ArticlePDF Available

Self-reported criminal and anti-social behavior on a dating site: the importance of cognitive ability

Authors:
  • Ulster Institute for Social Research

Abstract and Figures

The relationship between criminal and antisocial (CAS) behaviors and cognitive ability (CA) were examined in a large online sample of dating site users (complete sample n = 68,371). 12 question items were found that measured CAS to some degree. Of these, 11 showed a negative relation to CA. The answers to the CAS items were all positively related, suggesting the existence of a general factor of CAS behavior. Scores for this factor were estimated using multiple methods. The resulting scores were then subjected to a series of regression models to examine whether the link between CA and CAS would hold up in the presence of other predictors. The results showed that the link between CA and CAS scores was robust to model specifications with standardized betas of -.15 to -.20. Furthermore, a CA x sex interaction was found such that the CA x CAS relationship was stronger for men (r’s -.20 and -.13, for men and women, respectively).
Content may be subject to copyright.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
Published 10th January 2018
Submitted 22nd October, 2016
Self-reported criminal and anti-social
behavior on a dating site: the
importance of cognitive ability
Emil O. W. Kirkegaard1
Abstract
The relationship between criminal and antisocial (CAS) behaviors and cognitive ability (CA) were
examined in a large online sample of dating site users (complete sample n = 68,371). 12 question items
were found that measured CAS to some degree. Of these, 11 showed a negative relation to CA. The
answers to the CAS items were all positively related, suggesting the existence of a general factor of
CAS behavior. Scores for this factor were estimated using multiple methods. The resulting scores were
then subjected to a series of regression models to examine whether the link between CA and CAS
would hold up in the presence of other predictors. The results showed that the link between CA and
CAS scores was robust to model specifications with standardized betas of -.15 to -.20. Furthermore, a
CA x sex interaction was found such that the CA x CAS relationship was stronger for men (r’s -.20 and
-.13, for men and women, respectively).
Key words: cognitive ability, intelligence, IQ, crime, antisocial behavior, dating site, OKCupid, sexual
orientation
1. Introduction
Numerous studies show that criminal and antisocial (CAS) behaviors are negatively related to cognitive
ability (Ellis, Beaver, & Wright, 2009, p. 150; Frisell, Pawitan, & Långström, 2012; Herrnstein &
Murray, 1994; Hirschi & Hindelang, 1977; Høgh & Wolf, 1983; Levine, 2011; Schwartz et al., 2015).
1 Ulster Institute for Social Research. emil@emilkirkegaard.dk
Page 1 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
Although there does not seem to be a proper quantitative meta-analysis yet (but see Ttofi et al., 2016),
effect sizes tend to be around -.10 to -.20 on the Pearson correlation/standardized beta scale. As far as
the author knows, no previous study has examined self-reported CAS behaviors in an online dating
sample. As such, the purpose of the present study was to examine the validity of cognitive ability (CA)
in this population.
2. Data
The data came from the dating site OKCupid (www.okcupid.com) and is described in detail in
Kirkegaard & Bjerrekær (2016). Users on this site answer questions (multiple choice format with 2-4
options) in order to be better matched with potential mates using the site's algorithm. Most users
answer the questions in public, meaning that other users can see the selected answers, and they were so
obtained by web scraping the site (i.e. using a script that automatically visits users and saves their
information to a spreadsheet-like database). Most users were living in English-speaking countries, in
particular the United States (65%), the United Kingdom (12%) and Canada (3%). In total, there are
data for 68,371 users. However, since answering questions is voluntary, many users don’t answer any
or only a limited number. For this reason, most of the cells are missing (77%; 42 million cells with data
available).
CAS behaviors. To identify relevant items among the ~2,500 in the dataset, the following keywords
were used to search on the question text: crim, steal, stole, hit, kick, violen, police, arrest, prison.2 A
total of 12 plausible items were found:
1. Arrested (q252): Have you ever been arrested, even if just for a small crime or misdemeanor?
Yes
No
2. Prison (q1138): Have you ever been to prison?
Yes
No
Just to visit / I was working
3. Punched in face (q196): Excluding childhood fights, have you ever punched someone in the
face?
Yes
No
2 Drug use items were found but were excluded on purpose due to the special nature of such behavior.
Page 2 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
4. Cheated exam (q400): Have you ever cheated on an exam?
Yes
No
5. Would tax cheat (q180): Would you cheat on your taxes, if you were absolutely 100% sure you
could get away with it?
Yes
No
6. Stole glass from bar (q59919): Have you ever stolen a glass from a bar?
Yes
No
7. Used fake ID (q22569): Have you ever used a fake ID to do or acquire something you were
legally barred from as a result of your age?
Yes
No
8. Torture animal for fun (q19928): Honestly, did you ever torture a cat, dog, or any other furry
animal for pleasure?
Yes, but I regret it.
NO WAY!
Yeah, that's fun.
No, but I would do it.
9. Steal newspapers (q39226): You stop to pick up a newspaper and notice that the coin-operated
dispenser was not completely closed. No one is around so you have the opportunity to take a
paper without paying. Which of the following would you do?
Pay for a paper and close the dispenser.
Steal a paper and close the dispenser.
Steal a paper and leave the dispenser open.
Steal all of the remaining papers.
10. Litter (q17017): Do you litter?
Page 3 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
Often
Rarely
Never
11. Cigarette littering (q74381): Do you consider the act of leaving cigarette butts on the ground to
be littering?
