The Politics of Police Violence in Democratic Brazil
Ahnen, Ronald E.
Latin American Politics & Society, Volume 49, Number 1, Spring
2007, pp. 141-164 (Article)
Published by University of Miami
DOI: 10.1353/lap.2007.0000
For additional information about this article
http://muse.jhu.edu/journals/lap/summary/v049/49.1ahnen.html
Access Provided by Saint Marys College of California at 10/27/11 1:04AM GMT
The Politics of Police Violence
in Democratic Brazil
Ronald E. Ahnen
ABSTRACT
More than a decade after Latin America’s most recent turn to
democracy, unchecked police violence and torture continue and in
some cases have increased. This study examines police killings in
19 Brazilian states from 1994 to 2001 and finds that democracy has
not substantially reduced these types of human rights violations, for
two reasons. First, underlying social conflict has continued to exert
a significant impact on the lethal use of force by police officers.
Second, pro-order political coalitions, generally represented by
right-wing politicians, have blocked effective measures to control
police violence and have implemented public safety measures that
stress the use of force. The analysis emphasizes the nonteleological
nature of democratization processes and demonstrates the strength
of political forces working to maintain “illiberal democracy.”
S
ince the fall of the Brazilian military dictatorship and the return to
democracy in 1985, human rights conditions in that country can best
be described as very grave and, in some places, degenerating. Of
course, the human rights agenda no longer focuses on the rights of
politicians to criticize the state openly, but instead on the torture, disappearance, and assassination of common citizens by the military and
civil police forces or other state agents. Studies indicate that since the
return to democracy, the rate at which Brazilian police officers kill or
torture their fellow citizens has not only risen substantially in some
locales, but clearly constitutes excessive use of force and therefore
abuse of basic rights (Caldeira 2000; Chevigny 1995; Méndez et al. 1999;
NEV 1993). For example, in 1992, according to official statistics, São
Paulo police were responsible for 1,470 civilian killings, or roughly 16
percent of all homicides in that state (Holston and Caldeira 1998).
Both conventional wisdom and several cross-national studies suggest, however, that democratic governance is significantly associated
with less human rights abuse, especially in terms of the violation of a
person’s physical integrity by state agents (Davenport 1995, 1999; Fein
1995; Henderson 1991; Poe et al.1999; Poe and Tate 1994). At the heart
of this democratic proposition lies a simple logic: when given a
choice, voters will turn out of office leaders who do not rein in police
abuse against common citizens like them. In addition, democratic
institutions underscore negotiation, dialogue, and nonviolent resolu141
142
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
tion of conflict, which should also lessen human rights violations
(hereafter HRVs).
Brazil therefore presents a particularly puzzling case, because HRVs
have increased since redemocratization. This study demonstrates that
Brazil’s fluctuations in police violence occurred not in spite of its transition to democracy but partly because of it. Democracy gives citizens
the opportunity to elect leaders who offer a heavy hand (mão forte) in
dealing with the problem of crime and violence. When such candidates
take office, they often institute changes in public safety policy that
repress crime in ways that violate citizens’ basic rights.
This study examines the rate of police killings in 19 Brazilian states
from 1994 to 2001 in light of partisan electoral politics. The two major
police forces in each state are the military and the civil police. Each has
a different role. The military police are responsible for day-to-day policing of the streets, both preventing and repressing crime. The civil police
are also known as the investigative or judiciary police, for their role is
limited to criminal investigations. Because the police forces in Brazil are
organized administratively at the state level of government, this study
focuses on the political, ideological, and partisan identification of state
governors. It posits that the partisanship of government leaders is a key
explanatory variable in determining the rate that police officers employ
deadly force. The study’s findings show that the rate of police killings is
significantly lower in Brazilian states that are led by an executive from
the left or center left as opposed to the center, center-right, or right.
This study begins with a review of the literature on cross-national
studies of human rights violations in general, and studies of human
rights and police violence in Brazil in particular. It proposes a basic
model of police violence in Brazil that takes electoral incentives into
account. The model is tested using two estimates of the number and rate
(per one hundred thousand residents) of police killings, and an estimate
of the percentage of all reported homicides attributed to the police. The
study finds that partisanship remains a robust determinant of the rate of
killings by both the military and civil police forces.
RESEARCH
ON
HUMAN RIGHTS
In examining the literature on human rights abuse, the discussion here
is limited to violations of physical integrity of the person; that is, freedom from torture, disappearance, and extrajudicial killings. This literature can best be divided, according to methodology, into two camps:
cross-national analyses and case studies. In the past two decades, scholars conducting cross-national research on human rights abuse have
made significant improvements both in methodology and in greater coverage of countries and time periods (Davenport 1995, 1999; Fein 1995;
AHNEN: POLICE VIOLENCE IN BRAZIL
143
Henderson 1991; Poe et al. 1999; Poe and Tate 1994). As Davenport
points out, these studies are grounded in rational choice theory, and
conceptualize political repression as
a strategic choice that governments make among a number of
options in an attempt to bring about and maintain political quiescence. Governments are expected to employ repressive behavior
when the value of quiescence and the probability of success are
high and costs are low. Governments are less likely to violate human
rights, however, when the value of quiescence and the probability
of success are low and costs are high. (Davenport 1999, 95)
A key finding of these studies is that democratic governance has an
independent and significant impact on reducing human rights violations. It is interesting, however, that these studies differ substantially
regarding the theoretical relationship between these variables. At least
six different explanations have been offered to account for this positive
association. Democratic leaders are more responsive to citizen demands,
thus attenuating social conflict and eliminating the need for repression
(Henderson 1991). Democracy provides nonviolent channels to settle
disputes when they arise (Henderson 1991). Democracy allows for the
removal of leaders who would be ready to use repression against the
citizenry (Poe and Tate 1994). The civil liberties necessary for democracy allow for widespread revelations by the media of the activities of
human rights abusers, thus bringing the weight of national and international opinion to bear on government leaders and lowering the propensity to abuse basic rights (Poe and Tate 1994). Citizens in democracies
are socialized, and operate in a political culture that values nonviolence
in dispute settlements (Mitchell and McCormick 1988). Finally, civil
society in democracies is freer and better able to create mechanisms of
control over the state’s monopoly of the legitimate use of force (Chevigny 1995).
