See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/42583194

Predicting Trajectories of Offending over the Life

Course: Findings from a Dutch Conviction Cohort

Article  in  Journal of Research in Crime and Delinquency · September 2009

DOI: 10.1177/0022427809341939 · Source: OAI

CITATIONS READS

39 98

3 authors:

Bianca E. Bersani Paul Nieuwbeerta

University of Massachusetts Boston Leiden University
27 PUBLICATIONS   629 CITATIONS    300 PUBLICATIONS   4,182 CITATIONS   

SEE PROFILE SEE PROFILE

John H. Laub
University of Maryland, College Park
93 PUBLICATIONS   14,765 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Risky and Violent Behavior View project

Desistance, Turning Points, and the Life Course View project

All content following this page was uploaded by Bianca E. Bersani on 23 May 2014.

The user has requested enhancement of the downloaded file.

Article
Journal of Research in Crime and
Delinquency
Predicting 46(4) 468–494
© The Author(s) 2009

Trajectories of Reprints and permission: http://www.

sagepub.com/journalsPermissions.nav
DOI: 10.1177/0022427809341939
Offending over the http://jrc.sagepub.com

Life Course: Findings

from a Dutch
Conviction Cohort

Bianca E. Bersani1, Paul

Nieuwbeerta2, and John H. Laub1

Abstract
Distinguishing trajectories of criminal offending over the life course,
especially the prediction of high-rate offenders, has received considerable
attention over the past two decades. Motivated by a recent study by
Sampson and Laub (2003), this study uses longitudinal data on conviction
histories from the Dutch Criminal Career and Life-Course Study (CCLS)
to examine whether adolescent risk factors predict offending trajectories
across the life span. The CCLS is particularly well suited to study
developmental offending trajectories as it contains detailed information
on individual criminal offending careers for a representative sample of all
individuals convicted in the Netherlands in 1977 (n = 4,615) beginning at
12 years of age and continuing into late adulthood. To assess predictive

1
University of Maryland, College Park, MD, USA
2
Leiden University, The Netherlands; Utrecht University, The Netherlands

Corresponding Author:
Bianca E. Bersani, Department of Criminology and Criminal Justice, University of Maryland,
2220 LeFrak Hall, College Park, MD, 20742, USA
Email: [email protected]
Bersani et al. 469

ability, the authors employ two different analytical approaches. First, the
authors examine whether offending trajectories can be prospectively
differentiated by risk factors identified in adolescence. Second, the
authors use group-based trajectory analysis to retrospectively identify
distinct developmental offending trajectories and employ a cross-
validation technique to examine the ability to predict the probability of
an individual’s membership in a particular trajectory group. Overall, the
results support the notion that it is difficult to predict long-term patterns
of criminal offending using risk factors identified early in the life course.

Keywords
age and crime, trajectories, prediction, desistance

Over the past two decades, the developmental course of criminal offending
has received considerable theoretical and research attention. On one hand,
Gottfredson and Hirschi (1990) argued that the age-crime relationship is
invariant and that all offenders commit fewer offenses as they age. In con-
trast, Blumstein and colleagues (Blumstein, Cohen, and Farrington 1988;
Blumstein et al. 1986) maintained that the age-crime relationship varies
and that a relatively small group of chronic offenders do not desist from
crime with age (see also Piquero, Farrington, and Blumstein 2007). Along
similar lines, Moffitt (1993, 1994) contended that there is a dual taxonomy
of offenders: a life course–persistent group wherein offending does not
decline with age and an adolescence-limited group wherein offending is
concentrated during the teenage years (for similar typological approaches,
see Loeber and LeBlanc 1990; Patterson and Yoerger 1993).
To date, research using a variety of data collected from different
sociohistorical times and geographic locales has found evidence of con-
siderable heterogeneity in offending patterns over the life course (Ezell
and Cohen 2005; Laub and Sampson 2003; Piquero et al. 2007). One
question that has emerged from these studies is whether it is possible to
distinguish or predict these varying criminal offending trajectories,
especially the chronic high-rate offender group, using risk factors iden-
tified early in the life course. Using a particularly rich data set, Sampson
and Laub (2003) tackled this question of prediction by examining long-
term criminal trajectories for a sample of high-risk boys using a variety
of factors occurring in childhood and early adolescence. Contrary to the
work of Moffitt, among others, Sampson and Laub found that prediction
was at best problematic as there was no evidence that individual, child-
hood, and/or family risk factors could predict long-term trajectories of
470 Journal of Research in Crime and Delinquency 46(4)

criminal offending (see also Laub and Sampson 2003; Sampson and
Laub 2005).
Although Sampson and Laub’s (2003) findings are compelling, the
generalizability of their results is questionable. Specifically, we do not
know the extent to which their findings are specific to the sociohistori-
cal context of their study or the individuals in their analysis. Moreover,
the fact that these findings have significant implications for extant crim-
inological theory as well as policies that seek early intervention in
childhood to ward off serious adolescent and adult offending, a replica-
tion of their analysis is essential. The aim of the current research is to
assess the robustness of these important findings utilizing data gathered
in a more contemporary, cross-national context with a sample contain-
ing nearly 5,000 convicted individuals.
We wish to note the growing recognition of two salient factors within
criminology that are of particular relevance to the current study. First,
there is an increasing appreciation that research should not be restricted
to data from the United States. Although recognition of the need for
more cross-national comparative studies1 is not new (see Glueck 1964),
this concern has grown in recent years (see Adler 1996; Farrington and
Wikström 1994; LaFree 2007). For instance, in his presidential address
to the American Society of Criminology, Gary LaFree (2007:14) noted:
“Stating that you are in favor of more comparative cross-national
research in criminology is a bit like saying that you are opposed to pre-
meditated murder—hardly anyone will disagree with you.” Our study
adds to the small but growing collection of comparative cross-national
criminological research. To the extent that our findings are similar or
different to those of Sampson and Laub (2003), we garner an increased
understanding of the generalizability of the findings regarding the abil-
ity to predict long-term trajectories of criminal offending.
Second, whether it is through the replication of key findings or the
systematic statistical summary of information via meta-analysis, there is
a growing recognition of the importance of doing a better job of cumu-
lating knowledge (Gendreau 2001; Lowenkamp, Cullen, and Pratt
2003). Our article contributes to the need of replication studies in crimi-
nology and expands upon previous studies of prediction of high-rate
offenders. The research presented here is especially important given that
we are just beginning to understand the development of offending over
the life course and because the implications of predicting distinct offend-
ing trajectories hold promise for public policy initiatives. We begin with
a brief review of the relevant literature before moving on to a detailed
discussion of the research presented here.
Bersani et al. 471

