0); B (44.6 / 39.4); C (35.1 / 52.8).
309
A.5 The extent of nonseparable preferences in the DEU data set
310
This section provides the in Chap. 3 discussed NSP coding scheme. It comprises also a short description of every proposal and issue.
Table A.3: The extent of nonseparable preferences in the DEU data set
EXPLANATORY NOTE
The right six columns display the information on nonseparability. In general, it can be read as matrix A, with information on the salience in the main diagonal (s) and information on the
presumed direction of nonseparability in the secondary diagonal. A “+” indicates a positive and a “−“ indicates a negative nonseparability between two issues (row and column). The absolute
value bars “||” indicate the case of non-reciprocal nonseparability.
Sec. 2.4.5 emphasizes that allocation preferences may be conditionally dependent on the policies pursued but that this must not hold the other way around. In the table, the policy dimension is
displayed in the corresponding row, the allocation dimension in the column. For example, when looking at the first proposal in the table below, the preference to prolong or cancel the temporal
protection of refugees (issue 1 = “budget allocation”) depends on the duration of the temporal protection (issue 2 = “policy 1”) and the distribution of asylum seekers across member states
(issue 3 = “policy 2”).
COM ID Short Description Issue Label i i1 i2 i3 i4 i5 i6

On minimum standards for giving temporary decision rule to cancel the temporal protection of refugees 1 s
CNS00127 protection in the event of a mass influx of duration of the temporal protection 2 +;|| s
displaced persons way in which the asylum seekers have to be distributed among 3 +;|| s
EU member states
Value added tax VAT: length of application of the minimum of VAT standard rate 1 s +;||
CNS00223
current minimum standard rate time period of measure 2 s
duration of the extension of the current CMO regime 1 s
Proposal for a Council Regulation on the CMO
CNS00250 reduction of production quotas 2 +;|| s +
in the Sugar Sector
abolition of subsidies for storage costs 3 +;|| + s
extension of the current aid 1 s
use and labeling of mixtures of olive oil and vegetable oil 2 +;|| s +
CNS00358 Quality strategy for olive oil use of talc in the processing of olive oil 3 +;|| s
classification of different types of olive oil 4 +;|| + s
labeling of olive oil regarding the country of processing 5 +;|| s
use of cages 1 s
timing for the improvement of the general conditions of the cages 2 +;|| s
Minimum standards for the protection of laying minimum of cm2 of cage area for each hen 3 +;|| s +
CNS98092
hens timing for the introduction of a compulsory system of cm2 of 4 +;|| + s
cage area for each hen
timing for banding the cages after the first of January of 2009 5 +;|| s
imported eggs that do not fulfill with the EU rules 6 +;|| s
Common organization of the market in beef and reduction of the intervention price 1 s -
CNS98109
veal compensation for farmers 2 - s
Common organization of the market in milk and reduction of the intervention price 1 s -
CNS98110
milk products the future of the quota system 2 - s
EC/Turkey relations: implementation of amount of money 1 s
CNS98299
measures to intensify customs union minority rights 2 +;|| s
Agenda 2000: Financial instrument for fisheries size of the scrap-build penalty 1 s +;||
CNS98347
guidance linkage between the meeting of the objectives of the 2 s
“Multiannual Guidance Programme” and the allocation of
subsidies for fleet renewal and modernization
Community action program in the field of civil duration of the whole program in the field of civil protection 1 s +;||
CNS98354
protection amount of money assigned every year for this project 2 s
information to consumers 1 s
Common organization of the markets in fishery
CNS99047 market intervention (subsidies) 2 s -
and aquaculture products
trade with third countries (protection) 3 - s
Control measures covered by the Convention on implementation of the measures of regional fisheries 1 s -;||
CNS99138
Future Multilateral Co-operation in the organizations
North-East Atlantic Fisheries financing of these measures 2 s
Closer dialogue with the fishing industry and legal basis should be created for expenditure for the support of 1 s
CNS99163
groups affected by the common fisheries policy national level fisheries organizations
presence of national representatives in the Advisory Committee 2 +;|| s
on Fisheries
institutional place of the employment committee 1 s +;||
CNS99192 Setting up an employment committee
tasks that should be assigned to the committee 2 s
level of the Guaranteed National Quantities (GNQ) of Greece 1 s -;||
CNS99202 Council Regulation on production aid for cotton
and Spain
level of the penalties for surplus production 2 -;|| s
role of the MED Committee in deciding the projects 1 s +;||
CNS99214 Financial and technical measures to accompany
budget assigned by the Commission for financial aid to the 2 s
MEDA countries
import system to be adopted for the commerce and distribution 1 s +
CNS99235 Common organization of the market in bananas
of bananas within the EU countries
year in which will take into effect the new system for the 2 + s
commerce of bananas within the EU territory
311
312
Support system for producers of certain arable restriction of support scheme/SQ (long) 1 s +
CNS99236
crops (flax and hemp) restriction of support scheme/SQ (short) 2 + s
money 1 s
distribution 2 +;|| s
Audio-visual industry: development, distribution
CNS99276 project 3 +;|| s
and promotion of works
period 4 +;|| s
pilot 5 +;|| s
scope of the implementation of the public access to documents 1 s - -
Public access to documents of the European regimen of exceptions in order to provide public access to 2 - s
COD00032
Parliament, the Council and the Commission documents
public access to confidential documents 3 - s
whether the European rules should be applied to National 4 +;|| s
Authorities
threshold above which the resale right should apply 1 s - +;||
Intellectual property, original works of art: resale cap, or maximum amount of money an artist should receive as 2 s +;||
COD96085
right for the benefit of the author their resale right
degressivity of the resale right 3 - s +;||
date of implementation 4 s
use of vegetable fats other than cocoa butter in chocolate 1 s -
Directive on coca and chocolate products products
COD96112
intended for human consumption labeling of chocolate products that contain veg. fats other than 2 -;|| s
cocoa
derogation for the UK and Ireland regarding milk chocolate 3 s
timing of the impact study on developing countries 4 - s
copyright on internet 1 s
Directive on the harmonization of certain
COD97359 time shifting 2 +;|| s
aspects of copyright and related rights
exceptions 3 +;|| s
money 1 s
Education, training: Community action program
COD98195 revision clause 2 +;|| s
SOCRATES
terminology 3 s
Taking up, the pursuit and the supervision of the strength of regulation 1 s
COD98252
business of electronic money institutions derogations 2 +;|| s
EC/Turkey relations: implementation of amount of money 1 s
COD98300 measures to promote economic and social minority rights 2 +;|| s
development nuclear strategy 3 +;|| s
timing for the introduction of the first compulsory step 1 s
System for the identification and registration of timing for the introduction of the second compulsory step 2 s
COD99204
bovine animals level of detail on the label during the first stage of the system 3 -;|| s
level of detail on the label during the second stage of the system 4 -;|| s
scope of harmonization 1 s
Interoperability of the trans-European
COD99252 timing of the implementation of the Directive 2 +;|| s
conventional rail system
discretion available to the TSI committee 3 s
A.6 The magnitude of nonseparability at the proposal level
Figure A.3: Mean average error at different levels of nonseparability and model’s predic-
tive accuracy at the proposal level
EXPLANATORY NOTE
The figure depicts each model’s mean average error per proposal at different levels of nonseparability and 90% CI. The
graph is the result of a local polynomial smoother which uses an Epanechnikov kernel (Epanechnikov, 1969) with a degree
of 0. It is best interpreted as a moving average through the simulated data. The 41 issues categorized with reciprocal NSP
are depicted in the right and the 40 issues categorized with non-reciprocal NSP in the left figure.
mean average error (issue-level)
EXPLANATORY NOTE
The figure is the result of a local polynomial smoother which uses an Epanechnikov kernel (Epanechnikov, 1969) with a
degree of 0. It includes 90% CI and is best interpreted as a moving average through the simulated data. The 41 issues
categorized with reciprocal NSP are depicted in the right and the 40 issues categorized with non-reciprocal NSP in the left
figure.
error model 1 - error model 2
LABELS
313
A.7 Experimental instructions
The instructions for the participants in my laboratory experiment consisted of two parts.
The first part introduced the subjects to the general environment, such as that they are
taking part in an lab experiment, that they will face a collective decision-making, that
they will receive a payment for the points earned in the experiment after they have been
multiplied with an exchange rate, that they are already entitled to a show-up fee, etc. In
particular they were introduced to the payoff table and its significance.
The second part described the rules of the corresponding treatment as well as the screens
they would encounter when they are asked to make their decision. For that purpose, the
introductions included screenshots of the upcoming experiment.
Subjects were only given rules for one treatment at a time, the one design that was applied
subsequently. Nevertheless, subjects were told at the beginning of the experiment that
the decision-making rules would be altered during the experiment and that they would
receive the corresponding new instructions when this would happen.
The following pages show the instructions as they were handed out to the participants.
As the experimental sessions were conducted in Heidelberg the instructions are only
available in German.
After distributing the instructions they were also read aloud by a supervisor. Questions
were answered only privately. After that, the subjects went through the rules on their
screen again and had to pass some questions testing their understanding of the upcoming
game. The experiment did not start before all subjects had passed this test. During the
whole experiment communication between subjects was forbidden.
PART I
• general setup p. 315-317
PART II
• pooling p. 318-319
• simultaneous delegation p. 320-322
• sequential delegation p. 323-325
314
Figure A.4: General instructions
Erläuterungen
Herzlich willkommen zum heutigen Experiment! Falls Sie im Folgenden etwas nicht verstehen, eine Fragen haben oder
ein Problem auftaucht machen Sie bitte per Handzeichen auf sich aufmerksam. Wir werden dann zu Ihnen kommen.
Fragen bitte nicht laut stellen! Während des Experiments ist keine Kommunikation mit anderen Teilnehmern erlaubt.
Bitte stellen sie Ihre Handys lautlos oder schalten Sie sie aus.
In unserem Experiment können Sie Geld verdienen. Wie viel Sie erhalten, hängt von Ihren Entscheidungen im Experiment
ab.
AUSZAHLUNGEN
• Sie erhalten in jedem Fall eine "show-up-fee", für Ihr Erscheinen, in Höhe von 4 €.
• Für das Ausfüllen eines Fragebogens erhalten Sie weitere 2 €.
• Zusätzlich können Sie in den einzelnen Runden des Experiments Punkte sammeln.
– Am Ende des Experiments werden zufällig drei Runden ausgewählt. Die in diesen Runden gesammelten
Punkte werden Ihnen gutgeschrieben.
– Die Punkte werden im Verhältnis 5:1 in Euro umgerechnet. Für jeweils 5 gesammelte Punkte erhalten sie
also 1 € zusätzlich.
• Das Geld wird am Ende des Experiments bar ausgezahlt. Jeder Teilnehmer erfährt nur seinen eigenen
Auszahlungsbetrag.
ABLAUF
• Das Experiment ist nicht auf eine feste Anzahl Runden beschränkt.
• Das Experiment besteht aus 2 Phasen. Zuerst stimme Sie in einer Gruppe mit 6 Mitgliedern über mehrere Alter-
nativen ab. Die Gruppe muss dabei kollektiv ein Entscheidungsproblem lösen.
• Nach einigen Runden werden die Abstimmungsregeln verändert. Sie werden dann noch einmal auf die neuen
Regeln hingewiesen.
• Anschließend füllen Sie bitte noch einen Fragebogen aus. Darin werden wir Ihnen Fragen zum heutigen Experi-
ment und zu einem aktuellem politischen Problem stellen.
• Es ist sehr wichtig, dass alle Teilnehmer die Regeln verstehen. Deshalb werden wir diese ausführlich erklären. Sie
erhalten hierdurch aber keine geringeren Auszahlungen.
-1-
315
D IE A LTERNATIVEN Abbildung 1: Eine Alternative.
In jeder Runde müssen Sie zwischen verschiedenen Alternativen
wählen. Diese werden Ihnen am Bildschirm wie in Abbildung 1 (hier
rechts) angezeigt.
Jede Alternative weist jedem der 6 Spieler eindeutig eine bestimmte