This literature suffers from two serious problems. First, as noted
above, the relationship between democracy and human rights violations
is overdetermined. The difficulty is that even when a statistically significant association between the variables is found, the researcher is still
ignorant as to which of the six causal mechanisms (or which combination of the six) is responsible for generating the results. To resolve the
problem, researchers must utilize alternative indicators that more closely
approximate the specific theorized cause linked to democracy.
The second problem is that the opposite direction of the hypothesis
is rarely entertained in the literature on human rights, at least not theoretically; that is, that human rights conditions may deteriorate with the
onset of democracy. Nevertheless, several studies include theoretical
notions that support such a finding. One exception is Fein’s 1995 study
144
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
of human rights abuse and regime transitions. She finds that human rights
violations are greater under nascent democratic regimes than under either
established authoritarian or democratic regimes. Fein explains that the
transition to democracy opens new avenues (both legal and political) for
the excluded masses to rebel against the propertied class. Repression
increases as the latter fears the victory of a populist government, potentially directly threatening their interests. In addition, several recent studies
of democracy in Latin America demonstrate how democratic governance
in the region does not function adequately to ensure the protection of
basic freedoms enjoyed more widely in the postindustrial democracies
(Collier and Munck 2001; Collier and Levitsky 1997; Diamond 1996, 1999;
Dimenstein 1996; O’Donnell 1993, 1998, 1999).
Democracy may be associated with an increase of HRVs for several
reasons. As Fein theorizes, democracy may provide the opportunity for
the marginalized classes to gain a populist victory, thus creating incentives for the middle and upper strata to employ repression in efforts to
prevent it (Fein 1995). In a climate of growing crime and violence, citizens in newer democracies may vote for candidates who promise to
reestablish order through greater use of repression. Another reason is
that civil society in newer democracies is often more fragile because of
a lack of resources and leadership experience, and therefore less likely
to fulfill its role of holding abusive state agents accountable for their
actions. The deteriorating economic and social conditions that helped
usher in the transition to democracy can foster popular discontent and
the concomitant use of repression by the elite to maintain public order.
Furthermore, state institutions designed to protect individual rights (e.g.,
the judicial system) are often weak or only unevenly available to citizens, thereby allowing state agents to engage in repression with
impunity and encouraging such action in the future.
Given these several reasons that democracy should help to improve
human rights conditions and another several to explain why it might
not, I argue that the impact of democracy on respect for basic rights is
indeterminate a priori, and depends on the social, political, and economic context in which it operates.
This study seeks to make a significant contribution to the debate on
the causal mechanisms by which democracy affects human rights conditions by focusing on partisanship and electoral politics with respect to
public safety policies while controlling for important social and economic conditions. By employing variables that are more closely linked
to the theorized propositions, we can surmount the problem of overdetermination and deepen our understanding of the specific links between
democracy and human rights violations.
In addition to regime type, several structural variables have received
wide support in the literature on HRVs because they alter the costs and
AHNEN: POLICE VIOLENCE IN BRAZIL
145
benefits of repression in the eyes of government leaders. Lower levels
of economic development, for example, are associated with greater
repression because scarcity is higher, and therefore conflict between the
haves and have-nots is more acute (Poe and Tate 1994). Rapid rates of
economic development generally increase repression because such
growth is often uneven, generating swift economic dislocation and
greater conflict among citizens (Olson 1963). Both poverty and inequality exacerbate the poor’s jealousy of the rich, prompting the former to
rise up against the latter and generating incentives for greater repression
(Henderson 1991). Cano and Santos (2001) employ urbanization as a
key determinant of the general homicide rate in Brazilian states. It can
be argued, however, that higher levels of urbanization are expected to
be associated with greater police violence because the highly dense
urban areas are where most of the drug-trafficking networks and other
criminal activity that the police repress are found.
Turning to case studies, scholars focusing on police violence under
democracy in Brazil have, to date, put forth only broad sociopolitical
explanations of human rights abuse. Chevigny (1995), for example,
offers the pacification thesis. Here, police violence is viewed as the
product of the natural progression from a traditional society to a more
modern one. During an initial stage, citizens use force individually to
provide their own security. At a subsequent stage, these citizens become
horrified at the levels of violence in their society and decide to restrict
their ability to use force against one another (e.g., by adopting strict gun
laws), thereby granting a greater monopoly on the use of force to agents
of the state. Chevigny argues that citizens accept these restrictions only
in exchange for greater control over their police forces.
Pereira (2000) argues that Brazil’s bifurcated society has led to a
type of hybrid regime type he calls “elitist liberalism.” In his view,
democracy (and its concomitant respect for individual rights) is firmly in
place for the upper and middle classes of Brazil, but is only weak and
fragile for the poor masses. With highly limited access to the state institutions designed to protect their rights, the poor bear the brunt of police
violence, which the elite employ to obtain the poor’s quiescence in
Brazil’s extremely unequal society. Pereira’s argument suggests that
human rights conditions will improve only when elitist liberalism is
overthrown and the principles of democracy are extended to protect all
citizens. Neither of these theories, however, sheds light on how positive
transformations toward greater protection of rights might occur. Most
important, as phenomena of society as a whole, such theories cannot
account for the ample variation in police violence observed in Brazil.
Other case studies of police violence in Brazil emphasize how institutional and policy factors play a significant role in reducing such violence. These factors include improved police training, higher salaries,
146
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
stricter use of force guidelines, and reduced levels of impunity (Bicudo
1994; Cano 1997; de Amorin Ratton 1996; Mesquita Neto 1999; O’Donnell 1999; Pinheiro 1996). While these factors probably do shape the
propensity of police officers to use deadly force, reliable and comparable data on these variables across the Brazilian states are not available.
More fundamentally, however, these policy prescriptions are probably
tied to the political project of government leaders according to their ideological leanings.
PUBLIC SECURITY, ELECTORAL INCENTIVES,
AND PARTISANSHIP
Issues of crime, public safety, and the arbitrary use of (often lethal) violence are pressing electoral issues that one can reasonably expect voters
to consider when casting their vote. Indeed, then presidential candidate
José Serra declared public security the most important issue of the 2002
presidential campaign (Jornal do Brasil 2002). The key theoretical contribution of the present analysis is that it does not assume a strong
homogeneous concern across all citizens for greater protection of rights,
as the logic laid out by Davenport above suggests. Instead, the public
safety issue is problematized, conceptualized as a political partisan issue
that boils down to stressing one of two major goals.
