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

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:
School of Criminal Justice, Rutgers – Newark
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 19 articles hosted on the

SAGE Journals Online and HighWire Press platforms):
http://jrc.sagepub.com/cgi/content/refs/43/1/89
Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

© 2006 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
Journal of Research in Crime
Journal of Research
Article
10.1177/0022427805280066
Kazemian, Farrington
in /Crime
T wo Generations
and Delinquency
of Repeat Offenders
and Delinquency
Volume 43 Number 1
Exploring Residual Career February 2006 89-113
© 2006 Sage Publications
Length and Residual 10.1177/0022427805280066

http://jrc.sagepub.com
hosted at
Number of Offenses http://online.sagepub.com
for Two Generations

of Repeat Offenders
Lila Kazemian
David P. Farrington
University of Cambridge
Very few studies have explored residual career length (RCL) and residual num-
ber 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 con-
viction number for fathers. Over and above age on conviction, age of onset pre-
dicted 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 decision-
makers 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 resid-
ual criminal career length or residual number of offenses in criminal careers
(Piquero, Farrington, and Blumstein 2003; Von Hirsch 1988, 1998). Resid-
ual 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 age-
crime 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 individ-
uals who have been arrested or convicted in a given year and document
offenses committed in previous years. The second method involves the retro-
spective 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 individu-
als 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, retro-
spectively 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 pur-
poses: replication and “false desistance.” It seemed highly relevant to repli-
cate 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 trun-
Authors’ 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 age-
crime 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 theo-
retical 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 sen-
tences 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 fur-
ther 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 (break-
in), 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 remain-
ing 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 drop-
out 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 differ-
ences 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 resid-
ual 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 inca-
pacitation” (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 esti-
mation of career lengths; however, their analyses were limited to adult crimi-
nal careers and they had no information about juvenile onset. Another prob-
lematic 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 approxi-
mately 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 recidi-
vists. 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 vali-
dation 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 Delin-
quent 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 threat-
ening 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 explana-
tion 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 adoles-
cence. In contrast, the distributions of RCL and RNO are right-hand cen-
sored; 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, compari-
sons with fathers were carried out for two purposes: replication and address-
ing the issue of false desistance.
In all, 109 fathers were convicted of a total of 315 offenses; 48 were one-
time 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 con-
victions per father).5 The great majority of convicted offenses were nonvio-
lent 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 (juve-
nile onset: first conviction occurred before age 17; adult onset: first convic-
tion 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-offend-
ers (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 incar-
cerated 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
14
Mean RCL and RNO
12
10
0
10&
12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
11
RCL 15.6 12.1 12.8 11.6 10.1 11.6 9 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 distri-
butions 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 distri-
butions follow similar patterns, and RCL and RNO are significantly corre-
lated (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
25
20
Mean RCL and RNO
15
10
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 distribu-
tions 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 decreas-
ing 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
Residual Number of Offenses after Each Conviction
12
10
Mean RCL and RNO
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). Cor-
relation between serial conviction number and RNO: r = –0.77, p < .0001 (n = 16).
Figure 4
Residual Number of Offenses after Each Conviction
18
16
14
Mean RCL and RNO
12
10
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). Corre-
lation between serial conviction number and RNO: r = 0.27, ns (n = 10).

Figure 5
Residual Number of Offenses According to the Time Since
Last Conviction (n = 519 convicted offenses)
12
10
Mean RCL and RNO
0
1 2 3 4 5 6 7 8 9+
RCL 8.1 7.8 6.7 9 11 6.4 2.3 3.4 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). Aver-
age 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 con-
stant. 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 crimi-
nal 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 convic-
tion, 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
Residual Number of Offenses According to the Time Since
Last Conviction (n = 187 convicted offenses)
16
14
Mean RCL and RNO
12
10
0
1 2 3 4 5 6 to 9 10 +
RCL 14.6 7.3 12.3 10.6 8.6 4.8 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). Cor-
relation 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 lin-
ear (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 proba-
bility 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 pre-
dicts 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) 17 to 24 25 to 35 36 and Older Total
RCL
Juvenile onset 9.1 (n = 232) 4.2 (n = 118) — 7.5 (n = 350)
Adult onset 5.2 (n = 82) 4.0 (n = 38) — 4.8 (n = 120)
Total 8.1 (n = 314) 4.2 (n = 156) — 6.8 (n = 470)
RNO
Juvenile onset 5.6 (n = 232) 2.8 (n = 118) — 4.7 (n = 350)
Adult onset 2.1 (n = 82) 1.5 (n = 38) — 1.9 (n = 120)
Total 4.7 (n = 314) 2.5 (n = 156) — 4.0 (n = 470)
Fathers (n = 212) 17 to 24 25 to 35 36 and Older Total
RCL
Juvenile onset 14.7 (n = 28) 12.8 (n = 20) 6.1 (n = 22) 11.4 (n = 70)
Adult onset 12.3 (n = 43) 10.2 (n = 40) 3.7 (n = 59) 8.1 (n = 142)
Total 13.2 (n = 71) 11 (n = 60) 4.4 (n = 81) 9.2 (n = 212)
RNO
Juvenile onset 3.4 (n = 28) 4.1 (n = 20) 2.6 (n = 22) 3.3 (n = 70)
Adult onset 4 (n = 43) 2.6 (n = 40) 1.2 (n = 59) 2.4 (n = 142)
Total 3.8 (n = 71) 3.1 (n = 60) 1.6 (n = 81) 2.7 (n = 212)
RCL According to: Sons Fathers
1. Age on offense F = 22.7, df = 1, p < .0001 F = 15.7, df = 2, p < .0001