Punkteanzahl zu. Die für den Spieler relevante Punkteanzahl steht
immer in derselben Zeile wie seine Spieler-Nummer.
Im Beispiel: Spieler 1 erhält 7 Punkte, Spieler 2 erhält 28 Punkte, usw.
Insgesamt stehen 9 Alternativen zur Auswahl. Wie in der Abbildung 2 Abbildung 2: Die mögliche
(hier rechts) zu sehen entstehen die 9 Alternativen als Kombination der Ergebnisse für Spieler 4.
Spalten (Buchstaben A-C) und Zeilen (Ziffern 1-3). Somit sind A1, A2,
A3, B1, B2, B3, C1, C2 und C3 mögliche Kombinationen.
Um die Tabelle besser erklären zu können sind dort bisher nur die
Punkte für Spieler 4 eingetragen. Die für diesen Spieler relevante
Punkteanzahl steht für jede Alternative in derselben Zeile wie die
Spieler-Nummer.
Würde die Alternative B2 gewählt erhält der Spieler 4 1 Punkt. Falls
Alternative C3 gewählt würde, bekäme Spieler 4 17 Punkte.
Mit ihrer Abstimmung beeinflussen die Spieler das Ergebnis aller

Mitglieder der Gruppe. Für ihre eigene Punkteanzahl sind die Punkte
der anderen Spieler aber nicht relevant!
Alle 9 Alternativen bilden die komplette Auszahlungstabelle. In der

vollständigen Auszahlungstabelle ist in jeder Alternative für jeden
Spieler eine Punkteanzahl abgetragen. Auf der nächsten Seite sehen Sie
ein Beispiel.
-2-
316
D IE A USZAHLUNGSTABELLE Abbildung 3: Die komplette
Die Auszahlungstabelle bildet die Grundlage für alle im Experiment Auszahlungstabelle
getroffenen Entscheidungen. Um diesen wichtigen Aspekt zu
verdeutlichen hier ein Beispiel zu Abbildung 3:
Wird die Alternative B1 gewählt, erhält Spieler 1 eine Punkteanzahl von
2, Spieler 2 eine Punkteanzahl von 8, Spieler 3 eine Punkteanzahl von 12,
usw. Angenommen Sie sind Spieler 4, dann würden Sie eine
Punkteanzahl in Höhe von 77 erhalten.
B ITTE BEACHTEN S IE !
Hat die Gruppe eine Alternative gewählt, erhält jeder Spieler die dort
für ihn festgelegte Punkteanzahl. Die Punkte, die einem Spieler
gutgeschrieben werden, sind unabhängig davon, ob er auch für diese
Alternative gestimmt hat. Entscheidend ist, dass die Gruppe diese
Alternative gewählt hat.
In unserem Experiment werden Sie in einer Gruppe abstimmen. Bis sich