Prediction of Offending Trajectories

Although the prediction of future offending behavior has a long history
in criminology and criminal justice (see Gottfredson 1987; Harcourt
2006), it received increased popularity with the finding by Wolfgang,
Figlio, and Sellin (1972) that a small portion of the population was
responsible for a large portion of crime. With this finding came the
belief that if this small group of chronic offenders could be identified in
advance, then presumably a large amount of crime could be prevented
(Cohen 1983). However, identifying this small group of chronic offend-
ers in advance has been problematic due to a number of legal, ethical,
and methodological concerns (see Auerhahn 1999, 2006; Bernard and
Ritti 1991; Gottfredson and Moriarty 2006; Tonry 1987). Nonetheless,
the prediction of those most at risk for high-rate offending remains a
highly sought after goal.
The growth of developmental criminology and the accompanying
characterization of offender types add another dimension to the predic-
tion literature (e.g., Loeber and LeBlanc 1990; Moffitt 1993; Patterson
and Yoerger 1993). These researchers posit that there are distinct groups
of offenders whose developmental etiology can be linked to early risk
factors. Today, one of the most prominent theories in developmental
criminology is Moffitt’s (1993) dual taxonomy theory. Moffitt proposed
that there are two typologies of offenders: adolescence limited and life
course–persistent. The adolescence-limited offender follows a typical
path, initiating delinquent involvement in early adolescence with the
peak of involvement occurring in mid-adolescence followed by desis-
tance in young adulthood. Moffitt suggested that adolescence-limited
delinquent behavior represents a standard developmental sequence
where adolescents are caught in a “maturity gap” between childhood
and adulthood. Their behavior is thus a temporary declaration of auton-
omy and boundary testing. In fact, their involvement in delinquent acts
is deemed normative.
Conversely, life course–persistent offenders comprise a small per-
centage of individuals who initiate problematic behavior early in life
and remain highly delinquent/criminal throughout their lifetime. Moffitt
(1993) contended that neurological deficiencies, in tandem with defi-
cient environments in childhood, lead to the development of the life
course–persistent offender. She suggested that the prospective predic-
tion of the life course–persistent offender should be most effective when
employing measures of individual and family characteristics such as
472 Journal of Research in Crime and Delinquency 46(4)

gender, temperament, cognitive abilities, and family bonds. The predic-

tion of these individuals is of particular concern due to the likely
long-term detrimental consequences to both self and society.
In a recent article, Sampson and Laub (2003) examined whether they
could predict criminal trajectory patterns using data spanning virtually
the entire life course (ages 7–70) for a sample of nearly 500 high-risk
boys from the Gluecks’ classic study of juvenile delinquency and young
adult crime (Glueck and Glueck 1950; see also Laub and Sampson
2003; Sampson and Laub 2005). The authors addressed two key ques-
tions: whether there was a distinct offender group whose rates of crime
remained stable with increasing age and whether individual differences,
childhood characteristics, and family background could foretell long-
term trajectories of offending.
Three important findings were revealed in Sampson and Laub’s (2003)
analysis. First, when examined in the aggregate the pattern of offending that
emerged was one of a unimodal peak in crime in adolescence followed by a
decline through middle adulthood. Specifically, desistance from crime was
the norm even among a sample of high-risk boys (see also Blokland, Nagin,
and Nieuwbeerta 2005; Ezell and Cohen 2005; Piquero et al. 2007). Second,
they found evidence of considerable heterogeneity in individual age-crime
curves across the life span (see also Blokland et al. 2005; Ezell and Cohen
2005; Piquero et al. 2007). Finally, Sampson and Laub investigated whether
long-term patterns of offending could be predicted from a number of child
and adolescent risk factors, including cognitive ability, temperament, per-
sonality traits, childhood behaviors, early onset, and frequency of
involvement in crime and delinquency. Across an array of varying risk fac-
tors, the predictive power of individual differences, childhood factors, and
family background in identifying a distinct group of life course–persistent
offenders (Moffitt 1993) was not substantiated. Specifically, there was no
evidence that individual, child, and family risk factors could predict long-
term trajectories of criminal offending once conditioned on delinquency.

Current Study
The conclusions proffered by Sampson and Laub (2003) appear rather
dismal for criminological theory and public policy interventions rooted
in an early risk factor approach, but it would be imprudent to make rec-
ommendations based on these findings alone. Nevertheless, the
availability of data for conducting an investigation similar to the one
undertaken by Sampson and Laub are rare, as such an analysis requires
Bersani et al. 473

data covering the vast majority of an individual’s life course in addition

to the accessibility of information on important risk factors traced to
persistent offending in later life. Moreover, it is advantageous to have a
large sample size with a sufficient number of serious offenders. This
feature is particularly important in light of base rate problems that char-
acterize much prediction-based research (Gottfredson and Moriarty
2006). Data collected by the Netherlands Institute for the Study of Crime
and Law Enforcement (NSCR) offer a unique opportunity to conduct a
partial replication of the Sampson and Laub study as the data meet these
essential requirements. That is, the sample consists of more than 5,000
offenders who were convicted in 1977 with data spanning a large por-
tion of the life course (i.e., adolescence, young adulthood, and later
adulthood; for details, see Nieuwbeerta and Blokland 2003).
With this article, we aim to increase criminological knowledge by
adding to the discussion concerning the prediction of criminal careers.
Specifically, we test the robustness of Sampson and Laub’s (2003) find-
ings by replicating and extending their study regarding the prediction of
offending trajectories. First, we examine long-term patterns of offend-
ing (ages 18 to 55) in a cross-national context utilizing data gathered on
a large sample of convicted men and women in the Netherlands born
between 1907 and 1965. Due to the cross-national context and varying
sociohistorical periods, these data provide us with the unique opportu-
nity to assess the generalizability of Sampson and Laub’s findings.
Second, following the Sampson and Laub analysis, we examine whether
trajectories of criminal offending can be prospectively differentiated by
key risk factors identified in adolescence. Finally, we propose an alter-
native test of the prediction of offending trajectories. Specifically, we
use group-based trajectory analysis to identify retrospectively distinct
developmental offending trajectories and employ a cross-validation
technique to test our ability to predict the probability of an individual’s
membership in a particular group.

