1
Running head: SEXUAL INTEREST PROFILES
Indirect Measures of Sexual Interest in Child Sex Offenders: A Multi-Method Approach
Rainer Banse and Alexander F. Schmidt
University of Bonn
Jane Clarbour
University of York
2
Abstract
Although there is strong meta-analytical evidence that deviant sexual interest in children is
a major risk factor for recidivism in child-sex offenders, the assessment of deviant sexual
interest with self-report or phallometric measures is problematic. As an alternative
approach for assessment, the Explicit and Implicit Sexual Interest Profile (EISIP) is
introduced that features direct self-report and indirect latency-based measures (Implicit
Association Tests and viewing time measures) of sexual interest in adults and children. The
reliability and validity of the EISIP was investigated using a selected sample of child sex
offenders (n = 38), offender (n = 37) as well as non-offender (n = 38) controls. Among the
indirect measures, viewing time measures showed higher reliability, convergent, and
criterion validity than the IATs. However, the IATs independently accounted for criterion
variance in multivariate analyses. The combined indirect measures showed good
discriminative validity between child-sex offenders and controls.
(146 words)
Key words:
Deviant sexual interest, sexual preference, child sex offending, implicit measures, indirect
measures
3
Indirect Measures of Sexual Interest in Child Sex Offenders: A Multi-Method Approach
It is a common view – not only among forensic researchers – that sexually deviant
behaviors result from a favor of these over socially accepted sexual activity. This so called
sexual preference hypothesis (Freund & Blanchard, 1989) is corroborated by meta-analytic
evidence consistently showing that deviant sexual interest (e.g., in sex with children) is one
of the strongest risk factors for reoffending with effect sizes around d = .30 (Hanson &
Bussiѐre, 1998; Hanson & Morton-Bourgon, 2005). Recent aetiological models of sex
offending (e.g., Ward & Beech, 2006; Ward & Siegert, 2002) postulate that among
numerous clinical problems such as emotional dysregulation, social difficulties, and
cognitive distortions, deviant sexual interest is a primary causal factor for sex offences.
Assessing deviant sexual interest
Albeit “assessing the nature of the individual’s deviant sexual interests is often the
centerpiece of a sex offender evaluation” (Lanyon, 2001, p. 257) the assessment of
enduring sexual preference is fraught with difficulties, mainly due to the problematic
psychometric properties of the most commonly used measures (Kalmus & Beech, 2005).
This has led to general skepticism about the utility of assessing deviant sexual interest at all
(Marshall & Fernandez, 2003). Because neither the legal (based on sexual offences) nor the
clinical approach (diagnosis of pedophilia) allow for a valid inference on deviant sexual
interest (Marshall, 2007), conceptually more valid assessment tools are needed. Up to now
a range of quantitative, psychometric measures have been developed that can be divided
into direct and indirect measures.
Direct measures of sexual interest rely on the self-report of deviant sexual
preferences in questionnaires or card-sort procedures. Their validity is jeopardized by
impression management and deliberate faking. The general problem of transparency in
4
direct measures is all the more critical if disclosure of personal information is highly
embarrassing, socially undesirable, or has legal implications, as it is commonly the case in
forensic contexts (Kalmus & Beech, 2005).
To tackle the issues of socially desirable responding and simulation/dissimulation,
great effort has been invested in developing indirect measures of sexual interest. These are
supposed to be less susceptible to deliberate manipulation because test subjects are a) less
aware of the nature of the measure or the measurement principle, and/or b) because the
expression of the measured construct cannot be deliberately controlled. Based on their
methodological rationale, indirect measures either rely on physiological measures of sexual
arousal, or response latency measures reflecting information processing (in a wide sense).
Among the physiological measures penile plethysmography (PPG) or phallometry
is by far the most researched method and commonly used. An index of deviant sexual
interest derived from PPG assessment has repeatedly been shown to predict sexual
recidivism (Hanson et al., 1998; 2005). However, detailed reviews of the PPG literature
come to the conclusion that the main problems of PPG research and application result from
a) a lack of standardization of the procedures and stimulus materials, b) low retest
reliability, c) low specificity or discriminant validity, d) low response rates, and e) high
fakeability (e.g., Kalmus & Beech, 2005; Laws, 2003; Marshall & Fernandez, 2000;
Murphy & Barbaree, 1994).
In the last decade, a whole range of latency-based measures have been developed
in the very active area of implicit social cognition research. In recent reviews of indirect
measures these approaches have been classified as “attentional methodologies” (e.g., Gress
& Laws, 2009; Kalmus & Beech, 2005). However, for many indirect paradigms the
underlying processes are either unknown or are still subject of debate (e.g., Imhoff,
Schmidt, Nordsiek, Luzar, Young, & Banse, in press). Therefore we propose to use the
5
label latency-based measures for this class of instruments that refers to the nature of the
dependent variable and does not require any theoretical assumptions about underlying
processes.
Latency-based measures are usually based on rating, sorting, or detection tasks
that involve different classes of pictorial stimuli of potential sexual interest (e.g., adult
women, adult men, and children of either sex). The measurement rationale of indirect
paradigms either relies on the fact that certain categories are more strongly associated with
the concept of sex or sexual arousal than others (the Implicit Association Test or IAT; e.g.,
Gray, Brown, MacCulloch, Smith, & Snowden, 2005), or that task irrelevant sexually
preferred stimuli function as distractors that interfere with a primary task, such as the
Choice Reaction Time Task (e.g., Mokros, Dombert, Osterheider, Zappalà, & Santtila, in
press), the Emotional Stroop Task (Smith & Waterman, 2004) or the Attentional Blink
Task (Beech, Kalmus, Tipper, Baudouin, Flak, & Humphreys, 2008).
Viewing time (VT) measures – first described by Rosenzweig (1942) – exploit the
fact that photographs of sexually attractive individuals are inspected longer than sexually
less attractive individuals (e.g., Gress, 2005). In a typical viewing time task, participants
are asked to rate the sexual attractiveness of targets of both sexes and different age groups.
In addition to the explicit rating, the viewing or inspection time of each stimulus picture is
unobtrusively recorded, averaged for each target category, and used as an indirect indicator
of sexual interest. Viewing time measures have been used in forensic contexts to
successfully differentiate between child sex offenders and non-offenders (Laws & Gress,
2004; Harris; Rice, Quinsey, & Chaplin, 1996) and between different types of sex
offenders (Abel, Huffmann, Warberg, & Holland, 1998; Worling, 2006).