Yes
No
12. Hit significant other in anger (q81783): Have you ever hit a significant other in anger?
Yes
No
The items were recoded into dichotomous form to simplify the analysis, with 1 as indicating that the
user affirms having done the CAS behavior. For items with >2 response options, a judgment call was
made how to recode them. This was done based on the distribution of scores within the response
categories (not too small, and avoid heterogeneity). For instance, item 9 was coded to combine all the
responses that involved stealing.
Table 1 shows descriptive statistics for the items.
CAS var n Proportion
arrested 12456 0.24
prison 15236 0.02
punched in face 18128 0.34
cheated exam 5155 0.40
would tax cheat 15581 0.42
stole glass from bar 16747 0.44
used fake id 14067 0.28
torture animal for fun 10725 0.03
steal newspapers 2111 0.43
litter 35221 0.31
Page 4 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
cigarette littering 18983 0.11
hit SO in anger 3405 0.03
Table 1: Descriptive statistics for criminal or antisocial outcomes.
Age. Users state their age in their profiles. The mean age for users with data for cognitive ability was 32
with a standard deviation of 7.8. This is somewhat lower than the general population which is around
40, but substantially higher and more varied than typical college samples (Henrich, Heine, &
Norenzayan, 2010).
Sex/gender and sexual orientation. Users almost always state their sex/gender(s) in their profiles.
Because very few users selected a gender other than “Man” or “Woman”, data for these users were
excluded (0.24%). A previous study found that sexual orientation was a useful predictor of criminal
outcomes (K. Beaver et al., 2016). Because this variable likely interacts with sex/gender, the two were
combined yielding 6 combinations between hetero-, bi- and homosexual, and male/female. Table 2
shows the breakdown of the gender-sexual orientation variable.
Group Count Percent
Heterosexual male 16249 62.94
Heterosexual female 5859 22.70
Bisexual female 1426 5.52
Homosexual male 1236 4.79
Bisexual male 548 2.12
Homosexual female 301 1.17
(missing) 196 0.76
Table 2: Distribution of gender-sexual orientation.
Self-identified race/ethnicity (SIRE). Users report their SIRE on their profiles. Since it was possible to
select more than one, this presented a coding problem. Two different codings were used. In common
combinations, persons were classified as their chosen combination of SIREs. After this, the
combinations with less than 100 persons were recoded into ‘Other combos’. In dummy coding, a new
binary variable was created for each atomic SIRE in the data as well as ‘Multi SIRE’, which was a
dummy for whether the user had selected more than one option. This dummy was included based on
previous research indicating that multi-racial persons are at elevated risk for a variety of bad outcomes
(Choi, Harachi, Gillmore, & Catalano, 2006; Udry, Li, & Hendrickson-Smith, 2003). Table 3 shows the
breakdown of SIRE using the combinations coding.
Page 5 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
Group Count Percent
White 18261 70.74
(missing) 1741 6.74
Other combos 1197 4.64
Asian 958 3.71
Hispanic / Latin 779 3.02
Black 775 3.00
Other 606 2.35
Hispanic / Latin, White 454 1.76
White, Other 292 1.13
Native American, White 244 0.95
Indian 226 0.88
Asian, White 175 0.68
Black, White 107 0.41
Table 3: Distribution of self-identified race/ethnicity (SIRE) using the combinations coding.
Cognitive ability (CA). CA was estimated based on users' answers to 14 items as described in a
previous publication (Kirkegaard & Bjerrekær, 2016). The items cover a variety of domains (including
verbal, spatial, mathematical ability and general knowledge) and were taken as an estimate of general
cognitive ability/general intelligence/g (Jensen, 1998). The exact items (question text, response
options), sample sizes and pass rates can be found in the supplementary materials. While brief, scores
from this ad hoc test were previously shown to be related to known correlates, e.g. religious belief
(Dutton, 2014; Zuckerman, Silberman, & Hall, 2013), with typical effect sizes (Kirkegaard &
Bjerrekær, 2016). There were data for ~56k users, however, to avoid using unreliable CA estimates,
only users who answered at least 5 items were retained, yielding a sample size of 25,815. Figure 1
shows the distribution of CA in the reduced sample.
Page 6 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
As can be seen, despite the nature of the test, the distribution of scores was approximately normal
(skew = -0.54, kurtosis = -0.22). In terms of the original standardization on the complete dataset, this
subsample represented a selected group as the mean CA was 0.56 z (sd = 0.61). The CA scores were
then restandardized for this subsample.
3. Analyses
All analyses were done in R. An R notebook is available in the supplementary materials.
3.1. Group differences per item
The simplest approach to analyzing the data is to calculate the mean CA by CAS item. Table 4 shows
the results.
CAS d d lower d upper n
arrested -0.24 -0.19 -0.28 9653
prison -0.34 -0.19 -0.48 12336
punched in face -0.26 -0.23 -0.30 13784
cheated exam -0.13 -0.07 -0.20 4079
would tax cheat -0.15 -0.11 -0.18 11728
stole glass from bar -0.03 0.01 -0.06 12626
Page 7 of 20.