The first goal is greater control of crime and increased public safety.
Often, leaders confronted with rising crime reach toward a simple solution: increased (and often more violent) police action. Concern for individual rights (especially the rights of those being pursed by the police)
can weaken in such a context, even to the point of being nonexistent.
Politicians who take this stance are often labeled law and order candidates, though their policy prescriptions are most often heavy on order
and very light on law. This position can be called the public order position. Governor Luís Antonio Fleury of São Paulo (1990–94) promulgated
this view.
[t]he fact that this year there were more deaths caused by the [military police] means that they are more active. The more police in
the streets, the more chances of confrontations between criminals
and policemen. . . . From my point of view, what the population
wants is that the police act boldly. (Quoted in Holston and Caldeira
1998, 272)
The opposing viewpoint is that the problems of crime and violence
must not be solved by trampling the very rights that engendered the
struggle for democracy and against the military dictatorship. Candidates
who profess this view emphasize rooting out police corruption, retraining police officers, creating an ombudsman to hear civilian complaints
AHNEN: POLICE VIOLENCE IN BRAZIL
147
against the police, and so on as major public safety platforms. More
broadly, these candidates argue that social and economic reforms, especially more resources for education, health, and employment generation, hold the greatest promise for solving the crime problem. This can
be called the individual rights position.
More sophisticated versions of this latter position marry the twin
goals of police restraint and reduced crime, claiming that crime will be
reduced only when the police forces consist of honest individuals acting
in a professional manner, equipped with more sophisticated tools than
just a handgun. For example, in his book on public safety, which helped
launch his successful gubernatorial bid in Rio de Janeiro in 1998,
Anthony Garotinho declared, “We propose a broad coalition and
democracy against barbarity: 1) the law is the limit of authority, 2) with
respect to public safety, the trade-off between efficiency and civil rights
is false” (Garotinho et al. 1998). Whether such a trade-off actually exists
or whether it is possible to marry the two goals is irrelevant to the present analysis. The trade-off often appears real to candidates and voters
alike in considering their preferred positions and making appeals or in
making voting decisions.
The key hypothesis in this study is that governors on the right or
center-right favor the public order position and that therefore their states
exhibit higher rates of police killings, while politicians on the left or
center-left favor the individual rights position, and their states have
lower rates of police killings. The reasoning behind this hypothesis is
found in the historical roots of contemporary conservative and liberal
philosophies. Heywood (2003) notes, for example, that traditional conservative views stress that human nature is “imperfect” and that individuals are generally motivated to act based on their “fear [of] isolation and
instability” (75–76). Thus, conservatives emphasize the need for maintaining social order through a rigorous application of the law, even at
the expense of some liberty. Liberal philosophy, on the other hand,
posits the primacy of the individual to act as he or she sees fit. Left to
themselves, individuals will behave well, and the government would do
best to control the actions of individuals as little as possible (Heywood
2003, 28–30). Thus, the leftist approach to public safety policy emphasizes protecting the rights of individuals apprehended by the police and
the need to reduce and punish possible excess use of force against citizens by police officers.
It is interesting that these views do not overlap neatly with economic class. As Mainwaring et al. (2000) show, support for the left parties often comes from the middle class, while the conservative parties
have a strong base in the more underdeveloped areas of the country
(68–71). This finding echoes poll research indicating that the acceptance
of violence as a means of social control among the citizenry is stronger
148
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
in underdeveloped areas in Brazil. In 2002, for example, 29.4 percent of
Brazilian respondents overall indicated that it is either right or usually
right for a police officer to kill a suspect after being apprehended, but
the respective average for the north and northeastern regions of the
country was 36.4 percent and for the remaining regions 26.0 percent.
Likewise, when asked about whether the police could physically beat a
suspect in order to obtain a confession, 35.9 percent of all Brazilians felt
that this was right or usually right, but the percentage rose to an astonishing 45.1 percent in the northern and northeastern states and fell to
35.5 percent in the other states (PESB 2002).
CONTROL VARIABLES
Previous analyses have found support for a number of factors as significant determinants of human rights abuse, including lower levels of economic development, rapid rates of economic expansion, higher rates of
poverty and inequality, the presence of civil or international war, and
greater population density. This study uses five control variables:
•
•
•
•
•
Number of nonpolice homicides in the state
Illiteracy rate
Infant mortality rate
Level of inequality
Population density
Police violence is expected to increase in the face of higher homicide rates as governors feel the heat from voters to provide adequate
public security. The remaining four variables control for social conditions that the literature has widely suggested exhibit a positive association with police violence. High rates of illiteracy and infant mortality are
chosen as indicators of lower levels of economic development. We
would expect the use of deadly force by police officers to increase
where illiteracy and infant mortality are higher because social conflict
among classes is sharper at lower levels of development. In addition, a
less formally educated citizenry will be less capable of organizing itself
to serve as an oversight body of the police forces. More police killings
are also expected where inequality is higher because the poor become
jealous of the rich and rebel against the social system that has left them
materially behind. In response, the rich attempt to thwart their efforts
via repression.
With respect to population density, two hypotheses are possible.
First, we might expect more police killings in the less densely populated
areas of the country, where politicians often act in collaboration with
local judges and law enforcement to control what is essentially a small
AHNEN: POLICE VIOLENCE IN BRAZIL
149
fiefdom. On the other hand, we might expect more police killings in the
more densely populated areas as a reaction to a greater intensity of violent criminal activity. Thus the expected direction of the relationship
between population density and police violence remains undetermined.
DATA
AND
METHODOLOGY
The most consistent and reliable figures on police killings in Brazil come
from the Database on Criminal Violence (Banco de dados sobre violência, BDV) maintained by the National Human Rights Movement (Movimento Nacional de Direitos Humanos, MNDH), in Brasília. The BDV
consists of data about all homicides reported in several newspapers in
each state to ensure the most complete and comprehensive account of
homicides possible. Data collection is carried out by local human rights
organizations affiliated with the MNDH. Collection began in 1992 with
only 5 of 27 states participating that year and the next. Participation has
waxed and waned to the present, with some states joining the project
while others have dropped out. The original database contained information on 107,984 homicides, including 8,035 attributed to police officers and municipal guards from 1994 through 2004. Since the inclusion
of states and years is unbalanced, this study attempts to minimize bias
by including only those states that reported data for at least four years.