die anderen Teilnehmer entschieden haben kann es zu Wartezeiten
kommen.
Wir bemühen uns, diese so kurz wie möglich zu halten. Leider können
wir diese aber nicht vermeiden. Bitte warten Sie in diesem Fall, bis sich
alle Teilnehmer entschieden haben.
-3-
317
Figure A.5: Pooling instructions
D IE G RUPPEN
• Es werden Gruppen mit je 6 Spielern gebildet.
• Die Mitglieder einer Gruppe werden mit Spieler 1, Spieler 2, Spieler 3, Spieler 4, Spieler 5 und Spieler 6 bezeichnet.
• Jedes Gruppenmitglied verfügt über genau eine Stimme.
• Zu Beginn jeder Runde werden die Gruppen zufällig neu zusammengestellt. Sie kennen die Identität der anderen
Mitglieder Ihrer Gruppe nicht.
• Am Ende einer Runde erfahren Sie nur das Ergebnis Ihrer Gruppe, aber nicht die Ergebnisse der anderen Gruppen.
D IE R EGELN
• Alle Spieler stimmen über alle 9 Alternativen ab.
• Das Endergebnis der Runde ist die Alternative, die von einer absoluten Mehrheit - also mindestens 4 Spielern -
gewählt wird.
• Alle Spieler kennen alle möglichen Auszahlungen. Sie wissen also, wie viele Punkte Sie und jeder andere Spieler
bei der Wahl einer Alternative erhalten.
• Im Verlauf einer Runde wird so oft über eine Auszahlungstabelle erneut abgestimmt, bis eine Alternative min-
destens 4 Stimmen in einer Abstimmung erhält. Dann ist die Runde beendet und die Spieler erhalten ihre Punkte
zugewiesen.
• In jeder neuen Runde stimmen die Spieler über eine neue Auszahlungstabelle ab.
• Alle Mitglieder einer Gruppe stimmen in jeder Abstimmungsrunde gleichzeitig ab.
W ELCHE I NFORMATIONEN HABE ICH FÜR MEINE E NTSCHEIDUNG ?

Bevor die Teilnehmer abstimmen, sind sie über die Payoffs aller Mitglieder der Gruppe informiert.
Alle Mitglieder einer Gruppe stimmen in jeder Abstimmungsrunde gleichzeitig ab. Für welche Alternativen die anderen
Spieler derselben Gruppe gestimmt haben, erfahren Sie jeweils erst, wenn alle Mitglieder Ihre Stimme abgegeben haben.
-4-
318
D IE E NTSCHEIDUNGSFINDUNG
Abbildung 4 zeigt den Entscheidungsbildschirm wie Sie ihn auch später an Ihrem PC sehen werden.
Da Ihre Entscheidung auf der Auszahlungstabelle beruht, wird diese für Sie immer eingeblendet. Links davon sehen Sie
Ihre Spielernummer und die Nummer der aktuellen Abstimmungsrunde.
In das freie Textfeld können Sie ihre gewählte Alternative eingeben (Bitte achten Sie bei den Buchstaben auf Großschrei-
bung). Haben alle Mitglieder der Gruppe ihre Wahl getroffen, wird die Abstimmung ausgewertet. Sie erhalten dann
• entweder die Meldung „Ihre Gruppe hat sich entschieden!“ und Ihnen wird das Ergebnis mitgeteilt. Die Spieler
erhalten dann die entsprechende Punkteanzahl gutgeschrieben.
• oder die Meldung „Keine der Alternativen hat die erforderliche Mehrheit von mindestens 4 Stimmen erhalten!“
mit der Aufforderung erneut abzustimmen.
Die Abstimmung wird so lange fortgesetzt, bis eine Alternative die nötige Mehrheit von mindestens 4 Stimmen erreicht.
Sind mehrere Abstimmungen nötig, werden links unten die Ergebnisse früherer Abstimmungen eingeblendet.
Abbildung 4: Entscheidungsbildschirm
-5-
319
Figure A.6: Simultaneous delegation instructions
D IE G RUPPEN
• Es werden Gruppen mit je 6 Mitgliedern gebildet. Jedes Gruppenmitglied verfügt über genau eine Stimme.
• Jede Gruppe wird in zwei Teams aufgeteilt. Jedes Team hat 3 Mitglieder.
• Die Teams haben unterschiedlich Aufgaben. Jedes Team ist für einen Teil der Entscheidungsfindung verant-
wortlich.
– Ein Team muss sich auf die zu wählende Spalte (Buchstaben A - C) verständigen.
– Das andere Team muss sich auf die zu wählende Zeile (Ziffer 1 -3) verständigen.
D IE R EGELN
• Die Spieler können jeweils nur eine der beiden Entscheidungen selbst treffen!
• Es wird die Zeile / Spalte gewählt, die von einer absoluten Mehrheit eines Teams - also mindestens 2 von 3
Spielern - gewählt wird.
• Im Verlauf einer Runde wird so oft erneut abgestimmt, bis eine Zeile / Spalte mindestens diese 2 Stimmen erhält.
Dann ist die Abstimmung des Teams beendet.
• Zu Beginn jeder Runde werden die Gruppen und die Teams zufällig neu zusammengestellt.
W IE WIRD DAS E NDERGEBNIS ERMITTELT ?

Wenn Ihr Team sich geeinigt hat, ist der erste Teil der Entscheidungsfindung abgeschlossen. Für die Bestimmung des
Endergebnisses muss sich nun noch das andere Team entscheiden. Sobald sich beide Teams entschieden haben, wird das
Endergebnis bestimmt und den Spielern mitgeteilt.
-4-
320
Da Ihre Entscheidung auf der Auszahlungstabelle beruht, wird diese für Sie immer eingeblendet. Links davon sehen
Sie Ihre Spielernummer und die Nummer der aktuellen Abstimmungsrunde. Unter Ihrer Spielernummer sehen sie die
Angabe, ob Sie über die Zeilen oder die Spalten abstimmen werden. Zudem werden auch die beiden anderen Mitglieder
Ihres Teams genannt.
In das freie Textfeld können Sie ihre gewählte Alternative eingeben (Bitte achten Sie bei den Buchstaben auf Großschrei-
bung). Wie in Abbildung 4 unten zu sehen ist, dürfen Sie hier nur die Zeile (1, 2 oder 3) wählen. Das andere Team
entscheidet über die Spalte. Haben alle Mitglieder Ihres Teams ihre Wahl getroffen, wird die Abstimmung ausgewertet.
Sie erhalten dann entweder
• die Meldung „Ihr Team hat sich entschieden!“ und Ihnen wird das Ergebnis mitgeteilt oder
• die Meldung „Keine der Alternativen hat die erforderliche Mehrheit von mindestens 2 Stimmen erhalten!“ mit
der Aufforderung erneut abzustimmen.
Sind mehrere Abstimmungen nötig, werden links unten die Ergebnisse früherer Abstimmungen eingeblendet.
-5-
321
Bevor die Teilnehmer abstimmen, sind sie über die Payoffs aller Mitglieder des eigenen Teams und des anderen Teams
informiert. Dabei wissen die Spieler auch, welche beiden anderen Spieler im selben Team sind. Die übrigen drei Spieler
der Gruppe bilden dann das andere Team.
Alle Mitglieder eines Teams stimmen in jeder Abstimmungsrunde gleichzeitig ab. Für welche Alternativen die anderen
Spieler desselben Teams gestimmt haben, erfahren Sie jeweils erst, wenn alle 3 Teammitglieder Ihre Stimme abgegeben
haben. Für welche Alternativen das andere Team gestimmt hat, erfahren Sie erst, wenn sich beide Teams geeinigt haben.
B EISPIEL ZU A BBILDUNG 4
• Das Endergebnis hängt von den Abstimmungen in beiden Teams ab!
• Angenommen Team 1 wählt Zeile 2 und Team 2 wählt Spalte A. Dann ist Alternative A2 gewählt. I n diesem Fall
erhält Spieler 1 4 Punkte, Spieler 2 2 Punkte, Spieler 3 11 Punkte, usw.
• Wählt aber Team 2 statt Spalte A die Spalte B erhält Spieler 1 20 Punkte, Spieler 2 23 Punkt, Spieler 3 15 Punkte, usw.
• Die Punkteanzahl ändert sich also für alle Spieler, auch wenn nur ein Team seine Entscheidung verändert.
Hat das Team eine Zeile / Spalte gewählt, ist diese für alle Mitglieder festgelegt. Dies ist unabhängig davon, ob jedes
Mitglied des Teams auch für diese Zeile / Spalte gestimmt hat. Entscheidend ist, dass das Team diese Zeile / Spalte mit
min. 2 Stimmen gewählt hat.
-6-
322
Figure A.7: Sequential delegation instructions
D IE G RUPPEN
• Es werden Gruppen mit je 6 Mitgliedern gebildet. Jedes Gruppenmitglied verfügt über genau eine Stimme.
• Jede Gruppe wird in zwei Teams aufgeteilt. Jedes Team hat 3 Mitglieder.
• Die Teams haben unterschiedlich Aufgaben. Jedes Team ist für einen Teil der Entscheidungsfindung verant-
wortlich.
– Ein Team muss sich auf die zu wählende Spalte (Buchstaben A - C) verständigen.
– Das andere Team muss sich auf die zu wählende Zeile (Ziffer 1 -3) verständigen.
• Zuerst entscheidet das Team, welches über die zu wählende Spalte abstimmt.
• Das Ergebnis dieser Abstimmung wird allen 6 Mitgliedern der Gruppe (also beiden Teams) mitgeteilt.
• Erst danach stimmt das andere Team über die Zeile ab.
D IE R EGELN
• Die Spieler können jeweils nur eine der beiden Entscheidungen selbst treffen!
• Es wird die Zeile / Spalte gewählt, die von einer absoluten Mehrheit eines Teams - also mindestens 2 von 3
Spielern - gewählt wird.
• Im Verlauf einer Runde wird so oft erneut abgestimmt, bis eine Zeile / Spalte mindestens diese 2 Stimmen erhält.
Dann ist die Abstimmung des Teams beendet.
• Zu Beginn jeder Runde werden die Gruppen und die Teams zufällig neu zusammengestellt.
W IE WIRD DAS E NDERGEBNIS ERMITTELT ?