Figure 1: Distribution of cognitive ability.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
used fake id 0.02 0.06 -0.02 10811
torture animal for fun -0.35 -0.23 -0.48 8989
steal newspapers -0.26 -0.16 -0.35 1727
litter -0.38 -0.35 -0.41 23256
cigarette littering -0.49 -0.44 -0.55 14905
hit SO in anger -0.47 -0.24 -0.70 2876
Table 4: Mean cognitive ability by criminal/anti-social variable: cases and controls.
Of the 12 CAS items, 11 were negatively related to CA, and 10 beyond chance levels. Clearly, self-
reporting CAS behavior (or intentions) is quite consistently related to lower CA, even when done on a
dating site for other users to see.
3.2. An overall CAS score
A general CAS (reflective) factor was hypothesized based on analogies with other general factors such
as those of cognitive ability (Jensen, 1998) and psychopathology (Caspi et al., 2014). To determine
whether it was sensible to calculate an overall CAS score, the correlations between the items were
calculated. Because the items were coded as dichotomous, latent correlations were used to prevent
artificially low correlations (tetrachoric, see Uebersax, 2015). Of the 66 inter-item correlations, 100%
were positive, ranging from .02 to .72, with a mean of .22. The results were nearly identical for the full
sample (mean = .24, range .02 to .73), thus was not an artifact of subsetting by CA item coverage.
Despite the observed positive manifold among items, it is not straightforward to calculate an overall
score by person: there is massive, non-random missing data across items. This missingness both
reflects the fact that not all users take the time to answer thousands of questions on the site, and due to
systematic skipping of items (investigated in Section 3.4). The amount of missing data was judged too
large for reliable analysis and the sample must thus be further subsetted before. To decide on an optimal
amount of data to retain, subsets of the sample were created that had at least 13, ..., 1 CAS datapoints,
not necessarily the same items (e.g. one case might have items 1-5, while another might have 6-10, but
both have 5 items). Figure 2 shows the sample sizes by the item count.
Page 8 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
A steep drop is seen at item 7, and this was chosen at the minimum number of items needed for
analysis, yielding a sample of n = 7,882.
Two approaches were then used to score the CAS factor. In the first approach, the missing data were
noiselessly imputed using the IRMI algorithm (Templ, Kowarik, & Filzmoser, 2011). After this, the
data were scored using both unweighted summation and item-response theory (IRT) analysis. The IRT
approach used the 2-parameter normal model, as implemented in the psych package (Revelle, 2017). In
the second approach, the data were analyzed with IRT without any initial imputation. This is possible
because IRT analysis allows missing data (for details about how psych deals with missing data, see the
documentation for the scoreIrt function). Thus, in total, there were 3 sets of scores for each case. Table
5 shows the correlations between the CAS scores as well as CA.
Crime sum imp Crime score Crime score imp CA CAS complete
Crime sum imp 1.00 0.90 0.94 -0.18 0.04
Crime score 0.90 1.00 0.82 -0.13 0.05
Crime score imp 0.94 0.82 1.00 -0.16 0.09
CA -0.18 -0.13 -0.16 1.00 -0.03
CAS complete 0.04 0.05 0.09 -0.03 1.00
Table 5: Correlations between estimates of general criminal and anti-social (CAS) behavior scores,
Page 9 of 20.
Figure 2: Percent of cases by number of criminal and antisocial item
datapoints available.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
cognitive ability and number of CAS items answered. imp = based on imputed data.
There are several things of note. First, the correlations between the estimates of general CAS behavior
were quite strong, from .82 to .94. Thus, the exact method choice is unlikely to seriously distort results.
Second, all three estimates were negatively related to CA with quite typical effect sizes (r’s -.13 to
-.18). The strongest correlations were seen for methods that imputed the missing data beforehand.
Third, the number of answered CAS items was not strongly related to any other variable, suggesting
that bias from selective reporting was not strong. The scores from the simple summation were used for
further analysis because these had the highest correlations with the other general CAS behavior
estimates, the strongest relationship to CA, and the weakest relationship to the number of items
answered. Figure 3 shows the mean CA by each general CAS score.
A fairly linear trend was observed, in line with previous research (Frisell et al., 2012).
3.3. Multivariate analyses
The relationship seen between CAS score and CA might be inflated or deflated by the presence of
Page 10 of 20.
Figure 3: Mean cognitive ability by simple criminal and anti-social behavior score. n = 7,882.
Correlation = -.18. Error bars indicate 95% confidence intervals. The last two bars each contain only
a single individual and hence error bars are not shown.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
variation in other predictors. To examine whether this was the case, OLS regression was used to predict
CAS score from CA as well as the control variables. The primary model results are shown in Table 6.
Predictor Beta SE p
Cognitive ability -0.153 0.011 <0.0001
Heterosexual male (ref)
Bisexual male 0.057 0.079 0.476
Homosexual male -0.085 0.056 0.126
Homosexual female -0.096 0.108 0.375
Bisexual female -0.157 0.058 0.006
Heterosexual female -0.265 0.029 <0.0001
age -0.019 0.012 0.097
age' 0.010 0.081 0.904
age'' 0.158 0.330 0.632
age''' -0.410 0.423 0.332
White -0.027 0.049 0.586
Black -0.030 0.063 0.635
Asian -0.053 0.064 0.406
Hispanic 0.074 0.059 0.212
Native American -0.036 0.077 0.644
Indian -0.017 0.097 0.857
Middle Eastern 0.148 0.099 0.137
Pacific Islander 0.049 0.129 0.705
Other 0.206 0.063 0.001
Multi-SIRE 0.038 0.062 0.542
Table 6: Primary model. Outcome: general criminal and anti-social behavior score. n = 7,410. R adj.
