You are on page 1of 10

Journal of Abnormal Psychology Copyright 2001 by the American Psychological Association, Inc.

2001, Vol. 110, No. 4, 600-609 0021-843X/01/$5.00 DOT: 10.1037//0021-843X.110.4.600

Adolescence-Limited Versus Persistent Delinquency:


Extending Moffitt's Hypothesis Into Adulthood

Helene Raskin White, Marsha E. Bates, and Steven Buyske


Rutgers, The State University of New Jersey

The authors examined how neuropsychological, personality, and environmental risk factors and their
interactions were related to trajectories of delinquent behavior from adolescence to adulthood. Four
waves of longitudinal data from 698 male participants, ages 12-18 at Time 1 and ages 25-31 at Time 4,
were included in the analyses. Using a growth mixture model approach, 4 trajectories were identified:
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

nondelinquents, adolescence-limited delinquents, adolescence-to-adulthood-persistent delinquents, and


This document is copyrighted by the American Psychological Association or one of its allied publishers.

escalating delinquents. Five risk factors distinguished escalating from persistent delinquents and 5 also
distinguished nondelinquents from the 3 delinquency trajectories. Persistent delinquents scored signifi-
cantly higher than adolescence-limited delinquents on only one risk factor, disinhibition. Overall, few of
the factors that are related to childhood-to-adolescence persistence were associated with persistence in
delinquency beyond adolescence.

Some criminologists have argued that criminal propensity is deficits, especially in verbal skills and executive functions, bio-
attributable to stable individual differences established early in life logically based personality factors (such as an impulsive temper-
(Gottfredson & Hirschi, 1990). This view disregards the possibility ament), and environmental adversity. Adolescence-limited delin-
that trajectories may have different etiologies, change due to life quents, on the other hand, were hypothesized to engage in
circumstances, and be altered by current behavior (see Nagin & antisocial behaviors as a social group phenomenon.
Farrington, 1992; Laub & Sampson, 1993; Pasternoster & Brame, Knowledge of preadolescent behaviors has figured prominently
1997). In contrast, other criminologists have argued that there are in differentiating persistence of antisocial behavior from childhood
distinct groups of deviant individuals and that the causes of crime to adolescence, in part because adolescence-limited and persistent
differ over the life course (Blumstein, Cohen, & Farrington, 1988). delinquents purportedly have the same frequency and severity of
The preponderance of research on delinquency has supported this delinquency during adolescence (Moffitt, 1993a). The hypothe-
latter view and suggests that adolescent delinquents may represent sized neuropsychological, personality, and environmental risk
several types rather than a single entity (Bartusch, Jeglum, Lynam, factors have been useful in distinguishing early onset from esca-
Moffltt, & Silva, 1997; Loeber, 1988; Moffitt, 1993a; Nagin & lating delinquency and childhood-to-adolescence-persistent from
Land, 1993; Patterson & Yoerger, 1993; Simons, Wu, Conger, & adolescence-limited delinquency (Bartusch et al., 1997; Moffitt,
Lorenz, 1994). 1993a; Moffitt, Lyman, & Silva, 1994; Patterson & Yoerger, 1993;
Moffitt (1993a) has proposed that, for some individuals, antiso- Simons et al., 1994; Tibbetts & Piquero, 1999; Tremblay, Pihl,
cial behavior is stable and persistent, whereas for others it is Vitaro, & Dobkin, 1994; J. White, Moffitt, Earls, Robins, & Silva,
temporary. Her taxonomy distinguished two types of delinquents: 1990). Yet, little is currently known about the utility of this
adolescence-limited and life-course-persistent. Moffitt hypothe- typology for differentiating adolescence-limited from adolescence-
sized that life-course-persistent delinquents could be distinguished to-adulthood-persistent delinquency. Although studies have found
from adolescence-limited ones in terms of neuropsychological that many persistent offenders are also early-onset delinquents, the
correlation is not perfect (e.g., Farrington & Hawkins, 1991; Smith
& Brame, 1994). Moffitt and colleagues (Moffitt, Caspi, Dickson,
Silva, & Stanton, 1996) found that nearly half of childhood-onset
Helene Raskin White and Marsha E. Bates, Center of Alcohol Studies,
Rutgers, The State University of New Jersey; Steven Buyske, Department
delinquents were not seriously delinquent at age 18 years. The
of Statistics, Rutgers, The State University of New Jersey. aims of this study were as follows: (a) to extend the study of
{"reparation of this article was supported in part by grants from the individual differences in the persistence of delinquent behaviors by
National Institute on Drug Abuse (DA/AA-03395), the National Institute examining longitudinal patterns of antisocial behaviors from ado-
on Alcohol Abuse and Alcoholism (AA-11594), and the Rutgers Univer- lescence into adulthood, and (b) to examine how purported risk
sity Research Council. An earlier version of this article was presented at the factors were related to the persistence of antisocial behaviors
meeting of the American Society of Criminology, November 1997, San across the adolescence-to-adulthood transition.
Diego, California.
We are grateful to David Farrington, Cathy Widom, and Scott Lilienfeld
for their comments on an earlier draft of this article. Differentiating Limited and Persistent
Correspondence concerning this article should be addressed to Helene Trajectories of Delinquency
Raskin White, Center of Alcohol Studies, Rutgers, The State University of
New Jersey, 607 Allison Road, Piscataway, New Jersey 08854-8001. According to Moffitt (1993a), most adolescent delinquent be-
Electronic mail may be sent to hewhite@rci.rutgers.edu. haviors arise from peer and social environment factors, not from a
fion
ADOLESCENCE-LIMITED VERSUS PERSISTENT DELINQUENCY 601

