This document is copyrighted by the An can Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Sample Method COGNITIVE ABILITIES AND CRIME The data used in this investigation were part of a larger study designed to examine the criminal career patterns of violent offenders (Wenk, 1990). Data were collected for a total sample of 4,164 adolescent male offenders who were committed to the Reception Guidance Center at the Duel Vocational Institution in Tracy, California, between January 1964 and December 1965. This institution was a facility operated by the California Youth Authority (CYA) and was designed to provide treatment, education, and training to the oldest CYA wards during that time. The average age at admission to the CYA for this sample was 19 years. This is the age at which the tests of cognitive ability used in this investi- gation were administered. Because an overwhelming majority (n = 4,065, or 98%) of the original sample members were African American (n = 1,078), Caucasian (n = 2,214), or Hispanic (n = 773), only those three ethnic groups were included in the reported analyses. In our analyses, a subsample of this total was used because only individuals who were classifiable into our offender classification scheme and completed all of the measures of cognitive ability within a particular test battery were included. In all instances, sample sizes are noted. In 1984 and 1985, arrest records were collected for 3,652 of the ado- lescents who were initially assessed during the data collection in 1964 to 1965. These arrest histories were based on records supplied by the CYA and the California Bureau of Criminal Identification and Investigation. The records for the remaining 494 individuals were missing either because the records were purged (see Gottfredson & Gottfredson, 1994) or because the records were sealed in cases of death. For these individuals, the existing CYA records of crime and delinquency from 1964 to 1965 were used. Procedures Data were collected in three ways: self-reported questionnaires, psycho- logical tests, and official records. As a part of the initial assessment 397 process, the sample members were tested individually and in groups with an extensive battery of psychological tests. This information was collected by CYA clinical staff, researchers, and test administrators from other governmental agencies (e.g., proctors from the U.S. Department of Labor, who administered the General Aptitude Test Battery [GATB]) as part of routine intake processing. In a few sessions, fellow CYA wards assisted in collecting self-report questionnaires. Some individuals were occasionally not tested with certain tests for one of two reasons: (a) the institution was under lockup for security reasons, such as intense fog and other inclement weather conditions, or (b) holiday periods. Additionally, some individuals received only a subset of the measures for other reasons, including a failure to meet minimal compe- tency standards for a given test (such as minimal English language profi- ciency for tests of language ability). Measures Because many of the tests routinely used in the 1960s for cognitive abilities are no longer in widespread use, interested readers may wish to consult Anastasi (1976) for more background information on particular tests. California Achievement Test (CAT). This test was designed by Tiegs and Clark (1951) for use in California schools to gauge academic achieve- ment. It is similar to the Iowa Tests of Basic Skills and the Stanford Achievement Test (Anastasi, 1976). The CAT yields total scores in the academic domains of reading, arithmetic, language, and overall academic achievement, which were used in the reported analyses. In addition, the CAT includes tests of more specific abilities within each academic domain (e.g., spelling and language mechanics under the total language ability score) that were not included in our analyses because of their redundancy with the total scores. Scores on the CAT are expressed in grade-equivalent terms. California Test of Mental Maturity (CTTM). This test was designed by Sullivan, Clark, and Tiegs (1963) to test language and nonlanguage intel- ligence in a group setting. The CTTM is a multilevel test battery that yields a grade-equivalent score in the areas of verbal intelligence and nonverbal intelligence. General Aptitude Test Battery. This test was developed by the U.S. Department of Labor for use in employment counseling. It has been widely used in employment counseling since the late 1950s (Anastasi, 1976). There are several subtests in the GATB, including verbal, numeric, spatial, perceptual, and general intellectual abilities. In addition, the GATB yields several other scores in areas such as manual dexterity, finger speed, and clerical aptitude, which were not included in the reported analyses because it is not clear that they tap the same types of cognitive abilities as those measured by the other tests in these analyses. Raw test scores are converted to standard scores, with a mean of 100 and a standard deviation of 20. The GATB has demonstrated good test-retest reliability, with a correlation of about 80 (Anastasi, 1976). Raven Test of Progressive Mutrices. This test was developed by Raven (1960) as a measure of general intelligence. Respondents are presented with a series of 60 matrices that are each missing an element. Test takers must select from six to eight possibilities the response that correctly completes the matrix. Because it is a nonverbal intelligence test, it might be a more culturally fair measure of cognitive ability. The Raven test has demonstrated good test-retest reliability, with correlations of .70 to .90 (Aiken, 1996; Anastasi, 1976). It has also demonstrated good validity, with correlations of .58 to .70 with other measures of intelligence, such as the Standford-Binet Test and the Wechsler Test (Aiken, 1996). Criminal Career Classification We used arrest history information to classify the sample members into one of three mutually