Delinquency Journal of Research in Crime and - CiteSeerX

Journal of Research in Crime and Delinquency http://jrc.sagepub.com

Exploring Residual Career Length and Residual Number of Offenses for Two Generations of Repeat Offenders Lila Kazemian and David P. Farrington Journal of Research in Crime and Delinquency 2006; 43; 89 DOI: 10.1177/0022427805280066 The online version of this article can be found at: http://jrc.sagepub.com/cgi/content/abstract/43/1/89

Published by: http://www.sagepublications.com

On behalf of:

National Council on Crime and Delinquency

Additional services and information for Journal of Research in Crime and Delinquency can be found at: Email Alerts: http://jrc.sagepub.com/cgi/alerts Subscriptions: http://jrc.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations (this article cites 5 articles hosted on the SAGE Journals Online and HighWire Press platforms): http://jrc.sagepub.com/cgi/content/abstract/43/1/89#BIBL

Downloaded from http://jrc.sagepub.com at SAGE Publications on March 22, 2007 © 2006 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

Journal of Research Article 10.1177/0022427805280066 Kazemian, Farrington in /Crime T wo Generations and Delinquency of Repeat Offenders

Exploring Residual Career Length and Residual Number of Offenses for Two Generations of Repeat Offenders

Journal of Research in Crime and Delinquency Volume 43 Number 1 February 2006 89-113 © 2006 Sage Publications 10.1177/0022427805280066 http://jrc.sagepub.com hosted at http://online.sagepub.com

Lila Kazemian David P. Farrington University of Cambridge Very few studies have explored residual career length (RCL) and residual number of offenses (RNO), that is, the remaining time and number of offenses in criminal careers. This study uses conviction data from the Cambridge Study in Delinquent Development to investigate RCL and RNO, for a sample of British males and their fathers. The sons were followed up to age 40 and the fathers up to age 70. Distributions of RCL and RNO according to six different criteria are presented (age on offense, conviction number, time since the last conviction, age of onset, offense type, and number of co-offenders). There was a general decline in RCL and RNO with age. Although RCL declined steadily with each successive conviction for both sons and fathers, RNO did not decline with conviction number for fathers. Over and above age on conviction, age of onset predicted RCL and RNO for sons, but less so for fathers. The type of offense and the number of co-offenders did not predict RCL or RNO. Risk scores showed that the predictive power of these variables for RCL and RNO was statistically significant but not very high. This finding highlights the difficulties associated with predictions of criminal career outcomes based on information available in official records, which is the main source of information available to decisionmakers in the criminal justice system. Keywords: residual career length; residual number of offenses; criminal career

A

very large body of research has been dedicated to the prediction of recidivism. However, very few studies have attempted to estimate residual criminal career length or residual number of offenses in criminal careers (Piquero, Farrington, and Blumstein 2003; Von Hirsch 1988, 1998). Residual career length (RCL) refers to the remaining number of years in criminal careers until the last offense, whereas residual number of offenses (RNO)

89


90 Journal of Research in Crime and Delinquency

specifies the remaining number of offenses in criminal careers. This article addresses issues that are central to criminal career research, namely the agecrime distribution of active offenders, the time and number of offenses remaining in the career according to other criminal career indicators, and the ability to predict RCL and RNO based on information available in official records. Data on criminal careers in this study were collected prospectively. The alternative strategy is the retrospective method. There are two methods of collecting data retrospectively. First, some researchers may select all individuals who have been arrested or convicted in a given year and document offenses committed in previous years. The second method involves the retrospective selection of a cohort sample (e.g., all individuals who were born in a given year) and the gathering of information on offenses committed by them in subsequent years. Compared with the prospective follow-up of a cohort, the first method is limited by the fact that the sample would exclude individuals who had desisted from crime in previous years or those who had not been convicted during that given year. The second method is also problematic because a large proportion of the sample may be lost when the information is collected retrospectively (e.g., because of deaths, moving away, etc.), and researchers may have difficulty tracking down these individuals. Also, retrospectively defining a cohort requires high-quality criminal record data that are preserved over time, which may not be the case in all jurisdictions. Thus, a prospective follow-up of individuals seems to be the most valid method of advancing knowledge about criminal careers. Because our knowledge of RCL and RNO is limited, this article aims to provide basic information on an area of study that has been largely neglected by criminal-career research. This exploratory study seeks to examine the overall distributions of RCL and RNO for two generations of British males. Comparisons between sons and fathers were carried out for two main purposes: replication and “false desistance.” It seemed highly relevant to replicate the results with sons (who were the main subjects of interest) to assess the consistency in findings across two different samples. Comparisons with fathers also addressed the issue of “false desistance,” which refers to the false assumption that some individuals have ceased offending at the end of the observation period. Fathers were followed up to an age when virtually all criminal careers had ended (age 70), which made it possible to assess whether the overall distributions of RCL and RNO were affected by the trunAuthors’ Note: The authors wish to thank Alex Piquero and Andrew von Hirsch for comments on prior versions of the article. The authors especially wish to thank Ken Pease for his particular interest in this study and helpful comments.


Kazemian, Farrington / Two Generations of Repeat Offenders 91

cation of sons’ records at age 40. To our knowledge, these research questions have never been addressed using prospective longitudinal data. The study serves three main purposes. First, by examining distributions of RCL and RNO for the two samples, it seeks to test hypotheses about the agecrime curve of active offenders and explore issues relating to termination of criminal careers. Second, this research also aims to examine the distribution of RCL and RNO according to various criminal-career indicators that are available in official records (serial number of convictions, time since the last conviction, age of onset, co-offending, and offense type). Third, this study assesses the ability to predict RCL and RNO based on these variables (i.e., information available to decision-makers in the criminal justice system).

Why Study RCL and RNO? RCL and RNOs provide information on the time remaining in criminal careers and future rates of offending. Knowledge about RCL and RNO can potentially have important theoretical and policy implications. From a theoretical viewpoint, RCL and RNO reflect the age-crime curves of active offenders. In a follow-up to age 70, Sampson and Laub (2003) found that prevalence and incidence rates declined with age, even among serious and violent offenders. Blumstein, Cohen, and Hsieh (1982:11) pointed out that “The observation of declining population arrest rates with age . . . has led to the conventional wisdom that imprisonment after age 30 is not efficient because these older offenders are likely to be soon terminating their criminal careers.” Although the proportion of individuals who are active in offending after age 30 is relatively small, Blumstein et al. (1982:11) argued that “It is not clear . . . whether the expected future career length of those few who are still criminally active at age 30 is also small.” An assessment of RCL and RNO at each age could identify ages where active offenders are most likely to cease offending and ages where they are most likely to persist. Piquero et al. (2003:479) outlined the policy implications associated with estimates of RCL: Knowledge on career length and residual career length is perhaps one of the most critical areas of research that could best inform criminal justice policies because it deals directly with sentencing and incapacitation decisions, which are now so strongly driven by ideology rather than empirical knowledge. For example, if research shows that residual criminal-career lengths average around five years, then criminal justice policies advocating multi-decade sentences will waste scarce policy resources. Similarly, as offenders continue to be incarcerated in late adulthood when their residual career lengths have



diminished, then not only will incarceration space be wasted, but the costs of health care for such offenders will tend to increase, thereby exerting further strain on the scarce resources of an already taxed criminal justice system.