The resulting dataset contains information on 99,724 homicides, including 7,834 alleged killings by police officers reported in newspapers in
19 states between 1994 and 2001. The analysis uses a pooled time series
regression analysis with panel corrected standard errors on a dataset of
127 cases.
Previous research has demonstrated convincingly that official data
on police killings in Brazil are incomplete and unreliable. The federal
government has only recently begun to compile public security statistics
nationally (Ministry of Justice 2002). Unfortunately, even these federal
government tallies take for granted the reliability of the statistics
reported by states. While the federal government does maintain relatively reliable data on general homicides for all states, these figures are
not sorted by alleged perpetrator.
In his very careful study of military police violence in Rio de
Janeiro, Cano (1997) shows that the methodology employed to record
police killings is quite unreliable, especially due to duplicate paperwork
filed for many cases (15–18). Cano painstakingly corrected the accounting errors regarding homicides by the military police from 1993 to 1996
by reviewing each case of the use of force. His research provides the
most meticulously corrected and accurate data available on police
killings for Rio de Janeiro to date. Unfortunately, however, his data
cover only killings by the military police from 1993 to the first half of
150
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
1996, and data from the BDV overlap with his dataset for only three
years (1994–96). Nevertheless, the coefficient of correlation between
Cano’s data and the BDV is .972, further suggesting that the BDV data
are both accurate and reliable.1
The BDV dataset comprises more than 40 separate categories of
police officers, disaggregating, for example, instances in which a military or civil police officer is on or off duty, and cases in which the officer is acting with others, such as criminals, death squads, and private
security guards. Information about which police force (military or civil)
the alleged perpetrator belongs to is often unavailable. The current
analysis collapsed the BDV categories into only three classifications: military police (including any category that includes the military police per
se, whether on or off duty or acting with other agents), civil police
(including all categories that include civil police except those that combine military and civil police, which are included in the former), and
other police (where no indication of force is given). The analysis is separated into three parts: killings by the military police, the civil police,
and all police (includes military, civil, and the residual police categories
summed together).
The use of data gathered from newspapers has the danger of introducing a bias among the more populous states and thereby the question of reliability. Specifically, newspapers in such states report a far
lower percentage of homicides than those in the less populous states.
The explanation for this is clear: because newspapers in populous states
have so many homicides to cover, they limit their reporting to the more
newsworthy ones. Of course, one could argue that killings by police
officers are exactly the type of homicide that would not go unreported,
thus reducing if not eliminating this bias. Yet with rates of underreporting of homicides as high as two-thirds or more in some states, this
seems unlikely.2 A comparison of the official homicide statistics as
reported by the Ministry of Health with the BDV demonstrates that the
bias is statistically significant (see table 1).
This study uses a simple and straightforward correction to this possible bias and runs the analysis on both the raw and corrected data. The
correction is based on the assumption that the number of police killings
reported in newspapers is indeed underreported, but that because
police killings are more interesting to the public, the underreporting rate
is probably not as high as the general underreporting rate. Therefore I
split the difference and estimate the number of police killings left unreported to be 50 percent as high as the general underreporting rate. For
example, if newspapers in a particular state and year report 40 homicides by police and exhibit an overall reporting rate of homicides of 80
percent (underreporting rate of 20 percent), we may estimate 5 additional homicides, for a total estimate of 45 police killings.3 In addition
AHNEN: POLICE VIOLENCE IN BRAZIL
151
Table 1. Average Annual Rate of Homicides as Reported in
Newspapers, 19 Brazilian States, 1994–2001
Roraima
Acre
Tocantins
Sergipe
Amazonas
Federal District
Rio Grande do Norte
Alagoas
Piauí
Espírito Santo
Paraíba
Goiás
Pará
Ceará
Pernambuco
Rio Grande do Sul
Bahia
Rio de Janeiro
São Paulo
Pearson Correlation
Reported
Rate (%)
Population
122
130
78
59
82
52
50
78
151
116
140
52
84
92
50
33
102
30
31
197,919
343,837
1,114,498
1,691,489
1,897,778
1,934,875
2,630,959
2,693,274
2,718,521
2,905,256
3,357,516
4,769,428
5,768,451
6,939,375
7,534,110
9,889,365
12,881,865
13,703,318
35,398,542
–0.47
p = .02
Note: States are listed from least to most populous.
Sources: MNDH, BDV; IBGE estimates for 1998 population based on 1996 and estimated growth rates.
to the actual number, this analysis also calculates the rate of police
killings per one hundred thousand residents in the state and estimates
the model using it (both actual and estimated). Further, the model is
used to estimate the percentage of all homicides in each state that are
alleged to be committed by police officers.
Operationalization of Independent Variables
The key independent variable in this study is the position of the governor on the left-right spectrum of political ideology. Political parties in
Brazil are known for their relative weakness and lack of coherence.
Indeed, party switching is endemic; many politicians belong to as many
as seven different parties over the course of their political careers.
152
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
Because partisanship does not correlate well with specific public policy
positions taken by individual politicians, finding support for this hypothesis should not be easy. This variable is operationalized into three categories: right, center, or left. Right and center-right parties are based on
Mainwaring et al.’s classification (2000, 31–32) and Power’s surveys of
the Brazilian Congress throughout the 1990s (Power 2000). Considered
to be at the ideological center are governors from two parties: the Party
of the Brazilian Democratic Movement (PMDB) and the Party of Brazilian Social Democracy (PSDB). The rest of the governors are from parties considered to be of the left.4
For the remaining independent variables, the number of nonpolice
homicides is estimated by subtracting the number of police homicides
from the total. Data on homicides are available from the Brazilian Ministry of Health’s database on mortality (DATASUS 2002).5
Methodological Considerations
As is standard with pooled time series, heteroskedasticity across panels
is assumed here.6 Since this dataset contains more panels than time periods, autocorrelation is less of a problem, and therefore least squares
dummy variables regression is used. This consists of OLS regression
with panel corrected standard errors. The complete model is as follows:
depvar = ␣ + 1popden + 2nphom + 3right + 4rt2020
+ 5ill+ 6infmort +
Depvar is one of five dependent variables (number of police homicides,
number of estimated police homicides, police homicide rate, estimated
police homicide rate, and the percentage of all homicides that are attributed to the police) for each of the three police categories (military, civil,
and all). Popden is population density per square kilometer; right is the
party of the governor; rt2020 is the ratio of the top 20 percent of income
earners to the bottom 20 percent—an indicator of inequality; ill is the
rate of illiteracy; and infmort is the rate of infant mortality.