2. Age of onset: F = 10.5, df = 1, p < .01 F = 3.4, df = 1, p < .10
3. Age on offense *
age of onset: F = 8.0, df = 1, p < .01 F = 0.02, df = 2, ns
RNO According to: Sons Fathers
1. Age on offense: F = 16.5, df = 1, p < .0001 F = 5.9, df = 2, p < .01

2. Age of onset: F = 33.7, df = 1, p < .0001 F = 2.6, df = 1, ns
3. Age on offense *
age of onset: F = 7.4, df = 1, p < .01 F = 2.4, df = 2, p < .10
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) 7.59 4.55

1 (n = 166) 8.85 5.42
2 (n = 81) 9.38 5.41
3 (n = 31) 9.16 6.45
4 or more (n = 20) 8.4 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. Discrepan-
cies 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 sta-
tistically 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 co-

offenders are presented in Table 2 for sons. The average number of co-
offenders 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 17 to 24 25 to 35 Total
RCL
Nonviolent offenses 11.5 (n = 161) 8.0 (n = 261) 4.1 (n = 121) 8.1 (n = 543)
Violent offenses 14.6 (n = 16) 8.7 (n = 53) 4.3 (n = 35) 8.1 (n = 104)
Total 11.8 (n = 177) 8.1 (n = 314) 4.2 (n = 156) 8.1 (n = 647)
RNO
Nonviolent offenses 7.5 (n = 161) 4.7 (n = 261) 2.6 (n = 121) 5.0 (n = 543)
Violent offenses 9.2 (n = 16) 4.9 (n = 53) 2.3 (n = 35) 4.7 (n = 104)
Total 7.6 (n = 177) 4.7 (n = 314) 2.5 (n = 156) 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 co-
offenders; the univariate analyses of variance were not statistically signifi-
cant. 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 fig-
ures 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)
Low (%) High (%)
RCL Risk Score (0-7 years) (8 years or more) Total (%)
1 70 (n = 67) 30 (n = 29) 100 (n = 96)

2 56 (n = 108) 44 (n = 85) 100 (n = 193)
3 50 (n = 66) 50 (n = 66) 100 (n = 132)
4 35 (n = 34) 65 (n = 64) 100 (n = 98)
Total 53 (n = 275) 47 (n = 244) 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
Low (%) High (%)
RNO Risk Score (0-3 offenses) (4 offenses or more) Total (%)
1 74 (n = 71) 26 (n = 25) 100 (n = 96)

2 58 (n = 112) 42 (n = 81) 100 (n = 193)
3 35 (n = 46) 65 (n = 86) 100 (n = 132)
4 33 (n = 32) 67 (n = 66) 100 (n = 98)
Total 50 (n = 261) 50 (n = 258) 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 num-
ber, 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 devia-
tions from chance expectation and not the linearity of relationships.