Selective incapacitation policies aim to maximize incapacitation effects by targeting offenders during their years of active offending, by increasing the incarceration time of high-rate offenders and reducing the time served by low-rate offenders (see Greenwood and Abrahamse 1982; Piquero et al. 2003). This is especially relevant in the case of “3 Strikes Policies.” For example, in a California study, Stolzenberg and D’Alessio (1997) found that the mean age of a “third strike” individual being sentenced to a 25-year term was age 30; this type of policy does not necessarily target violent offenders but is rather aimed at any type of repeat offender. What evidence is there that such an individual is likely to continue offending for 25 years?

Estimating the Length of Criminal Careers Despite the theoretical and practical importance of this topic, very few projects have focused on estimating the time remaining in criminal careers; Blumstein et al.’s (1982) study was one of the first systematic attempts to assess residual criminal career lengths. Using data on arrests rather than arrestees, they estimated total and residual criminal career lengths for index offenses recorded during 1973 in Washington, D.C. The authors identified three important periods in the criminal career. During the first period (breakin), dropout rates (i.e., termination rates) declined steadily and the mean RCLs increased with age from 5 to 10 years. The stable period that followed was characterized by low dropout rates and stable mean residual career lengths, which peaked during this period (around age 30). The time remaining in criminal careers in this period was estimated at 10 years, regardless of the past length of criminal careers. The wear-out period occurred around age 40; during this last period, declining residual career lengths and greater dropout rates were observed. Thus, RCL first increased (up to age 30), remained relatively stable at a high level (between ages 30 and 40), and finally decreased after age 40. Blumstein et al.’s (1982) results showed that the overall average duration of criminal careers was about 5 years (approximately 4 years for property offenders and 7 years for personal offenders). They found significant differences in residual lengths according to the type of offense, with person offenders having longer residual lengths. Property offenders were not as likely to persist in crime, but among those who did persist in their 30s, the



average RCL was approximately 10 years. Property offenders had their peak residual length many years into their criminal careers, whereas person offenders (those convicted of murder and rape) “begin at their maximum residual career length of 9.6 and 5.9 years, respectively, and remain there for about 25 years” (p. 58). At all ages, there is a mix of persistent offenders with long RCLs and temporary offenders with shorter RCLs. Blumstein et al. (1982) also found that offenders who began their careers at age 18 (or earlier) and who remained active in their 30s were characterized by the longest residual length of criminal careers (about 10 years). The authors concluded that “those adult index offenders who started index careers at age 18 and who continue to be criminally active between the ages of 30 and 42 are seen to be the most persistent offenders, and so represent a prime target group for incapacitation” (p. 38). This pioneering study was not without limitations, and efforts were undertaken to address these shortcomings in the present study. First, Blumstein et al. (1982) argued that age of onset was a key variable in the estimation of career lengths; however, their analyses were limited to adult criminal careers and they had no information about juvenile onset. Another problematic feature of their analysis was the assumption that the probability of arrest was stable across age. Also, the use of retrospective arrest data resulted in complex calculations and estimates. Some of their analyses of career length were based on the age distribution of arrestees for one single year, rather than on a prospective follow-up of a single cohort of individuals. Other studies have also generated estimates of career length and explored the implications for incapacitation. Greene (1977) used a life-table approach (survival models) in his estimates of career length, which were again based on the age distribution of offenders arrested in a single year. He found that the average adult career length was 12 years. As pointed out by Blumstein et al. (1982:10), this method relies on the questionable assumptions that: (a) all active offenders have the same probability of being arrested at least once in a year, (b) all offenders onset at age 18, and (c) the number of offenders remains stable at each age over time. Greenberg (1975) estimated that the average “index career” was approximately five years. Using data from the Rand Inmate Survey, Spelman (1994) estimated that the total criminal career length was about six or seven years (seven to 10 years for property offenders, and seven to nine years for violent offenders). Shinnar and Shinnar (1975) estimated that the average duration of criminal careers was five years for all offenders, and 10 years for recidivists. Farrington, Lambert, and West (1998) excluded one-time offenders from their analyses and found that the average career length was 10 years up to age 32. Piquero, Brame, and Lynam (2004) found that the average career



length for a sample of serious offenders was 17 years, ranging from four to 30 years.

The Present Study Cohen (1983:72) argued that “There is also need for prospective validation of results that have so far been based entirely on retrospective arrest histories.” The present study explores the distribution of RCL and RNO for two samples of repeat offenders, followed up from adolescence to adulthood. These analyses will provide a better insight on the changes occurring in RCL and RNO over time and across the life course. Unlike Blumstein et al.’s (1982) research, which limited its analyses to adult criminal careers, this study assesses RCL and RNO for juvenile and adult onsetters.

Data Data used in this study were collected in the Cambridge Study in Delinquent Development, which is a prospective longitudinal survey of 411 males from a working-class area of London. The males were mainly Caucasian, of British origin, and working class; detailed descriptions of the sample can be found in a recent publication (Farrington 2003). They were first contacted in 1961 to 1962, when they were eight to nine years old, and were interviewed on eight subsequent occasions. Official records of convictions were obtained between ages 10 and 40 for the following offenses: shoplifting, theft from vehicles, theft of vehicles, joyriding, theft of cycles, theft from machines, theft from work, other theft, burglary, fraud, receiving, suspicious behavior (loitering with intent), robbery, assault and wounding, insulting or threatening behavior, carrying an offensive weapon, sex offenses, drug offenses, arson, vandalism, and disqualified driving. Estimates of RCL and RNO in this study are based on convictions.1 Piquero et al. (2004:417) argued that the estimates of career duration inevitably rely on “the characteristics of the person-years” rather than characteristics of offenders. Due to the nature of the research question at hand, the analyses carried out in the present study are based on person-years. The analyses focus on both the length of official criminal careers and the residual number of convictions. RCL refers to the number of years remaining in the criminal career up to the last recorded conviction, whereas RNO is defined as the number of offenses remaining in the criminal career. For instance, if a respondent is convicted for one offense in each year at ages 16, 24, 30, and 38, his respective residual career lengths would be 22, 14, 8, and 0



years, and his residual number of offenses would be 3, 2, 1, and 0. From the initial sample of 411 boys, 164 had at least one conviction, of which 6 died before age 35. Of the remaining 158 offenders, 49 were convicted only once and were excluded from the analyses.2 Thus, estimates of RCL and RNO could be derived for 109 sons, who committed a total of 676 convicted offenses up to age 40. Results were only presented up to age 35 (see explanation below), resulting in a total of 647 convictions. The average age at first conviction was 16.5 years (21.8 years for violent offenses and 16.2 years for nonviolent offenses) and the median age of onset was 15 years (19 years for violent offenses and 15 years for nonviolent offenses). Most of these convicted offenders had their first conviction during adolescence (from 10 to 16 years old, n = 70, 65 percent). Both average and median ages at the last conviction were approximately 26 years old and the average total career length for these repeat offenders was 9.5 years (median: 11 years). On average, offenders were convicted for 6.2 offenses up to age 40 (median: 4 offenses). The vast majority of convicted offenses consisted of nonviolent crimes (83 percent). Because studies of crime rates in relation to age have traditionally relied on official data, these distributions have been left-hand censored; first arrests and convictions rarely occur in childhood and rather tend to occur in adolescence. In contrast, the distributions of RCL and RNO are right-hand censored; observations are cut off at a given age, even though offenders may not have ceased offending at this point in the life course (see Piquero et al. 2003). Although criminal career research has generally found that a relatively small proportion of offenders remain criminally active after age 40 (Blumstein and Cohen 1987; Le Blanc and Fréchette 1989), estimates of residual criminal career lengths may have been affected by false desistance (see Blumstein et al. 1986). It has been suggested that absolute desistance can only occur when offenders have died (Blumstein et al. 1982). In the present study, the sons’ records of convictions were available up to age 40, which may have underestimated the number of years and offenses remaining in the criminal careers of individuals who persisted in crime after this period. Thus, sons with a conviction at age 39 could not have a residual criminal career length greater than one year, regardless of whether they persisted in offending for many subsequent years.3 Two steps were undertaken to address this issue. First, results were only presented for offenses committed up to age 35 in order to minimize the biases associated with false desistance. The maximum possible RCL at age 35 is five years; this cutoff point was used on the basis that a five-year crime-free period is better evidence of career termination than a one-year crime-free period. Second, estimates of RCL and RNO have also been computed for the