Page 11 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
= .207. Apostrophes denote restricted cubic spline terms.
The primary model produced a slightly smaller beta for CA (.153) than the bivariate correlation (.177).
The addition of the control variables did not do much to improve predictive validity, as CA was almost
equally good to the combined set of predictors (r = .177 vs. R adj. = .207). A weak trend was observed
between gender-sexual orientation involving non-heterosexuals and CAS such that crime level were
highest for heterosexual men, lowest for heterosexual women and intermediate in roughly
monotonically falling fashion between (except for bisexual men, who were slightly above heterosexual
men). Thus, the findings are roughly in line with previous research (K. Beaver et al., 2016). However,
the sample size of the present study was too small to allow for precise results.
To examine robustness, a number of model variations were tried. First, the alternative coding of SIRE,
common combinations, was used instead. This produced a slight decrease in model fit and essentially
no change in the beta for CA. Second, a model was fit only on the White (only SIRE) sub-sample. This
produced a slight decrease in model fit (R adj. = .197) and a slight decrease in the beta for CA (-.163).
Third, a model was fit with the inclusion of an interaction between CA and gender-sexual orientation.
This fit slightly better (R adj. = .209), owning to an interaction between heterosexual female and CA
(0.07, p = .011), indicating that CA was a less useful predictor for female heterosexuals (or females in
general). This was confirmed in a simple subgroup analysis: the CA x CAS correlations were -.20 and
-.13, for men and women respectively. The model also found a slightly stronger main effect of CA
(-.171). Fourth, a model was fit by excluding persons with a CA score below -2. This group of persons
likely constitute people who refuse to answer IQ-like questions on a dating site rather than people with
a particularly low level of ability, and they are thus likely to disrupt the pattern in the data. This model
fit slightly better (R adj. = .211) and further strengthened the effect of CA (-.200). Fifth, finally, a
model with a nonlinear effect of CA was fit, using a restricted cubic spline (same as for age). This fit
slightly better still (R adj. = .218), though at the price of being less interpretable. Figure 4 shows a
comparison between the two model predictions for the effect of CA based on simulated data (the other
variables were set to their mean or modal value: White-only, male, heterosexual, sample mean age
[33.7]).
Page 12 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
As can be seen, the nonlinear model fit suggested that the influence of CA on CAS was greater at
higher levels of CA.
3.4. Systematically skipping of CAS items
The results in Table 4 indicated that non-random skipping of CAS items was not a big issue, though it
may be a small issue. To further investigate this, a logistic model was fit for each CAS item, with the
dependent outcome being whether the item was skipped or not. The predictors in the model were the
total number of questions answered by the user, their CA level, their crime score as estimated by IRT
based on the non-imputed data, as well as their age and gender-sexual orientation. Theoretically, this
should allow one to spot whether subjects’ CA or crime scores are related to skipping particularly
items, holding their overall number of items answered constant. Table 7 shows the results.
CAS Questions answered CA Crime score R2 adj.
arrested -0.90 0.14 0.26 0.09
prison -1.13 -0.42 -0.54 0.22
Page 13 of 20.
Figure 4: Model predictions for linear (red) and nonlinear (blue) models.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
punched in face -1.10 0.19 0.10 0.09
cheated exam -2.17 0.02 0.07 0.40
would tax cheat -1.19 0.35 0.31 0.16
stole glass from bar -1.10 0.19 -0.07 0.10
used fake id -1.10 0.20 0.18 0.11
torture animal for fun -1.54 -0.37 -0.06 0.25
steal newspapers -2.48 -0.06 -0.07 0.44
litter -0.29 -0.03 0.08 0.03
cigarette littering -0.99 0.08 0.34 0.09
hit SO in anger -0.83 -0.30 0.01 0.12
Table 7: Results of non-random item skipping models. Numbers in columns 2-4 are betas from a
logistic regression. Betas for age and gender-sexual orientation not shown.
There was no consistent pattern in the data. For some items, higher CA/crime score predicted that
persons would avoid disclosure, while for others, lower CA/crime score predicted avoiding disclosure.
The predictability of the non-random skipping varied strongly by CAS item with the most predictable
being whether one would cheat on exams or steal newspapers from a stand. The strongest results for the
two predictors of interest were for having been imprisoned, where it was found that persons with lower
CA and lower crime scores were more likely to not disclose their status; and the second strongest were
seen for tax evasion, where both persons with higher CA and higher crime scores were more likely to
not disclose their intentions.
4. Discussion and conclusion
The present study observed small to medium negative relationships (betas around -.15 to -.20) between
most criminal and antisocial (CAS) behaviors and cognitive ability (CA). These relationships were
robust to the addition of age (nonlinear), gender-sexual orientation, and self-identified race/ethnicity
(SIRE) predictors. An interaction between sex was found such that CA tended to be a stronger predictor
for males. A number of alternative model specifications were tried and produced essentially identical
results. As such, the findings are highly congruent with the literature which reports a robust relation
between lower CA and various measures of CAS behaviors (Frisell et al., 2012; Herrnstein & Murray,
1994; Høgh & Wolf, 1983; Levine, 2011; McGloin & Pratt, 2003).