Wenn das erste Team sich für eine Spalte entschieden hat, ist der erste Teil der Entscheidungsfindung abgeschlossen. Für
die Bestimmung des Endergebnisses muss sich nun noch das zweite Team für eine Zeile entscheiden.
-4-
323
Da die Teams nicht mehr gleichzeitig abstimmen wird Ihnen mitgeteilt, ob ihr Team zuerst abstimmt oder das andere
Team sich bereits entschieden hat. Dann sehen Sie deren Entscheidung. In Abbildung 4 hat das andere Team sich bereits
für Spalte B entschieden. Sie stimmen nun noch über die zu wählende Zeile ab. Haben alle Mitglieder der Gruppe ihre
Wahl getroffen, wird die Abstimmung ausgewertet. Sie erhalten dann
• entweder die Meldung „Ihre Gruppe hat sich entschieden!“ und Ihnen wird das Ergebnis mitgeteilt. Die Spieler
erhalten dann die entsprechende Punkteanzahl gutgeschrieben.
• oder die Meldung „Keine der Alternativen hat die erforderliche Mehrheit von mindestens 2 Stimmen erhalten!“
mit der Aufforderung erneut abzustimmen.
Sind mehrere Abstimmungen nötig, werden links unten die Ergebnisse früherer Abstimmungen eingeblendet. Die Run-
den die das andere Team zur Einigung benötigt hat werden leer angezeigt, da Ihr Team hier noch keine Wahl getroffen
hat.
-5-
324
Bevor die Teilnehmer abstimmen, sind sie über die Payoffs aller Mitglieder des eigenen Teams und des anderen Teams
informiert. Dabei wissen die Spieler auch, welche beiden anderen Spieler im selben Team sind. Die übrigen drei Spieler
der Gruppe bilden dann das andere Team.
Alle Mitglieder eines Teams stimmen in jeder Abstimmungsrunde gleichzeitig ab. Für welche Alternativen die anderen
Spieler desselben Teams gestimmt haben, erfahren Sie jeweils erst, wenn alle 3 Teammitglieder Ihre Stimme abgegeben
haben.
B EISPIEL ZU A BBILDUNG 4
• Das Endergebnis hängt von den Abstimmungen in beiden Teams ab!
• Angenommen Team 1 hat sich bereits für Spalte A entschieden und ihr Team wählt nun Spalte 2. Dann ist Alterna-
tive A2 gewählt. In diesem Fall erhält Spieler 1 59 Punkte, Spieler 2 11 Punkte, Spieler 3 9 Punkte, usw.
• Wählt aber ihr Team statt Zeile 2 die Zeile 3 ist Alternative A3 gewählt und Spieler 1 erhält 2 Punkte, Spieler 2 erhält
31 Punkt, Spieler 3 erhält 21 Punkte, usw.
• Die Punkteanzahl ändert sich also für alle Spieler, auch wenn nur ein Team seine Entscheidung verändert.
Hat das Team eine Zeile / Spalte gewählt, ist diese für alle Mitglieder festgelegt. Dies ist unabhängig davon, ob jedes
Mitglied des Teams auch für diese Zeile / Spalte gestimmt hat. Entscheidend ist, dass das Team diese Zeile / Spalte mit
min. 2 Stimmen gewählt hat.
-6-
325
A.8 All payoff tables used in the experiment
326
EXPLANATORY NOTE
Below each table it is indicated if the payoff table constitutes a constant-sum or non-constant-sum game (cf. Sec. 4.2.4) and if a core alternative for the different procedures exists (cf. Sec. 5.2).
Table A.5: Payoff tables used in the experiment
A B C A B C A B C A B C
Player 1 7 2 1 Player 1 9 4 3 Player 1 10 30 20 Player 1 9 29 19
1 1 1 1
2 2 2 2
3 3 3 3
Payoff table 1 Payoff table 2 Payoff table 3 Payoff table 4
Non-constant-sum table Non-constant-sum table Non-constant-sum table Non-constant-sum table
Core: - Core: - Core: SEQ {A1} Core: SEQ {A1}
1 1 1 1
2 2 2 2
3 3 3 3
Constant-sum table Constant-sum table Constant-sum table Constant-sum table
Core: SEQ {A1} Core: - Core: POL, SIM and SEQ {B2} Core: POL {A3}, SIM and SEQ {C1}
1 1 1 1
1 1 1 1
1 1 1 1
Constant-sum table Constant-sum table Constant-sum table Constant-sum table
327
Core: SEQ {A1} Core: - Core: POL, SIM and SEQ {B2} Core: POL {A3}, SIM and SEQ {C1}
328
1 1 1 1
2 2 2 2
3 3 3 3
Core: POL {A1}, SIM and SEQ {C3} Core: POL and SIM {C2} Core: POL {B3}, SEQ {A3} Core: POL {C2}, SEQ {B3}
1 1 1 1
2 2 2 2
3 3 3 3
Core: POL {A1}, SIM and SEQ {C3} Core: POL and SIM {C2} Core: POL {B3}, SEQ {A3} Core: POL {C2}, SEQ {B3}
A.9 Payoff table characteristics
Table A.6: Payoff table characteristics

EXPLANATORY NOTE
The table describes the distribution of the payoff table characteristics across the decision-making procedures. For each procedure tables with and without a core alternative as well as constant-
sum and non-constant-sum tables exist. Also, every combination of the two characteristics exists for every procedure. The core concept refers under the pooling procedure to the “standard”
deterministic core. For simultaneous and sequential delegation it indicates the existence of the credible core. Due to these differences the table classification differs between procedures.
CORE ALTERNATIVE
CONSTANT- SUM TABLE CONSTANT- SUM TABLE
yes no yes no
pooling 9-12 13-20 5-8 1-4
PROCEDURE simultaneous delegation 9-12 13-16 5-8 1-4 and 17-20
sequential delegation 5, 6, and 9-12 3, 4, 13, 14 and 17-20 7-8 1, 2, 15 and 16
A.10 Frequency of use of each payoff table
Table A.7: Frequency of use of each payoff table