Current study
6
Despite the growing interest in indirect measures of sexual interest and the recent
development of a range of conceptually different latency-based measures, there seems to be
very little research comparing the reliability and criterion validity of different latency-based
measures of sexual interest. To the best of our knowledge, all published studies include
comparisons between and among different combinations of direct measures, behavioral
offence data, and/or PPG measures (e.g., Abel, Jordan, Hand, Holland, & Phipps, 2001;
Letourneau; 2002), but as yet there are no published empirical studies comparing different
latency-based measures of sexual interest in forensic samples. However, empirical data of
this kind are required to investigate to what extent latency-based measures could improve
the assessment of sexual preference over and above direct measures.
In order to address these issues, the general objective of the present study was to
introduce and validate the Explicit and Implicit Sexual Interest Profile (EISIP) featuring
four direct self-report measures of sexual interest, three different IATs, and four VT
measures. It was the first aim of the present study to investigate the reliability and
convergent validity of the direct and indirect measures of the EISIP. Second, as a more
direct test of the discriminant validity of the EISIP, it was explored to what extent the
different measures contributed to a discrimination between child sex-offenders and
controls. Third, the incremental validity of the measures featured in the EISIP was
investigated by contrasting the rates of correct classification and ROC-analyses. Fourth and
finally, all EISIP measures were validated against the Screening Scale for Pedophilic
Interests (Seto & Lalumiѐre, 2001).
Although not all child sex offenders show deviant sexual preferences and not all
men showing deviant sexual preferences do actually offend against children (Seto, 2008),
we expected that child sex offenders will show stronger sexual interest in children on direct
and indirect measures than controls. Moreover, sex offenders’ preferences should be related
7
to the gender profile of their victims – girl-only and boy-only offenders would be expected
to show relatively stronger sexual interest in targets corresponding to the gender of their
victims. Offenders with mixed victims would be expected to show stronger sexual interest
in both male and female children than control groups. Most male controls would be
expected to show exclusively sexual interest in women, a few in men, but virtually none in
children. Regarding child molesters’ sexual interest in adults as compared to interest in
children, no clear-cut predictions could be made, because child sex offenders quite
commonly also show some level of, or even exclusively, sexual interest in adults (e.g.,
Worling, 2006).
Method
Sample. Inclusion criteria for all participants were: age over 21, white ethnic
origin (to exclude possible interactions between ethnic origin of participants and target
stimuli), IQ of at least 80 (in case of offenders IQ was extracted from prison files of prior
psychological assessments; controls were assumed to have IQs above the threshold due to
their level of functioning as prison personnel, professionals or students), and general
reading ability.
The sample consisted of N = 113 participants (38 child sex offenders, 75 controls).
Offenders were recruited in four different prison establishments in the UK. The local prison
records were used to identify and contact all child sex offenders in each of the four
institutions who met the inclusion criteria. Offender and non-offender control groups of
equal size were recruited. Potential participants were approached and subsequently
informed that the study investigated the usefulness of new measures of sexual preferences.
Participation was voluntary. Participants were informed that there were no negative
consequences for not participating or withdrawing consent in the course of the study. Child
sex offender groups (boy victims only n = 14, girl victims only n = 16, mixed victims n =
8
8) consisted of child sex offenders (most, but not all, repeat offenders) convicted of handson offences with extra-familiar victims below 12 years of age (clearly prepubescent
children). Prior to this study all child sex offenders had completed either the Rolling, Core,
or Extended Sex Offender Treatment Programme of HM Prison Service (SOTP; Beech,
Oliver, Fisher, & Beckett, 2005). Offender controls (n = 37) had convictions for non-sex
related offences only. Non-offending controls (n = 38) consisted of a community sample
including prison officers, other prison personnel and men from the community. The age of
child sex offenders ranged from 28 to 74 years, for the controls from 21 to 63 years.
Controls were on average younger than child-sex offenders, F (4, 108) = 10.16, p < .001
(first row of Table 4).
Direct EISIP measures. The Explicit Sexual Interest Questionnaire (ESIQ, Table
1) was developed for the purposes of the present study. The ESIQ features two subscales
assessing sexual behavior (e.g., “I have enjoyed orally stimulating a
man/woman/boy/girl.”) and sexual fantasy (e.g., “I have daydreamed of having sex with a
man/woman/boy/girl”) with 5 items each. The ten fantasy and behavior items are combined
with the four types of targets (man, woman, boy, or girl). Items were responded to in a
dichotomous yes/no format. Corresponding scale frequency scores are calculated from the
amount of “yes” responses divided by the number of scale items. With only 40 items it is
highly economic and conceptually directly comparable with the indirect measures assessing
sexual interest in men, women, and prepubescent boys and girls.
Indirect EISIP measures. Four different viewing time measures regarding sexual
interest in men, women, girls and boys were used. Participants were asked to rate the
sexual attractiveness of target stimuli on a 5-point Likert-scale ranging from 1 (“sexually
unexciting”) to 5 (“sexually very exciting”) without time constraints. The 20 stimulus
pictures were presented on the PC monitor until the response was given. The viewing time
9
was recorded unobtrusively. To optimize the assessment of individual differences (i.e.,
person and group effects) as opposed to main effects, the same random order of stimuli was
used for all participants. Individual response latencies were truncated at 10,000 ms (i.e.,
latencies longer than 10,000 ms were recoded to that value), and averaged across the five
stimuli belonging to each target category. No suspiciously short response latencies were
observed (all latencies ≥ 440 ms).
Three different IATs with the object categories Man-Woman, Girl-Woman, and
Boy-Man, and the attribute categories sexually exciting-unexciting were used. These were
derived from Ahlers et al.’s (2006) suggestion to disentangle sexual orientation (hetero- vs.
homesexual) and sexual age preference (children vs. adults). The former is measured with
the Men-Women IAT and the latter with both the Girls-Women and the Boys-Men IATs.
Each IAT consisted of five blocks following standard procedures as described in
Greenwald, McGhee, and Schwartz (1998). The stimulus pictures were the same as in the
VT task. The first block of 40 trials comprised a discrimination task of 10 words that had to
be classified as sexually exciting (erotic, exciting, lustful, sensual, orgasm) or unexciting
(dull, bland, indifferent, unexciting, boring). In the second block of 40 trials, 10 pictures
had to be assigned to the categories man and woman (girl vs. woman, boy vs. man,
respectively) by pressing the left or right response key, respectively. In the third block, both
tasks were mixed in alternating order. Four practice trials preceded 80 test trials. The left
response key had to be pressed for items belonging to the categories “Man” or “sexually
unexciting”, the right response key for items for “Woman” or “sexually exciting” (and
accordingly for the Boys-Men and Girls-Women IATs). The fourth block of 40 trials was
similar to the second, but the key assignment was reversed. In the fifth and final block of
4+80 trials, both tasks were again combined. In this final block the left response key had to
be pressed for items relating to the categories “Man” or “sexually exciting”, and the right
10
response key for items relating to “Woman” or “sexually unexciting”. Incorrect responses
were indicated by an error message throughout all blocks. The IAT was scored by
calculating the difference between the mean response latencies of the critical third and fifth
block, divided by the pooled standard deviation of response latencies (Greenwald, Nosek,
& Banaji, 2003). Only trials with a correct answer were used, error trials were discarded.