study males’ fathers. Conviction reports were available for fathers up to age 70, for those who had not died.4 As mentioned in the introduction, comparisons with fathers were carried out for two purposes: replication and addressing the issue of false desistance. In all, 109 fathers were convicted of a total of 315 offenses; 48 were onetime offenders and were excluded from the analyses. Two individuals (who were convicted for 14 offenses) were excluded for reasons of death (see note 4), resulting in a final sample of 253 convicted offenses (59 recidivist fathers with convictions up to age 70), with an average of 4.3 and a median of three convictions per father (up to age 40: average of 3.3 and median of two convictions per father).5 The great majority of convicted offenses were nonviolent crimes (91 percent).

Analyses The distributions of RCL and RNO according to six different variables will be presented; most of these variables have been used in previous studies estimating the length of criminal careers or recidivism. These variables are age on offense6 (respondent’s age at the time of the offense; Blumstein et al. 1982; Carney 1967; Silver, Smith, and Banks 2000), conviction number (nth convicted offense; Ashford and LeCroy 1988; Blumstein et al. 1982; Carney 1967; Horwitz and Wasserman 1980; Scarpitti and Stephenson 1971; Silver et al. 2000), the time since the last convicted offense (in years; Barnett, Blumstein and Farrington 1989; Blumstein et al. 1982), age of onset (juvenile onset: first conviction occurred before age 17; adult onset: first conviction occurred at age 17 or later; Ashford and LeCroy 1988; Blumstein et al. 1982, 1986; Carney 1967; Farrington et al. 1998; Piquero et al. 2004; Scarpitti and Stephenson 1971; Silver et al. 2000), the number of co-offenders (Le Blanc and Fréchette 1989; Reiss and Farrington 1991), and offense type (Blumstein et al. 1982; Carney 1967; Horwitz and Wasserman 1980). Offense type is a dichotomous variable that contrasts violent (robbery, physical assault, wounding, insulting or threatening behavior, possession of an offensive weapon, and sex offenses) and nonviolent offenses (shoplifting, theft of vehicles, theft from vehicles, joyriding, theft of cycles, theft from machines, theft from work, other theft, burglary, fraud, receiving, suspicious behavior, drug offenses, arson, damage, and disqualified driving). It would have been interesting to study the impact of incarceration on patterns of RCL (possible deterrent or criminogenic effects); however, the number of incarcerated individuals was small, and it was not possible to include this variable in the analyses.7



Figure 1 Sons’ Average Residual Career Length (in years) and Residual Number of Offenses at Each Age, 10 to 35 Years Old (n = 647 convicted offenses) 18 16

Mean RCL and RNO

14 12 10 8 6 4 2 0

10& 12 11

13

14

15

16

RCL 15.6 12.1 12.8 11.6 10.1 11.6

17 9

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

8.8 8.6 8.7 6.7 5.9 6.6 7.7 6.1 5.9 6.9 5.5 4.4 3.1 1.1 3.4 3.1 2.4 0.5

RNO 13.1 8.1 8.5 6.2 6.7 7.4 5.9 5.4 5.1 4.9 3.3 3.4 2.6 3.1 3.8 2.9 3.9 2.1 2.4 2.3 0.44 2.6 2.6 2.3 0.67

Age

Note: Correlation between age and RCL: r = –0.96, p < .0001 (n = 25). Correlation between age and RNO: r = –0.89, p < .0001 (n = 25). Mean residual length: 8.1 years. Mean residual residual number of offenses: 5 offenses.

Results Distributions of RCL and RNO Age on Offense. Figure 1 shows the distribution of RCL and RNO for sons. Two striking observations emerge from this figure. First, there is a steady drop in RCL and RNO with age. The impressive degree of linearity of these distributions is noteworthy (RCL: r = –.96, p < .0001; RNO: r = –0.89, p < .0001). Second, the fluctuations occurring in both RCL and RNO distributions are very similar, and they are both significantly correlated (r = 0.69, p< .0001). Figure 2 shows the distributions of RCL and RNO for fathers. The fathers’ distributions of RCL and RNOs also decline linearly (RCL: r = –0.93, p < .0001; RNO: r = –0.84, p < .01), replicating the results for sons. Both distributions follow similar patterns, and RCL and RNO are significantly correlated (r = 0.66, p < .0001). Despite the fact that the average time remaining in careers is almost always higher for fathers than for sons (overall average



Figure 2 Fathers’ Average Residual Career Length (in years) and Residual Number of Offenses at Each Age, 10 to 70 Years Old (n = 253 convicted offenses) 25

Mean RCL and RNO

20

15

10

5

0

9 to 14

15 to 19

20 to 24

24 to 29

30 to 34

35 to 39

40 to 44

45 to 49

50 +

RCL

21.4

15.7

12.4

10.2

13.2

7.1

3.1

6.5

1.4

RNO

5

4.4

3

2.5

4

2

0.9

2.5

1

Age

Note: Correlation between age and RCL: r = –0.93, p < .0001 (n = 9). Correlation between age and RNO: r = –0.84, p < .01 (n = 9). Mean residual length: 10.9 years. Mean residual number of offenses: 3 offenses.

RCL: 10.9 and 8.1 years, respectively), RNO generally tends to be higher for sons (overall average RNO: five offenses vs. three offenses). Serial Number of Conviction. Figure 3 shows that both RCL and RNO tend to decline for sons after each successive conviction, and both distributions display a considerable degree of linearity (RCL: r = –0.91, p < .0001; RNO: r = –0.77, p < .0001). Declines with successive convictions are less steep than declines with age. Despite the high degree of linearity, the RNO remain relatively stable up to the 11th conviction. Why is RNO not decreasing as steadily as RCL with each successive conviction? It may be that the sample of individuals convicted at each serial conviction number includes an increasing proportion of persisters as opposed to desisters (see Blumstein, Farrington, and Moitra 1985). In other words, there may be an increasing proportion of individuals who continue committing offenses at high rates. The fathers’ distributions of RCL and RNO are not as linear as the sons’ distributions (RCL: r = –0.75, p < .05; RNO: r = 0.27, ns). Again, declines with successive convictions are less steep than declines with age. Overall, RCL tends to decrease with each successive conviction, but this pattern is not



Figure 3 Sons’ Average Residual Career Length (in years) and Residual Number of Offenses after Each Conviction (n = 647 convicted offenses) 12

Mean RCL and RNO

10

8

6

4

2

0

1

2

3

4

5

6

RCL

10.5

8.2

7.9

8.7

9

7.7

8.7

7.2

5.8

6.3

6.8

5.4

6.4

4.8

3.9

2.5

RNO

5.2

5

4.7

5.5

5.6

5.5

6.1

5

4.5

4.7

5.3

4.1

4.1

3.1

2.8

2.8

Serial number of convictions

Note: Correlation between serial conviction number and RCL: r = –0.91, p < .0001 (n = 16). Correlation between serial conviction number and RNO: r = –0.77, p < .0001 (n = 16).