high-risk developmental trajectory (see also Moffitt et al., 1996). Although neuropsychological functioning and personality char-
However, for a minority group of life-course-persistent delin- acteristics have been found to differentiate early-onset from late-
quents (5%-10%), antisocial behavior is hypothesized to have onset delinquents, few studies have examined their ability to
pathological roots. Specifically, neurological damage is proposed differentiate those individuals who persist in delinquency beyond
to disrupt early development by fostering neuropsychological adolescence from those who do not. Nagin, Farrington, and Moffitt
problems and a difficult and undercontrolled temperament, which (1995) found no evidence that neurocognitive and personality
in turn increase vulnerability to negative environmental influences. factors—including attention problems, low 1Q, reading failures,
and impulsivity—could differentiate adolescence-limited from
chronic offenders. On the other hand, Donnellan, Ge, and Wenk
Neuropsychological Dysfunction
(2000) found that some cognitive abilities were able to discrimi-
The sources of early neuropsychological dysfunction are hy- nate White and Hispanic but not African American persistent
pothesized to include disruption of brain development in utero, delinquents from adolescence-limited delinquents. However, in
neonatal deprivation, birth complications, and genetic influences both of these studies, persistence groups were based on official
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

(Moffitt, 1993a). There is some evidence that proxy measures of records of arrests or criminal convictions. Thus, some subjects
This document is copyrighted by the American Psychological Association or one of its allied publishers.

neurocognitive deficits, such as low birth weight and low scores on were assigned to the adolescence-limited group even though they
achievement tests, are related to age of delinquency onset (Tibbetts may have continued to engage in undetected criminal behavior in
& Piquero, 1999). Deficits in verbal abilities and executive func- adulthood. Additional research separating those who continued
tions consistently have been related to antisocial behavior (Gian- criminal behavior from those who stopped, regardless of convic-
cola, 1995; Giancola, Moss, Martin, Kirisci & Tarter, 1996; Ly- tion status, might reveal differences that were masked by these
nam, Moffitt, & Stouthamer-Loeber, 1993; Moffitt, 1993a; Seguin, group definitions.
Pihl, Harden, Tremblay, & Boulerice, 1995).
Deficits in verbal skills that may promote delinquency include Environmental Adversity
the failure to use verbal mediation for self-control, a present-
oriented cognitive style that does not promote delay of gratifica- According to Moffitt (1993a), the co-occurrence of a neurolog-
tion, poor comprehension, and poor communication skills evoking ically impaired infant in an adverse rearing environment sets the
negative interactions (Moffitt, 1993b). The term executive cogni- stage for life-course-persistent deviance. She suggested that the
tive functions is often used to encompass the interrelated operation impact of neuropsychological impairment would be exacerbated in
and control of attention, abstract reasoning, cognitive flexibility, disadvantaged homes because impaired children create a real chal-
strategic planning, use of environmental feedback, working mem- lenge for parents, and inadequate parents would not be able to
ory, goal selection, and response inhibition (Benton, 1994; Duffy compensate for these problems. Family adversity has been defined
& Campbell, 1994; Luria, 1973). Executive impairment is thought to include such familial environment factors as low parental edu-
to promote undercontrolled behavior that is disinhibited and ag- cation and income, single-parent status, and measures of a dis-
gressive (Giancola, 1995; Hoaken, Giancola, & Pihl, 1998; Mof- rupted family social environment (e.g., Moffitt & Silva, 1988).
fitt, 1990, 1993a, 1993b; Seguin et al., 1995). Impairment may
disrupt interpersonal relations and social problem solving by lim- Present Study
iting one's ability to generate alternative solutions to problems,
inhibit aggressive responses, appropriately respond to environmen- The present study extended previous research by examining
tal cues, and appreciate behavioral consequences (see Loeber & trajectories of delinquent behaviors from early adolescence into
Hay, 1997). adulthood. On the basis of previous studies of childhood-to-
adolescence persistence, we hypothesized that three different tra-
jectories would be identified: nondelinquents, adolescence-limited
Personality delinquents, and adolescence-to-adulthood-persistent delinquents.
A substantial literature supports the idea that the risk for per- We also hypothesized that adolescence-limited and adolescence-
sistence of antisocial behavior is compounded by temperamental to-adulthood-persistent delinquents would differ on selected mea-
characteristics (Moffitt, 1993a). These characteristics may be sec- sures of neuropsychological functioning, personality risk, and en-
ondary to neuropsychological deficits, come about through focal vironmental risk. As described above, Moffitt (1993b) postulated
neurological injury, or have other etiological roots. As infants and that neurocognitive dysfunction would moderate the association of
young children, those with difficult temperaments are described as other intrapersonal and interpersonal factors to delinquency. Al-
overactive and irritable. In adolescence, such youths tend to re- ternatively, there is also support for direct associations between
main impulsive and disinhibited. The personality characteristics of these risk factors and antisocial behavior (e.g., Patterson &
impulsivity and disinhibition have been linked to delinquency in Yoerger, 1993; Pennington & Bennetto, 1993). We therefore ex-
many studies (Block, 1995; Tremblay et al., 1994; H. White, amined both main effect and interaction models.
Labouvie, & Bates, 1985; J. White et al., 1994). Deficits in
impulse control interfere with one's ability to inhibit behavior and Method
are often accompanied by tactlessness and a lack of appreciation
for the consequences of one's behavior. Negative emotionality Design and Sample
(e.g., alienation) and low constraint (e.g., low traditionalism, harm Data were collected as part of the Rutgers Health and Human Develop-
avoidance, and control) have also been linked to delinquency (e.g., ment Project (HHDP), a prospective cohort-sequential, longitudinal study
Caspi et al., 1994). of adolescent development. Adolescents were originally identified by a
602 WHITE, BATES, AND BUYSKE