Data and Measures

In this study, we use data from the Criminal Career and Life-Course
Study (CCLS), a large-scale longitudinal study carried out at the NSCR
(Nieuwbeerta and Blokland 2003). The CCLS is based on a representa-
tive sample of 4 percent of all cases of criminal offenses that were tried
in the Netherlands in 1977. The total sample consists of 5,164 convicted
individuals. The principal investigators for the CCLS were able to trace
474 Journal of Research in Crime and Delinquency 46(4)

89.4 percent of the original sample, resulting in a final sample size of

4,615 (4,187 men and 428 women). The characteristics of these 4,615
individuals are similar to the total sample consisting of 5,164 persons
and therefore can be regarded as representative of all offenders in 1977.2

Subsample
In addition to analyzing the full CCLS sample, we also carry out sepa-
rate analyses on a random subset of male offenders (n = 689) from the
original CCLS data. Research supports that factors such as low intelli-
gence and psychological instability are important predictors of chronic
offending trajectories (DeLisi 2005; Fergusson and Horwood 2002;
Moffitt, Lynam, and Silva 1994). To measure the predictive utility of
these factors, we utilize data collected from the Ministry of Defense,
which contain information on individual traits compiled from psycho-
logical and physical assessments conducted when the men were 18 years
of age and drafted for military service.3

Criminal Career Data

The criminal careers of the offenders in the CCLS sample were recon-
structed using abstracts from the General Documentation Files (GDF)
of the Criminal Record Office (“rap sheets”). The GDF contain informa-
tion on every criminal case registered by the police at the Public
Prosecutor’s Office.4 These abstracts were supplemented with informa-
tion that normally would not be included due to statutory limitations.
That is, in the Netherlands a person is not given a “blank sheet” upon
becoming an adult. Therefore, the data used here contain information on
both juvenile and adult official offenses. The standard classification
system used in the Netherlands groups offenses into the following cat-
egories: violent offenses (i.e., sexual offenses, robbery), property
offenses, vandalism and offenses against the public order, drug offenses,
offenses of the Firearms Act, and other criminal law offenses (e.g.,
drunk driving, hit and run).5
This data set is not only sizable but also covers a large portion of the
adult life course. Individual offending rates are measured annually
beginning when the offenders were aged 12 up to the year 2002. Because
the data pertain to a sample of all convicted persons in 1977, the par-
ticipants range in age from 12 to 79 years old. Thus, although data for
everyone in the sample are available beginning at age 12, depending on
Bersani et al. 475

a
1.0
0.9
Mean Rate of Total Criminal

0.8
0.7
Convictions

0.6
0.5
0.4
0.3
0.2
0.1
0.0
12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age

b
0.5
Mean Rate of Violent Criminal

0.4
Convictions

0.3

0.2

0.1

0.0
12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age

Figure 1. Mean rate of offenses by criminal conviction type (n = 4,615)

1a. Mean rate of total criminal convictions (ages 12–55)
1b. Mean rate of violent criminal convictions (ages 12–55)

one’s age at conviction in 1977, the amount of follow-up data available

for any one individual varies.6 To avoid problems associated with having
only a small number of individuals defining offending trajectories at the
oldest ages, we limit the analysis to ages for which data are available on
476 Journal of Research in Crime and Delinquency 46(4)

a substantial number of individuals. This restriction censors our offend-

ing trajectories at 55 years of age. In addition, taking into account
mortality, each observation is censored after the time of death.
The data contain 177,137 person-by-age crime counts from ages 12 to
55. We present the actual mean conviction rate for all crimes and for vio-
lent crimes in Figures 1a and 1b, respectively. Clearly, for both the total
crime and violent crime measures, the age-crime pattern is asymmetric,
representing 21,024 person-years in which one or more crime convictions
were recorded. These figures illustrate that the crime trends peak in adoles-
cence and decline through middle adulthood.

Risk Factors
To examine whether trajectories of offending can be differentiated by
risk factors identified in adolescence, we employ four risk factors: two
measures of criminal history (i.e., early onset, chronic offending) and
two measures of individual traits believed to be related to crime (i.e.,
low intelligence, psychological instability).
Early onset of criminal offending has been identified as one of the
strongest predictors of long-term offending (Blumstein et al., 1986;
DeLisi 2005; Farrington et al. 1990; Loeber and LeBlanc 1990; Piquero
et al. 2007). Research suggests that individuals who begin offending at
early ages are likely to have longer, more varied, and more serious crim-
inal careers than those who initiate their criminal behavior at a later age
(for thorough reviews, see LeBlanc and Loeber 1998; Thornberry and
Krohn 2003). In the present study, we capture early onset with a mea-
sure of age at first conviction obtained from official reports. Those
respondents who had a criminal conviction history at 15 years of age or
younger are categorized as having an early onset of offending (early
onset = 1, comprising 17 percent of the sample).
Our second risk factor captures the frequency of official convictions
during adolescence (i.e., prior to age 18). Beginning in the early 1970s,
criminal career researchers identified a small subgroup of individuals,
“chronic offenders,” who account for a disproportionately large number
of crimes (Wolfgang et al. 1972). This research suggests that individuals
identified as chronic offenders have an increased likelihood of maintain-
ing their criminal behavior into adulthood (Farrington 2003; Piquero
et al. 2007). Therefore, by placing an additional restriction on our desig-
nation of at risk, including only those individuals with a history of five
Bersani et al. 477

or more criminal convictions prior to 18 years of age, we garner

increased assurance that our risk factor captures those individuals who
fit the criminal career persister and life course–persistent profiles
(Blumstein, Farrington, and Moitra 1985; Moffitt 1993). Individuals
with the chronic offender classification comprise nearly 5 percent of our
sample (chronic offending = 1), which is consistent with the criminal
career literature.
Information pertaining to intelligence and psychological instability is
only available for the subsample (n = 689). In the Netherlands, men
aged 18 were physically and psychologically tested for military service.
This information is archived with the Dutch Ministry of Defense and is
available for all men born after 1940. The examination consisted of
three parts and evaluated an array of factors, including general physical
condition, practical insight, intelligence, psychological stability, and
mathematical and language skills. Based on these assessments, the mili-
tary examination committee classified the man as fit for military service,
not fit due to physical limitations, not fit due to low intelligence, and not
fit due to psychological instability. We use the last two outcomes as
indicators of two additional risk factors: low intelligence and psycho-
logical instability. Both factors are believed to differentiate lifelong
trajectories of criminal offending (DeLisi 2005; Fergusson and Horwood
2002; Ge, Donnellan, and Wenk 2001; Moffitt 1993; Moffitt et al. 1994;
Piquero, Brame, and Lynam 2004; Piquero et al. 2007; Stattin and
Klackenberg-Larsson 1993). Specifically, individuals with low intelli-
gence and/or psychological instability have a greater likelihood of
becoming involved in criminal behavior and are more likely to have a
high rate of criminal involvement over the life course. In our sample,
102 of the 689 men (15 percent) were deemed not fit for military service
due to low intelligence (coded 1) and 97 of the 689 men (14 percent)
were declared not fit due to psychological instability (coded 1).

Prospectively Defined Groups

Our study starts by using a similar strategy as Sampson and Laub (2003)
to analyze the predictability of long-term patterns of offending.
Specifically, we examine whether trajectories of offending can be pro-
spectively differentiated by risk factors identified in adolescence. The
basic question posed by Sampson and Laub and examined here asks
whether the age-crime trajectory follows a different pattern across the life
course of offenders distinguished by adolescent risk factors.
478 Journal of Research in Crime and Delinquency 46(4)

To address this question, we analyze a series of predicted probability

models. These models estimate the probability of a criminal conviction
at each age from which a mean probability is calculated. Because of the
relative rarity of criminal convictions in each age period, we estimate
logistic regression models with three age terms (age, age2, age3) and
present the findings for any criminal conviction and violent criminal
conviction at each age from 18 to 55.7 Although we are limited in cap-
turing the wide range of predictors assessed by Sampson and Laub, we
employ four key risk factors here: early onset, chronic offending in ado-
lescence, low intelligence, and psychological instability.