A trend towards a gender continuum finding was found, as previously reported by Beaver et al (2016),
Page 14 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
such that e.g. homo- and bisexual females tended to have higher CAS scores than heterosexual females.
However, the model estimates were too imprecise for one to be able to draw any strong conclusions
from this study.
It is perhaps somewhat surprising that many people are willing to admit to serious crimes or poor
behaviors on a dating site for potential partners to see. It's possible that a tendency to admit this is
related to CA, which would bias the estimate of the relationship. If smarter people are more likely to
admit having been arrested given that they have been arrested, this would bias the correlation towards
1. If the admit tendency x CA relationship was the other way, the bias would be towards -1. It has been
found that smarter people are more honest (not just self-report more honesty), so this suggests the bias
is towards 1, not -1 (Paulhus & Dubois, 2015; Ruffle & Tobol, 2016), which would thus tend to deflate
any CA x CAS relation. However, when non-random skipping behavior was examined, no consistent
pattern was found. Rather, it seems that smarter people deliberately avoided answering some CAS
items, while the tendency was reversed for other items. In general, the study of biases in self-reported
CAS data deserve further study.
The study has a number of limitations. First, the final dataset used consisted of a subset of a self-
selected online sample. Persons wanting to date are not a random subset of the population, but are
younger and much more likely to be single. In general, people who use the internet tend to be brighter,
so the sample is probably also somewhat selected for CA. This would tend to lower the observed
correlations due to reduced variance (Hunter & Schmidt, 2004). Prisoners usually do not have access to
dating sites3, so they would tend to be missing from the sample. This would reduce the variance for
some of the criminal outcomes and thus also reduce the observed correlations. This problem was
further exacerbated by the subsetting of the sample for higher quality CA estimates. One particular
interesting finding was that self-identified race/ethnicity did not seem to be much related to CAS,
despite the known relationships based on other data sources (K. M. Beaver, Barnes, & Boutwell, 2015;
Herrnstein & Murray, 1994; New Century Foundation, 2005). This may be related to differential
lying/non-disclosure by SIRE (Hindelang, 1981; Piquero, Schubert, & Brame, 2014), the self-selection
of the sample, or reflect biases in the justice system (Alexander, 2014; K. Beaver et al., 2013).
Second, some of the CAS items were unclear in their interpretation. Punching someone in the face
(item 3) might be in self-defense, part of a consensual fight (e.g. boxing), or as part of the job (e.g.
police officer). Wanting to cheat on taxes if it could never be found out is a hypothetical, not an actual
action, and thus may never be instantiated given that the chance of discovery is never 0% in real life.
Despite these problems of interpretation, all the outcomes were positively related to each other. Thus,
there seems to be a general CAS factor. Such a pattern has also been found at the aggregate-level in a
study of London boroughs (Kirkegaard, 2016), where the rates of different violations were all
positively related.
3 It should be said that there are many active dating sites for dating inmates, but they involve sending letters since prison
inmates usually do not have internet access (Wilson, 2013).
Page 15 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
Third, the quality of the cognitive data was low to medium. While the cognitive test scores seem to
function well based on correlations with known correlates (i.e. calibration), the test is very brief based
on a max of 14 items (Kirkegaard & Bjerrekær, 2016). The test has not been validated on another
sample, so it's test-retest reliability is not known, and it is not known whether it suffers from
measurement bias. The probable low reliability of the test would tend to decrease the observed
correlations. It’s not clear which direction, if any, measurement bias would distort the relationship to
CAS scores in. The lower reliability, however, would certainly cause the relation to appear weaker than
it is.
Overall, though the study has a number of problems, these would mainly tend to decrease the observed
relationships. The fact that despite these problems negative relationships can still be seen testifies to the
robustness of the relationship between CAS behaviors and CA.
Supplementary material and acknowledgments
Peer review thread: https://openpsych.net/forum/showthread.php?tid=294
Study files: https://osf.io/7htpz/
Thanks to Gerhard Meisenberg and John P. Wright for reviewing.
References
Alexander, S. (2014, November 26). Race and Justice: Much More Than You Wanted To Know.
Retrieved October 17, 2017, from http://slatestarcodex.com/2014/11/25/race-and-justice-much-
more-than-you-wanted-to-know/
Beaver, K., Connolly, E. J., Schwartz, J. A., Boutwell, B. B., Barnes, J. C., & Nedelec, J. L. (2016).
Sexual Orientation and Involvement in Nonviolent and Violent Delinquent Behaviors: Findings
From the National Longitudinal Study of Adolescent to Adult Health. Archives of Sexual
Behavior, 45(7), 1759–1769. https://doi.org/10.1007/s10508-016-0717-3
Beaver, K., Delisi, M., Paul Wright, J., Boutwell, B., Barnes, J. C., & Vaughn, M. (2013). No evidence
of racial discrimination in criminal justice processing: Results from the National Longitudinal
Study of Adolescent Health. Personality and Individual Differences, 55.
https://doi.org/10.1016/j.paid.2013.01.020
Beaver, K. M., Barnes, J. C., & Boutwell, B. B. (Eds.). (2015). The nurture versus biosocial debate in
Page 16 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
criminology: on the origins of criminal behavior and criminality. Thousand Oaks, California:
SAGE.
Caspi, A., Houts, R. M., Belsky, D. W., Goldman-Mellor, S. J., Harrington, H., Israel, S., … Moffitt, T.