To optimize the measurement of individual differences, and not main effects, the order of
the two combined IAT blocks was kept constant for all participants.
Other measures. To control for an influence of socially desirable responding (SD),
the Balanced Inventory of Desirable Responding (BIDR; Paulhus, 1998) was included.
This questionnaire has proven to be valid in offender samples (Kroner & Weekes, 1996).
As a short screening tool of pedophilic sexual interest the Screening Scale for Pedophilic
Interests (SSPI; Seto & Lalumiѐre, 2001) was used. It is an actuarial four-item scale
summarizing offenders’ sexual victim characteristics (any male victims, more than one
victim, any victims < 12 years, any unrelated victims). This index has been shown to
predict phallometrically assessed sexual arousal to children and serious violent or sexual
reoffending in adult male child sex offenders (Seto, Harris, Rice, & Barbaree, 2004) and
was used as a validation measure.
Materials. The pictures of target persons used for the indirect measures were
selected from the Not-Real-People picture set A and B (Pacific Psychological Assessment
Corporation, 2004). The picture set features categories of sexual maturation for individuals
(all white Caucasian) according to Tanner (1978), ranging from Tanner categories 1 to 3
(prepubescent children), Tanner category 4 (adolescents) to Tanner category 5 (adults). For
men, women, boys and girls, five pictures showing the head and full body in swimming
clothes were used. Pictures of children were taken from Tanner categories 1 to 3, and
pictures of adults from Tanner category 5. No pictures of adolescents (Tanner 4) were used.
11
Procedure. All experimental protocols and data collection methods were approved
by the Department of Psychology’s Ethics Committee at the University of York, and the
prison establishments. All assessments were run on an IBM-compatible laptop in individual
sessions in a separate and quiet room. To familiarize participants with the 20 stimulus
pictures, these were presented at the beginning of the session one by one. Then the
participants worked through the different tasks in the following order: Men-Women IAT,
Girls-Women-IAT, Boys-Men IAT, Viewing Time Task, BIDR, and the ESIQ explicit
sexual interest questionnaires. After completing the assessments, each participant was
thanked and debriefed.
Results
Psychometric Data
Reliabilities. Psychometric data of the self-report ESIQ are presented in Table 1.
Internal consistencies of the eight subscales and the four combined scales of the ESIQ were
satisfactory (.86 < α < .97). For the purposes of the present study only the combined sexual
interest scales will be reported. The VT-indices based on truncated raw data showed
satisfactory reliabilities. With the exception of viewing time for young girls (α = .77)
internal consistency was .85 or better (Table 2). An exploration of the scoring technique
used by Gress (2005) based on intra-individual z-transformation (i.e., ipsatization)
diminished the Cronbach’s alphas of the VT-indices to .30 < α < .66 due to an inflation of
small intrapersonal differences that reduced interpersonal variance.
The reliability results for the IATs were mixed. After discarding the IAT scores of
nine participants with error rates ≥ 35% in at least one combined block, the Man-Woman
and Girl-Woman IATs showed nearly satisfactory reliability (α = .79). The Boys-Men IAT
showed an unsatisfactory low α of .65. This weak reliability coefficient is likely to be due
to a lack of variability in the sexual preference of boys over men, because the sample did
12
not include a sizeable proportion of homosexual men who were not child sex offenders.
This means that virtually all participants were either sexually attracted to men and boys, or
to neither of them. In consequence, the IAT as a relative measure of sexual interest for one
target category over the other could not detect substantial individual differences in the BoyMan IAT, and hence, reliability was low. The BIDR total score showed a satisfactory
internal consistency of α = .84.
Convergent and discriminant validity of sexual interest measures. Correlations
between all sexual interest measures were calculated to investigate convergent validity
(Table 3). To enable comparisons with the three relative IAT measures, conceptually
analogous difference scores were also calculated for the ESIQ and viewing time measures.
All intra-method intercorrelations of the ESIQ showed coefficients in the expected
directions. The correlations between three out of four VT-indices and the corresponding
explicit ESIQ measures were significant and in the expected directions. Only VT for
women did not correlate with the explicit measure. The correlations outside the main
diagonal were low or negative with the exception of high correlations between VT-men and
VT-boys that are again due to the fact that virtually all participants with a sexual preference
for boys also showed a sexual preference for men and vice versa. A demonstration of
discriminant validity of these VT-scales (and the corresponding explicit scales) would
require a sizeable proportion of homosexual offender or non-offender controls. In this
sample only 2 (2.7%) men in the control groups reported more sexual interest in men than
in women in the corresponding ESIQ categories compared to 12 (31.6%) child sex
offenders. Nine out of 12 (75%) gay child sex offenders had boy victims only, the
remaining 3 (25%) had abused boys and girls. Interestingly, the VT-boy correlated with the
VT-girl (indicating interest in children in general), whereas VT-men did correlate with VT-
13
boys, but not significantly with VT-girl, indicating that homosexual orientation as such was
more strongly related to sexual interest in boys than in girls in this sample.
The correlations of the three IATs and the explicit ESIQ correlations indicate a
substantial convergent validity for the Men-Women IAT and the Girls-Women IAT and the
corresponding explicit difference scores. However, no significant correlation was obtained
for the Boys-Men IAT. No unexpected correlations between the IATs and the ESIQ scales
emerged, thus providing evidence for the discriminant validity of the measures. Regarding
the convergent validity of the VT and IAT measures, a significant correlation emerged only
for the Girls-Women IAT.
SD as measured by the averaged BIDR showed only a negative correlation with
self-reported interest in women (r = -.22). Participants with high SD scores showed a
tendency to downplay explicit sexual interest in women (Table 3).
Correlations of all EISIP measures and the SSPI pedophilic interests score were
consistently significant for sexual interest in men and men over women. The correlation
levels up to .60 (Table 3), were surprisingly strong taken into consideration that the SSPI is
only defined for child sex offenders. Therefore, the variability in this sample was quite
restricted (M = 3.9; SD = 1.2; range 2-5, for group levels see Table 4) due to selection
criteria and small sample size of the child sex offenders. In the trade-off between criterion
group purity in a known-group validation approach and group heterogeneity needed to
optimize the test of convergent validity we opted for a clearly selected child sex offender
sample. Due to this only two of the four actuarial SSPI-items showed variability in this
study. Therefore, to calculate the internal consistency of the SSPI under these
circumstances is meaningless. Nevertheless, the VT Boys-Men measure showed good
convergent validity with the SSPI (r = -.49). All other correlations had the expected sign.