Full Sample Analysis

We begin by estimating the predicted probabilities for any criminal con-
viction by early onset of offending (Figure 2a). Throughout young
adulthood, those in the early onset group have a higher mean level of
offending than those who do not have an early onset. It is evident, how-
ever, that although the two groups have different levels of offending, the
shape of the offending trajectories across the life course is very similar.
That is, offending in both groups peaks in early adulthood followed by
a gradual decline with age.
Because our any criminal conviction measure includes a diverse array
of crimes, it may be possible that this measure obscures important patterns
in the data. Specifically, Moffitt (2006) contended that individuals dis-
playing life course–persistent patterns of offending show the greatest
divergence from adolescent-limited offenders within the realm of vio-
lence. Therefore, we restrict our analysis to focus on violent criminal
convictions (see Figure 2b). The findings for violent criminal conviction
by early onset reveal a remarkably similar pattern to those using any crim-
inal conviction. The early onset risk group has a higher mean probability
of a violent conviction in early adulthood, but by mid-adulthood their
probability of offending is indistinguishable from that of the low risk
group. Moreover, by age 40 the two groups are nearly identical.
To capture the small subset of chronic offenders who may be more
likely to display stable patterns of offending throughout the life course, we
repeat the analysis using our second measure of risk: chronic offending in
adolescence. This measure captures the small portion of our sample that
has a history of five or more criminal convictions prior to age 18. We pres-
ent the findings for any and violent criminal convictions in Figures 2c and
2d, respectively. Again, we find that although the groups differ in their
Bersani et al. 479

a c
1.0 1.0

Mean Predicted Probability of

Mean Predicted Probability of

0.9 Early Onset = 1 0.9 Chronic = 1

Any Criminal Conviction

Any Criminal Conviction

0.8 Early Onset = 0 0.8 Chronic = 0

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age Age
b d
0.5 0.5
Mean Predicted Probability of a

Mean Predicted Probability of a

Violent Criminal Conviction

Chronic = 1

Violent Criminal Conviction

Early Onset = 1
0.4 0.4 Chronic = 0
Early Onset = 0

0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age Age

Figure 2. Mean predicted probability of conviction for risk factors: early onset,
chronic offending (ages 18–55; n = 4,615)
2a. Any criminal conviction by early onset
2b. Violent criminal conviction by early onset
2c. Any criminal conviction by chronic offending
2d. Violent criminal conviction by chronic offending

rates of offending, the overall long-term pattern remains unchanged.

Regardless of whether individuals are designated as a chronic offender
or not, a declining pattern of criminal offending across adulthood is
evidenced.8
Because there is no agreed upon marker identifying an “early onset
offender” or a “chronic adolescent offender” in the extant literature, we
repeat the analysis four additional times imposing different restrictions
on the composition of the at-risk group. First, we restrict the early onset
risk indicator to capture only those with a criminal conviction at 13 years
of age or younger (4 percent of the sample). Second, we restrict the early
onset indictor to represent those with a violent criminal conviction at 15
years of age or younger (3 percent of the sample). Third, we restrict the
chronic offender indicator to represent those with two or more violent
criminal convictions up to 17 years of age or younger (2 percent of the
sample). Even with these additional restrictions, we find a general
pattern of desistance among those at greatest risk for continued offending
(results not shown).
Fourth and finally, to isolate the most at-risk individuals in our sample,
we create a risk factor that captures those youth who have both an early
480 Journal of Research in Crime and Delinquency 46(4)

a
1.0
0.9 Early and Chronic = 1
Mean Predicted Probability of

Early and Chronic = 0

Any Criminal Conviction

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age

b
0.5
Early and Chronic = 1
Mean Predicted Probability of a
Violent Criminal Conviction

Early and Chronic = 0

0.4

0.3

0.2

0.1

0.0
18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age

Figure 3. Mean predicted probability of conviction for risk factor: early onset and
chronic offending (ages 18–55; n = 4,615)
3a. Any criminal conviction by early onset and chronic offending
3b. Violent criminal conviction by early onset and chronic offending

onset of offending (prior to 15 years of age) and a history of chronic offend-

ing in adolescence (five or more criminal convictions prior to 18 years of
age). This group represents arguably the most at-risk youth in the sample
(4 percent). We present the findings for this group looking at both any and
violent criminal convictions in Figures 3a and 3b. In general, although
Bersani et al. 481

a c
1.0 1.0

Mean Predicted Probability of

Mean Predicted Probability of

0.9 Low IQ = 1 0.9 Psychological Instability = 1

Any Criminal Conviction

Any Criminal Conviction

Low IQ = 0 Psychological Instability = 0

0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age
Age

b d
0.5 0.5

Mean Predicted Probability of a

Mean Predicted Probability of a

Low IQ = 1 Psychological Instability = 1

Violent Criminal Conviction

Violent Criminal Conviction

Low IQ = 0 Psychological Instability = 0

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age Age

Figure 4. Mean predicted probability of conviction for risk factors: low IQ and
psychological instability (ages 18–55; n = 689)
4a. Any criminal conviction by low IQ
4b. Violent criminal conviction by low IQ
4c. Any criminal conviction by psychological instability
4d. Violent criminal conviction by psychological instability

there are magnitude differences during early adulthood, we find yet again
that all groups evidence a sharp declining pattern with age.

Subsample Analysis
Using a random subsample of cases, we investigate the predictive abil-
ity of two additional individual difference measures (low intelligence
and psychological instability) understood to be important predictors for
future criminal behavior. The analytical strategy mirrors that of the full
sample described previously. We begin with the measure of intellectual
ability and examine both any criminal conviction and violent criminal
conviction (see Figures 4a and 4b, respectively). Unlike the magnitude
differences displayed previously using official statistics to capture early
onset and chronic offending, the differences depicted in these graphs
appear negligible. Next, we examine differences in long-term offending
trajectories taking into account psychological instability. We present the
findings for any criminal conviction and violent criminal conviction in
Figures 4c and 4d, respectively. Similar to the previous analyses, the
general pattern remains substantively the same.9
482 Journal of Research in Crime and Delinquency 46(4)

In sum, the findings from our prospective analyses are consistent and
suggest that the ability to prospectively predict a group of high-rate, chronic
offenders defined by a flat rate of offending over the life course based on
key adolescent risk factors is problematic. Regardless of crime type or risk
predictor employed, the substantive results presented here do not change
demonstrating the robustness of the findings. The pattern among all groups
of high-risk offenders reveals a high conviction rate in young adulthood
followed by a significant declining pattern with age. Overall then, although
magnitude differences are apparent, the criminal offending trajectories
derived from the CCLS follow a general path that declines over time.