E. (2014). The p Factor One General Psychopathology Factor in the Structure of Psychiatric
Disorders? Clinical Psychological Science, 2(2), 119–137.
https://doi.org/10.1177/2167702613497473
Choi, Y., Harachi, T. W., Gillmore, M. R., & Catalano, R. F. (2006). Are Multiracial Adolescents at
Greater Risk? Comparisons of Rates, Patterns, and Correlates of Substance Use and Violence
Between Monoracial and Multiracial Adolescents. The American Journal of Orthopsychiatry,
76(1), 86–97. https://doi.org/10.1037/0002-9432.76.1.86
Dutton, E. (2014). Religion and intelligence: an evolutionary analysis. London: Ulster Institute for
Social Research.
Ellis, L., Beaver, K. M., & Wright, J. P. (2009). Handbook of crime correlates. Amsterdam; Boston:
Elsevier/Academic Press.
Frisell, T., Pawitan, Y., & Långström, N. (2012). Is the Association between General Cognitive Ability
and Violent Crime Caused by Family-Level Confounders? PLoS ONE, 7(7).
https://doi.org/10.1371/journal.pone.0041783
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world? The Behavioral
and Brain Sciences, 33(2–3), 61-83-135. https://doi.org/10.1017/S0140525X0999152X
Herrnstein, R. J., & Murray, C. A. (1994). The bell curve: intelligence and class structure in American
life (1st Free Press pbk. ed). New York: Simon & Schuster.
Hindelang, M. J. (1981). Variations in Sex-Race-Age-Specific Incidence Rates of Offending. American
Sociological Review, 46(4), 461–474. https://doi.org/10.2307/2095265
Hirschi, T., & Hindelang, M. J. (1977). Intelligence and Delinquency: A Revisionist Review. American
Sociological Review, 42(4), 571–587. https://doi.org/10.2307/2094556
Page 17 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
Høgh, E., & Wolf, P. (1983). Violent Crime in a Birth Cohort: Copenhagen 1953–1977. In K. T. V.
Dusen & S. A. Mednick (Eds.), Prospective Studies of Crime and Delinquency (pp. 249–267).
Springer Netherlands. Retrieved from http://link.springer.com/chapter/10.1007/978-94-009-
6672-7_14
Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis correcting error and bias in research
findings. Thousand Oaks, Calif.: Sage. Retrieved from http://site.ebrary.com/id/10387875
Jensen, A. R. (1998). The g factor: the science of mental ability. Westport, Conn.: Praeger.
Kirkegaard, E. O. W. (2016). Inequality among 32 London Boroughs: An S factor analysis. Open
Quantitative Sociology & Political Science, 1(1). Retrieved from https://openpsych.net/paper/45
Kirkegaard, E. O. W., & Bjerrekær, J. D. (2016). The OKCupid dataset: A very large public dataset of
dating site users. Open Differential Psychology. Retrieved from https://openpsych.net/paper/46
Levine, S. Z. (2011). Elaboration on the association between IQ and parental SES with subsequent
crime. Personality and Individual Differences, 50(8), 1233–1237.
https://doi.org/10.1016/j.paid.2011.02.016
McGloin, J. M., & Pratt, T. C. (2003). Cognitive Ability and Delinquent Behavior among Inner-City
Youth: A Life-Course Analysis of Main, Mediating, and Interaction Effects. International
Journal of Offender Therapy and Comparative Criminology, 47(3), 253–271.
https://doi.org/10.1177/0306624X03047003002
New Century Foundation. (2005). The Color of Crime: Race, Crime, and Justice in America. New
Century Foundation. Retrieved from http://www.colorofcrime.com/2005/10/the-color-of-crime-
2005/
Paulhus, D. L., & Dubois, P. J. (2015). The link between cognitive ability and scholastic cheating: A
meta-analysis. Review of General Psychology, 19(2), 183–190.
https://doi.org/10.1037/gpr0000040
Piquero, A. R., Schubert, C. A., & Brame, R. (2014). Comparing Official and Self-report Records of
Page 18 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
Offending across Gender and Race/Ethnicity in a Longitudinal Study of Serious Youthful
Offenders. Journal of Research in Crime and Delinquency, 51(4), 526–556.
https://doi.org/10.1177/0022427813520445
Revelle, W. (2017). psych: Procedures for Psychological, Psychometric, and Personality Research
(Version 1.7.8). Retrieved from https://cran.r-project.org/web/packages/psych/index.html
Ruffle, B. J., & Tobol, Y. (2016). Clever Enough to Tell the Truth (SSRN Scholarly Paper No. ID
2656755). Rochester, NY: Social Science Research Network. Retrieved from
https://papers.ssrn.com/abstract=2656755
Schwartz, J. A., Savolainen, J., Aaltonen, M., Merikukka, M., Paananen, R., & Gissler, M. (2015).