14
All EISIP measures of homosexual interest in men indicated pedophilic
preferences. However, across methods, all three Boys-Men difference scores were not at
all, or even negatively associated with the SSPI score. It has to be noted though, that the
association between homosexual orientation and deviance is due to the distribution of gay
participants in the sample, thus confounding age preference and sexual orientation.
Group Differences and Classification
Group differences across sexual interest variables. To test whether sexual interest
variables differentiated between child sex offenders and control groups, a series of oneway
ANOVAs was conducted (Table 4). For all ESIQ measures except the Boys-Men
difference measure, F < 1, a significant group effect was found, 3.54 < F < 10.67, p < .05.
All but one of the VT categories showed group effects 2.82 < F < 10.05, p < .05 and for the
IATs, the Men-Women, Girls-Women, and averaged Children-Adults IATs produced
group effects, 2.91 < F < 6.19, p < .05, but not the Boys-Men IAT, F(4, 102) = 1.01.
Post-hoc comparisons confirmed the hypothesis that in the ESIQ scales both
control groups showed strong sexual interest in women only. Child sex offenders’ interest
for children was significantly more pronounced than controls’, but interestingly, child
molesters also reported high levels of interest in adults – especially in women. As expected,
boy-only sex offenders showed the strongest interest in boys and hardly any interest in
girls; this finding was reversed for girl-only offenders and the difference was statistically
significant. The mixed victim group reported interest in girls and boys. Offenders with any
boy victims showed strong explicit sexual interest in both men and boys, differing
significantly from the other three groups. Sex offenders with any girl victims reported high
explicit interest in women, not significantly differing from the controls, but from the boyonly offenders.
15
VT data showed that there were no group differences in VT for women. Boy-only
sex offenders produced the highest VTs for men and boys, but only differed significantly
from the offender controls. Offenders with any girl victims had the highest latencies in both
adult categories, indicating primary interest in adult sex objects. No significant differences
were observed between offender groups with girl only and mixed victims. In the IAT
results the only group differences emerged for the Girl-Women and the averaged ChildrenAdults IATs. In these measures child sex offenders with any boy victims differed
significantly from non-offending control groups.
Univariate discrimination. A series of ROC-analyses was conducted to test the
criterion validity of all single measures used in this study (Table 2). Comparison groups
were all controls or offender controls versus the groups of all child sex offenders or boy- or
girl-only victim groups. As implied by the effect sizes reported in Table 2, nearly all
measures discriminated above chance levels between child sex offenders and all controls.
Only the explicit Boys-Men difference score and the VT Children-Adults score did not
show significant criterion validity. VT measures in general had good classificatory power
in discriminating child sex offender groups from offender controls. The IATs performed
generally less well. Only the Children-Adult IAT consistently discriminated between
controls and all offender groups with boy victims. Particularly high AUCs were found for
the VT categories of Men and Boys (.78 < AUC < .90) indicating that long VTs for boys
and men were associated with prior sex offending against children in general – even for
the girl victims only group. The explicit measures discriminated well, too, especially for
the categories involving child-related items, indicating a relatively open self-report of
deviant sexual interest in the sample.
Multivariate analyses. Sex offenders showed higher levels of general sexual
interest than controls across the full range of ESIQ and VT measures as evidenced by
16
significant group main effects in the corresponding 5 (Group) x 4 (Sexual interest
categories) mixed model MANOVAs, F (4, 108) = 15.83, p < .001, ηp2 = .37 and F (4, 107)
= 8.79, p < .001, ηp2 = .25, respectively. Regarding the IAT measures, the same significant
group main effect emerged for the sex offenders in an omnibus 5 (Group) x 3 (IAT
measures) mixed model MANOVA, F (4, 99) = 6.24, p = .000, ηp2 = .20. These results
suggest that sex offenders generally have stronger sexual interest or in case of the IATs
stronger homosexual or child preferences across all the EISIP measures.
Binary logistic regression and ROC analyses. To test whether a profile of
conceptually different direct and indirect measures of sexual interest would generate
incremental validity, we conducted block-wise binary logistic regression analyses
combining groups of measures to classify controls versus child sex offenders. All
regressors were z-standardized. The resulting probability estimates from multivariate
binary logistic regressions were used as test-scores in corresponding ROC-analyses
(controls vs. child sex offenders).
The IAT measures alone explained the smallest fraction of variance in the sample,
but criterion validity was acceptable (Table 5). VT measures alone performed better. The
combination of all indirect measures explained 55% of variance and showed a very good
criterion validity of AUC = .88. The explicit ESIQ scales accounted for 63% of the
variance and showed the same criterion validity (.88) as all indirect measures taken
together. The whole profile including all explicit and indirect predictors explained 75% of
variance and showed excellent discriminative power (AUC = .95).
To ascertain that sample specific characteristics did not produce an artificial
overestimation, all binary logistic regressions were cross-validated via the leave-one-outmethod (Efron, 1983). This statistical procedure allows for estimating the size of sampledependent estimation errors in the binary logistic regression function – the so called
17
optimism. Optimism rates were low ranging between 2.0% for all indirect measures and
3.5% for the explicit ESIQ self-reports only (Table 5).
Discussion
The results of this study provide first preliminary evidence that the EISIP test
battery is a reliable and valid measure of sexual interest in child sex offenders, non sexrelated offenders, and non-offenders. Using an univariate approach, specific IATs and VT
measures of the EISIP have been found to discriminate child sex offenders from controls
nearly as well as the explicit measures in this largely non-denying sample. This finding is
in line with previous research that has mostly considered single indirect measures only
(e.g., Abel et al., 1998; Gray et al., 2005; Gress, 2005; Harris et al., 1996). Abel et al.
(2001) further compared VT measures with phallometric assessments and found both
indirect measures to be valid. However, the present study is the first to compare the
criterion validity of different latency-based indirect measures of sexual interest using a
forensic sample.
