Patterns of residual criminal careers among a sample of adjudicated French-Canadian males.

• Author: Kazemian, Lila

Introduction

Residual career length (RCL) is defined as the number of years remaining in criminal careers up to the last offence. Knowledge about RCL is still limited, despite its crucial importance to penal policy. This study aims to provide estimates of RCL in relation to general criminal career parameters (age, number of prior convictions, time since the previous conviction, age of onset, and offence type). It further seeks to estimate the predictability of RCL based on offending information available in official records, and hence assess the value of these predictors for decision making in criminal justice. The analyses carried out in this paper will also determine whether distributions of self-reported RCL (SRCL) differ from those of official RCL (ORCL), a question that has not yet been addressed in the criminal career literature. The data used in this study were collected prospectively, and the benefits of this approach over retrospective collection have been highlighted in a previous publication (Kazemian and Farrington 2006).

The current study extends the analysis of RCL carried out in a recent study using two samples of British males (see Kazemian and Farrington 2006), by studying self-reports alongside official records. This replication is useful to determine whether age-crime distributions of active offenders and predictions of RCL diverge with different types of sample (representative British males versus high-risk French-Canadian males) and also to investigate whether distributions of self-reported RCL differ from those obtained using official records of offending. The value of comparing career duration estimates between offender samples and individuals who are representative of the general population has been discussed in a recent publication by Piquero, Brame, and Lynam (2004). The comparative approach is not only useful for replication purposes but also allows the strengths of one sample to compensate for the limitations of the other (e.g., the use of self-reports in the Montreal study and the more representative sampling of the Cambridge study). These two studies (MTSLS and CSDD) have complementary strengths and limitations.

The potential theoretical and policy relevance of RCL has been discussed by Kazemian and Farrington (2006). From a theoretical viewpoint, RCL reflects the age-crime distributions of active offenders. There has been some debate regarding the interpretation of the age-crime curve; namely, whether declines in the aggregate curve are driven by changes in prevalence (participation) or incidence (frequency) rates, or both. Although few individuals remain active in crime after age 30, it is not clear whether the number of years remaining in the criminal careers of those who do remain active after this age is negligible (Blumstein, Cohen, and Hsieh 1982). In other words, is the RCL of individuals in their thirties higher than that of individuals in their twenties?

Piquero (2004) underlined the relevance of Moffitt's (1993) taxonomy in the explanation of the age-crime curve: "Unlike traditional explanations of criminal behavior, Moffitt's developmental taxonomy attempts to account for the variation in offending patterns that underlie the aggregate age--crime curve" (111). It may also be that variations in the distributions of RCL are a result of the intermittent offending patterns of life-course-persistent offenders.

The potential of measures of past criminal behaviour (i.e., age of onset, past number of offences, and time since the previous conviction) to predict future offending are also explored in this paper. In Moffitt's (1993) taxonomy, the career length of life-course-persistent offenders is said to be linked to other criminal career dimensions, namely age of onset. This paper will explore to what extent it is possible to predict RCL based on early measures of antisocial behaviour.

The main policy implications of RCL are associated with sentencing and incapacitation decisions (see Kazemian and Farrington [2006] for a more thorough discussion of this topic). When offenders are arrested and convicted, sentencers must decide whether the offender should be incarcerated and, if so, the length of the term that should be imposed (Cohen 1983). Discretionary release decisions also incorporate presumptions about RCL. Error (in terms of incapacitation at least) is introduced when people have residual criminal careers extending beyond their release from prison or when they remain incapacitated after the end of their criminal careers.