Figure 4 Fathers’ Average Residual Career Length (in years) and Residual Number of Offenses after Each Conviction (n = 253 convicted offenses) 18

Mean RCL and RNO

16 14 12 10 8 6 4 2 0

1

2

3

4

5

6

7

8

9

10 +

RCL

16.2

9.2

10.3

10.6

11.5

9.2

6.2

8.3

5

8.4

RNO

3.3

2.2

3.1

3.3

3.4

3.1

2.9

3.6

2.3

4

Serial number of convictions

Note: Correlation between serial conviction number and RCL: r = –0.75, p < .05 (n = 10). Correlation between serial conviction number and RNO: r = 0.27, ns (n = 10).



Figure 5 Sons’ Average Residual Career Length (in years) and Residual Number of Offenses According to the Time Since Last Conviction (n = 519 convicted offenses) 12

Mean RCL and RNO

10

8

6

4

2

0

1

2

3

4

5

6

7

8

RCL

8.1

7.8

6.7

9

11

6.4

2.3

3.4

9+ 1

RNO

5.5

5

4.4

4.9

7.2

1.6

0.86

1.5

0.55

Time since last conviction (in years)

Note: Correlation between the time since the last conviction and RCL: r = –0.73, p < .05 (n = 9). Correlation between the time since the last conviction and RNO: r = –0.77, p < .05 (n = 9). Average time since the last conviction: 2.7 years. First offenses were excluded from this distribution (time since last offense = 0).

observed for RNO; the fathers’ distribution of RNO remains relatively constant. The positive, nonsignificant association between the fathers’ RNO and serial conviction number (as opposed to the negative, significant association observed for RCL) suggests that although the number of years remaining in criminal careers tends to decline in a relatively uniform manner for all offenders, this is not the case for the number of offenses remaining in criminal careers. Time Since the Last Conviction. The sons’ distributions of RCL and RNO according to the time between the last and current convictions are presented in Figure 5. These distributions are characterized by increased fluctuations in comparison with previous figures, but they still remain linear (RCL: r = –0.73, p < .05; RNO: r = –0.77, p < .05). Once again, the similarity of patterns between the RCL and RNO distributions is striking. Both distributions show a considerable increase between three and five years since the last conviction, followed by a substantial decrease. Despite these fluctuations, RCL and RNO tend to decline as the time lag between the last and current convicted offense increases, which suggests that individuals who are convicted at rela-



Figure 6 Fathers’ Average Residual Career Length (in years) and Residual Number of Offenses According to the Time Since Last Conviction (n = 187 convicted offenses) 16

Mean RCL and RNO

14 12 10 8 6 4 2 0

1

2

3

4

5

6 to 9

RCL

14.6

7.3

12.3

10.6

8.6

4.8

10 + 3.6

RNO

4.6

2

4.2

3

2.2

1.5

1.3

Time since last conviction (in years)

Note: Correlation between the time since the last conviction and RCL: r = –0.53, ns (n = 7). Correlation between the time since the last conviction and RNO: r = –0.48, ns (n = 7). Average time since the last conviction: 5.6 years. First offenses were excluded from this distribution (time since last offense = 0).

tively short time intervals tend to have longer RCL and RNO; these are likely to be high-rate offenders. Figure 6 shows that the fathers’distributions of RCL and RNO are less linear (RCL: r = –0.53, ns; RNO: r = –0.48, ns). Both RCL and RNO decline as the time lag between the last and current convicted offenses increases, which suggests that long time intervals identify low-rate offenders. Again, findings for fathers and sons are similar, but the overall average time since the last convicted offense is higher for fathers than for sons (5.6 and 2.7 years, respectively). Overall, these results are consistent with Blumstein’s (1994: 403) findings: “Since all careers have to have a last crime, the inferred probability that a particular career has terminated increases as the interval since the last crime becomes longer . . . while the career is active, long intervals are less likely when the offending frequency is higher.” Age of Onset. Past research has repeatedly shown that an early onset predicts the length and intensity of future criminal activity (Blumstein et al. 1982; Farrington et al. 1990; Le Blanc and Fréchette 1989; Moffitt 1993; Piquero et al. 2004). The analyses carried out in this section explore whether



Table 1 Sons’ and Fathers’ Average Residual Career Length and Residual Number of Offenses According to Age on Offense and Age of Onset, 17 to 70 Years Old Age on Offense Sons (n = 470) RCL Juvenile onset Adult onset Total RNO Juvenile onset Adult onset Total Fathers (n = 212) RCL Juvenile onset Adult onset Total RNO Juvenile onset Adult onset Total

17 to 24

25 to 35

36 and Older

Total

9.1 (n = 232) 5.2 (n = 82) 8.1 (n = 314)

4.2 (n = 118) 4.0 (n = 38) 4.2 (n = 156)

— — —

7.5 (n = 350) 4.8 (n = 120) 6.8 (n = 470)

5.6 (n = 232) 2.1 (n = 82) 4.7 (n = 314)

2.8 (n = 118) 1.5 (n = 38) 2.5 (n = 156)

— — —

4.7 (n = 350) 1.9 (n = 120) 4.0 (n = 470)

17 to 24

25 to 35

36 and Older

Total

14.7 (n = 28) 12.3 (n = 43) 13.2 (n = 71)

12.8 (n = 20) 10.2 (n = 40) 11 (n = 60)

6.1 (n = 22) 3.7 (n = 59) 4.4 (n = 81)

11.4 (n = 70) 8.1 (n = 142) 9.2 (n = 212)

3.4 (n = 28) 4 (n = 43) 3.8 (n = 71)

4.1 (n = 20) 2.6 (n = 40) 3.1 (n = 60)

2.6 (n = 22) 1.2 (n = 59) 1.6 (n = 81)

3.3 (n = 70) 2.4 (n = 142) 2.7 (n = 212)

RCL According to:

Sons

Fathers

1. Age on offense 2. Age of onset: 3. Age on offense * age of onset:

F = 22.7, df = 1, p < .0001 F = 10.5, df = 1, p < .01

F = 15.7, df = 2, p < .0001 F = 3.4, df = 1, p < .10

F = 8.0, df = 1, p < .01

F = 0.02, df = 2, ns

RNO According to:

Sons

1. Age on offense: 2. Age of onset: 3. Age on offense * age of onset:

F = 16.5, df = 1, p < .0001 F = 33.7, df = 1, p < .0001

F = 5.9, df = 2, p < .01 F = 2.6, df = 1, ns

F = 7.4, df = 1, p < .01

F = 2.4, df = 2, p < .10

Fathers

the relationship between RCL and RNO and age varies according to age of onset. Table 1 shows the sons’ and fathers’ average RCL and RNO according to age on offense and juvenile versus adult onset. For sons, in both age groups



Table 2 Sons’ Average Residual Career Length and Residual Number of Offenses per Number of Co-offenders (n = 631 convicted offenses) Number of Co-offenders

RCL

RNO

0 (n = 333) 1 (n = 166) 2 (n = 81) 3 (n = 31) 4 or more (n = 20)

7.59 8.85 9.38 9.16 8.4

4.55 5.42 5.41 6.45 5.7

RCL according to the number of co-offenders: F = 1.83, df = 4, ns RNO according to the number of co-offenders: F = 2, df = 4, ns Average number of co-offenders: 0.8 Note: RCL and RNO lengths according to the number of co-offenders could not be computed for fathers, due to small sample sizes. Most offenses committed by fathers (n = 146, 87 percent) did not involve any co-offenders.