Intelligence and criminal behavior in a total birth cohort: An examination of functional form,
dimensions of intelligence, and the nature of offending. Intelligence, 51, 109–118.
https://doi.org/10.1016/j.intell.2015.06.001
Templ, M., Kowarik, A., & Filzmoser, P. (2011). Iterative stepwise regression imputation using
standard and robust methods. Computational Statistics & Data Analysis, 55(10), 2793–2806.
https://doi.org/10.1016/j.csda.2011.04.012
Ttofi, M. M., Farrington, D. P., Piquero, A. R., Lösel, F., DeLisi, M., & Murray, J. (2016). Intelligence
as a protective factor against offending: A meta-analytic review of prospective longitudinal
studies. Journal of Criminal Justice, 45, 4–18. https://doi.org/10.1016/j.jcrimjus.2016.02.003
Udry, J. R., Li, R. M., & Hendrickson-Smith, J. (2003). Health and Behavior Risks of Adolescents with
Mixed-Race Identity. American Journal of Public Health, 93(11), 1865–1870.
Uebersax, J. S. (2015, September 8). Introduction to the Tetrachoric and Polychoric Correlation
Coefficients. Retrieved from http://john-uebersax.com/stat/tetra.htm
Wilson, M. (2013, May 16). Why Did a Nice Girl Like Me Date a Guy in Jail? Retrieved November 25,
2016, from
http://www.slate.com/articles/double_x/doublex/2013/05/dating_an_inmate_why_would_a_nice
Page 19 of 20.
E. O. W. Kirkegaard, 2018 Open Differential Psychology
_girl_like_me_make_that_decision.html
Zuckerman, M., Silberman, J., & Hall, J. A. (2013). The Relation Between Intelligence and Religiosity:
A Meta-Analysis and Some Proposed Explanations. Personality and Social Psychology Review,
17(4), 325–354. https://doi.org/10.1177/1088868313497266
Page 20 of 20.
... This demonstrated the validity of our methods of estimating chronotype and cognitive ability and allowed further fine-grained global analyses about the possible moderating effects of geography, sex and age. Importantly, the validity of our chronotype and cognitive ability measures complement similar recent research using the same dataset and the same statistical approach (Figueroa, 2018;Kirkegaard, 2018;Kirkegaard & Lasker, 2020) suggesting that using IRT on non-targeted questions may be a generally valid way of measuring psychological phenotypes. ...
... First, we did not use psychometric tools to assess cognitive ability or chronotype, but instead relied on questionnaire responses. While we did not validate these directly against psychometric tools, we successfully replicated the geographic and demographic correlates of the phenotypes in question, which together with previous results from the same dataset (Figueroa, 2018;Hauser, 2018;Kirkegaard, 2018;Kirkegaard & Lasker, 2020) demonstrate the validity of our method. Second, our database was not representative and the correlation between our variables of interest and likelihood of participation in the database (i.e., the use of OKCupid to find romantic partners) carried the risk of collider bias (Munafò et al., 2017). ...
Article
Full-text available
Chronotype and cognitive ability are two human phenotypes with an uneven geographic distribution due to both selective migration and causal environmental effects. In our study, we aimed to examine the relationship between geographic variables, cognitive ability and chronotype. We used a large anonymized sample (n = 25,700, mostly from the USA, UK, Canada and Australia) of dating site users to estimate chronotype and cognitive ability from questionnaire responses using item response theory. We matched each user to geographic coordinates and city size using the reported locations and geographic databases. In line with previous research we found that male sex, younger age, residence in a more populous locale, higher cognitive ability and more westward position within the same time zone were associated with later chronotype. Male sex, younger age, residence in a more populous locale, later chronotype and higher latitude were associated with higher cognitive ability, but the effect of population on chronotype and latitude on cognitive ability was only present in the USA. The relationship between age and chronotype was stronger in males, and the relationship between chronotype and cognitive ability was stronger in males and in older participants. Population density had an independent association with cognitive ability, but not chronotype. Our results confirm the uneven geographic distribution of chronotype and cognitive ability. These findings generalize across countries, but they are moderated by age and sex, suggesting both biological and cultural effects.
... We used a previously compiled collection of fourteen questions used to measure cognitive ability [20]. While the full set of questions included in this cognitive ability measure was small, it has been found to be related to variables with known relationships to cognitive ability, including crime/antisocial behavior [22] and political interest/participation [20]. Paraphrased, these items included 1. ...
... However, the medium may also result in social desirability bias in responding; this response bias is probably more likely to be reflected in the answers to questions about one's religion than in answers to the cognitive ability-related questions unless cheating on these questions reflects social desirability bias. A previous study using this dataset for criminal and antisocial behavior did not indicate that social desirability bias was strong enough to remove expected criterion relationships [22]. Second, as an extension of the first limitation, the data were not particularly representative of the national populations they were drawn from but instead reflected mainly younger persons looking for love online. ...
Article
Full-text available
We sought to assess whether previous findings regarding the relationship between cognitive ability and religiosity could be replicated in a large dataset of online daters (maximum n = 67k). We found that self-declared religious people had lower IQs than nonreligious people (atheists and agnostics). Furthermore, within most religious groups, a negative relationship between the strength of religious conviction and IQ was observed. This relationship was absent or reversed in nonreligious groups. A factor of religiousness based on five questions correlated at −0.38 with IQ after adjusting for reliability (−0.30 before). The relationship between IQ and religiousness was not strongly confounded by plausible demographic covariates (β = −0.24 in final model versus −0.30 without covariates).
... First, we did not use psychometric tools to assess cognitive ability or chronotype, but instead relied on questionnaire responses. While we did not validate these directly against psychometric tools, we successfully replicated the geographic and demographic correlates of the phenotypes in question, which together with previous results from the same dataset [51,[67][68] demonstrates the validity of our method. Second, our database was not representative and the correlation between our variables of interest and likelihood of participation in the database (that is, the use of OKCupid to find romantic partners) carried the risk of collider bias [70]. ...