Table 6
Fathers’ Distribution of Residual Career Length
Low (%) High (%)
RCL Risk score (0-8 years) (9 years and higher) Total (%)
1 78 (n = 14) 22 (n = 4) 100 (n = 18)

2 76 (n = 58) 24 (n = 18) 100 (n = 76)
3 46 (n = 37) 54 (n = 44) 100 (n = 81)
4 25 (n = 3) 75 (n = 9) 100 (n = 12)
Total 60 (n = 112) 40 (n = 75) 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
Low (%) High (%)
RNO Risk Score (0-2 offenses) (3 offenses or more) Total (%)
1 50 (n = 9) 50 (n = 9) 100 (n = 18)
2 75 (n = 57) 25 (n = 19) 100 (n = 76)
3 53 (n = 43) 47 (n = 38) 100 (n = 81)
4 42 (n = 5) 58 (n = 7) 100 (n = 12)
Total 61 (n = 114) 39 (n = 73) 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 crimi-
nal 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 peri-
ods 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 rela-
tively 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 pre-
dicted 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 pre-
dictive 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 relat-
ing to employment, public housing, welfare, and so on. In England, the
Rehabilitation of Offenders Act of 1974 allows for some criminal convic-
tions 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 con-
viction 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 non-
offenders), and that RCL and RNO may be distinct criminal career parame-
ters. 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 with-
out 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 propor-
tion 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 self-
reported measures of offending and explore which variables included in offi-
cial 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 risk-
assessment 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 age-
crime 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 convic-
tions 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 lat-
ter 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 han-
dling 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 bicy-
cle 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 occur-
ring 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 individu-
als 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 indi-
viduals 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-

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 con-
cerned 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 underesti-
mation 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 con-
viction 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 con-
victed 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 Wis-
consin 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 Crimi-
nal 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.

Blumstein, Alfred, Jacqueline Cohen, Jeffrey A. Roth and Christy A. Visher. Eds. 1986. Crimi-
nal 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 Pit-
falls.” 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 Termi-
nation.” Criminology 32:1-21.
Elliott, Delbert S., David Huizinga, and Scott Menard. 1989. Multiple Problem Youth: Delin-
quency, 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: Uni-
versity of Chicago Press.
⎯⎯⎯. 1997. “Human Development and Criminal Careers”. Pp. 361-408 in The Oxford Hand-
book 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.” Stud-
ies 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. “Advanc-
ing 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 Serious-
ness 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. Unpub-
lished 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 Circum-
stances.” 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.

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 Behav-
ior: 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 Para-
digm.” 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 Pre-
dicting 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 Delin-
quency 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 Sci-
ences’Report on Criminal Careers and ‘Career Criminals’.” Criminal Justice Ethics 7: 19-35.
⎯⎯⎯. 1998. “Selective Incapacitation: Some Doubts”. Pp. 121-27 in Principled Sentenc-
ing: 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.


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

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

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

Uploaded by

Copyright:

Available Formats

Journal of Research in Crime and Delinquency

The online version of this article can be found at:

Email Alerts: http://jrc.sagepub.com/cgi/alerts

Citations (this article cites 19 articles hosted on the

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Length and Residual 10.1177/0022427805280066

for Two Generations

Keywords: residual career length; residual number of offenses; criminal

A very large body of research has been dedicated to the prediction of

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

specifies the remaining number of offenses in criminal careers. This article

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Why Study RCL and RNO?

RCL and RNOs provide information on the time remaining in criminal

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Selective incapacitation policies aim to maximize incapacitation effects

Estimating the Length of Criminal Careers

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

The Present Study

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

The distributions of RCL and RNO according to six different variables

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Distributions of RCL and RNO

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Serial number of convictions

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

RCL 8.1 7.8 6.7 9 11 6.4 2.3 3.4 1

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

RCL 14.6 7.3 12.3 10.6 8.6 4.8 3.6

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Fathers (n = 212) 17 to 24 25 to 35 36 and Older Total

RCL According to: Sons Fathers

1. Age on offense F = 22.7, df = 1, p < .0001 F = 15.7, df = 2, p < .0001

RNO According to: Sons Fathers

1. Age on offense: F = 16.5, df = 1, p < .0001 F = 5.9, df = 2, p < .01

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

0 (n = 333) 7.59 4.55

Co-offending. Average RCL and RNO in relation to the number of co-

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

RCL according to:

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

1 70 (n = 67) 30 (n = 29) 100 (n = 96)

1 74 (n = 71) 26 (n = 25) 100 (n = 96)

Can Risk Scores Predict RCL and RNO?

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

1 78 (n = 14) 22 (n = 4) 100 (n = 18)

ROC curves assess the predictive efficacy of classification schemes, when

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

score is not linearly related to the outcome (A = .582, SD = .046, AROC =

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

Downloaded from http://jrc.sagepub.com by Francisco Bernabeu on May 25, 2008

You might also like