FINDINGS
The first important finding is that the number of police homicides varies
significantly by police force (see table 2). That is, military police officers
are responsible for five times as many homicides as civil police officers.
Of course, the military police force in each state is generally larger, but
usually in a ratio of about three to one. Therefore, even for their respective size, the military police are more likely to use deadly force than the
AHNEN: POLICE VIOLENCE IN BRAZIL
153
Table 2. Homicides by Police Officers in 19 States, 1994–2001
Military Police
Civil Police
Other Policea
All Police
Number
Percent
6,021
1,067
747
7,835
76.8
13.6
9.5
100.0
a
Includes police officers not identified by force, and may include both civil and military police officers.
Source: MNDH, BDV.
civil police. This finding was expected, however, given the nature of
respective policing duties noted above.
A second finding is that the level of police homicides varies substantially across Brazilian states, as measured in two different ways. First,
the percentage of all homicides attributed to the police is very high in
three states with percentages above 15 percent: Rio de Janeiro, Pará, and
Bahia (table 3). The state with the fourth-highest percentage of reported
police homicides (Rio Grande do Norte) has only about half those percentages (7.8 percent); from there the percentage declines slowly to 2.9
percent for Roraima. With respect to the rate of police homicides and
estimated police homicides, Rio de Janeiro and Bahia again demonstrate
accentuated levels, while Pará falls in ranking to eighth place and Acre
jumps to second place (figure 1, p. 155). If the high rankings of Rio de
Janeiro and Bahia with respect to police homicides per one hundred
thousand residents bear out the conventional wisdom that police violence takes place primarily in large urban centers, the ranking of the two
smallest states of Acre and Roraima does not confirm that notion. With
respect to the rate, these two states have the second- and seventh-most
violent police forces among the states considered, while Rio Grande do
Sul, a very populous state with a very large capital city, has the least.
Examining the trends of police violence across years from 1994 to
2001 reveals both good and bad news (figure 2, p. 156). The good news
is that rate of homicides by all police officers has declined generally
over the period in question. The overall rate of estimated police homicides decreased by almost 9 percent, although the trend does show substantial fluctuations. For example, estimates show that the rate at which
the military police employed deadly force grew steadily for two years
through 1996, dropped for one year, then peaked in 1998. Sharp
declines in reported police homicide rates are evident in 1999 and 2000,
but the trend moves sharply upward again in 2001. Figure 2 also
demonstrates that the rate for total police killings is clearly driven by
homicides by military police officers. No strong patterns of homicides
154
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
Table 3. Homicides by Police Officers as Percentage of All Homicides
as Reported in State Newspapers, 19 States, 1994–2001
Rio de Janeiro
Pará
Bahia
Rio Grande do Norte
Piauí
São Paulo
Alagoas
Federal District
Pernambuco
Paraíba
Sergipe
Goiás
Acre
Tocantins
Rio Grande do Sul
Ceará
Amazonas
Espirito Santo
Roraima
Military
Police
Civil
Police
Other
Police
All
Police
15.2
9.1
8.1
4.8
5.1
5.9
4.4
4.7
4.4
4.2
2.2
3.4
2.7
3.4
3.2
3.3
2.0
2.8
0.6
1.7
3.9
4.7
1.1
1.2
0.1
1.2
0.9
1.5
1.0
2.2
0.7
1.3
0.7
0.4
0.5
0.6
0.2
1.5
0.2
2.5
2.7
1.8
1.0
0.8
1.0
0.9
0.4
0.5
1.1
0.5
0.4
0.2
0.6
0.2
1.0
0.2
0.8
17.0
15.4
15.4
7.8
7.2
6.8
6.6
6.6
6.3
5.6
5.5
4.6
4.4
4.3
4.2
4.0
3.6
3.2
2.9
Note: Based on available data.
Source: MNDH, BDV.
by the civil police emerge from the data. The rate of homicides by the
civil police is much lower than that of the military police—again, more
than the difference in size of the two forces would account for. Reported
civil police homicides declined by nearly 45 percent from 1994 to 1997,
but the trend has been upward since then.
We can now examine the rate of police killings using a pooled time
series regression analysis.7 The first results correspond to five different
modes of the dependent variable with respect to the military police
(table 4, p. 157). When predicting the number or estimated number of
reported homicides attributed to the military police, all the control variables except inequality and illiteracy are statistically significant and in
the expected direction. The coefficient for illiteracy does not reach statistical significance, and inequality lies in the opposite direction from
that hypothesized above. The overall model explains roughly 70 percent
of the variance in the number of police homicide victims across states
and years. The substantive finding with respect to political partisanship
AHNEN: POLICE VIOLENCE IN BRAZIL
155
Figure 1. Rate of Reported and Estimated Police Homicides
in 19 States, 1994–2001
is that the number of police homicides in a given state and year
increases by an average of 24 to 43 when the governor comes from a
party of the right. The standardized (beta) coefficients show that partisanship is not as important as population density and nonpolice homicides in predicting military police homicides.
With respect to the rate of police homicides, the overall model
accounts for much less of the variance among states when only the
reported rates are used (R2 = .23), but does about twice as well when
the estimates are used (R2 = .44). The political variable is again significant, and lies in the expected direction in both models. An increase of
0.1 in the rate of killing per 100,000 when the governor is of the right
may not sound high, but in a state like São Paulo, this suggests an extra
33 police homicide victims every year, after controlling for the other
variables. The beta weights are stronger here with respect to the other
variables, however, and roughly on par with nonpolice homicides and
about half to a third as important as population density. All variables
reach statistical significance (p < .05) in the final model, but the surprise
here is the negative association with the nonpolice homicide variable.
Again, the substantive conclusion with this model is that the police
account for 1.2 percent more homicides in states that are governed by
the right, controlling for the other variables.