EXPLANATORY NOTE
The table shows the number of observations for each payoff table. Each observation refers to one collective agreement on an alternative. The observations are separated according to the
implemented decision-making procedure. Overall, I conducted 221 rounds of which 76 used pooling, 85 used simultaneous delegation and 60 used sequential delegation. Due to the random
assignment rule the frequency of use differs and not all tables are used under all procedures. This does not constitute inevitably a problem as multiple tables share the same properties; e.g., if a
tables contains a core alternative (cf. Tab. A.6). Sec. A.11 takes a closer look at the actual data structure.
NUMBER OF PAYOFF TABLE
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Σ
pooling 4 0 3 2 4 0 5 0 7 2 7 2 4 4 6 4 7 4 5 6 76
simultaneous delegation 0 10 0 3 0 8 2 3 2 9 0 11 2 9 0 9 2 7 0 8 85
PROCEDURE
sequential delegation 0 4 0 2 2 2 0 2 6 2 4 4 4 4 2 6 2 6 4 4 60
Σ 4 14 3 7 6 10 7 5 15 13 11 17 10 17 8 19 11 17 9 18 221
329
A.11 Data structure
In my 13 experimental sessions I collected, overall, 76 pooling, 85 simultaneous delegation and 60 sequential delegation
observations. Despite the random assignment rule (cf. Sec. 4.2.4) I obtained observations for every possible combination
of table characteristics. However, across all procedures the number for some specifications is rather small and it must
be admitted that the distribution is skewed. Pooling and sequential delegation show a preponderance of core alternative
tables and under simultaneous delegation more observations were collected as first decision rule. In my analysis of
the experimental results in Chap. 5 and Chap. 6, I always state the respective number of observations to secure a valid
investigation.
Table A.8: Data structure according to decision rule and payoff table properties
EXPLANATORY NOTE
The table below separates the 221 collected observations due to procedure, first or second decision rule, if the table con-
tained an equilibrium alternative and if the table represented a constant-sum or non-constant-sum game.
CORE - ALTERNATIVE
DECISION
PROCEDURE
RULE
CONSTANT- SUM TABLE CONSTANT- SUM TABLE
yes no yes no
first 7 11 5 5
pooling second 11 29 4 4
Σ 18 40 9 9
first 15 13 11 21
simultaneous
second 7 7 2 9
delegation
Σ 22 20 13 30
first 10 20 2 6
sequential
second 8 10 2 2
delegation
Σ 18 30 4 8
A.12 Trend analysis of the collective results

The following tables show the results of the trend analysis for the measures obtained in Sec. 5.3.1 regarding decision-
making efficiency, social welfare allocation, distribution of wealth and approval rate. As discussed in Sec. 5.4.1, I analyzed
the development over the single rounds by using STATA’s somersd package (implemented for STATA by Newson, 2002), a
nonparametric rank-statistic. The package performs a maximum likelihood fit to obtain association measures.
Table A.9: Trend analysis

EXPLANATORY NOTE
In each table I list the Somers’ D coefficient, its z-score and 95% CI. The observations are separated by procedure and for
being the first or second decision rule of a session. Also, for approval rate the two delegation procedures show the results
on the delegation level. Statistically significant (two-tailed) at the 0.1 level *, at the 0.05 level ** and at the 0.01 level ***.
Independent variable: NUMBER OF ROUND Dependent variable: APPROVAL RATE
N D z-score 95% CI
pooling 76 0.00 0.74 -0.01 0.01
simultaneous delegation 170 -0.01 -0.16 -0.07 0.06

PROCEDURE first decision rule 120 -0.02 -0.46 -0.11 0.07
sequential delegation 120 0.07* 1.66 -0.01 0.14

330
Independent variable: NUMBER OF ROUND Dependent variable: DECISION - MAKING EFFICIENCY
N D z-score 95% CI
all 221 0.16*** 3.24 0.06 0.26
pooling 76 0.15** 1.98 0.00 0.30

second decision rule 48 0.20* 1.7 -0.03 0.44
PROCEDURE delegation level 170 -0.4 -0.78 -0.14 0.07
second decision rule 50 0.05 0.58 -0.13 0.23
sequential delegation 60 0.09 1.06 -0.08 0.27

Independent variable: NUMBER OF ROUND Dependent variable: SOCIAL WELFARE ALLOCATION

N D z-score 95% CI
pooling 49 -0.02 -0.19 -0.23 0.19
delegation level 100 -0.05 -0.60 -0.21 0.11
PROCEDURE
sequential delegation 38 -0.16 -1.09 -0.44 0.13

Independent variable: NUMBER OF ROUND Dependent variable: DISTRIBUTION OF WEALTH

N D z-score 95% CI
pooling 76 0.08 0.96 -0.08 0.24
delegation level 170 -0.02 -0.37 -0.14 0.09
PROCEDURE
first decision rule 120 -0.14* -1.77 -0.28 0.01

Independent variable: NUMBER OF ROUND Dependent variable: GINI COEFFICIENT

N D z-score 95% CI
pooling 76 0.09 0.95 -0.09 0.26
PROCEDURE

331
A.13 Gini coefficients of the collective results
In Sec. 5.4 I measure the distribution of wealth in the experiment by using the standard deviation and the Gini coefficient of
a selected alternative. The comparison of these two measures increased the validity of my analysis. As the Gini coefficients
mirrored the results of the standard deviation I include it here.
Table A.10: Gini coefficient

EXPLANATORY NOTE
The table shows the Gini coefficient of the collectively selected alternative for the different procedures. The observations
are separated by procedure and for being the first or second decision rule of a session. In addition, I differentiate under
SIM for group and delegation level and under SEQ for first and second stage. The table also shows the Gini coefficient
which would result if the group had always chosen the core alternative. Obviously this includes only the observations
made with tables containing such equilibrium. Please note that the different procedures lead to different core alternatives.
The unit of observation are session averages of the collective decisions per procedure.
pooling 10 0.44 0.13 0.19 0.64
always core 9 0.51 0.20 0.08 0.84