Ideally, incapacitation should be applied to offenders during their years of active offending. In a perfect system, efforts would be invested in increasing the incarceration time of high-rate offenders and reducing the time served by low-rate offenders (Greenwood and Abrahamse 1982; Piquero, Farrington, and Blumstein 2003). Similarly, it would not be powerfully crime reductive to incapacitate individuals with a low RCL. Such principles may be useful in addressing the issue of prison overcrowding or necessary prison building. If the objective sought by incarceration is incapacitation, there is little relevance in locking up individuals after they have ceased offending (Blumstein et al. 1982; Piquero et al. 2004). On the other hand, the release of offenders during periods in which they present a high risk of recidivism compromises public safety. A better understanding of the distribution of RCL is necessary to inform sentencing and discretionary release policy, in general, and selective incapacitation, in particular. However, such policy implications rely on the assumption that it is possible to make accurate predictions about the number of years and offences remaining in criminal careers. Although official records are an incomplete source of information on offending, they are all that is routinely available to decision makers. The use of self-report information may perhaps result in greater accuracy in predictions, but this is seldom available at the sentencing stage. Moreover, considering that such information is likely to affect sentences, offenders are unlikely to provide accurate self-reports.

Past studies of residual career length

Various studies have offered estimates of total career length (Greenberg 1975; Shinnar and Shinnar 1975; Greene 1977; Blumstein et al. 1982; Spelman 1994; Farrington, Lambert, and West 1998; Piquero et al. 2004). However, few have focused on RCL or residual number of offences (RNO) (Piquero et al. 2003), despite their potential relevance for decision making in the criminal justice system. Blumstein et al.'s (1982) study has precedence. Although a detailed summary of this study can be found in Kazemian and Farrington (2006), one key finding in particular is worthy of mention. In identifying three periods in the criminal career, Blumstein et al. (1982) found that RCL first increased (up to age 30), then remained relatively stable at a high level (between ages 30 and 40), and finally decreased after age 40. These results suggest that the age--RCL distribution differs from the age--crime curve. Their findings also suggest that offenders who remained active in their thirties were characterized by the longest RCLs (about 10 years), thus being a "prime target group for incapacitation" (38). Some of the limitations of this ground-breaking study were discussed in Kazemian and Farrington (2006).

Drawing on Blumstein et al.'s (1982) study, Kazemian and Farrington (2006) estimated RCL and RNO for two generations of British males. Their analyses were based on official records only; information about convictions was available up to age 40 for the sons and up to age 70 for the fathers. Comparisons between sons and fathers were carried out for the purpose of replication (i.e., to assess the consistency in findings across different samples) and to address the issue of false desistance. False desistance refers to the erroneous assumption that individuals have ceased offending at the end of a given observation period. Fathers were followed up to an age when virtually all criminal careers had ended (age 70), which made it possible to assess how the overall distributions of RCL and RNO were affected by the truncation of sons' records at age 40.

Five key findings emerged from Kazemian and Farrington's (2006) analyses. First, most age--RCL distributions were characterized by a remarkable degree of linearity, and this was particularly true for age. The number of years and offences remaining in criminal careers declined at a remarkably steady pace with age, and this was observed across both samples (sons and fathers).

Second, distributions of RCL generally displayed a greater degree of linearity than those of RNO. Results from both samples suggested that RCL tends to decline more consistently than RNO, which suggests intermittent patterns in criminal careers and different paths to desistance from crime. Furthermore, the fathers' distributions of RNO did not decrease as consistently as for the sons, suggesting that the number of offences remaining in criminal careers does not decline uniformly for all offenders across life or with longer follow-up periods.

Third, early onset was a significant predictor of RCL and RNO for sons but not for fathers. This finding may suggest that early onset has an attenuating effect when the follow-up extends past mid-life (Cline 1980; Sampson and Laub 2003) or may be an indication of cohort effects. Since these analyses were based on small numbers, further research is needed in order to draw definite conclusions regarding this issue.

Fourth, results relating to the relationship between offence type and RCL/RNO were inconsistent, and co-offending was not significantly associated with RCL and RNO.

Fifth, risk scores were significantly but generally not highly predictive of RCL and especially not of RNO. These findings highlight the difficult task that must face decision makers in assessing future offending...