(17 to 24 and 25 to 35), RCL and RNO were higher for individuals who had first been convicted between 10 and 16 years old (overall RCL for juvenile onsetters: 7.5; overall RCL for adult onsetters: 4.8; overall RNO for juvenile onsetters: 4.7; overall RNO for adult onsetters: 1.9). Thus, for different ages on offenses, age of onset is negatively related to RCL and RNO. Discrepancies between the RCL and RNO of juvenile and adult onsetters are greater for offenses committed between ages 17 and 24, which suggests a greater effect of age of onset on RCL and RNO during this period (in contrast to the 25 to 35 period); this interaction between the age on offense and age of onset is statistically significant for both RCL and RNO. Table 1 also shows the fathers’ distributions of RCL and RNO according to age on offense and age of onset. Once again, the overall average RCL is higher for juvenile onsetters than for adult onsetters (11.4 vs. 8.1 years), and the same is true for RNO (3.3 vs. 2.4 offenses). Despite this general tendency for juvenile onsetters to have higher RCL and RNO, univariate analyses of variance show that the effect of onset on these two variables is not significant, nor is the interaction between age on offense and age of onset. Thus, age of onset predicts RCL and RNO for sons but not for fathers. Nonetheless, it seems that the discrepancies between the RCL and RNO of juvenile versus adult offenders are more pronounced in older ages. The nonsignificant tests may be a result of small sample sizes in the fathers’ analyses. Co-offending. Average RCL and RNO in relation to the number of cooffenders are presented in Table 2 for sons. The average number of cooffenders was 0.8, with 53 percent of offenses involving no co-offenders at



Table 3 Sons’ Average Residual Career Length and Residual Number of Offenses According to Age on Offense and Offense Type, 10 to 35 Years Old (n = 645 convicted offenses) Age on Offense 10 to 16 RCL Nonviolent offenses Violent offenses Total RNO Nonviolent offenses Violent offenses Total

17 to 24

25 to 35

Total

11.5 (n = 161) 14.6 (n = 16) 11.8 (n = 177)

8.0 (n = 261) 8.7 (n = 53) 8.1 (n = 314)

4.1 (n = 121) 4.3 (n = 35) 4.2 (n = 156)

8.1 (n = 543) 8.1 (n = 104) 8.1 (n = 647)

7.5 (n = 161) 9.2 (n = 16) 7.6 (n = 177)

4.7 (n = 261) 4.9 (n = 53) 4.7 (n = 314)

2.6 (n = 121) 2.3 (n = 35) 2.5 (n = 156)

5.0 (n = 543) 4.7 (n = 104) 5.0 (n = 647)

RCL according to: 1. Age on offense: F = 39.8, df = 2, p < .0001 2. Offense type: F = 3.4, df = 1, p < .10 3. Age on offense * offense type: F = 1.2, df = 2, ns RNO according to: 1. Age on offense: F = 33.7, df = 2, p < .0001 2. Offense type: F = 1.2, df = 1, ns 3. Age on offense * offense type: F = 0.93, df = 2, ns Note: Residual lengths according to offense type could not be computed for fathers, due to small numbers of convictions for violent offenses (n = 14, 8 percent).

all. RCL and RNO did not vary significantly according to the number of cooffenders; the univariate analyses of variance were not statistically significant. These findings suggest that the number of co-offenders does not predict RCL and RNO. Offense Type. Table 3 shows the distribution of RCL and RNO according to the age on offense and offense type (violent vs. nonviolent). Although the overall average RCL and RNO were similar for both offense types, these figures are generally higher for violent offenses committed at the youngest ages (10 to 16). This suggests that the effect of offense type on RCL and RNO is attenuated with age. Univariate analyses of variance show that offense type and the interaction between age and offense type did not have a significant effect on RCL and RNO. However, significance tests may have been affected by the small number of convictions for violent offenses in comparison to nonviolent offenses.



Table 4 Sons’ Distribution of Residual Career Length According to Risk Scores (n = 519 convicted offenses) RCL Risk Score 1 2 3 4 Total

Low (%) (0-7 years)

High (%) (8 years or more)

70 (n = 67) 56 (n = 108) 50 (n = 66) 35 (n = 34) 53 (n = 275)

30 (n = 29) 44 (n = 85) 50 (n = 66) 65 (n = 64) 47 (n = 244)

Total (%) 100 (n = 96) 100 (n = 193) 100 (n = 132) 100 (n = 98) 100 (n = 519)

2

Note: χ = 25.2, df = 3, p < .0001. AROC = .262, p < .05.

Table 5 Sons’ Distribution of Residual Number of Offenses According to Risk Scores (n = 519 convicted offenses) RNO Risk Score

Low (%) (0-3 offenses)

High (%) (4 offenses or more)

1 2 3 4 Total

74 (n = 71) 58 (n = 112) 35 (n = 46) 33 (n = 32) 50 (n = 261)

26 (n = 25) 42 (n = 81) 65 (n = 86) 67 (n = 66) 50 (n = 258)

Total (%) 100 (n = 96) 100 (n = 193) 100 (n = 132) 100 (n = 98) 100 (n = 519)

2

Note: χ = 50.9, df = 3, p < .0001. AROC = .362, p < .05.

Can Risk Scores Predict RCL and RNO? To investigate the predictability of RCL and RNO based on variables available in criminal records, risk scores were computed for sons and fathers based on the four most influential variables: age on offense, conviction number, time since the last conviction, and age of onset. The variables that were not dichotomous were dichotomized (using the median as the cut-off point), and a cumulative score ranging from 0 to 4 was created by summing the four scores.8 The predictive validity of the risk scores was investigated according to their ability to predict RCL and RNO (see Tables 4 to 7); RCL and RNO were dichotomized in low and high categories. Results show that individuals with higher risk scores tend to have more years remaining in their criminal careers, and to commit more offenses. All four chi-squared tests were significant, but chi-squared measures only deviations from chance expectation and not the linearity of relationships.



Table 6 Fathers’ Distribution of Residual Career Length According to Risk Scores (n = 187 convicted offenses) RCL Risk score 1 2 3 4 Total

Low (%) (0-8 years)

High (%) (9 years and higher)

78 (n = 14) 76 (n = 58) 46 (n = 37) 25 (n = 3) 60 (n = 112)

22 (n = 4) 24 (n = 18) 54 (n = 44) 75 (n = 9) 40 (n = 75)

Total (%) 100 (n = 18) 100 (n = 76) 100 (n = 81) 100 (n = 12) 100 (n = 187)

2

Note: Χ = 23.8, df = 3, p < .0001. AROC = .424, p < .05.

Table 7 Fathers’ Distribution of Residual Number of Offenses According to Risk Scores (n = 187 convicted offenses) RNO Risk Score

Low (%) (0-2 offenses)

High (%) (3 offenses or more)

1 2 3 4 Total

50 (n = 9) 75 (n = 57) 53 (n = 43) 42 (n = 5) 61 (n = 114)

50 (n = 9) 25 (n = 19) 47 (n = 38) 58 (n = 7) 39 (n = 73)

Total (%) 100 (n = 18) 100 (n = 76) 100 (n = 81) 100 (n = 12) 100 (n = 187)

2

Note: χ = 11.2, df = 3, p < .05. AROC = .164, p < .10.

