Criminal Careers

The criminal careers research of the 1980s and early 1990s, begun with several seminal longitudinal cohort studies, has been expanded with the addition of life-course or developmental criminology, as well as evidence supporting multiple age-crime curves or classes/categories of criminal careers. According to Loeber and Stouthamer-Loeber, studying criminal careers from a life-course or developmental perspective implies three major goals: (1) encouraging the description of within-individual changes in offending across time; (2) identifying the causal factors of the longitudinal course of offending; and (3) studying the impact of life transitions and relationships on offending behavior.

Support for a developmental perspective on criminal careers is abundant. Bartusch and others find that an individual's age is indeed important in the development of offending, which it would not be in a more static model. They suggest that "identical antisocial behaviors [e.g., delinquency] represent somewhat different underlying constructs when the behaviors are measured during childhood versus during adolescence" (p. 39). Paternoster and Brame find more limited support for the developmental perspective. While association with delinquent peers and prior offending behavior—both of which change over time—are important predictors of serious delinquent activity, there also remains an element of time-invariance of between-individual differences in offending. Paternoster and Brame suggest that crime is best studied taking into account both static and dynamic models by assuming a "theoretical middle ground" (p. 49). Simons and others find a dynamic and reciprocal causal chain of events that support a developmental perspective of offending, as opposed to a direct relationship between childhood and adolescent misconduct, which would be suggestive of a latent trait or static approach.

The most recent criminal careers research has examined the possibility of heterogeneity in the unimodal age-crime curve. What this unimodal curve may mask is the possibility of heterogeneity between different kinds of offender groups that is lost at the aggregate level. Research by Nagin and Land and by D'Unger and others has suggested that the aggregated age-crime curve for the population may mask the existence of qualitatively different classes of criminal careers with distinct trajectories and differing ages of offending onset, peak ages and rates of offending. In addition, the assumption of a singular age-crime curve may hamper empirical research, particularly if different variables predict membership into different classes of offenders.

Theoretical developments in criminology have suggested that there are many different types of offenders, and empirical evidence for this assertion has been found. In particular, the development of latent class analysis (in the form of semiparametric mixed Poisson regression models) has allowed researchers to disaggregate offenders groups with multiple pathways of offending over time (Nagin and Land; Land, McCall, and Nagin; Land and Nagin; D'Unger et al.). The first major study that conjoined the criminal careers paradigm and latent class analysis of delinquent/criminal careers cohort data was Nagin and Land. Nagin and Land determined that four distinct categories of delinquent/criminal careers could be identified in the Cambridge cohort (West and Farrington, 1973, 1977): nonoffenders (i.e., those who did not have any recorded convictions), individuals whose offending was limited predominantly to the teen years, and two categories of chronic offenders—one with a low-rate and the other with a high-rate of offending.

The research was some of the first to provide empirical evidence of multiple types of offending trajectories with distinct patterns of offending over time. However, because it contained only a limited number of covariates, it was not able to establish the predictors of group membership. Subsequent latent class analysis of criminal careers offered further support for the notion of multiple types of delinquent/criminal offenders across many different types of data. D'Unger and others provide evidence that multiple categories of offenders are present across several data sets with a high level of consistency. Using all-male data from the Cambridge cohort, the 1958 Second Philadelphia Birth Cohort, and the Racine 1942, 1949, and 1955 Birth Cohorts, D'Unger and colleagues concluded:

Rather than merely representing a discrete approximation to an underlying continuous distribution of unobserved delinquent/criminal propensity, the small number of latent offending categories estimated in [the] models may represent distinct classifications of cohort offenders with respect to age trajectories of offending that are meaningful in and of themselves (emphasis added) (p. 1622).

Across all samples from various cohorts and with differing measures of offending (e.g., arrests, convictions) as the dependent variable, several basic patterns emerged: the adolescent-peaked offender, the chronic offender, and the nonoffender. These categories sometimes bifurcated into "higher-" and "lower-" rate groups, but the shape of the offending trajectories across samples was remarkably consistent. However, this research still did not answer the question: What predicts membership in various offender/nonoffender classes?

In another attempt to answer the above question, Nagin, Farrington, and Moffitt looked at the impact of several types of factors on the four offending categories initially delineated by Nagin and Land in the Cambridge cohort. Using both self-report and conviction data on the Cambridge cohort, several significant differences were found between the predictors of the four criminal career categories. In particular, evidence was found in support of the distinctiveness of the high-rate chronic offenders. At their peak, these offenders were more likely to be engaged in violent behavior, smoking cigarettes, using drugs, and having sexual intercourse. By age thirty-two, all three of the offending groups (high-rate chronic, low-rate chronic, and adolescence-limited) were more likely to be fighting, using drugs, and abusing alcohol than the nonoffenders, based upon their self-reported offending. All three offender groups also suffered in the job market. Using data on males from the National Youth Survey, McDermott and Nagin also attempted to delineate the predictors of offending-class membership. It was determined that three classes of offenders exist in this sample, analogous to findings from other latent class research: nonoffenders, an adolescent-peaked group that rises to a peak early but then slows down and reaches zero by the age of twenty four, and a higher-rate group that exhibits an increasing rate of offending until the age of eighteen—characteristic of chronic offenders—but then acts erratically between the ages of eighteen and twenty-four. McDermott and Nagin determined that variables such as involvement with delinquent peers and deviant labels imposed by parents were most significant for predicting changes in the offending index, while measures of social control offered conflicting evidence.