Intelligence and Crime - Is R = -.20 A Meaningful Correlation Size?

While studies have frequently found that IQ and crime correlate at around r = -.20, they disagree about how to interpret the size of this correlation coefficient. At one extreme, some studies have dismissed the IQ-crime correlation as being simply too small to matter. Menard and Morse concluded that "the association between IQ and delinquent behavior is so weak as to be negligible," and so it "contributes nothing to existing delinquency theory" (pp. 1374, 1347). Likewise, a task force of the American Psychological Association figured that since a correlation of r = -.20 produces an explained variance of only 4 percent (r² =.04), the IQ-crime correlation is "very low" (Siegel, p. 174). At the other extreme, some studies have identified IQ as a critical, if not the fundamental, correlate of crime. Herrnstein and Murray argued that the effect of IQ on crime, as well as other social problems, is so strong that "much of the attention now given to problems of poverty and unemployment should be shifted to . . . coping with cognitive disadvantage" (p. 251).

In between these two extremes is a more sensible interpretation of the IQ-crime correlation as moderately strong. One way to gauge the strength of the IQ-crime correlation is to compare it to other correlates of crime. A study by Wright and others (1999a) looked at social bonds and crime in late adolescence and early adulthood. They found that some social bonds correlated with crime much more strongly (in absolute value) than r = -.20; for example, delinquent friends correlated with crime at r =.40, and living with one's parents correlated at r = -.32. Other social bonds correlated less strongly, for example, full-time employment (r = -.13) and romantic partnerships (r = -.13). Still other social bonds correlated right at r = -.20, including educational achievement, occupational aspirations, and months unemployment. These comparisons show IQ to be a moderately strong, though neither the strongest nor weakest, correlate of crime.

Another way to gauge the IQ-crime correlation is to restate it in more intuitive terms. Rosenthal and Rubin allow for this with their binomial effect size display (BESD), a procedure that translates simple correlations into equivalent experimental results. In this approach a correlation of r = -.20 is equivalent to an experimental intervention that reduces subjects' success rates from 60 percent to 40 percent. Hypothetically, then, randomly assigning high IQs to low-IQ individuals would decrease their criminal behavior by about 30 percent (i.e., from 60 percent to 40 percent)—certainly a meaningful change.