ROC curves assess the predictive efficacy of classification schemes, when the outcome variable has two categories. In other words, the ROC curve “plots the probability of a ‘hit’ versus the probability of a ‘false positive’,” and is a measure that is “unaffected by changes in sample size and row and column totals” (Farrington et al. 1996:515). The area under the ROC curve is a better measure of predictive efficiency in 4 × 2 tables than chi-squared because it measures the linearity of the relationship. Farrington et al. (1996) developed a measure (“AROC”) that varies from 0 to 1 (i.e., from chance to perfect prediction): AROC= 2 * (A – 0.5), where A = area under the ROC curve. AROC obtained for the sons’ RCL and RNO were both significant (RCL: A = .631, SD = .025, AROC = .262, p < .05; RNO: A = .681, SD = .024, AROC = .362, p < .05). Although these predictions are significantly better than chance, neither is greater than AROC = 0.5, which suggests that they are nearer to chance than to perfect prediction on the 0 to 1 scale. The same is true for fathers’ RCL (A = .712, SD = .042, AROC = .424, p < .05). The prediction for the fathers’ RNO is not quite significant because the risk



score is not linearly related to the outcome (A = .582, SD = .046, AROC = .164, p < .10). In short, these results suggest that these risk scores are significant but not highly predictive of RCL and RNO. This finding highlights the difficulties associated with predictions based on information included in official records (i.e., information that is most often available to decision-makers in the criminal justice system).

Discussion One of the striking findings of this study is the remarkable linearity of the distributions of RCL and RNO, particularly according to the age on offense. As offenders got older, their number of remaining years of active offending declined; similar results were observed for the RNO. This general decline is consistent with the age-crime curve, which suggests increasing dropout rates with increasing age. Sampson and Laub (2003:569) found that the traditional age-crime distribution also applied to their sample of serious and persistent offenders, and concluded that “Aging out of crime is thus the norm—even the most serious delinquents desist.” The replication of findings for sons and fathers is impressive, because their conviction careers spanned different time periods. The distributions of RCL and RNO reflect age-crime tendencies of active offenders. Some authors have argued that the relationship between age and crime reveals changes in prevalence (participation) rather than incidence (frequency) of offending. In other words, the number of active offenders peaks in late adolescence and declines thereafter, but individuals who remain active in offending tend to do so at a relatively stable rate across various periods of the life course (Blumstein, Cohen, and Farrington 1988; Farrington 1986). Our results showed that distributions of RCL generally displayed a greater degree of linearity than those of RNO, particularly in relation to offense number. The fact that the fathers’ distribution of RNO remained relatively stable after each successive conviction suggested that the number of offenses remaining in criminal careers does not decline uniformly for all offenders across the life course, especially when the follow-up extends past mid-life. In their follow-up of Glueck men up to age 70, Sampson and Laub (2003:584) argued that “life-course-persistent offenders are difficult, if not impossible, to identify prospectively using a wide variety of childhood and adolescent risk factors.” It was interesting to observe that age of onset predicted RCL and RNO independently of age on offense for sons but not for



fathers. One could assume that this result suggests that the impact of age of onset on RCL and RNO is reduced when the follow-up extends past mid-life. However, despite the nonsignificance of results obtained for fathers, it still remained that discrepancies between the RCL and RNO of juvenile versus adult offenders were more pronounced in older ages. Due to the small sample sizes, further analyses are required before reaching any definite conclusions regarding the predictive power of age of onset on the outcome of criminal careers past mid-life. In the present analyses, risk scores were significantly but not highly predictive of RCL and RNO. This finding underlines the difficulty of making predictions based on information available in official records, and supports the need for the use of self-reports of offending and other features of social background as predictors of future criminal behavior. Although this study has focused on implications for sentencing, criminal history information may be used for other purposes, such as decisions relating to employment, public housing, welfare, and so on. In England, the Rehabilitation of Offenders Act of 1974 allows for some criminal convictions to be “spent” or overlooked after a so-called “rehabilitation period.” This period refers to a given period of time since the last conviction, after which an individual would be relieved of the obligation to divulge the conviction to potential employers or when applying for loans or insurance, for instance. Thus, information about criminal history extends beyond the realm of sentencing and is relevant to other social policies. Exact estimates of RCL and RNO should be interpreted with caution for both sons and fathers. For fathers, average RCL and RNO are generally based on small numbers due to the relatively low prevalence of convictions. For sons, these figures are likely to be an underestimation of the actual RCL and RNO due to the effects of truncation and false desistance. The purpose of this study was not to offer perfectly accurate estimates of RCL and RNO (which would be an impossible task, considering the diversity of offending trajectories across criminal careers), but rather to present basic information about RCL and RNO across developmental stages. More research is needed on RCL and RNO using more extensive prospective longitudinal data. Our results have shown that the highest RCL and RNO are observed in early adolescence, but it would obviously be undesirable to incapacitate young offenders for more than 10 years. Blumstein et al. (1988) suggested alternative solutions based on supportive interventions. Individuals who are convicted during adolescence may benefit from increased supervision, nonstigmatizing intervention programs, and a more intensive follow-up. In adulthood, they may benefit from job-training programs and other forms of intervention that promote successful reintegration into society.



How does this article add to existing results on the Cambridge Study? Although the prevalence of convictions in this sample of British males has been found to peak around age 17 (Farrington 1997), this does not appear to be the case for the distributions of RCL and RNO. These contrasting results suggest that age-crime tendencies tend to be distinct for active offenders in contrast to the general population (including one-time offenders and nonoffenders), and that RCL and RNO may be distinct criminal career parameters. Age of onset has been found to be predictive of an increased duration of criminal careers (Farrington 1997; Farrington et al. 1990), but our results showed that it remains unclear whether this association is also relevant to criminal career outcomes past mid-life. Finally, although past research has demonstrated the link between age and co-offending (Reiss and Farrington 1991), we have demonstrated that co-offending is a weak predictor of the time and number of offenses remaining in criminal careers. The same is true for offense type, supporting the hypothesis that “violent offences occurred at random in criminal careers” (Farrington 1997:380). It would be unwise to base policy recommendations on our results without further replication. One of the limitations of the analyses conducted in this study relates to the fact that termination is measured as the last convicted offense. Because officially recorded offenses comprise only a small proportion of all committed offenses (e.g., Elliott 1994), results found in this study can only be extended to official criminal careers, rather than actual criminal careers. Future studies should attempt to integrate both official and selfreported measures of offending and explore which variables included in official records predict self-reported RCL. Self-reported estimates of remaining career lengths will undoubtedly be higher than official RCL (Le Blanc and Fréchette 1989). Our findings are based on samples of repeat offenders, that is, we are measuring criminal persistence among offenders who commit crime at rates high enough to register. It is important to acknowledge that the results may be different with a more representative sample. Also, because most convicted offenses by sons and fathers were nonviolent, it is not clear to what extent these findings can be generalized to violent offenders. Finally, the risk prediction of RCL and RNO was of descriptive nature only, and this issue needs to be explored more in-depth, using more sophisticated riskassessment measures. The main objective of this study was to advance knowledge about a topic that has been largely neglected by criminal career research. Although RCL and RNO can potentially have important policy implications, namely for sentencing and incapacitation policies, further replication is needed. Future studies should explore RCL and RNO using self-reports of offending, with larger representative samples followed up throughout various periods of the



life course, and with samples of violent offenders. The potential implications of RCL and RNO are, in our view, at least as important as those of the agecrime curve.