In the present study, the VT measures outperformed the IATs not only in terms of
discriminatory power but also with respect to reliability, convergent, and discriminant
validity. With regard to their psychometric properties, the VT measures reached the level of
reliability and validity that is otherwise only known for direct assessment methods such as
questionnaires. The reliability and validity of the IATs can be considered as moderate to
satisfactory at best. However, it has to be noted that the sample characteristics of the
present study may have obscured the psychometric quality of the Boys-Men IAT which
performed less well than the other two IATs. Given the relative nature of this measure, this
particular IAT can only perform well if participants substantially differ in their relative
sexual preference for men and boys. The present sample showed only very little variability
in this respect. The critical test of the sensitivity of the Boys-Men IAT can only be
18
conducted with a community sample with a sizeable proportion of homosexual men. This
has still to be examined in future studies. Importantly, even though the IATs were
psychometrically less satisfactory than the VT-measures, they still provided additional
information and showed a significant correlation with the SSPI scores of pedophile
interest.
The newly developed explicit self-report questionnaire ESIQ was found to be
psychometrically sound. Besides the good psychometric quality and economy, the ESIQ
has the advantage of providing a direct assessment of sexual interest that is conceptually
directly comparable to the information provided by the VT measures (using absolute ESIQscores) and the IAT measures (using relative ESIQ difference scores).
Importantly, by combining conceptually different measurement approaches to
profiles of sexual interest, the predictive validity could be increased. Taken together, the
indirect measures used in this study explained nearly the same amount of between-group
variance as the self-report measures. The combination of all direct and indirect measures
could further increase the prediction to the level of nearly perfect criterion validity (AUC =
.95) and the very large amount of 75% of explained variance. The evidence suggests that
the different measures tap into slightly different domains of sexual interest with specific
variance (e.g., based on more automatic or more controlled processes).
This differentiation is beneficial because one has to be very cautious when
interpreting effects of indirect and/or implicit measures at single-case level. The scores of
indirect measures may be influenced by numerous personal and contextual factors (De
Houwer, Teige-Mocigemba, Spruyt, & Moors, 2009). Therefore it is questionable – if not
dubious – to interpret any single indirect measurement as an absolute index of a specific
psychological attribute. Of course, this note of caution applies to direct measures as well. A
solution to this problem lies in the general diagnostic principle of convergence.
19
Conclusions about deviant or non-deviant sexual interest can be drawn with greater
confidence if they are based on several conceptually different, convergent, and valid,
indirect measures (De Houwer et al., 2009).
Limitations of the study
Some limitations of this study have to be taken into account regarding the
conclusions that can be drawn from the results. First, the limited number of child sex
offenders does not allow for strong inferences about differences between child sex offender
subgroups with different victim characteristics. Second, as already mentioned, this study
lacks a balanced proportion of homosexual men in the control sample. Hence, offender
status is confounded with sexual orientation. This may lead to underestimating the criterion
validity of deviant sexual interest measures. It may also result in overestimating the validity
of measures of sexual orientation for discriminating between offenders and controls.
However, an additional series of of hierarchical multiple regressions revealed that measures
of sexual interest in children (mainly the Boys-Men and Girls-Women IATs) independently
accounted for criterion variance. This finding corroborates that the criterion validity of the
EISIP was not only based on the homosexual-heterosexual discrimination. Nevertheless,
future studies should use balanced control samples.
Implications for the practical use of the EISIP
The results of the present study suggest that the Explicit and Implicit Sexual
Interest Profile (EISIP) could be useful as an additional diagnostic tool in several respects:
First, the EISIP is an economical (with respect to both assessment time and monetary costs)
computer-based assessment tool to assess socially problematic and/or inappropriate sexual
interest. Participants took 35 minutes on average to complete the EISIP test battery. The
assessment is non-intrusive and ethically acceptable due to the use of sexually non-explicit,
clothed stimuli not showing identifiable, real individuals. From a practical point of view the
20
EISIP appears to be an attractive and affordable complement to be used alongside
established PPG procedures. Incremental validity might be gained from converging data of
conceptually different measurement paradigms that tap into differential domains of deviant
sexual interest. First evidence into this direction is indicated by the associations with the
phallomerically validated SSPI scores across relevant EISIP categories. Hence, comparing
EISIP with PPG data would be an important next research step.
Second, because deviant sexual interest has proven to be one of the best predictors
of sexual recidivism (Hanson et al., 1998; 2005) the EISIP may prove useful to assess
sexual interest in the context of criminal prognosis – especially in settings where ongoing
sex offender risk assessment and -management is mandatory. It could also be used to
provide additional diagnostic information in cases where little actuarial and detailed file
data are available. In these instances, the EISIP may be useful in detecting deviant sexual
preferences that call for more intense monitoring and further dynamic risk assessment
strategies. Whether the EISIP measures are valid predictors of reoffending is at this point
an open empirical question that requires further research.
Interestingly, the sex offender sample not only showed higher levels of deviant
sexual interests than the control groups in this study, but also a generally higher level of
socially accepted sexual interests in adults (female and/or male). The first finding is in line
with the so-called sexual preference hypothesis (Freund & Blanchard, 1989). The latter
result of even stronger sexual interest in adults – at least at group level – is a little more
puzzling, although it is not totally unexpected from a clinical perspective. It is rather
common among child sex offenders to show adult sexual interest (e.g, Worling, 2006). Seto
and Lalumiѐre (2001) found that among the child sex offenders who scored lowest on the
SSPI (score 0-1) approximately 80% showed greater sexual arousal for adults than for
children; among those who scored highest (SSPI score 5) this pattern was still obtained in
21
28%. Hence, a large proportion of child sex offenders is genuinely sexually interested in
adults. It can be speculated whether sexual preoccupation – empirically a powerful risk
factor for sexual recidivism (Hanson, 2006) – is the underlying cause of the observed
increased level of sexual interest scores.
Last but not least, the EISIP may be useful in a therapeutic context such as to
confront denying child sex offenders with their deviant sexual interests in order to initiate a
process of questioning self-relevant assumptions about their child related sexual
preferences. In addition, the EISIP may also be used to identify socially acceptable sexual
interest that can be used as a resource in sex offender therapy.
In summary, the present research is a first step on the way of developing a multimethod measuring approach of sexual interest that can be further extended with other,
conceptually different measures ranging from phallometry to new latency-based measures
of sexual interest. With an increasing number of related but conceptually different
measures diagnostic decisions benefit from the aggregation principle that renders the
assessment more accurate, and makes it more difficult to fake. This is of great importance
since in the age of the internet it is easy to retrieve information about the underlying
rationale of any specific measurement method used in sex offender assessment. The
availability of a test battery of indirect measures of sexual interest with promising
psychometric properties now calls for a thorough evaluation of the usefulness of this
approach for child sex offender treatment, management, and the prediction of reoffending
especially in less select samples of more denying and treatment refusing child sex
offenders.
22
References
Abel, G. G., Jordan, A., Hand, C. G., Holland, L., & Phipps, A. (2001). Classification
models of child molesters utilizing the Abel Assessment of Sexual Interest. Child
Abuse & Neglect, 25, 703-718.