Turning to the civil police, the overall model performs much less
well, which is not surprising, given the low variation noted in civil
156
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
Figure 2. Rates of Reported and Estimated Police Homicides by Year,
1994–2001
police homicides (see table 5, p. 158). Still, only one variable achieves
statistical significance in all five estimations and lies in the expected
direction: the political variable. With respect to the number of police
homicides, the results indicate that states with governors from parties of
the right report between 3 and 5 more homicides by the civil police
when controlling for the other variables. An examination of the beta
weights of the partisan variable reveals substantial variation and leads
to no straightforward conclusions. Probably most interesting here is that
the nonpolice homicide variable performs much less well in these
models, and in two of them does not reach statistical significance. This
result indicates that lethal violence by the civil police is much less linked
to the overall crime and violence exhibited in the state.
An analysis of total reported police killings yields results similar to
the findings for the military police alone. The major difference is that
the coefficient for the political variable rises; to wit, states governed by
executives from the right exhibit between 31 and 51 more police homicide victims than those governed by a center-left or left party. In addition, the coefficient for the rate of police killings rises between .23 and
.33 per 100,000. Again, in a large state such as São Paulo, this signifies
between 76 and 100 more police homicides per year when the state is
governed by the right.
Table 4. Regression Analysis of Homicides by the Military Police as Reported in Statewide Newspapers, 1994–2001
Estimated
Military Police Homicides
_________________________
Coeff.
Std. Err.
Beta
20.7
.368***
.067***
24.0***
–2.04***
65.5*
.901***
.128***
43.8***
–5.84***
14.7
.039
.004
5.67
.549
.400
.620
.186
–.149
28.2
.064
.007
13.0
1.06
127
.70
127
.71
Estimated Military Police Homicides
per 100,000
_____________________________________
Coeff.
Std. Err.
Beta
Constant
Population density
Nonpolice homicides
Right-wing governor
Inequality
Illiteracy
N
R-square
.943***
.007***
2.9e–04***
.294*
–.031**
.459
.558
.160
–.177
.226
.001
8.9e–04
.129
.011
127
.44
.593
.210
.172
–.149
Military Police Homicides
per 100,000
_________________________
Coeff.
Std. Err.
Beta
.367***
.003***
1.5e–04***
.206*
.044
4.0e–04
3.0e–05
.081
.412
.254
.219
127
.23
Military Police Homicides as
Percentage of All Reported Homicides
_____________________________________
Coeff.
Std. Err.
Beta
.045***
.0002***
8.73e–06***
.012***
–.001***
.001**
.011
1.86e–05
2.24e–06
.002
4.2e–04
2.57e–04
127
.29
157
Notes to tables 4, 5, and 6: *p < .05, **p < .01, ***p < .001; only trimmed models shown. Sources: MNDH BDV; IBGE; DATASUS.
.496
.187
.218
–.208
.215
AHNEN: POLICE VIOLENCE IN BRAZIL
Constant
Population density
Nonpolice homicides
Right-wing governor
Inequality
Illiteracy
N
R-square
Military Police Homicides
_________________________
Coeff.
Std. Err.
Beta
.001
127
.13
.252
Estimated Civil Police Homicides
per 100,000
____________________________________
Coeff.
Std. Err.
Beta
.438*
.190
8.1e–04*
3.5e–04
.226
–.015*
.007
–.254
.092***
.025
.187
.006***
.001
127
.12
127
.07
Civil Police Homicides as
Percentage of All Reported Homicides
____________________________________
Coeff.
Std. Err.
Beta
.018***
.005
–4.6e–04**
.003**
3.0e–04***
1.6e–04
.001
5.2e–05
–.019
.014
.020
.194
49: 1
Constant
Population density
Inequality
Right-wing governor
Illiteracy
Infant mortality
Nonpolice homicides
N
R-square
.005***
Estimated
Civil Police Homicides
Civil Police Homicides
per 100,000
_________________________
_________________________
Coeff.
Std. Err.
Beta
Coeff.
Std. Err.
Beta
13.0*
5.70
.133**
.048
.104**
.035
.312
–.981***
.274
–.176
3.20*
1.37
.069
.102***
.017
.263
.387***
.092
.125
.160*
.055
.003*
.001
.123
.006***
.002
.155 –3.5e–05*** 1.0e–05
–.108
127
127
.13
.12
LATIN AMERICAN POLITICS AND SOCIETY
Constant
Population density
Inequality
Right-wing governor
Illiteracy
Infant mortality
Nonpolice homicides
N
R-square
Civil Police Homicides
_________________________
Coeff.
Std. Err.
Beta
4.65
3.87
.034**
.013
.199
–.406*
.172
–.142
4.47**
1.56
.188
.395***
.068
.250
158
Table 5. Regression Analysis of Homicides by the Civil Police as Reported in Statewide Newspapers, 1994–2001
Table 6. Regression Analysis of Homicides by All Police as Reported in Statewide Newspapers, 1994–2001
All Police Homicides
per 100,000
_________________________
Coeff.
Std. Err.
Beta
.708***
.059
.003***
4.0e-04
.364
1.0e–04* 4.9e-05
.105
.331***
.071
.290
127
.69
127
.71
127
.23
Estimated All Police Homicides
per 100,000
____________________________________
Coeff.
Std. Err.
Beta
1.67***
.320
.009***
.001
.642
2.7e–04**
8.5e–05
.185
.230*
.091
.118
–.053***
.012
–.227
127
.46
127
.29
All Police Homicides as
Percentage
of All Reported Homicides
____________________________________
Coeff.
Std. Err.
Beta
.088***
.013
2.5e–04***
2.0e–05
.579
.018***
–.003***
.001***
.003
6.1e–04
3.6e–04
.251
–.348
.209
159
Constant
Population density
Nonpolice homicides
Right-wing governor
Inequality
Illiteracy
N
R-square
Estimated
All
Police
Homicides
_________________________
Coeff.
Std. Err.
Beta
82.5*
33.0
.978***
.083
.439
.149***
.006
.576
50.8***
12.0
.164
–6.52***
1.26
–.175
AHNEN: POLICE VIOLENCE IN BRAZIL
Constant
Population density
Nonpolice homicides
Right-wing governor
Inequality
Illiteracy
N
R-square
All Police Homicides
_________________________
Coeff.