Retrospectively Defined Groups

In the next stage of our study, we employ a group-based trajectory anal-
ysis that defines offending trajectories retrospectively and examine
whether risk factors identified in adolescence can prospectively differ-
entiate these trajectory groups. We begin by predicting trajectory group
membership. Conceptually, the group-based trajectory approach identi-
fies groups of individuals who display similar behavioral trajectories
over the life course (Nagin 2005). This analytic strategy is an advance-
ment over the previous prospective analysis in that rather than examining
average trajectories, group-based trajectory analysis allows for a within-
group examination of life course offending trajectories, increasing our
ability to isolate distinct pathways.
We use Nagin and Land’s (1993) semiparametric group-based mod-
eling approach (see also Nagin 2005) and estimate a zero inflated
poisson form of a group-based trajectory model:

lnðljit Þ ¼ bj0 þ bj1 ðageit Þ þ bj2 ðage2it Þ þ bj3 ðage3it Þ;

where ln(λitj ) is the natural logarithm of the number of total convictions

for persons i in group j at each age t. The equation is specified to follow
a cubic function of age (age, age2, and age3). This analysis results in the
identification of a number of different groups of individuals who
display similar behavioral trajectories from 18 to 55 years of age. We
find that a four-group model provides the best representation of the
conviction histories when considering parsimony and comprehensibility.
Due to the large number of cases and long observation period under
investigation in our data, the Bayesian Information Criterion (BIC)
score continues to increase as more groups are added. The five- and
Bersani et al. 483

six-group models merely differentiate among those with zero or a

negligible number of convictions and those repeatedly convicted during
the observation period.
Moreover, the four-group model achieves the most important criteria
for the adequacy of the trajectory models (Nagin 2005:88). Across the
groups, the average posterior probability of group membership ranges
from .85 to .95—these values are well above the .7 threshold recom-
mended when determining whether the groups are distinct and the model
fits well. Nagin (2005:88–90) also recommended three other criteria for
judging model adequacy, including the calculation of the odds of correct
classification for the distinguished groups, that the proportion assigned
to each group closely corresponds to the probability of group member-
ship for that group, and that the confidence intervals on group
membership probabilities are reasonably tight. The four-group model
presented here performs well by all these model accuracy criteria.
We present these four trajectories in Figure 5. Individuals are classified
in the trajectory group for which their posterior probability is the highest.
The first trajectory group (sporadic offenders) is made up of nearly
70 percent of the sample and evidences a near zero conviction rate
throughout adulthood. The second trajectory group (low-rate offenders;

2.5
Chronic Offender
Mean Rate of Total Criminal Convictions

Classic Desister
2.0 Low-rate Offender
Sporadic Offender

1.5

1.0

0.5

0.0
18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54
Age

Figure 5. Latent offending trajectories for total criminal convictions (ages 18–55)
484 Journal of Research in Crime and Delinquency 46(4)

15 percent of the sample) follows a path that rises steadily through early
adulthood and begins declining in the mid- to late 30s. The third group
(classic desisters; 11 percent of the sample) follows the classic age-crime
curve with conviction rates peaking in early adulthood and declining
steadily thereafter. The final group, chronic offenders (4 percent of the
sample), demonstrates a high rate of convictions throughout the 20s and
30s, followed by a declining pattern beginning in the late 30s.
With the trajectory groups identified, we examine whether these groups
can be prospectively differentiated by adolescent risk factors. We display
the means for our risk factors and individual characteristics (i.e., gender,
ethnicity, marital status at age 18, and parental status at age 18) by group
membership for total criminal convictions in Table 1. Significant mean

Table 1. Comparison of Adolescent Risk Factors by Trajectory Group

Membership
Sporadic Low-rate Classic Chronic
Offender Offender Desister Offender
Full Sample (N = 4,615) n = 3,288 n = 646 n = 487 n = 194 F statistic
Risk factors
  Early onset .091a .143a .279 .321 72.129*
  Chronic adolescent .007a .014a .081a .176 116.046*
offender
Personal characteristics
  Female .140a .058 .024 .018 34.264*
  Non–Dutch .084a .141 .090a .193 13.965*
  Married at age 18 .030 .025 .023 .014 .819
  Child at age 18 .020 .035 .021 .022 2.064
Subsample (n = 618) n = 407 n = 73 n = 105 n = 33
Risk factors
  Early onset .126a .216a .242a .483 11.400*
  Chronic adolescent .012a .026a .054a .174 10.925*
offender
  Low IQ .100 .175 .200 .212 3.933*
  Psychological instability .043a .122 .176 .202 9.818*
Personal characteristics
  Non–Dutch .034 .122 .078 .144 5.065*
  Married at age 18 .001 .010 .030 .010 3.168*
  Child at age 18 .000 .019 .017 .010 2.262
a. Significant mean difference at the p < .05 level; the comparison group is the chronic group.
*ANOVA f test is significant at the p < .05 level.
Bersani et al. 485

differences are estimated using an ANOVA test with a post hoc comparison
to determine which groups significantly differed from one another. We
report the significant differences using chronic offenders as our compari-
son group. The results illustrate some significant differences in the means
across the four groups of offenders. For example, classic desisters and
chronic offenders are more often characterized by early onset. In addition,
as defined in the data, chronic offenders are characterized by chronic
offending in adolescence. In contrast, we find few significant differences
across the four groups when looking at individual characteristics. Overall,
although there are some mean differences across the four trajectory groups,
individuals identified as being at risk for high-rate, chronic offending in
adulthood (i.e., those defined by early onset, chronic offending in adoles-
cence, low intelligence, and/or psychological instability) are dispersed
across all four trajectory groups. Moreover, there is no consistent pattern-
ing of risk factors distinguishing the chronic offender group.

Cross-Validation of Latent Classes of Criminal Behavior

In the final stage of the analysis, we use a cross-validation technique to
examine our ability to predict offending patterns throughout adulthood.
This strategy involves three steps. First, we randomly select half of the
cases in our sample and estimate the parameters of a prediction model
for trajectory group membership. To do so we use a multinomial logistic
regression model in which the dependent variable is the trajectory group
to which each person belongs (i.e., sporadic offenders, low-rate offend-
ers, classic desisters, or chronic offenders). The independent variables
include the four adolescent risk factors (i.e., early onset, chronic offend-
ing in adolescence, low intelligence, and psychological instability) and
a number of individual characteristics (i.e., gender, ethnicity, marital
status at age 18, and parental status at age 18).
In the second step, we use the estimated parameters of the prediction
model obtained in the first step10 to predict the trajectory group member-
ship for the cases in the second half of our data set. Specifically, for each
person in the second half of our sample we take the values on the risk
factors and individual characteristics and impute them into the predic-
tion model with the parameters as estimated in the first step. Then, based
on this prediction model, we calculate the probability that a person
belongs to each of the four distinguished trajectory groups. Individuals
are predicted to be a member of the trajectory group for which their
calculated probability of group membership is the highest.
486 Journal of Research in Crime and Delinquency 46(4)