Abel, G. G., Huffmann, J., Warberg, B., & Holland, C. L. (1998). Visual reaction time and
plethysmography as measures of sexual interest in child molesters. Sexual Abuse. A
Journal of Research and Treatment, 10, 81-95.
Ahlers, C. J., Schaefer, G. A., & Beier, K. M. (2006). Das Spektrum der Sexualstörungen
und ihre Klassifizierbarkeit in DSM-IV und ICD-10 [The spectrum of sexual
disorders and their classifiability in ICD-10 and DSM-IV]. Sexuologie [Sexuology],
12, 120-152.
Beech, A. R., Kalmus, E., Tipper, S. P., Baudouin, J. Y., Flak, V. & Humphreys, G. W.
(2008). Children induce an enhanced attentional blink in child molesters.
Psychological Assessment, 20, 397-402.
Beech, A., Oliver, C., Fisher, D., & Beckett, R. (2005). STEP 4: The Sex Offender
Treatment Programme in prison: Adressing the offending behaviour of rapists and
sexual murderers. London: HM Prisonservice. Retrieved November 20, 2009, from
http://www.hmprisonservice.gov.uk/assets/documents/100013DBStep_4_SOTP_repo
rt_2005.pdf
De Houwer, J., Teige-Mocigemba, S., Spruyt, A., & Moors, A. (2009). Implicit measures:
A normative analysis and review. Psychological Bulletin, 135, 347-368.
Efron, B. (1983). Estimating the error rate of a prediction rule: Improvement on
crossvalidation. Journal of the American Statistical Association, 78, 316-331.
Freund, K., & Blanchard, R. (1989). Phallometric diagnosis of pedophilia. Journal of
Consulting and Clinical Psychology, 57, 100-105.
23
Gray, N. S., Brown, A. S., MacCulloch, A. J., Smith, J., & Snowden, R. J. (2005). An
implicit test of the associations between children and sex in pedophiles. Journal of
Abnormal Psychology, 114, 304-308.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual
differences in implicit cognition: The Implicit Association Test. Journal of
Personality and Social Psychology, 74, 1464-1480.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding and using the
Implicit Association Test: I. An improved scoring algorithm. Journal of Personality
and Social Psychology, 85, 197-216.
Gress, C. L. Z. (2005). Viewing time measures and sexual interest: Another piece of the
puzzle. Journal of Sexual Aggression, 11, 117-125.
Gress, C. L. Z., & Laws, D. R. (2009). Measuring sexual deviance: Attention-based
measures. In A. R. Beech, L. A. Craig, & K. D. Browne (Eds.). Assessment and
treatment of sexual offenders: A handbook (pp. 109-128). West Sussex, UK: Wiley.
Hanson, R. K. (2006). Stability and change: Dynamic risk factors for sexual offenders. In
W. L. Marshall, Y. L. Fernandez, L. E. Marshall, & G. A. Serran (Eds.). Sexual
offender treatment. Controversial issues (pp. 18-31). New York, NJ: Wiley
Hanson, R. K., & Bussiѐre, M. T. ( 1998). Predicting relapse: A meta-analysis of sexual
offender recidivism studies. Journal of Consulting and Clinical Psychology, 66, 348362.
Hanson, R. K., & Morton-Bourgon, K. E. (2005). The characteristics of persistant sexual
offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical
Psychology, 73, 1154-1163.
24
Harris, G. T., Rice, M. R., Quinsey, V. L., & Chaplin, T. C. (1996). Viewing time as a
measure of sexual interest among child molesters and normal heterosexual men.
Behavior Research and Therapy, 34, 389-394.
Imhoff, R., Schmidt, A. F., Nordsiek, U., Luzar, C., Young, A. W. & Banse, R. (in press).
Viewing time effects revisited: Prolonged response latencies for sexually attractive
targets under restricted task conditions. Archives of Sexual Behavior.
Kalmus, E., & Beech, A. R. (2005). Forensic assessment of sexual interest: A review.
Aggression and Violent Behavior, 10, 193-218.
Kroner, D. G. & Weekes, J. R. (1996). Balanced Inventory of Desirable Responding:
Factor structure, reliability, and validity with an offender sample. Personality and
Individual Differences, 21, 323-333.
Laws, D. R. (2003). Penile plethysmography: Will we ever get it right?. In T. Ward, D. R.
Laws, & S. M. Hudson (Eds.). Sexual deviance – Issues and controversies (pp. 82102). London: Sage.
Laws, D. R., & Gress, C. L. Z. (2004). Seeing things differently: The viewing time
alternative to penile plethysmography. Legal and Criminal Psychology, 9, 183-196.
Lanyon, R. I. (2001). Psychological assessment procedures in sex offending. Professional
Psychology: Research and Practice, 32, 253-260.
Letourneau, E. J. (2002). A comparison of objective measures of sexual arousal and
interest: Visual reaction time and penile plethysmography. Sexual Abuse: A Journal
of Research and Treatment, 14, 207-223.
Marshall, W. L. (2007). Diagnostic issues, multiple paraphilias, and comorbid disorders in
sexual offenders: Their incidence and treatment. Aggression and Violent Behavior,
12, 16-35.
25
Marshall, W. L., & Fernandez, Y. L. (2000). Phallometric testing with sexual offenders:
Limits to its value. Clinical Psychology Review, 20, 807-822.
Marshall, W. L., & Fernandez, Y. L. (2003). Sexual preferences: Are they useful in the
assessment and treatment of sexual offenders. Aggression and Violent Behavior, 8,
131-143.
Mokros, A., Dombert, B., Osterheider, M., Zappalà, A., & Santtila, P. (in press).
Assessment of pedophilic sexual interest with an attentional choice reaction time
task. Archives of Sexual Behavior.
Murphy, W. D., & Barbaree, H. E. (1994). Assessments of sex offenders by measures of
erectile response: Psychometric properties and decision making. Brandon, VT: Safer
Society.
Pacific Psychological Assessment Cooperation (2004). The Not-Real People Stimulus Set
for Assessment of sexual interest. Victoria, BC: Author.
Paulhus, D. L. (1998). The Balanced Inventory of Desirable Responding. Toronto/Buffalo:
Multi-Health Systems.
Rosenzweig, S. (1942). The photoscope as an objective device for evaluating sexual
interest. Psychosomatic Medicine, 4, 150-157.
Seto, M. C. (2008). Pedophilia and sexual offending against children: Theory, assessment,
and intervention. Washington, DC: American Psychological Association.