always core 8 0.21 0.08 0.08 0.31

always core 8 0.30 0.07 0.19 0.37
first stage 8 0.37 0.13 0.27 0.66
second stage 8 0.35 0.18 0.11 0.67
332
A.14 Standard experimental games used to measure social preferences
EXPLANATORY NOTE
The table shows standard experimental games used to investigate social preferences. For every game it provides a short description of set-up, typical findings and interpretation. The table
is based on Levitt and List (2007, p. 155, table 1). Another collection can be found in Camerer and Fehr (2004, p. 62). In addition, Ert et al. (2011, p. 257) provided a proper summary of main
deviations from rational choice theory in experimental games.
Table A.11: Experimental games to measure social preferences
Game Summary Typical finding Interpretation
Ultimatum A two-stage game where two people, a proposer and a responder, bargain over a fixed Proposer: Majority of offers in the range Proposer: Fairness
game amount of money. In the first stage, the proposer offers a split of the money, and in the of 25-50% of fixed amount. Few offers Responder: Punish unfair
second stage, the responder decides to accept or reject the offer. If accepted, each player below 5%. Responder: Frequently reject offers: negative reciprocity,
receives money according to the offer; if rejected, each player receives nothing. offers below 20% of fixed amount. fairness, inequity aversion
Dictator A variant of the ultimatum game: strategic concerns are absent as the proposer simply states Above 60% of subjects pass a positive Altruism; fairness
game what the split will be and the proposer has no veto power, rendering the proposed split as amount of money, the mean transfer is preferences, such as
effective. roughly 20% of the endowment. inequity aversion.
Trust game A sequential prisoner’s dilemma game wherein the first mover decides how much money to Proposer: Average transfer of roughly Proposer: Trust; foresee
pass to the second mover. All money passed is increased by a factor, f> 1, and the second 50% of endowment. Responder: positive reciprocity
mover then decides how much money to return to the first mover. In this light, the second Repayment is increasing in transfer. Responder:
mover is a dictator who has been given his endowment by the first mover. Average repayment rate is nearly 50% Trustworthiness, positive
of transfer. reciprocity
Gift Similar to the trust game, but the money passed by the first mover (often labeled the "wage" Proposer: “Wage" or "price" offer is Proposer: Trust; foresee
exchange or "price" offer), is not increased by a factor, rather it represents a pure lump-sum transfer. typically greater than the minimum positive reciprocity
game Also, the first mover requests a desired effort, or quality, level in return for the "wage" or allowed. Responder: Effort or quality Responder:
"price" offer. The second mover then chooses an effort or quality level that is costly to increases in "wage" or "price" offer. Trustworthiness, positive
provide, but increases the first mover’s payoff. reciprocity
Public Generalization of the prisoner’s dilemma. N group members decide simultaneously how Contribution to public good is roughly Altruism; fairness
goods game much to invest in a public good. The payoff function is given by Pi = e − gi + β ∑n g j where 50% of endowment in one-shot games. preferences, conditional
e represents initial endowment; g are the tokens which subject i places in the group account; Many contributions approach 0% in reciprocity
β is the marginal payoff of the public good; and ∑n g j is the sum of the n individual latter rounds of multi-period games.
contributions. By making 0 < β < 1 < n , the dilemma follows.
333
A.15 Predicted probabilities for core solutions at different levels of rationality
334
The figures compare predictions of the sophisticated and the sincere voting model for different levels of random errors. In the sincere model the probability of individual choices follows the
standard conditional logit model. In the sophisticated model the probability of the collective outcome is the likelihood by which at least four players chose the same alternative.
Figure A.8: Predicted probabilities under pooling
Table 1 sincere Table 1 sophisticated

335
336

337
338

339
340

Figure A.9: Predicted probabilities under simultaneous delegation
341
342

343
344

345
346

347
348
Figure A.10: Predicted probabilities under sequential delegation

349
350

351
352

353
354

A.16 Coding guide for the qualitative content analysis
The following table depicts the coding guide used for the qualitative content analysis in Chap. 7. Its draft followed the approach suggested by Mayring (2002). The guide is only available in
German as both the experimental sessions and inquiries were conducted in Heidelberg.
The table states four types of information for every free input field of the post-experiment inquiry. These information are: i) “Kategorie” (German for category), which classifies the type of
quotes found in this field; ii) “Definition” which describes the category; iii) “Ankerbeispiel” (German for anchoring example) which plainly exemplifies what kind of quotes are classified into
this category; iv) “Kodier-Regel” (German for encoding rule) which points out the norms a quote must fulfill to be sorted into this category.
The information was collected by asking these four questions: i) “Please describe your decision rule with your own words?”; ii) “Which players did you focus on?”; iii) “Which characteristic of
an alternative did you consider?”; iv) “What changed when the decision-making procedure split up?”. Tabelle A.15 presents the data grouped according to those questions.
Tabelle A.15: Coding guide for the free input fields
Kategorie Definition Ankerbeispiel Kodier-Regel
Free input field “Please describe your decision rule with your own words?”
“meine Payoffs” Der eigene Payoff ist entscheidend. „Damit meine Auszahlung möglichst hoch wird.“ Nur der eigene Payoff wird explizit
„Ich habe in auf die Entscheidung insistiert, die für mich maximale genannt.
Punktzahl bedeutete.“
“Erfolgs- Der eigene Payoff wird mit der „Überlegt was die anderen wählen würden und danach meine Wahl Der strategische Aspekt muss eindeutig
wahrscheinlichkeit Wahlwahrscheinlichkeit gewichtet, ausgerichtet.“ benannt werden. Das Ziel ist ein hoher
meiner Payoffs” mit welcher die anderen Spieler „Meine beste Auszahlungsmöglichkeit. Dann habe ich berücksichtigt, Payoff.
auch diese Alternative wählen. wie wahrscheinlich es wäre diese auch zu bekommen, wenn man sich
die Möglichkeiten der anderen anschaut.“
“mein Risiko Risikominimierung der eigenen „Möglichst die Spalte gewählt, in welcher für mich keine niedrigen Der strategische Aspekt muss eindeutig
minimieren” Auszahlung. Punkte waren.“ benannt werden. Das Ziel ist kleine Payoffs
zu vermeiden.
“Konsens” Eine Alternative mit fairer „. . . ein Mittelmaß der Auszahlung zu finden, mit dem eigentlich alle Es geht nicht darum das Quorum zu
Verteilung, der alle zustimmen Spieler gut gestellt sind.“ erfüllen (4 von 6). Alle 6 Spieler sollen dem
können. „Alternativen, die gerecht für alle wären.“ Ergebnis zustimmen können.
“Hohe Summe aller Maximierung der sozialen „Die Summe zu maximieren.“ Nicht die Verteilung, sondern das Aggregat
Payoffs” Wohlfahrt. „Maximaler Nutzen für alle.“ ist wichtig.
355
356
Free input field “Which players did you focus on?”
“auf mich” Der eigene Payoff ist entscheidend. „Ich habe versucht, für mich selbst zuerst zu maximieren.“ Nur der eigene Payoff wird explizit
„Ich habe versucht viele Punkte zu sammeln und deswegen habe ich auf genannt.
mich geachtet.“
“auf mein Team” Die Payoffs des eigenen Teams „Hauptsächlich auf die Leute in meiner Gruppe.“ Nur das eigene Team wird genannt.
werden berücksichtigt. „Ich habe versucht mit meiner Entscheidungen eine Wahl zu treffen, bei
der mein Team gut punktet.“
“auf das andere Team” Die Payoffs des anderen Teams „Ich habe geguckt, was die zweite Gruppe wahrscheinlich wählt.“ Nur das andere Team wird genannt.
werden berücksichtigt. „Ich habe darauf geachtet welche Zeile das andere Team ausgehend von
ihren Gewinnchancen wohl wählen wird.“
“alle sechs Spieler” Die Payoffs aller Spieler werden „Die Auswahl bezog sich in erster Linie auf meine Gruppe. Danach sah Beide Teams und / oder alle sechs Spieler
berücksichtigt. ich mir möglichen Entscheidungen der anderen Gruppe an.“ werden genannt.
“genug Spieler in Einzelne Spieler in beiden Teams „Ich habe jeweils die Strategien der anderen eingeschätzt.“ Das Ziel ist das Erreichen des Quorums (2
beiden Teams, welche werden genannt. „Welche Präferenzen andere Spieler haben werden.“ mal 2 von 3 Spielern).
für diese Alternative
stimmen”
Free input field “Which characteristic of an alternative did you consider?”