Std. Err.
Beta
24.9
19.8
.371***
.030
.350
.080***
.005
.652
31.2***
6.18
.213
–2.03**
.743
–.115
160
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
CONCLUSIONS
This study advances our understanding of the relationship between
democracy and the protection of basic rights in several ways. First, scholars must be ready to buck the conventional wisdom regarding the relationship between human rights and democracy, and to examine the conditions under which democracy (in its minimalist electoral form) might
also lead to more human rights violations. Second, this analysis brings to
light a subtle and important assumption that runs through much of the
democratic consolidation literature: that members of a society do not necessarily share an equal interest in promoting the effective guarantee of
physical integrity rights for all citizens. In short, the politics of democratization are distinct from the politics of democratic consolidation. This is
especially true if what we mean by democratic consolidation is not only
the growing assurance that democratic governance will not be reversed,
but the deepening of democratic principles so that state institutions treat
all citizens in ways that protect their individual rights no matter what.
Understanding partisan politics is key to understanding outcomes of
human rights conditions in nascent democracies. The analysis here
demonstrates that governors on the right end of the spectrum are associated with significantly greater rates of killing by the military and civil
police forces. Partisanship was shown to be a robust explanatory variable
in predicting rates of police violence using both raw data and superior
estimates, though not the only or even the most important determinant.
If partisan politics is key to understanding the lethal use of police
force, the analysis also clearly demonstrates that controlling or diminishing key social maladies could greatly contribute to fighting this problem. Lethal police violence significantly increases in densely populated
areas and areas where the level of homicides is already high. Key indicators of poverty, such as the levels of infant mortality and illiteracy,
often proved significant determinants of police violence, as expected.
When controlling for these variables, however, inequality was often significant but consistently opposite to the hypothesized direction. This
research demonstrates, then, the link between economic policy and the
use of lethal violence by state agents.
The good news here is that police violence can be reduced once
politicians find political support for such a policy among the electorate.
Ultimately, however, this is bad news for those who believe in the basic
principles of liberal democracy, for the protection of individual rights
should not depend to such a large extent on the partisanship of the
executive. After all, the rights guaranteed by the constitution do not
change with the arrival of a new governor. Instead, supporters of a
strong human rights agenda must implement rights protection mechanisms that are either legally difficult or politically costly to change. In
AHNEN: POLICE VIOLENCE IN BRAZIL
161
this way, the protection of fundamental rights for all citizens will slowly
become less a partisan issue, as members of different classes come to
understand the inherent value of liberty for all. Such institutionalization
of rights protection is, of course, itself a partisan political issue in the
short run. Thus, we might still expect parties on the left to take the lead
in implementing such rights-enhancing reforms.
Finally, there is no guarantee that the right political conditions exist
to achieve such strong democratic consolidation. As many scholars have
noted, the democratic regimes we observe in Latin America are wanting
in important respects—so important that many scholars refuse to classify them as democracies. Research on the politics of human rights protection can help us understand the extent to which achieving a more
inclusive democratic governance might be possible.
NOTES
Research for this project was supported by a grant from the Faculty Development Fund of St. Mary’s College of California. The National Human Rights
Movement of Brazil also generously provided access to its database on violence,
without which this project could not have been completed. I wish to thank
Jonathan Hartlyn, Claudio Fuentes, Susan Weissman, and four anonymous
reviewers for very helpful comments on various drafts of the manuscript. All
errors or omissions are my own.
1. P-value for the correlation is .075 for a one-tailed test. The average
number of police killings across the three years is 280 in the BDV and 318 in
Cano’s data, a difference of 38, or less than 14 percent. As expected, the BDV
data undercount Cano’s reported number of military police killings.
2. It is curious that newspapers in some states report a total number of
homicides greater than the official government figure for that year. Nevertheless,
the Health Ministry notes that its statistics are not complete, and typically include
only about 80 percent of all deaths, though the number should theoretically also
be higher for homicides. For greater discussion on this point see Cano and
Santos 2001; Amorin Ratton 1996.
3. Another plausible assumption is that police officers have more resources
with which to hide their misdeeds, and therefore the underreporting of killings
by police officers should not differ from the general underreporting rate. This
estimate is also calculated, but analysis of the results shows that the substantive
conclusions of this study do not change under this assumption.
4. The coding is right or center-right = 1, center = 0, left or center-left = 1.
Data available from Nicolau (1998) and TSE (2002).
5. As per Cano and Santos (2001), this analysis recognizes the incompleteness of Ministry of Health data. Especially noteworthy is the undercounting of
homicides; many deaths from external causes are listed as “indeterminate”
(meaning unsure if they were accidents, suicides, or homicides) because of
police and other pressure on coroners’ offices. After carefully studying data on
deaths from both the Ministry of Health and the Public Safety Office in Rio de
162
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
Janeiro, Cano and Santos’s solution is to count as intentional homicides 10 percent of the total number of deaths due to external causes but listed as indeterminate. This study likewise calculates this figure employing data from the Ministry of Health (DATASUS 2005) and then subtracts the number of reported
police homicides to arrive at the number of estimated nonpolice homicides. All
data by state and year on the other control variables are available from IBGE.
6. Substantive results do not change when first order panel specific autocorrelation is assumed.
7. For each estimation, a full model was run, and then trimmed by excluding variables one at a time that did not reach the .05 level of statistical significance. Only trimmed models are presented here.
REFERENCES
Amorin Ratton, Jr., José Luiz de. 1996. Violência e crime no Brasil contemporâneo: homicídios e políticas de segurança pública nas décadas de 80 e 90.
Brasília: Movimento Nacional de Direitos Humanos.
Bicudo, Hélio. 1994. Violência: O Brasil cruel e sem maquiagem. São Paulo: Editora Moderna.
Brazil. Ministry of Justice. 2002. National Public Security Plan 2002.
<www.mj.gov.br/Senasp/senasp/home_estatisticas.htm>. Accessed February 15, 2003.
Caldeira, Teresa. 2000. City of Walls: Crime, Segregation, and Citizenship in São
Paulo. Berkeley: Unversity of California Press.
Cano, Ignácio. 1997. Letalidade da ação policial no Rio de Janeiro. Rio de
Janeiro: ISER.