Seto, M. C., Harris, G. T., Rice, M. E., & Barbaree, H. E. (2004). The Screening Scale for
Pedophilic Interests and recidivism among adult sex offenders with child victims.
Archives of Sexual Behavior, 33, 455-466.
Seto, M. C., & Lalumiѐre, M. L. (2001). A brief screening scale to identify pedophilic
interests among child molesters. Sexual Abuse: A Journal of Research and
Treatment, 13, 15-25.
26
Smith, P., & Waterman, M. (2004). Processing bias for sexual material: The Emotional
Stroop and sexual offenders. Sexual Abuse: A Journal of research and Treatment, 16,
163-171.
Tanner, J. (1978). Foetus into man: Physical growth from conception to maturity.
Cambridge, MA: Harvard University Press.
Ward, T., & Beech, T. (2006). An integrated theory of sexual offending. Aggression and
Violent Behavior, 11, 44-63.
Ward, T., & Siegert, R. J. (2002). Toward a comprehensive theory of child sexual abuse: a
theory knitting perspective. Psychology, Crime and Law, 9, 319-351.
Worling, J. (2006). Assessing sexual arousal with adolescent males who have offended
sexually: Self-report and unobtrusively measured viewing time. Sexual Abuse: A
Journal of Research and Treatment, 18, 383-400.
27
Author Note
We wish to thank the Governors and personnel of HMP Aklington, Frankland, Full Sutton,
and Wakefield for help and support in conducting this study; John Fisher of HM Prison
Service Directorate of High Security for his invaluable support; and Sharon Griffiths,
Leona Hunter, Emma Longfellow, Ben Malinksi, Rachel Woodward, Ruth Philips, and
Emma Watson for their help in collecting the data as part of their dissertation work for the
MSc in Applied Forensic Psychology. Also, we are grateful to Marieke De Witte, Richard
Laws, and Bruno Verschuere for providing critical comments on an earlier draft.
Correspondence should be sent to Rainer Banse, Universität Bonn, Institut für Psychologie,
Kaiser-Karl-Ring 9, 53111 Bonn, Germany; or by email to banse@uni-bonn.de.
28
Biographical Sketches
Rainer Banse, PhD, is Professor for Social & Legal Psychology at the Department of
Psychology, University of Bonn. His main research interests are indirect assessment
methods in social and legal psychology, aggression, personal relationships, and personality
psychology.
Alexander F. Schmidt, PhD, is currently a post-doctoral researcher at the Department of
Psychology, Social & Legal Psychology, University of Bonn. His research interests include
forensic assessment, offender therapy, indirect measures, and self-regulation processes.
Jane Clarbour, PhD, is a Senior Lecturer at the University of York and a chartered Forensic
Psychologist. Her research interests are psychometrics, personality and antisocial
behaviour, as well as emotional development.
29
Table 1: Items and psychometric properties of the Explicit Sexual Interest Questionnaire
(ESIQ).
Part-whole corrected
item-total correlations
Man
Woman
Boy
Girl
I have enjoyed orally stimulating a ...
.91
.66
.62
.64
I have sexually caressed a ...
.92
.82
.82
.76
I have sexually penetrated a ... with my tongue or finger
.76
.80
.66
.73
I have sexually touched a ...
.92
.80
.71
.90
I have sexually penetrated a ... with my penis
.81
.82
.68
.62
.95
.91
.86
.89
I find it erotic if I see a ...’s beautiful chest
.93
.87
.87
.76
I have daydreamed of having sex with a ...
.85
.82
.66
.76
I find it erotic to see a ...’s body through the clothes
.90
.73
.75
.65
I find it erotic to see a ...’s beautiful legs or bottom
.84
.73
.77
.71
I get excited when I imagine that a ... stimulates me
.89
.77
.66
.75
Cronbach’s α
.96
.91
.89
.88
Cronbach’s α
.97
.94
.88
.90
Item
Behavior
Cronbach’s α
Fantasy
Total Score
Note. In the instruction it is stated, that “boy” and “girl” refers to prepubescent children
below the age of 12.
30
Table 2. Psychometric properties of social desirability, direct, and indirect measures of sexual preference.
Reliability (α)
Effect sizea (r)
Criterion Validity (ROC-analysis)
Child Sex Offenders
Boy Victims Only
Girl Victims Only
(n = 38)
(n = 14)
(n = 16)
All Controls
Offender Controls
Offender Controls
(n = 75)
(n = 37)
(n = 37)
Explicit Sexual Interest
Men
.97
.45
.41
.66
.74
Boys
.88
.50
.50
.72
.89
Women
.94
.37
-.37
.29
.24
Girls
.90
.49
.79
.45
.69
Men-Women
n/a
.79
.59
.42
.74
Boys-Men
n/a
-.16
.42
.36
.55
Girls-Women
n/a
.59
.83
.78
.83
Children-Adults (average)
n/a
.52
.71
.77
.87
Viewing Times
Men
.85
.54
.82
.89
.78
Boys
.85
.49
.80
.90
.86
Women
.86
.08
.56
.63
.74
Girls
.77
.44
.76
.81
.73
Men-Women
n/a
.46
.33
.72
.82
Boys-Men
n/a
.48
-.34
.29
.22
Girls-Women
n/a
.51
.43
.20
.61
Children-Adults (average)
n/a
.02
.51
.33
.46
IATs
Men-Women
.79
.13
.57
.63
.34
Boys-Men
.65
.16
.62
.60
.57
Girls-Women
.79
.67
.56
.37
.72
Children-Adults (average)
n/a
.60
.32
.71
.71
n/ab
SSPI
.85
BIDR averaged
Note. a Effect size Child Sex Offenders vs. All Controls; b see results section; n/a = not applicable; bold coefficients are significant at p < .05; values
below .50 can be turned into their complement (1-value below .50) when testing direction is recoded.