exclusive classifications of criminal careers: life- 398 DONNELLAN, GE, AND WENK course-persistent offenders, adolescent-limited offenders, and an omnibus "other" criminal career category. The categories of life-course-persistent offender and adolescent-limited offender were intended to capture Mof- fitt's (1993a) taxonomy. Life-course-persistent offenders. An individual was classified as a life-course-persistent offender if three conditions were satisfied: (a) the individual was first arrested on or before age 17, (b) the individual had one or more arrests after age 25, and (c) the individual had 19 or more total arrests. The first condition was designed to capture the early-onset or early- starter (Patterson, DeBaryshe, & Ramsey, 1989) aspect of Moffitt's char- acterization of life-course-persistent offenders. The age of 17 was selected as the specific cutoff point for a number of reasons. First, age 17 is below the modal age of first arrest for this sample (which was age 18), and Moffitt suggests that life-course-persistent offenders will have started their crim- inal careers before the modal age of onset of criminal behavior. Second, individuals are probably involved in delinquent acts well before they are first arrested. Therefore, selecting ages 17 and below ensures that we will be able to include a large number of individuals who started committing antisocial acts at an early age, well before they were first arrested. We tried alternative operationalizations of life-course-persistent offend- ers that used a younger age at first arrest for this condition. When ages 16 and younger were used, we found that the patterns of results were similar to those reported here. However, when ages 15 and younger were used for this condition, there was no ability to distinguish differences between the two types of offenders because the sample size for this category was so drastically reduced. Therefore, we believe that the cutoff point of age 17 and below is the best compromise between capturing the essence of Moffitt's work and maximizing statistical power in our sample. The second condition was designed to operationalize Moffitt's notion that life-course-persistent offenders commit criminal acts past adolescence and young adulthood. By including individuals who commit criminal acts past their mid-20s, this condition ensures that truly life-course-persistent offenders are included in this category. The third condition of having 19 or more total arrests was designed to capture the chronic and severe nature of the life-course-persistent category. Having 19 or more arrests placed an individual in the top 75th percentile of frequency of arrests for this sample. Taken together, these three condi- tions ensure that only offenders who were chronic life-course-persistent offenders were included in this category. Adolescent-limited offenders. An individual was designated an adolescent-limited offender if the first arrest occurred after age 17 and commission of criminal offenses stopped by the age of 25. The justification for the onset condition after age 17 and the carly-adulthood offset condition follows from Moffitt's suggestion that transient criminal activity limited to adolescence and possibly young adulthood is the hallmark of the adolescent-limited offender (see Moffitt, 1993a, 1993b). Although it could be argued that age 21 would be a better cutoff point, it is important to note that this is a sample of delinquent juveniles who may have had an "extended" adolescence due to incarceration and, consequently, age 25 should be considered a more fair cutoff point. Moreover, this category includes only individuals who engaged in crime for a very specific part of their life span and, thus, cannot be properly termed life-course-persistent offenders. Using this offender classification scheme, we were able to classify 888 offenders into the adolescent-limited category and 774 offenders into the life-course persistent category (21.8 and 19% of the sample, respectively). Accordingly, 1,662 offenders were included in the final analyses. The remaining 2,403 offenders (about 59.1% of the initial sample) were clas- sified as "other" and not included in the subsequent analyses. The ethnic breakdown of the sample used in the final analyses was 27% African American (n = 448), 52% Caucasian (n = 864), and 21% Hispanic (n=350), which closely mirrors the ethnic breakdown of the initial sample (26.5. 54.5, and 19%, respectively). The average age of the offenders included in the final analyses was 19, which was also the average age of the offenders in the initial sample. Although the ethnic breakdowns of the two samples were approximately equivalent, there were ethnic differences in the offender classification scheme. The ethnic breakdown of the adolescent-limited category was 17% African American (n=151), 68% Caucasian (n = 603), and 15% Hispanic (n=134), whereas the breakdown of the life-course-persistent category was 38% African American (n=297), 34% Caucasian (n = 261), and 28% Hispanic (n=216). Thus, ethnic minority groups are overrepresented in the life-course-persistent category (p < .001, (2, N = 1,662) = 195.27. This issue will be revisited in the Discussion. = We further explored the validity of our classification system by com- paring the two groups on the average total number of arrests. Consistent with expectations and category design, life-course-persistent offenders had an average of 27.74 arrests (SD = 8.03), whereas adolescent-limited offenders had an average of 2.54 arrests (SD = 2.84). This difference was statistically significant, F(1, 1660) 7,646.41, p < .001. Similarly, life-course-persistent offenders were first arrested at the average age of 15.08 years (SD = 1.55), whereas adolescent-limited offenders were first arrested at the average age of 18.76 (SD = 0.91). This difference was also highly significant, F(1, 1660) = 3,588.86, p < .001. These differences underscore the dramatic differences in criminal careers between these two groups.