In the final step, we address our main research question and test the
extent to which group membership in a trajectory group can be ade-
quately predicted based on risk factors identified in adolescence and
individual characteristics. That is, we examine whether a person’s pre-
dicted group membership corresponds with or differs from a person’s
observed trajectory group. We present these results in Table 2.
We discuss only the results for the full sample (upper table) because
the results for the subsample (bottom table) are very similar. Overall, the
findings reveal that we accurately predict group membership for 71 per-
cent of the cases (in bold in table). Although the general predictive
accuracy of our model is informative, a more meaningful analysis for
criminologists interested in criminal careers involves an assessment of
the model’s predictive accuracy within groups.
Due to their continued involvement in crime into later adulthood, we
believe the two groups of greatest substantive interest are the low-rate
and chronic offender groups. When we inspect the findings from our
estimation analysis, we find that of the 328 cases observed to be mem-
bers of the low-rate offender group, only 5 are predicted to be a member

Table 2. Comparison of Predicted Versus Observed Trajectory Group

Membership

Predicted Trajectory Groupa

Sporadic Low-rate Classic Chronic
Offender Offender Desister Offender
Second half of the full sample
(n = 2,275)
Observed trajectory group
   Sporadic offender 1,598 2 5 0
   Low-rate offender 322 5 1 0
   Classic desister 238 6 14 0
   Chronic offender 68 2 12 2

Second half of the subsample

(n = 310)
Observed trajectory group
   Sporadic offender 200 1 2 1
   Low-rate offender 33 2 0 0
   Classic desister 54 0 3 0
   Chronic offender 13 0 0 1
Note: Bold figures represent the accurately predicted trajectory group membership.
a. Multinomial logistic model predicted trajectory group.
Bersani et al. 487

of that trajectory group. Consequently, we find that our prediction model

accurately predicts slightly less than 2 percent of the low-rate offender
cases. The results for the chronic offender group are very similar—only
2 percent (n = 2) of the cases predicted to be members of the chronic
offender group are indeed members of that trajectory group. The remain-
ing 98 percent of cases are predicted to belong to a less criminally active
group. Overall then, when we disaggregate our data by trajectory group
and examine the predictive ability of our model, our results illustrate
poor predictive power. That is, excluding the sporadic offenders, our
model accurately predicts less than 10 percent of the low-rate offender,
classic desister, and chronic offender cases combined.11

Concluding Remarks
This article contributes to the empirical literature on prediction by ana-
lyzing long-term patterns of offending (ages 18 to 55) for a large sample
of convicted men and women in the Netherlands. Data collected by the
Netherlands Institute for the Study of Crime and Law Enforcement offer
a unique opportunity to conduct a partial replication and extension of
the Sampson and Laub (2003) study. Two important findings stand out
in our analysis. First, regardless of the type of crime or the specific risk
predictor assessed, our findings indicate that although magnitude differ-
ences are readily apparent, criminal offending trajectories derived from
the CCLS follow a general path that declines over time for all convicted
offenders. These results lend support to Sampson and Laub’s argument
for a general process of desistance from crime. Second, despite the use
of data covering a large portion of the life course, containing a large
sample of serious offenders, and the application of an advanced statisti-
cal technique, our results are consistent with previous findings regarding
the difficulty of predicting high-rate offenders prospectively (see e.g.,
Auerhahn 1999; Gottfredson and Gottfredson 1994; Laub and Sampson
2003; Sampson and Laub 2003).
We acknowledge two important limitations of the current study.
First, due to data limitations we are only able to examine four adoles-
cent risk factors. Moreover, the indicators of low intelligence and
psychological instability are not optimal as they were gathered from
judgments made by a military examination committee, which may be
influenced by bias. It is important to note though that the patterns for
low intelligence and psychological instability match those found for
the early onset and chronic offending analyses. Second, our analysis
488 Journal of Research in Crime and Delinquency 46(4)

relies on official records of offending, which suffer from a number of

important limitations but are recognized as a valid indicator of more
serious crimes (Blumstein et al. 1986; Gove, Hughes, and Geerken
1985; Hindelang, Hirschi, and Weis 1979). To assess the robustness
of our findings and identify inconsistencies that may have resulted
from bias in the official data, we analyzed two conviction outcomes:
any criminal conviction and violent criminal conviction. We believe
the findings are noteworthy for their overall similarities across crime
type.
The findings presented here have modest implications for extant
criminological theory, future empirical research, and public policy. Our
results lend support to the position that it is difficult to predict long-term
patterns of criminal offending using risk factors identified in adoles-
cence. Moreover, the fact that our data come from the Netherlands
suggests that the results generated by Sampson and Laub (2003) are not
unique to the United States or, for that matter, unique to Boston boys
born in the late 1920s. As discussed earlier in this article, a number of
criminological theories have focused on the developmental course of
offending over time (e.g., Blumstein et al., 1986; Gottfredson and
Hirschi 1990; Moffitt 1993). Moreover, these theories focus extensive
attention on risk factors in childhood and adolescence. Our findings add
a note of caution to theories that maintain that long-term offending pat-
terns can be predicted using adolescent risk factors identified in advance.
With respect to future empirical research, while it is the case that much
current research focuses on identifying risk factors in childhood and
adolescence, we believe our results suggest expanding the research
focus to include an examination of factors and mechanisms in adoles-
cence and adulthood that may redirect offending trajectories toward
desistance from crime. This line of research is especially necessary
among samples that have been identified as having a high risk of offend-
ing throughout their life course. We believe research that can identify
the mechanisms that generate both behavioral stability and change
within individuals will move the field forward.
Finally, our findings question public policies that rest on our ability
to predict criminal behavior over the long haul. Although this idea has
enormous appeal, the empirical reality suggests that prediction is at best
problematic. Moreover, we are skeptical that better measures of risk factors
or new statistical techniques will lead to substantial improvement in our
capability to predict long-term behavior in advance. As demonstrated
Bersani et al. 489

throughout the history of criminology, our capacity to predict and distin-

guish a small group of high-rate offenders remains elusive.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interests with respect to the author-
ship and/or publication of this article.

Funding
The author(s) received no financial support for the research and/or authorship of this
article.