31
Table 3. Correlations of direct and indirect measures of sexual interest, social desirability, and the Screening Scale for Pedophilic Interests.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
Explicit Sexual Interest
1. in Men
2. in Women
-.70
3. in Boys
.65
-.62
4. in Girls
.06
.08
.01
.42
-.45
.32
.16
.47
-.31
.48
.28
-.16
.17
-.18
.08
-.17
.07
-.09
-.04
.34
.23
-.29
.24
.27
.28
-.11
.41
.35
.79
.47
.13
-.10
.11
.51
.12
-.08
.40
.37
.55
.48
17. Men-Women
.36
-.15
.28
.19
.15
-.11
.19
.12
18. Boys-Men
.12
-.04
.11
.09
.11
-.15
.14
.06
.13
.27
-.22
.32
.19
.23
-.23
.27
.14
.44
.17
.26
-.17
.29
.18
.21
-.25
.26
.11
.40
.73
.84
-.05
.14
-.04
.16
.14
.06
-.01
.05
-.04
.05
-.04
-.02
-.03
.16
.09
.07
.11
-.24
.60
-.27
.29
.07
.42
-.06
.19
.03
.42
-.49
.09
-.19
.39
.12
.17
.21
5. Men-Womena
6. Boys-Mena
7. Girls-Womena
8. Children-Adultsb
Viewing Times
9. Men
10. Women
11. Boys
12. Girls
.62
a
13. Men-Women
14. Boys-Mena
15. Girls-Womena
16. Children-Adultsb
IATs
19. Girls-Women
b
20. Children-Adults
BIDR averaged
.05
-.22
.04
SSPI
.56
-.50
.58
a
b
Note. Difference scores based on single interest measures; Based on differences in mean combined single interest measures; only meaningful correlations are reported; bold sizes are significant at p < .05, two-tailed.
32
Table 4. Mean differences and standard deviations of age, BIDR, SSPI, direct, and indirect
measures of sexual preference across child sex offender and control groups.
Measures
Age
SSPIac
BIDR averaged
Child sex offenders
Boy
Girl
Mixed
victims Victims victims
only
only
(n = 14) (n= 16) (n = 8)
47.2 AB
50.9 A
48.4 AB
(12.3)
(9.1)
(10.2)
4.79 A
2.62 B
5.00 A
(0.43)
(0.50)
(0.00)
3.21A
3.05 A
2.98 A
(0.59)
(0.46)
(0.41)
Offender
Controls
(n= 37)
35.2 C
(9.1)
Nonoffender
Controls
(n = 38)
41.8 BC
(8.8)
10.20***
(108)
n/a
n/a
32.01***
2.91 A
(0.34)
2.79 A
(0.51)
2.38
(108)
5.88***
(30.37)
Group effect
F(4, (df2))
Explicit Sexual Interest
in Menbc
in Boysac
in Womenbc
in Girlsbc
Men-Womenbc
Boys-Menbc
Girls-Womenbc
Children-Adultsbc (average)
0.50 A
(0.46)
0.33 A
(0.23)
0.51 A
(0.45)
0.01 A
(0.05)
-0.01 A
(0.77)
-0.17 A
(0.41)
-0.50 A
(0.44)
-0.34 AB
(0.25)
0.02 B
(0.08)
0.00 B
(0.00)
0.93 B
(0.09)
0.31 B
(0.34)
-0.91 B
(0.14)
-0.02 A
(0.08)
-0.61 A
(0.36)
-0.32 A
(0.16)
0.45 AB
(0.38)
0.28 AB
(0.32)
0.73 AB
(0.34)
0.35 AB
(0.29)
-0.28 AB
(0.61)
-0.18 A
(0.32)
-0.39 A
(0.31)
-0.28 AB
(0.27)
0.05 B
(0.18)
0.00 B
(0.00)
0.95 B
(0.15)
0.01 A
(0.05)
-0.90 B
(0.32)
-0.05 A
(0.18)
-0.94 B
(0.16)
-0.49 B
(0.04)
0.03 B
(0.16)
0.01 B
(0.02)
0.94 B
(0.18)
0.02 A
(0.07)
-0.91 B
(0.33)
-0.02 A
(0.16)
-0.92 B
(0.19)
-0.47 B
(0.06)
4035 A
(1933)
2656 A
(948)
3404 A
(1774)
2438 A
(872)
631 A
(1448)
-1380 A
(1104)
-967 A
(1161)
-1173 A
(918)
2679 A
3459 A
(863)
(1190)
2291 A 2385 AB
(830)
(945)
3595 A
3460 A
(1181)
(2085)
2516 AB 2573 AB
(1045)
(1480)
-1 AB
-916 BC
(2777)
(1063)
-388 AB -1074 AB
(603)
(816)
-1079 A -887 A
(1391)
(2010)
-734 A
-980 A
(773)
(1072)
1826 B
(770)
1356 B
(529)
2637 A
(1362)
1604 B
(671)
-811 BC
(1174)
-469 AB
(623)
-1033 A
(1208)
-751 A
(719)
2048 AB
(660)
1815 A
(636)
3795 A
(1624)
1654 B
(553)
-1746 C
(1551)
-233 B
(548)
-2141 A
(1310)
-1187 A
(634)
73.57***
3.54*
(30.07)
5.14**
(28.78)
6.20***
(30.55)
0.98
(30.51)
10.67***
(28.05)
7.14***
(26.90)
Viewing Times (ms)
Menbc
Boysbc
Women
Girlsbc
Men-Women
Boys-Menbc
Girls-Women
Children-Adults
(average)
IATs
8.47***
(28.69)
10.05***
(28.90)
2.82*
(107)
4.86**
(28.47)
7.59***
(107)
4.78**
(29.49)
4.53**
(107)
2.19
(107)
0.03 A
-0.42 A -0.15 A
-0.21 A
-0.44 A 2.91*
(0.52)
(0.36)
(0.61)
(0.41)
(0.54)
(99)
-0.01 A -0.11 A
0.02 A
-0.13 A
-0.19 A 1.01
Boys-Men
(102)
(0.30)
(0.36)
(0.16)
(0.28)
(0.37)
-0.06 A -0.20 AB 0.01 A
-0.31 AB
-0.52 B 6.19***
Girls-Women
(0.36)
(0.31)
(0.46)
(0.40)
(0.34)
(102)
-0.06 A -0.19 AB 0.00 A
-0.23 AB
-0.35 B 5.02***
Children-Adults (average)
(0.28)
(0.31)
(0.25)
(0.27)
(106)
(0.26)
Note. * = p < .05; ** = p < .01; *** = p < .001. Group means with different subscripts in one row are
statistically different (Student-Newman-Keuls-Test, p < .05). a = Non-parametric Kruskal-Wallis-Test (exact
χ2) b = Welch-Test c = Dunnett’s C post-hoc comparisons, p < .05. n/a = not applicable.
Men-Women
33
Table 5. Criterion validity (ROC-analysis) after crossvalidated binary logistic regression
(Controls vs. Child Sex Offenders).
Correct
Crossvalidated
Measures
AUC
Nagelkerke’s R2
Classifications
Optimism
Direct + Indirect
90%
2.0%
.95***
.75
Direct
87%
3.5%
.88***
.63
Indirect
85%
2.0%
.88***
.55
Viewing Time
80%
2.7%
.86***
.48
IAT
77%
2.9%
.77***
.23
Note. *** = p < .001; all regression models p < .001, all Hosmer-Lemeshow-Tests are
non-significant.