Preprint
Full-text available
Chronotype and cognitive ability are two psychological phenotypes with an uneven geographical distribution due to both selective migration and causal environmental effects. In our study we aimed to unravel the relationship between geographical variables, cognitive ability and chronotype. We used a large anonymized sample (N=25700) of dating site users to estimate chronotype and cognitive ability from questionnaire responses using item response theory. We matched each user to geographical coordinates and city size using the reported locations and geographical databases. In line with previous research we found that male sex (β=0.029), younger age (β=-0.178), residence in a more populous locale (β=0.02), higher cognitive ability (β=0.05) and more westward position within the same time zone (β=-0.04) was associated with later chronotype. Male sex (β=0.065), younger age (β=-0.04), residence in a more populous locale (β=0.149), later chronotype (β=0.051) and higher latitude (β=0.03) was associated with higher cognitive ability, but the effect of population on chronotype and latitude on cognitive ability was only present in the United States. The relationship between age and chronotype was stronger in males, and the relationship between chronotype and cognitive ability was stronger in males and in older participants. Population density had an independent association with cognitive ability, but not chronotype. Our results confirm the uneven geographical distribution of chronotype and cognitive ability. Country-wise analyses distinguish universal cultural/biological and country-specific effects. The moderating effect of age on the cognitive ability-chronotype relationship suggests that cultural rather than biological effects underlie this relationship.
Article
Full-text available
A very large dataset (N=68,371, 2,620 variables) from the dating site OKCupid is presented and made publicly available for use by others. As an example of the analyses one can do with the dataset, a cognitive ability test is constructed from 14 suitable items. To validate the dataset and the test, the relationship of cognitive ability to religious beliefs and political interest/participation is examined. Cognitive ability is found to be negatively related to all measures of religious belief (latent correlations -.26 to -.35), and found to be positively related to all measures of political interest and participation (latent correlations .19 to .32). To further validate the dataset, we examined the relationship between Zodiac sign and every other variable. We found very scant evidence of any influence (the distribution of p-values from chi square tests was flat). Limitations of the dataset are discussed.
Article
Full-text available
This study examined the association between sexual orientation and nonviolent and violent delinquency across the life course. We analyzed self-reported nonviolent and violent delinquency in a sample of heterosexual males (N=5220–7023) and females (N=5984–7875), bisexuals (N=34–73),gay males (N=145–189), and lesbians (N=115–150) from the National Longitudinal Study of Adolescent to Adult Health (Add Health). The analyses revealed, in general, that bisexuals were the most delinquent of the sexual orientation categories for both males and females. Additional analyses revealed that heterosexual males reported significantly higher levels of both violent and nonviolent delinquency than gay males, whereas lesbians reported more involvement in nonviolent delinquency and, to a lesser extent, violent delinquency relative to heterosexual females. Analyses also revealed that lesbians reported significantly more delinquent behavior, particularly for nonviolent delinquency, than gay males. Future research should explore the mechanisms that account for these observed patterns and how they can be used to more fully understand the etiology of delinquency.
Article
Full-text available
A dataset of 30 diverse socioeconomic variables was collected covering 32 London boroughs. Factor analysis of the data revealed a general socioeconomic factor. This factor was strongly related to GCSE (General Certificate of Secondary Education) scores (r's .683 to .786) and and had weak to medium sized negative relationships to demographic variables related to immigrants (r's -.295 to -.558). Jensen's method indicated that these relationships were related to the underlying general factor, especially for GCSE (coefficients |.48| to |.69|). In multiple regression, about 60% of the variance in S outcomes could be accounted for using GCSE and one variable related to immigrants.
Book
The Nurture Versus Biosocial Debate in Criminology: On the Origins of Criminal Behavior and Criminality takes a contemporary approach to address the sociological and the biological positions of human behavior by allowing preeminent scholars in criminology to speak to the effects of each on a range of topics. The text aims to facilitate an open and honest debate between the more traditional criminologists who focus primarily on environmental factors and contemporary biosocial criminologists who examine the interplay between biology/genetics and environmental factors.
Article
Drawing on the emerging life-course paradigm in criminological theory, this study examines the relationship between cognitive ability and delinquent behavior within a sample of inner-city youth. The results indicate that net of statistical controls, cognitive ability maintains a robust inverse relationship with the likelihood of the onset of delinquency, the early onset of delinquency, and the persistence of delinquency during the 18-year period covered by the dataset. Furthermore, cognitive ability mediates the effect of concentrated disadvantage on both the onset and early onset of delinquency. Overall, the results of this study, suggest that cognitive ability is an important criminogenic risk factor that has important implications for both correctional interventions and the continued development of structural and multilevel theories of crime.
Chapter
During the last 45 years or so, since Robert Merton (in 1939) first published his now classic article on “Social Structure and Anomie” (1957), undogmatic hypothesizing about crime and social status seems to have favored one or the other of two main, and partly competing, directions: One (that of Merton and others) has to do with frustrated ambitions and the effects of blocked opportunities. The other may be characterized and perhaps originate from Genevieve Knupfer’s article from 1947: The “Portrait of the Underdog”, and it tends to view individual criminality as one among several effects of a person’s low social status origin.