Notes 1. Self-reports of offending were not available for all ages; for instance, at age 32, sons were asked whether they had committed particular offenses in the past five years. Because of gaps in self-reported offending information, analyses were limited to official records. Data on convictions were used because information on arrests was not available. Blumstein, Farrington, and Moitra (1985) found similar estimates of recidivism probabilities between two different cohorts (Philadelphia and London), despite the fact that the former data was based on arrests and the latter on convictions. Although the list of convicted offenses includes minor crimes, these tend to have the lowest conviction rates (sons: theft of cycle: n = 8, 1 percent; theft from work: n = 14, 2 percent; theft from machines: n = 16, 2.5 percent; fathers: theft of cycle: n = 8, 3 percent; theft from work: n = 16, 6 percent; theft from machines: n = 6, 2 percent). Burglary was the most common convicted offense for sons (n = 106, 16 percent) and other thefts were most common among fathers (n = 59, 23 percent). Thus, the distributions of convictions are not dominated by trivial offenses. It may be argued that if most committed offences are minor, this may result in informal handling and thus, an underestimation of the actual number of offenses. However, although most trivial crimes (i.e., theft of cycle, theft from machines, etc.) are unlikely to lead to a conviction, individuals may be convicted for more than one offense in a year. The samples used in this study are repeat offenders who are probably not strangers to the criminal justice system; thus, one bicycle theft may not result in a conviction, but a combination of this offense with other offenses may do so. The underestimation of convictions may be less substantial for repeat offenders because the probability of conviction for any offense is likely to increase once the individual is known to the criminal justice system. 2. Overall, 49 sons and 48 fathers had committed only one offense and were excluded from the analyses (out of 647 and 253 offenses, respectively); these are small figures considering the extended observation periods (from age 10 to 40 for sons, and from age 10 to 70 for fathers). If the criminal career is defined as the “longitudinal sequence of crimes committed by an individual offender” (Blumstein et al. 1986:12), then it follows that the analysis of patterns of change occurring in criminal careers is not relevant to one-time offenders. Estimating the duration (or residual duration) of criminal careers inevitably requires at least 2 offenses. The idea that the analysis of patterns of change is not relevant to occasional offenders has been expressed by other researchers (Laub and Sampson 2001; Le Blanc 1993; Le Blanc and Loeber 1998). Some may argue that the exclusion of one-time offenders results in an overestimation of RCL and RNO. Only about one third of one-time offenders were convicted during adolescence and thus, because the convictions of one-time offenders are not concentrated in a specific period and seem to be more or less evenly distributed across the life course, the exclusion of these individuals did not have a substantial effect on the correlation coefficients and analysis of variance results. In fact, the analyses were repeated, including the one-time offenders, and coefficients remained similar. Thus, although the exclusion of one-time offenders may have slightly inflated the average RCL and RNO, it did not affect the overall distributions. 3. Although we do not wish to minimize the problem of false desistance, following up individuals over a 30-year span is still a considerably long observation period; past studies have based their analysis of termination on much shorter observation periods (Ayers et al. 1999: three-


Kazemian, Farrington / Two Generations of Repeat Offenders 111 year follow-up during adolescence; Elliott, Huizinga, and Menard 1989: seven-year follow-up during adolescence). 4. One issue that required particular attention in the analyses of fathers’ convictions concerned deaths. Overall, almost 60 percent of fathers had died by the year 1993. Because some of these individuals may have persisted in crime had they been alive, deaths may cause an underestimation of RCL and RNO. A total of 99 percent of the fathers’convictions occurred before age 60. Among the three individuals who were convicted after age 60, two were still alive in 2001, and the other passed away seven years after his last conviction (age 69). The potentially problematic cases refer to convicted fathers who died before age 60 (n = 8). In six of these cases, the last conviction occurred at least 10 years before the date of death; thus, it is unlikely that these individuals would have been reconvicted if they had not died. Of the remaining two fathers, the first was convicted one year before he died (age 55) and the other five years before death (age 45). Because these two fathers would have been more likely to persist in crime if they had not died, they were excluded from the analyses in order to minimize the risk of underestimating actual RCL. 5. Figures are presented up to age 40 in order to make comparisons with the sons’averages. 6. This variable refers to the offender’s age at the time of the offense and not at the time of the conviction. 7. It might be thought that the assessment of RCL and RNO needs to take account of periods when individuals were not at risk of offending (using crimes committed per “year free,” see Blumstein et al. 1986; Horney, Osgoos, and Marshall 1995; Piquero et al. 2001; Sampson and Laub 2003; Visher 1986) (i.e., periods of incarceration). Between ages 10 and 40, 30 males had been incarcerated for a total period of more than three months, and all but two individuals had been incarcerated for a cumulative period of less than two and a half years. Because these periods are relatively short, they are not likely to greatly impact estimates of RCL and RNO and were thus not taken into account. 8. The reliability of the risk scores was assessed using Cronbach’s alpha measure (sons: n = 519, alpha = 0.46; mean = 44.38, SD = 10; fathers: n = 187, alpha = 0.56, mean = 62.78, SD = 22.64).

References Ashford, José B. and Craig LeCroy. 1988. “Predicting Recidivism: An Evaluation of the Wisconsin Juvenile Probation and Aftercare Risk Instrument.” Criminal Justice and Behavior 15:141-51. Ayers, Charles D., James Herbert Williams, J. David Hawkins, Peggy L. Peterson, Richard F. Catalano, and Robert D. Abbott. 1999. “Assessing Correlates of Onset, Escalation, Deescalation, and Desistance of Delinquent Behavior.” Journal of Quantitative Criminology 15:277-306. Barnett, Arnold, Alfred Blumstein, and David P. Farrington. 1989. “A Prospective Test of a Criminal Career Model.” Criminology 27:373-87. Blumstein, Alfred. 1994. “Next Steps in Criminal Career Research.” Pp. 401-8 in Cross-National Longitudinal Research on Human Development and Criminal Behavior, edited by Elmar G. M. Weitekamp and Hans-Jürgen Kerner. Dordrecht, the Netherlands: Kluwer Academic. Blumstein, Alfred and Jacqueline Cohen. 1987. “Characterizing Criminal Careers.” Science 237:985-91. Blumstein, Alfred, Jacqueline Cohen, and David P. Farrington. 1988. “Longitudinal and Criminal Career Research: Further Clarifications.” Criminology 26:57-74. Blumstein, Alfred, Jacqueline Cohen, and Paul Hsieh. 1982. The Duration of Adult Criminal Careers: Final Report to National Institute of Justice. Pittsburgh, PA: Carnegie-Mellon University.