Notes
  1. We note that our study is not a conventional illustration of a comparative cross-
national study (see Farrington and Wikström 1994). However, we undertake a
replication of a prediction study by Sampson and Laub (2003) using data from
the Netherlands and compare the findings within an international framework.
Although somewhat limited, this form of a quasi–cross-national comparative
work is not new in criminology (see Farrington 1999).
  2. For more information on the full Criminal Career and Life-Course Study
(CCLS) sample, we refer readers to the CCLS codebook (Nieuwbeerta and
Blokland 2003) and earlier publications based on this data set (Blokland,
Nagin, and Nieuwbeerta 2005; Blokland and Nieuwbeerta 2005).
  3. The selection of the 689 men was carried out as follows. First, because the
archive of the Ministry of Defense does not hold detailed records for per-
sons born prior to 1940, the sample was restricted to men born after 1940
(N = 3,515). Second, due to resource limitations, data collection efforts were
restricted to gathering information for one-fifth of the cases in the total sample
(N = 736). These men were randomly selected from the full sample. Finally,
information for 47 of the men in the randomly selected subset could not be
traced. In sum, files containing full medical and military records are available
for the remaining group of 689 men. The final group is representative of men
convicted in 1977 between the ages of 12 and 37.
  4. While the General Documentation Files contain information on all offenses
that lead to any type of judicial interference, here we use only information
on those offenses that were either followed by a conviction or a prosecutorial
disposition due to policy reasons, thereby excluding cases that resulted in an
acquittal or a prosecutorial disposition due to insufficient evidence. The results
of analyses conducted with these cases in the data are substantively similar.
  5. Given its prevalence in 1977, the sample for driving under the influence was
confined to 2 percent. Less common, serious offenses were oversampled, including:
490 Journal of Research in Crime and Delinquency 46(4)

25 percent of all robbery, public violence, and battery cases; 100 percent of all
cases involving murder (including attempts), offenses against decency, rape,
child molesting, and other sexual assaults; and 17 percent of all drug offenses.
In addition, because the sample was one of cases not people, offenders who had
two or more adjudications in 1977 were more likely to be included in the study.
To deal with this sampling strategy we include a weight in all analyses so that
the weighted sample is representative of the distribution of offense types and
individuals tried in 1977.
  6. The CCLS data is a conviction cohort consisting of persons who were tried
in 1977. Therefore, unlike birth cohort studies, the age range in the sample
is broad and skewed, ranging from 12 to 79 with a peak at age 18. This has
two implications. First, the convictions recorded for the sample cover a long
period—from 1924 to 2002 (when data collection concluded). Second, indi-
viduals were not randomly sampled from the entire population; they were all
criminally active in 1977. To examine the extent to which historical changes
and sampling properties affected our results, we divided the sample into three
cohorts based on the individuals’ age in 1977. The first cohort was comprised
of offenders aged between 12 and 21 in 1977, the second of those aged 22 to
31, and the third of those aged 32 and up. Because the substantive findings
reported in the following remain the same, we do not report the results by
cohort here.
  7. Negative binomial regression models using a measure of the frequency of con-
victions were also analyzed. The results mirror those obtained using the logistic
regression models. We also graphed the raw data trajectories and found that the
substantive findings did not change. These results are available upon request.
Due to space considerations, we present only the smoothed trajectories based
on the logistic regression models.
  8. The slight up-tick in the mean predicted probability of conviction at older ages
does not represent a trend in the observed data but is an artifact of the model.
  9. Similar to our full sample analysis, we repeat the analysis with two alternative
risk factor conceptualizations capturing those with an early onset of offending
and a low intelligence or psychologically unstable designation. Of the sub-
sample, 7 percent meet this criterion. We find that these results remain substan-
tively the same.
10. These estimated parameters of the prediction model are available upon request.
11. The trajectory group and cross-validation analyses were also conducted using
violent criminal convictions as the dependent variable. We found a three-group
model best fits for trajectories of violent convictions. The results of the predic-
tion analysis are substantively similar to those reported using total criminal
convictions as the dependent variable. Specifically, excluding the sporadic
Bersani et al. 491

offender group, our model accurately predicts less than 7 percent of the cases
(results available upon request).

References
Adler, Freda. 1996. “Our American Society of Criminology, the World and the State
of Art.” Criminology 34:1–9.
Auerhahn, Kathleen. 1999. “Selective Incapacitation and the Problem of Prediction.”
Criminology 37:703–34.
Auerhahn, Kathleen. 2006. “Conceptual and Methodological Issues in the Prediction
of Dangerous Behavior.” Criminology & Public Policy 5:771–78.
Bernard, Thomas J. and R. Richard Ritti. 1991. “The Philadelphia Birth Cohort
and Selective Incapacitation.” Journal of Research in Crime and Delinquency
28:33–54.
Blokland, Arjan A. J., Daniel Nagin, and Paul Nieuwbeerta. 2005. “Life Span
Offending Trajectories of a Dutch Conviction Cohort.” Criminology 43:919–54.
Blokland, Arjan A. J. and Paul Nieuwbeerta. 2005. “The Effects of Life Circums-
tances on Longitudinal Trajectories of Offending.” Criminology 43:1203–40.
Blumstein, Alfred, Jacqueline Cohen, and David P. Farrington. 1988. “Criminal
Career Research: Its Value for Criminology.” Criminology 26:1–35.
Blumstein, Alfred, Jacqueline Cohen, Jeffrey Roth, and Christy Visher. 1986. Criminal
Careers and Career Criminals. Washington, DC: National Academy Press.
Blumstein, Alfred, David P. Farrington, and Soumyo Moitra. 1985. “Delinquency
Careers: Innocents, Desisters, and Persisters.” Pp. 187–219 in Crime and Justice: An
Annual Review of Research, Vol. 6, edited by M. Tonry and N. Morris. Chicago:
University of Chicago Press.
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. Vol. 5, edited by M. Tonry and N. Morris. Chicago: University of
Chicago Press.
DeLisi, Matt. 2005. Career Criminals in Society. Thousand Oaks, CA: Sage.
Ezell, Michael E. and Lawrence E. Cohen. 2005. Desisting from Crime: Continuity
and Change in Long-term Crime Patterns of Serious Chronic Offenders. New
York: Oxford University Press.
Farrington, David P. 1999. “A Criminological Research Agenda for the Next
Millennium.” International Journal of Offender Therapy and Comparative
Criminology 43:154–67.
Farrington, David P. 2003. “Key Results from the First Forty Years of the Cambridge
Study in Delinquent Development.” Pp. 137–184 in Longitudinal Research in
the Social and Behavioral Sciences: An Interdisciplinary Series, edited by: T. P.
Thornberry and M. D. Krohn. New York: Kluwer Academic/Plenum Publishers.
492 Journal of Research in Crime and Delinquency 46(4)