Cano, Ignácio, and Nilton Santos. 2001. Violência letal, renda e desigualidade
social no Brasil. Rio de Janeiro: Letras.
Chevigny, Paul G. 1995. Edge of the Knife: Police Violence in the Americas. New
York: New Press.
Collier, David, and Gerardo L. Munck. 2001. Introduction: Regimes and Democracy in Latin America. Studies in Comparative International Development
36, 1: 3–6.
Collier, David, and Steven Levitsky. 1997. Democracy with Adjectives: Conceptual Innovation in Comparative Research. World Politics 49: 430–51.
Davenport, Christian. 1995. Multi-Dimensional Threat Perception and State
Repression: An Inquiry into Why States Apply Negative Sanctions. American Journal of Political Science 39: 683–713.
––––. 1999. Human Rights and the Democratic Proposition. Journal of Conflict
Resolution 43, 1: 92–117.
Departamento de Informática do Sistema Unico de Saude (DATASUS). 2002. Sistema de Informação sobre Mortalidade (SIM). <www.datasus.gov.br>
Accessed January 25, 2002.
Diamond, Larry. 1996. Democracy in Latin America: Degrees, Illusions, and
Directions for Consolidation. In Beyond Sovereignty: Collectively Defending
Democracy in the Americas, ed. Tom Farer. Baltimore: John Hopkins University Press. 52–104.
AHNEN: POLICE VIOLENCE IN BRAZIL
163
Diamond, Larry, Jonathan Hartlyn, Juan Linz, and Seymour Martin Lipset. 1999.
Democracy in Developing Countries: Latin America. 2nd ed. Boulder:
Lynne Rienner.
Dimenstein, Gilberto. 1996. Democracia em pedaços: direitos humanos no
Brasil. São Paulo: Schwarca.
Fein, Helen. 1995. More Murder in the Middle: Life-Integrity Violations and
Democracy in the World. Human Rights Quarterly 17: 170–91.
Garotinho, Anthony, and Luiz Eduardo Soares. 1998. Violência e criminalidade
no estado do Rio de Janeiro: diagnóstico e propostas para uma política
democrática de segurança pública. Rio de Janeiro: Hama.
Henderson, Conway W. 1991. Conditions Affecting the Use of Political Repression. Journal of Conflict Resolution 35, 1: 120–42.
Heywood, Andrew. 2003. Political Ideologies. 3rd ed. New York: Palgrave
Macmillan.
Holston, James, and Terry P. R. Caldeira. 1998. Democracy, Law, and Violence:
Disjunctions of Brazilian Citizenship. In Fault Lines of Democracy In PostTransition Latin America, ed. Felipe Agüero and Jeffrey Stark. Coral
Gables: North-South Center Press.
Instituto Brasileiro de Geografia e Estatística (IBGE). Data on social indicators.
<www.ibge.gov.br>
Jornal do Brasil (Rio de Janeiro). 2002. Serra elege a segurança pública como
prioridade. June 17: A1.
Mainwaring, Scott, Rachel Meneguello, and Timothy Power. 2000. Partidos conservadores no Brasil contemporâneo: quais são, o que defendem, quais so
suas bases. Rio de Janeiro: Paz e Terra.
Méndez, Juan, Guillermo O’Donnell, and Paulo Sérgio Pinheiro, eds. 1999. The
(Un)rule of Law and the Underprivileged in Latin America. Notre Dame:
University of Notre Dame Press.
Mesquita Neto, Paulo. 1999. Violência policial no Brasil: abordagens teóricas e
prácticas de controle. In Cidadania, justiça e violência, ed. D. Chaves Pandolfi et al. Rio de Janeiro: Fundação Getúlio Vargas.
Mitchell, Neil J., and James M. McCormick. 1988. Economic and Political Explanations of Human Rights. World Politics 40, 4: 476–98.
Movimento Nacional de Direitos Humanos, Banco de Dados sobre Violênica
(MNDH DBV). Database on violence. Brasília: MNDH.
Nicolau, Jairo Marconi. 1998. Dados eleitorais do Brasil (1982–1996). Rio de
Janeiro: Editora Revan.
Núcleo de Estudos sobre a Violênica (NEV). 1993. Direitos humanos no Brasil.
São Paulo: Universidade de São Paulo.
O’Donnell, Guillermo. 1993. On the State, Democratization, and Some Conceptual Problems: A Latin American View with Glances at Some Postcommuist
Countries. World Development 21, 8: 1355–69.
––––. 1998. Horizontal Accountability in New Democracies. Journal of Democracy 9, 3: 112–26.
––––. 1999. Polyarchies and The (Un)Rule of Law in Latin America: A Partial
Conclusion. In Méndez et al. 1999.
Olson, Mancur. 1963. Rapid Growth as a Destabilizing Force. Journal of Economic History 23: 529–52.
164
LATIN AMERICAN POLITICS AND SOCIETY
49: 1
Pereira, Anthony W. 2000. An Ugly Democracy? State Violence and the Rule of
Law in Postauthoritarian Brazil. In Democratic Brazil: Actors, Institutions,
and Processes, ed. Timothy J. Power and Peter R. Kingstone. Pittsburgh:
University of Pittsburgh Press.
Pesquisa Social Brasileira (PESB). 2002. Study conducted by the Universidade
Federal Fluminense, coord. Alberto Carlos Almeida and Zairo Cheibub.
Funding: Ford Foundation.
Pinheiro, Paulo Sérgio. 1996. Prefácio: O passado não está morto: nem passado
ainda. In Dimenstein 1996.
Poe, Steven C., and Neal Tate. 1994. Repression and Human Rights to Personal
Integrity in the 1980s: A Global Analysis. American Political Science Review
88: 853–72.
Poe, Steven C., C. Neal Tate, and Linda Camp Keith. 1999. Repression of the
Human Right to Personal Integrity Revisited: A Global Cross-National Study
Covering the Years 1976–1993. International Studies Quarterly 43: 291–313.
Power, Timothy J. 2000. The Political Right in Postauthoritarian Brazil: Elites,
Institutions, and Democratization. University Park: Pennsylvania State University Press.
Tribunal Superior Eleitoral (TSE). 2002. Eleições 2002. <www.tse.gov.br/eleicoes/
index.html> Accessed March 10, 2002.