112 Journal of Research in Crime and Delinquency Blumstein, Alfred, Jacqueline Cohen, Jeffrey A. Roth and Christy A. Visher. Eds. 1986. Criminal Careers and Career Criminals (volume 1). Washington, DC: National Academy Press. Blumstein, Alfred, David P. Farrington, and Soumyo D. Moitra. 1985. “Delinquency Careers: Innocents, Desisters, and Persisters.” Pp. 187-219 in Crime and Justice: An Annual Review of Research (volume 6), edited by Michael Tonry and Norval Morris. Chicago: University of Chicago Press. Carney, Francis J. 1967. “Predicting Recidivism in a Medium Security Correctional Institution.” Journal of Criminal Law, Criminology and Police Science 58:338-48. Cohen, Jacqueline. 1983. “Incapacitation as a Strategy for Crime Control: Possibilities and Pitfalls.” Pp. 1-84 in Crime and Justice: An Annual Review of Research (volume 4), edited by Michael Tonry and Norval Morris. Chicago: University of Chicago Press. Elliott, Delbert S. 1994. “Serious Violent Offenders: Onset, Developmental Course, and Termination.” Criminology 32:1-21. Elliott, Delbert S., David Huizinga, and Scott Menard. 1989. Multiple Problem Youth: Delinquency, Substance Use, and Mental Health Problems. New York: Springer-Verlag. Farrington, David P. 1986. “Age and Crime.” Pp. 189-250 in Crime and Justice: An Annual Review of Research (Volume 7), edited by Michael Tonry and Norval Morris. Chicago: University of Chicago Press. ⎯⎯⎯. 1997. “Human Development and Criminal Careers”. Pp. 361-408 in The Oxford Handbook of Criminology (2d edition), edited by Mike Maguire, Rod Morgan, and Robert Reiner. Oxford: Clarendon Press. ⎯⎯⎯. 2003. “Key Results from the First Forty Years of the Cambridge Study in Delinquent Development.” Pp. 137-83 in Taking Stock of Delinquency: An Overview of Findings from Contemporary Longitudinal Studies, edited by Terence P. Thornberry and Marvin D. Krohn. New York: Kluwer Academic/Plenum. Farrington, David P., Sandra Lambert, and Donald J. West. 1998. “ Criminal Careers of Two Generations of Family Members in the Cambridge Study in Delinquent Development.” Studies on Crime and Crime Prevention 7:85-106. Farrington, David P., Rolf Loeber, Delbert S. Elliott, J. David Hawkins, Denise B. Kandel, Malcolm W. Klein, Joan McCord, David C. Rowe, and Richard E. Tremblay. 1990. “Advancing Knowledge about the Onset of Delinquency and Crime.” Pp. 283-342 in Advances in Clinical Child Psychology (volume 13), edited by Alan E. Kazdin and Benjamin Lahey. New York: Plenum. Farrington, David P., Rolf Loeber, Magda Stouthamer-Loeber, Welmoet B. Van Kammen, and Laura Schmidt. 1996. “Self-Reported Delinquency and a Combined Delinquency Seriousness Scale Based on Boys, Mothers, and Teachers: Concurrent and Predictive Validity for African-Americans and Causasians.” Criminology 34:493-517. Greenberg, David F. 1975. “The Incapacitative Effect of Imprisonment: Some Estimates.” Law and Society 9:541-80. Greene, M. A. 1977. The Incapacitative Effect of Imprisonment on Policies of Crime. Unpublished Ph.D. thesis. Pittsburgh, PA: Carnegie-Mellon University. Greenwood, Peter W. and Allan Abrahamse. 1982. Selective Incapacitation. Santa Monica, CA: Rand. Horney, Julie, D. Wayne Osgood, and Ineke Haen Marshall. 1995. “Criminal Careers in the Short-Term: Intra-Individual Variability in Crime and Its Relation to Local Life Circumstances.” American Sociological Review 60:655-73. Horwitz, Allan and Michael Wasserman. 1980. “Some Misleading Conceptions in Sentencing Research: An Example and a Reformulation in the Juvenile Court.” Criminology 18:411-24.


Kazemian, Farrington / Two Generations of Repeat Offenders 113 Laub, J. H. and Sampson, R. J. 2001. “Understanding Desistance from Crime.” Pp. 1-69 in Crime and Justice (volume 28), edited by Michael Tonry. Chicago: University of Chicago Press. Le Blanc, Marc. 1993. “Late Adolescence Deceleration of Criminal Activity and Development of Self- and Social Control.” Studies on Crime and Crime Prevention 2:51-68. Le Blanc, Marc and Marcel Fréchette. 1989. Male Criminal Activity from Childhood Through Youth: Multilevel and Developmental Perspectives. New York: Springer-Verlag. Le Blanc, Marc and Rolf Loeber. 1998. “Developmental Criminology Updated.” Pp. 115-98 in Crime and Justice (volume 23), edited by Michael Tonry. Chicago: University of Chicago Press. Moffitt, Terrie E. 1993. “‘Life-Course Persistent’ and ‘Adolescence-Limited’ Antisocial Behavior: A Developmental Taxonomy.” Psychological Review 100: 674-701. Piquero, Alex, Alfred Blumstein, Robert Brame, Rudy Haapanen, Edward P. Mulvey, and Daniel S. Nagin. 2001. “Assessing the Impact of Exposure Time and Incapacitation on Longitudinal Trajectories of Criminal Offending.” Journal of Adolescent Research 16:54-74. Piquero, Alex, Robert Brame, and Donald Lynam. 2004. “Studying Criminal Career Length through Early Adulthood Among Serious Offenders.” Crime and Delinquency 50:412-35. Piquero, Alex, David P. Farrington, and Alfred Blumstein. 2003. “The Criminal Career Paradigm.” Pp. 359-506 in Crime and Justice: A Review of Research (volume 30), edited by Michael Tonry. Chicago: University of Chicago Press. Reiss, Albert J. and David P. Farrington. 1991. “Advancing Knowledge about Co-Offending: Results from a Prospective Longitudinal Survey of London Males.” Journal of Criminal Law and Criminology 82:360-95. Sampson, Robert J. and John H. Laub. 2003. “Life-Course Desisters: Trajectories of Crime Among Delinquent Boys Followed to Age 70.” Criminology 41:555-92. Scarpitti, Frank R. and Richard M. Stephenson. 1971. “Juvenile Court Dispositions: Factors in the Decision-Making Process.” Crime and Delinquency 17:142-51. Shinnar, Shlomo and Reuel Shinnar. 1975. “The Effects of the Criminal Justice System on the Control of Crime: A Quantitative Approach.” Law and Society 9:581-611. Silver, Eric, William R. Smith, and Steven Banks. 2000. “Constructing Actuarial Devices for Predicting Recidivism: A Comparison of Methods.” Criminal Justice and Behavior 27: 733-64. Spelman, William. 1994. Criminal Incapacitation. New York: Plenum. Stolzenberg, Lisa and Stewart J. D’Alessio. 1997. “Three Strikes and You’re Out’: The Impact of California’s New Mandatory Sentencing Law on Serious Crime Rates.” Crime and Delinquency 43:457-69. Visher, Christy A. 1986. “The Rand Inmate Survey: A Reanalysis.” Pp. 161-211 in Criminal Careers and “Career Criminals” (volume 2), edited by Alfred Blumstein, Jacqueline Cohen, Jeffrey A. Roth, and Christy A. Visher. Washington, DC: National Academy Press. Von Hirsch, Andrew. 1988. “Selective Incapacitation Reexamined: The National Academy of Sciences’Report on Criminal Careers and ‘Career Criminals’.” Criminal Justice Ethics 7: 19-35. ⎯⎯⎯. 1998. “Selective Incapacitation: Some Doubts”. Pp. 121-27 in Principled Sentencing: Readings on Theory and Policy, edited by Andrew Ashworth and Andrew Von Hirsch. Oxford: Hart.

Lila Kazemian’s, Ph.D., research interests include offending across the life course, comparative criminology, and desistance from crime. David P. Farrington’s, Ph.D., research interests include criminal career research, juvenile delinquency, violent offending, and crime prevention.
