Wednesday 15 July 2015

Intelligence, crime and getting caught

Quips encapsulate an observation, and the well-established association between low ability and crime provokes the dismissive observation that duller minds are not more criminal, just less able to avoid capture. Perhaps so.

Capture/recapture methodologies allow us to estimate the size of populations on the basis of two or three random samples having been tagged. The overlap between samples (showing up in the second sample as having already been tagged on the first occasion) makes close estimation of population totals possible.

Measures of intelligence taken in school aged children provide a great insight into their achievements and difficulties in adult life. These prior measures often invalidate other contemporary measures and explanations, showing that habitual traits predominate over situational variables.

Having set the scene, what if we “tag” children who were in middle or high school in 1994-5, administer various tests including IQ tests (a short version of the Peabody Picture Vocabulary), and then interview them about their behaviour, including violence, as young adults in 2001-2.

Kevin M. Beaver, Matt DeLisi, John Paul Wright, Brian B. Boutwell, J.C. Barnes, and Michael G. Vaughn. No evidence of racial discrimination in criminal justice processing: Results from the National Longitudinal Study of Adolescent Health. Personality and Individual Differences 55 (2013) 29–34

Overall, the final analytic sample size ranged between N = 1308 and 3506 and varied as a function of missing data and the unique restrictions placed on the data for some of the statistical models. Respondents were asked to indicate whether they had ever been arrested (0 = no, 1 = yes) and whether they had ever been incarcerated (0 = no, 1 = yes). In addition, respondents who indicated that they had been arrested were asked to report the length of their sentence in total months. To assess frequency of antisocial behavior, a self-reported life-time violent behavior scale was created. For each wave, items were identified that measured involvement in acts of serious physical violence and then summed to develop a lifetime violence scale that consisted of twenty-two items across all four waves of data (a = .81). Respondents completed an abbreviated version of the Peabody Picture Vocabulary Test-Revised (PPVT-R), known as the Picture Vocabulary Test (PVT) during waves 1 and 3 of data collection. The PVT measures verbal abilities and has been used extensively as a measure of IQ among researchers using the Add Health data.

Table 1 demonstrates that, as expected, African-American males are more likely to be arrested, incarcerated, and receive longer criminal sentences than White males. Importantly, however, the results of the t-tests in Table 1 also reveal significant racial differences with African Americans self-reporting more violent behavior over their life course and Whites scoring significantly higher on the composite IQ measure.


It should be pointed out that the lifetime violence figures come from self report at interview, not capture by Police. African American males were 43% more likely to be arrested than Whites, so at first sight this is confirmation of race-based bias against young black men.

After controlling for life-time violence and verbal IQ, however, the effect of race on the probability of being arrested dropped from statistical significance, though it is still different. Fig. 1 further illustrates the finding in that the predicted probability of being arrested in the baseline model for Whites was 0.41 and for African Americans was 0.49. After controlling for self-reported lifetime violence and verbal IQ, however, the difference was not statistically significant with the White predicted probability being 0.41 and the African American predicted probability being 0.44.



The same reductions when using the control model are found for incarceration. Is it appropriate to apply these controls? I think so. Self-reported lifetime violence is unlikely to be exaggerated, since it is given in confidence to a health survey, not a law enforcement agency. It is a relevant prior behaviour which might lead to being violent, and thus arrested in future. Verbal IQ is a good measure of general ability, which has been showing to be related to offending behaviour, in that duller people may not understand the need for regulations or why they apply to them.

The authors say: Without including control variables for potential alternative explanations, the results were consistent with previous research indicating that African American males are more likely to be arrested and incarcerated compared to their White counterparts. After introducing control variables for self-reported lifetime violence and verbal IQ (to rule out alternative explanations), the association between race and being processed through the criminal justice system was reduced to non-significance. Taken together, analysis of data from the Add Health strongly suggest that research examining racial disparities in the criminal justice system must include covariates for self-reported criminal involvement and perhaps even for verbal IQ or they are likely misspecified. The most likely result of this misspecification is an upwardly biased race effect that purportedly indicates that African American males are treated more harshly than White males due to a biased criminal justice system.

The authors ask for a replication, so as to test the soundness of their results.

Here is another approach, using a different capture method. Suppose in a city or some other defined location you collect victim and bystander descriptions of assailants, and limit yourself to those in which there is agreement about the race of the perpetrator. (You can also look at the race of the victim, but that is not the main point here). Then compare the racial rates of assault with the proportions of the relevant races in that population. That gives you a quick indication whether rates of assault are in proportion to population numbers. Then look at the proportions of the witnessed assailants who get arrested, tried and if convicted, for how long. These will be relatively raw data, uncorrected for previous violence and mental ability, but will provided a rule of thumb check of how the numbers come out as accused persons go through the legal system.


  1. I assume that inherited personality/behavior type also has a great deal to do with criminality. Is there a correlation between personality type and IQ?
    Anecdotally, I have noticed that my smarter kids seem to have a particular personality type (serious, cautious, conscientious, etc.) that’s often lacking in my duller kids (N=7). One of my sons (age 7) has an impulsive personality and he’s the only one who’s had a petty shoplifting (candy, gum, etc.) problem.

    1. Some data in this study

  2. The kinds of studies you sketch in the last paragraph have been done. See pp. 327-329 here:

    1. Very interesting study. Thank you for directing me to it.

  3. (i) Fascinating post; thanks, doc.

    (ii) "Respondents were asked to indicate whether they had ever been arrested (0 = no, 1 = yes) and whether they had ever been incarcerated (0 = no, 1 = yes). In addition, respondents who indicated that they had been arrested were asked to report the length of their sentence in total months." My instinct is to rebel, and demand not self-reported arrests and sentences, but figures from the records.

  4. I wonder whether something could be done with military personnel, whose arrests and sentences would presumably be much cheaper to check than civilians'?

  5. Is it smart to admit violent behavior in a health survey? It's very trusting, I think, but I would not describe it as a wise choice.

    Not all crime is violent. White-collar crime can wipe people out, without any violence at all. Think of Madoff and his accomplices.

    Conviction and incarceration are not necessarily proof of guilt. Fewer than 3% of cases go to trial right now in the US. The rest are disposed of through plea bargaining, in which the prosecutors hold most of the evidence and power.

    Imagine two defendants accused of the same crime. Assume they committed the same acts--breaking and entering or vandalism. One has a public defender; the other has a private lawyer hired by his wealthy parents. If the first is convicted, and the second has charges dismissed after a time of good behavior, does that mean the first is provably more prone to criminal acts? Or does it reflect other characteristics associated with intelligence, such as family affluence, education, and the availability of lawyers and advisors to help the accused fare as well as possible in the system of negotiation known as plea bargaining?

  6. Self report about violence given in confidence is very probably a more reliable measure of actual crime than conviction rates. They differ by race, which is instructive. The point of the paper is that conviction and incarceration rates come down almost to the norm when adjustment is made for two factors which account for actual violence: prior violence and low ability.