“meine Payoffs” Der eigene Payoff ist entscheidend. „Ich habe auf die Höhe meines Payoffs geachtet.“ Nur der eigene Payoff wird explizit
„Das eigene Ergebnis war entscheidend und nicht das der anderer.“ genannt.
“mindestens das Verzicht auf den höchsten Payoff „. . . es sei denn es gab eine Alternative, die für mich nur eine gering Der strategische Verzicht muss eindeutig
Zweitbeste meiner um einen hohen Payoff für sich zu schlechtere Auszahlung bedeutete.“ benannt werden.
Möglichkeiten” sichern. „Damit die Wahrscheinlichkeit einen hohen Payoff zu erhalten am
größten war, habe ich nicht immer den höchsten Payoff gewählt.“
“mein Risiko Risikominimierung der eigenen „. . . nicht zu hohes Verlustrisiko.“ Der strategische Aspekt muss eindeutig
minimieren” Auszahlung. „Ich habe geschaut, was für mich dabei rausspringt, aber auch, dass ich benannt werden. Das Ziel ist einen geringen
keinen großen Verlust machen konnte.“ Payoff zu vermeiden.
“Erfolgs- Der eigene Payoff wird mit der „Zwecks Entscheidungsfindung geschaut, wie die Payoffs für alle wären Der strategische Aspekt muss eindeutig
wahrscheinlichkeit Wahlwahrscheinlichkeit gewichtet, und welche Alternative so die am wahrscheinlichste Wahl wird.“ benannt werden. Das Ziel ist das Erreichen
meiner Payoffs” mit welcher die anderen Spieler eines hohen Payoffs.
auch diese Alternative wählen.
“mindestens vier hohe Erreichen des Quorums. „Ich habe eine Möglichkeit gewählt, bei der mindestens 4 Personen hohe Die Verteilung und Summe aller Payoffs ist
Payoffs” Auszahlungen hatten.“ nicht entscheidend. Es wird auf das
„Ich habe geschaut, welche Alternative für mich die Beste gewesen Quorum fokussiert.
wäre, und dann, ob sie denn auch für 3 andere noch in Frage käme“.
“Differenz” Differenz zwischen dem höchsten „Einer ganz hohen und geringen Auszahlung bin ich möglichst aus dem Höchster und kleinster Payoff sind
und dem kleinstem Payoff. Weg gegangen.“ entscheidend.
“Verteilung” Möglichst gleiche Verteilung. „. . . aber möglichst niemand extrem wenige Punkte bekommt.“ Bei der Verteilung werden alle sechs Spieler
„. . . versucht geringe Auszahlungen für Spieler zu vermeiden.“ berücksichtigt.
Free input field “What changed when the decision-making procedure split up?”
“es änderte sich nichts” Keine Veränderung. „kein Unterschied“ Nur der eigene Payoff wird explizit
„Gleiches Prinzip“ „Viel mehr auf meine eigenen Payoffs, vor allem ein genannt.
möglichst hohes Minimum.“
“mehr Egoismus” Der eigene Payoff wird wichtiger. „Da habe ich hauptsächlich nur auf meine Punkte geachtet, da die Veränderung muss explizit genannt
Entscheidung eingeschränkt wurde.“ werden. Betonung des eigenen Payoffs.
“mehr Kompromiss” Die Zustimmung aller Spieler wird „Es ging weniger um den eigenen Vorteil.“ Veränderung muss explizit genannt
wichtiger. „Ich habe eher Kompromisse gemacht.“ werden. Konsens muss betont werden.
“mehr mein Team im Die Teammitglieder sind jetzt „Mehr auf die Auszahlung meiner Teammitglieder.“ Veränderung muss explizit genannt werden.
Vordergrund” wichtiger. „. . . habe ich mehr auf meine Teammitglieder bzw. deren Punktzahlen Abgrenzung des eigenen Teams nötig.
geachtet.“
357
A.17 References of experiments conducted online and in the
laboratory
EXPLANATORY NOTE
In December 2012 Israel Waichman (Department of Economics, Heidelberg University) conducted a survey via the Eco-
nomic Science Association (ESA) Google Group discussion list “ESA Experimental Methods Discussion” 461 . He asked
all subscribers for references to studies comparing between experiments conducted in the laboratory and via the internet.
The listing below resembles his posted results (Google Group entry from Wednesday 2nd January 2013 13:15).
Amir, O., D. G. Rand, and Y. K. Gal. 2012. “Economic games on the internet: The effect
of $1 stakes”. PLOS ONE 7.
Anderhub, V., R. Müller, and C. Schmidt. 2001. “Design and evaluation of an economic
experiment via the internet”. Journal of Economic Behavior and Organization 46:227-
247.
Bellemare, C. and S. Kröger. 2007. “On Representative Social Capital”. European Eco-
nomic Review, 51:183-202.
Birnbaum, M. H. 2000. Psychological Experiments on the Internet. Academic Press.
Bolton, G., B. Greiner, and A. Ockenfels. 2012. “Engineering Trust: Reciprocity in the
Production of Reputation Information”. Management Science.
Bosch-Domènech, A., J. García-Montalvo, R. Nagel, and A. Satorra. 2002. “One, Two,
(Three), Infinity. . . : Newspaper and Lab Beauty-Contest Experiments”. American
Economic Review 92:1687-1701.
Bosch-Domènech, A., and R. Nagel. 1997 “Cómo se el da la bolsa,” Expansión, June 4:40.
Bosch-Domènech, A., and R. Nagel. 1997. “El juego de adivinar el numero X: una expli-
cación y la proclamación del vencedor,” Expansión, June 16:42-43.
Bosch-Domènech, A., and R. Nagel. 1997. “Guess the Number: Comparing the FT’s and
Expansion’s Results,” Financial Times, Section Mastering Finance 8, June 30:14.
Buchanan, T. and J. L. Smith. 1999. “Using the internet for psychological research: per-
sonality testing on the World Wide Web”. British Journal of Psychology 90:125-144.
Cassar, A. 2007. “Coordination and cooperation in local, random and small world net-
works: Experimental evidence”. Games and Economic Behavior 8:209-230.
Cabrales, A., and R. Nagel. 2002. “Ein Gewinnspiel zur Entscheidung unter Unsicher-
heit”. Spektrum der Wissenschaft (German issue of Scientific American) Nov.
Cabrales, A., and R. Nagel. 2003. “Entscheidung unter Unsicherheit – die Ergebnisse”.
Spektrum der Wissenschaft (German issue of Scientific American) May.
Charness, G., E. Haruvy, and D. Sonsino. 2007. “Social distance and reciprocity: An
internet experiment”. Journal of Economic Behavior & Organization 63:88-103.
Chesney, T., S.-H. Chuah, and R. Hoffmann. 2009. “Virtual world experimentation: An
exploratory study”. Journal of Economic Behavior & Organization 72:618-635.
Cleave, Blair L., N. Nikiforakis, and R. Slonim. 2012. “Is there Selection Bias in Labora-
tory Experiments? The Case of Social and Risk Preferences”. Experimental Economics
(forthcoming).
461 Thediscussion can be viewed after prior registration under https://groups.google.com/d/msg/
esa-discuss/-/iTvrjXqCod4J.
358
Coffman, L. C. 2011. “Intermediation reduces punishment (and reward)”. American Eco-
nomic Journal: Microeconomics 3:77-106.
Costa, Gomes M., V. Crawford, and R. Nagel. 2007. “Experiment teilzunehmen”. Spek-
trum der Wissenschaft (German issue of Scientific American) July:96-97.
Costa, Gomes M., V. Crawford, and R. Nagel. 2008. “Ergebnisse zum Preisausschreiben“.
Spektrum der Wissenschaft (German issue of Scientific American) February:76-80.
Dürsch, P., J. Oechssler, and B. C. Schipper. 2009. “Incentives for subjects in internet
experiments”. Economics Letters 105:120-122.
Dutcher, E. Glenn. 2012. “The effects of telecommuting on productivity: An experimen-
tal examination. the role of dull and creative tasks”. Journal of Economic Behavior &
Organization 84:355-363.
Dutcher, E. Glenn and Krista Jabs Saral. 2012. “Does team telecommuting affect produc-
tivity? An experiment”. mimeo. https://sites.google.com/site/glenndutcher/
Home/research?pli=1.
Eckel, C. C., and R. K. Wilson. 2006. “Internet Cautions: Experimental Games with
Internet Partners”. Experimental Economics 9:53-66.
Egas, M. and A. Riedl. 2008. “The economics of altruistic punishment and the mainte-
nance of cooperation”. Proceedings of the Royal Society B 275:871-878.
Fiedler, M. and E. Haruvy. 2009. “The lab versus the virtual lab and virtual field - an ex-
perimental investigation of trust games with communication”. Journal of Economic
Behavior and Organization 72(2):716-724.
Füllbrunn, S., K. Richwien, and A. Sadrieh. 2011. “Trust and trustworthiness in anony-
mous virtual worlds”. Journal of Media Economics 24:48-63.
Gosling, Samuel D., Simine Vazire, Sanjay Srivastava, and Oliver P. John. 2004. “Should
We Trust Web-Based Studies? A Comparative Analysis of Six Preconceptions about
Internet questionnaires”. American Psychologist 59: 93-104.
Jerit, J., J. Barabas, and S. Clifford. 2012. “Comparing Contemporaneous Laboratory
and Field Experiments on Media Effects”. mimeo. Department of Political Science
at the Florida State University. http://scottaclifford.com/wp-content/uploads/
2012/12/JeritBarabasClifford_Comparing_Experiments.pdf.
Hergueux, J. and N. Jacquemet. 2012. “Social preferences in the online laboratory: A
randomized experiment”. CES working papers 2012 70.
Ho, B., J. Taber, G. Poe, and A. Bento. 2012. “Moral Licensing and Moral Cleansing
in Contingent Valuation and Laboratory Experiments of Willingness to Pay to Re-
duce Negative Externalities”. mimeo. Vassar College, Economics Department. http://
idei.fr/doc/conf/bee2012/Culpability%20CV%20experimental%206-8-12.pdf.
Horton, J. J., D. G. Rand, and R. J. Zeckhauser. 2011. “The online laboratory: conducting
experiments in a real labor market”. Experimental Economics 14:399-425.
Lyons, B., G. Menzies, and D. Zizzo. 2012. “Conflicting evidence and decisions by
agency professionals: an experimental test in the context of merger regulation”.
Theory and Decision 73:465-499.
Nagel, R., and R. Selten. 1997. “1000 DM zu gewinnen”. Spektrum der Wissenschaft (Ger-
man issue of Scientific American) Nov.
359
Nagel, R., and R. Selten. 1998 “Das Zahlenwahlspiel - Hintergruende und Ergebnisse”
Spektrum der Wissenschaft (German issue of Scientific American) Febr.:16-22.
Paolacci, G., J. Chandler, and P. G. Ipeirotis. 2010. “Running experiments on Amazon
mechanical turk”. Judgment and Decision Making 5:411-419.
Potters, J. and S. Suetens. 2009. “Cooperation in experimental games of strategic com-
plements and substitutes”. Review of Economic Studies 76:1125-1147.
Rand, D. G. 2012. “The promise of Mechanical Turk: How online labor markets can help
theorists run behavioral experiments”. Journal of Theoretical Biology 299: 172-179
Rand, D. G., J. D. Greene, and M. A. Nowak. 2012. “Spontaneous giving and calculated
greed”. Nature 489:427-430.
Reips, U.-D. 2001. “The Web Experimental Psychology Lab: Five years of data collection
on the Internet”. Behavior Research Methods, Instruments, and Computers 33(2):201-11.
Schmelz, K. and A. Ziegelmeyer. 2012. “Reactions to control with(out) background
control: Evidence from the internet and the laboratory”. mimeo. Working paper.
Max Planck Institute of Economics. http://www.econ.mpg.de/files/2011/staff/
schmelz/Background_Control_Schmelz_Ziegelmeyer.pdf.
Schulte-Mecklenbeck, M. and O. Huber. 2003. “Information search in the laboratory
and on the web: With or without an experimenter”. Behavior Research Methods,
Instruments, and Computers 2:227-235.
Shavit, T., D. Sonsino, D., and U. Benzion, U. 2001. “A comparative study of lotteries-
evaluation in class and on the web”. Journal of Economic Psychology 22:483-491.
Suri, S. and D. J. Watts. 2011. “Cooperation and contagion in web-based, networked
public goods experiments”. PLOS ONE 6:e16836.
von Gaudecker, H.-M., A. van Soest, and E. Wengstroem. 2012. “Experts in experiments:
How selection matters for estimated distributions of risk preferences”. Journal of
risk and uncertainty 45:159-190.
360