Farrington, David P., Rolf Loeber, Delbert S. Elliott, J. David Hawkins, Denise Kandel,
Malcolm 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. Vol. 13, edited by B. B. Lahey and A. E.
Kazdin. New York: Plenum.
Farrington, David P. and Per-Olof H. Wikström. 1994. “Criminal Careers in London
and Stockholm: A Cross-National Comparative Study.” Pp. 65–89 in Cross-
National Longitudinal Research on Human Development and Criminal Behavior,
edited by E. G. M. Weitekamp and H-J. Kerner. Dordrecht, the Netherlands:
Kluwer Academic Publishers.
Fergusson, David M. and L. John Horwood. 2002. “Male and Female Offending
Trajectories.” Development and Psychopathology 14:159–77.
Ge, Xiaojia, M. Brent Donnellan, and Ernst Wenk. 2001. “The Development of Persistent
Criminal Offending in Males.” Criminal Justice and Behavior 28:731–55.
Gendreau, Paul. 2001. “We Must Do a Better Job of Cumulating Knowledge.”
Canadian Psychology 43:205–10.
Glueck, Sheldon. 1964. “Wanted: A Comparative Criminology.” Pp. 304–22 in
Ventures in Criminology, edited by S. Glueck and E. Glueck. Cambridge, MA:
Harvard University Press.
Glueck, Sheldon and Eleanor Glueck. 1950. Unraveling Juvenile Delinquency. New
York: The Commonwealth Fund.
Gottfredson, Don M. 1987. “Prediction and Classification in Criminal Justice
Decision Making.” Pp. 1–20 in Prediction and Classification: Criminal
Justice Decision Making, edited by D. M. Gottfredson and M. Tonry. Chicago:
University of Chicago Press.
Gottfredson, Michael R. and Travis Hirschi. 1990. A General Theory of Crime.
Stanford, CA: Stanford University Press.
Gottfredson, Stephen D. and Don M. Gottfredson. 1994. “Behavioral Prediction and
the Problem of Incapacitation.” Criminology 32:441–74.
Gottfredson, Stephen D. and Laura J. Moriarty. 2006. “Statistical Risk Assessment:
Old Problems and New Applications.” Crime & Delinquency 52:178–200.
Gove, Walter R., Michael Hughes, and Michael Geerken. 1985. “Are Uniform
Crime Reports a Valid Indicator of the Index Crimes? An Affirmative Answer
with Minor Qualifications.” Criminology 23:451–501.
Harcourt, Bernard E. 2006. Against Prediction: Profiling, Policing and Punishing in
an Actuarial Age. Chicago: University of Chicago Press.
Hindelang, Michael J., Travis Hirschi, and Joseph G. Weis. 1979. “Correlates of
Delinquency: The Illusion of Discrepancy between Self-report and Official
Measures.” American Sociological Review 44:995–1014.
Bersani et al. 493

LaFree, Gary. 2007. “Expanding Criminology’s Domain: The American Society of

Criminology 2006 Presidential Address.” Criminology 45:1–31.
Laub, John H. and Robert J. Sampson. 2003. Shared Beginnings, Divergent Lives:
Delinquent Boys to Age 70. Cambridge, MA: Harvard University Press.
LeBlanc, Marc and Rolf Loeber. 1998. “Developmental Criminology Updated.” Pp.
115–198 in Crime and Justice: A Review of Research, Vol. 23, edited by M.
Tonry and N. Morris. Chicago: University of Chicago Press.
Loeber, Rolf and Marc LeBlanc. 1990. “Toward a Developmental Criminology.”
Pp. 375–473 in Crime and Justice. Vol. 12, edited by M. Tonry and N. Morris.
Chicago: University of Chicago Press.
Lowenkamp, Christopher T., Francis T. Cullen, and Travis C. Pratt. 2003.
“Replicating Sampson and Grove’s Test of Social Disorganization Theory:
Revisiting a Criminological Classic.” Journal of Research in Crime and
Delinquency 40:351–73.
Moffitt, Terrie E. 1993. “Adolescent-limited and Life-course-persistent Antisocial
Behavior: A Developmental Taxonomy.” Psychological Review 100:674–701.
Moffitt, Terrie E. 1994. “Natural Histories of Delinquency.” Pp. 3–66 in Cross-
national Longitudinal Research on Human Development and Criminal Behavior,
edited by E. G. M. Weitekamp and H-J. Kerner. Dordrecht, the Netherlands:
Kluwer Academic Publishers.
Moffitt, Terrie E. 2006. “Life-course-persistent Versus Adolescent-limited Antisocial
Behavior.” Pp. 570–598 in Developmental Psychopathology, 2nd ed., edited by
D. Cicchetti and D. Cohen. New York: John Wiley.
Moffitt, Terrie E., Donald R. Lynam, and Phil A. Silva. 1994. “Neuropsychological
Tests Predict Persistent Male Delinquency.” Criminology 32:101–24.
Nagin, Daniel S. 2005. Group-based Modeling of Development. Cambridge, MA:
Harvard University Press.
Nagin, Daniel and Kenneth Land. 1993. “Age, Criminal Careers, and Population
Heterogeneity: Specification and Estimation of a Nonparametric, Mixed Poisson
Model.” Criminology 31:327–62.
Nieuwbeerta, Paul and Arjan A.J. Blokland. 2003. Criminal Careers and Adult
Dutch Offenders (Codebook and Documentation). Leiden, the Netherlands:
NSCR.
Patterson, Gerald R. and Karen Yoerger. 1993. “Developmental Models for
Delinquent Behavior.” Pp. 140–172 in Mental Disorder and Crime, edited by S.
Hodgins. Newbury Park, CA: Sage.
Piquero, Alex R., Robert Brame, and Donald Lynam. 2004. “Studying Criminal
Career Length through Early Adulthood among Serious Offenders.” Crime &
Delinquency 50:412–35.
494 Journal of Research in Crime and Delinquency 46(4)

Piquero, Alex R., David P. Farrington, and Alfred Blumstein. 2007. Key Issues in
Criminal Career Research: New Analyses of the Cambridge Study in Delinquent
Development. New York: Cambridge University Press.
Sampson, Robert J. and John H. Laub. 2003. “Life-course Desisters? Trajectories of
Crime among Delinquent Boys Followed to Age 70.” Criminology 41:301–40.
Sampson, Robert J. and John H. Laub. 2005. “A Life-course View of the Development
of Crime.” Annals of the American Academy of Political and Social Sciences
602:12–45.
Stattin, Håkan and Ingrid Klackenberg-Larsson. 1993. “Early Language and
Intelligence Development and Their Relationship to Future Criminal Behavior.”
Journal of Abnormal Psychology 102:369–78.
Thornberry, Terence P. and Marvin D. Krohn. 2003. Taking Stock of Delinquency:
An Overview of Findings from Contemporary Longitudinal Studies. Longitudinal
Research in the Social and Behavioral Sciences: An Interdisciplinary Series.
New York: Kluwer Academic/Plenum Publishers.
Tonry, Michael. 1987. “Prediction and Classification: Legal and Ethical Issues.” Pp.
367–413 in Prediction and Classification: Criminal Justice Decision Making,
edited by D. M. Gottfredson and M. Tonry. Chicago: University of Chicago
Press.
Wolfgang, Marvin, Robert Figlio, and Thorsten Sellin. 1972. Delinquency in a Birth
Cohort. Chicago: University of Chicago.

Bios
Bianca E. Bersani is a doctoral candidate in the Department of Criminology and
Criminal Justice at the University of Maryland, College Park.

Paul Nieuwbeerta was a senior researcher at the Netherlands Institute for the Study
of Crime and Law Enforcement (NSCR) when writing this article. Currently he is a
professor in the Department of Criminology at Leiden University and in the
Department of Sociology at Utrecht University in Utrecht.

John H. Laub is distinguished university professor in the Department of

Criminology and Criminal Justice at the University of Maryland, College Park.

View publication stats