Fleig NSP Final A

Uploaded by

Copyright:

Available Formats

Fleig NSP Final A

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fleig NSP Final A

Uploaded by

Copyright:

Available Formats

Collective Decisions on

ZUR ERLANGUNG DES AKADEMISCHEN GRADES

FAKULTÄT FÜR WIRTSCHAFTS- UND SOZIALWISSENSCHAFTEN

Heidelberg, Juni 2013

Prof. Dr. Jürgen Eichberger, Department of Economics, University of Heidelberg

Prof. Dr. Daniel Finke, Department of Political Science, University of Heidelberg

Prof. Christoph Vanberg, Ph.D., Department of Economics, University of Heidelberg

Date of Disputation: 04.10.2013

• Fleig, Andreas. 2011a. The Merits of Adding Complexity: Conditional Preferences

• Fleig, Andreas. 2011b. Pooling vs. Delegation: Collective decision-making under

• Fleig, Andreas. 2012. Motivated by the Process: How decision-making procedures

Analytical politics investigates collective decision-making in political systems. Such vot-

Die Analytische Politikwissenschaft befasst sich mit kollektivem Entscheidungsverhalten

List of Tables vii

2 What are nonseparable preferences? 23

3 The merits and costs of incorporating nonseparable preferences 53

5 The influence of institutional rules on collective decisions 117

6 The determinants of individual choices 151

7 Post-experiment survey 193

8 Complex reality and limited models 207

2.1 Separable and nonseparable preference orders . . . . . . . . . . . . . . . . 25

3.1 Extent of nonseparability in EU law-making . . . . . . . . . . . . . . . . . 67

4.1 Treatment characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

5.1 Two-way error components model . . . . . . . . . . . . . . . . . . . . . . . 120

6.1 Co-variates for the random utility mixture model . . . . . . . . . . . . . . 175

7.1 Results of the binary questionnaire . . . . . . . . . . . . . . . . . . . . . . . 195

8.1 Methods of data collection for investigating nonseparability . . . . . . . . 211

9.1 Pros and cons of internet experiments . . . . . . . . . . . . . . . . . . . . . 231

A.2 Structure of the DEU expert interviews . . . . . . . . . . . . . . . . . . . . 305

2.1 A simple spatial model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.1 Model predictions for COM1999/163 with separable preferences . . . . . 61

4.1 Payoff scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5.1 Example of the determination of the core alternative . . . . . . . . . . . . . 126

6.1 Robustness of the core alternative . . . . . . . . . . . . . . . . . . . . . . . . 169

7.1 WORDFISH : latent discrepancy dimension . . . . . . . . . . . . . . . . . . . 201

A.1 Three-dimensional contour plots for Council regulation COM1999/163 . . 306

a Single element of the matrix of saliences

ACP African, Caribbean and Pacific Group of States

AIC Akaike information criterion, a measure of the relative goodness of fit of a

ALDE Alliance of Liberals and Democrats for Europe

AMT Amazon Mechanical Turk

APM Accumulated payoff mechanism

BW German Federal Armed Forces (translation of Deutsche Bundeswehr)

CP-nets Conditional preference networks

CRRA Constant relative risk aversion

D Somers’ D, an ordinal measure of association introduced by Somers (1962)

D+ A measure for advantageous inequality in the inequality averversion model

D- A measure for disadvantageous inequality in the inequality averversion model

DEU Decision-making in the European Union

DGP Data-generating process

DVPW German association for political science (translation of Deutsche Vereinigung

ECHA European chemicals agency

ECPR European consortium for political research

EDE Experimenter demand effect

ERC Equity, reciprocity and competition

ESA Economic Science Association

ESM European stability mechanism

FBNE Feedback Nash equilibrium