random telephone survey in New Jersey (NJ). Following the telephone survey, the self-report data, questionnaires were administered individually by a
field staff interviewed interested participants and their parents in their homes. trained interviewer assigned to a participant for the length of the testing
Subsequently, participants came to the test site for a full day of testing. day. Participants were instructed not to put their names on any question-
Families were paid for their participation, and informed written consent was naire, were advised of the protections granted by the Certificate of Confi-
obtained from both the participants and their legal guardians. (For greater dentiality obtained from the National Institutes of Health, and were repeat-
detail on initial participant recruitment, see Horwitz & White, 1987.) edly assured of the confidentiality of all data, especially with regard to
The HHDP samples included 698 male and 682 female participants who parents, teachers, and public authorities.
were tested initially between 1979 and 1981 (Time 1, Tl). Three cohorts Delinquency. At each test occasion, we measured the number of times
were tested: The youngest cohort was age 12 at Tl, the middle cohort was (5-point scale, from never to more than 10 times) that participants had
age 15 at Tl, and the oldest cohort was age 18 at Tl. These participants engaged in the following behaviors in the last 3 years: petty theft, grand
returned for retest 3 years later in 1982-1984 (Time 2, T2), again in theft, motor vehicle theft, breaking and entering, avoiding payment, van-
1985-1987 (Time 3, T3), and finally in 1992-1994 (Time 4, T4). A total dalism, assault, and armed robbery. These frequency scores were summed
of 91% (n = 1,257) of the original sample returned at T4._The original to form an overall scale of delinquent behavior at each test occasion.2 The
sample was predominantly White (90%; compared with the NJ population, level of delinquency reported by our participants at Tl was similar to the
which was 83% White at the time; U.S. Bureau of the Census, 1981). Half
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

amount found in other surveys of predominantly middle-class adolescents


This document is copyrighted by the American Psychological Association or one of its allied publishers.

of the participants were Catholic, 30% were Protestant, 9% were Jewish, collected at the same time (e.g., Linden, 1978). (For rates, see Pandina &
and the final 11% were another or no religion, analogous to the religious White, 1984.) A scale of aggressive behavior was also constructed at each
composition of NJ. The median family income of participants' parents at test occasion, which was the sum of the frequencies for armed robbery,
Time 1, between $20,000 and $29,000, was also comparable to that of the assault, and fighting with weapons or in a gang.3
entire state at that time (U.S. Bureau of the Census, 1981). Participants Risk factors. A proxy measure of early neuropsychological problems
were comparable to those who refused at Tl in demographic characteristics (labeled birth risk; BIRISK) was developed on the basis of two birth risk
and selected behaviors sampled during the original telephone survey, factors that have been related to neuropsychological disturbance: low birth
except that participants displayed slightly higher parental occupation and weight (<6 Ibs. [<2.72 kg]) and premature birth (Moffitt, 1993a; Tibbetts &
education. Overall, the sample of participants is most representative of Piquero, 1999; see Footnote 4). These two variables were coded 1 (true) and 0
White adolescents living in a metropolitan, working-class and middle-class (not true) and summed to form a Birth Risk scale (alpha reliability = .72).
environment. (For more extensive details of this study, see Pandina, Labou- In addition to the proxy measure of early neuropsychological dysfunc-
vie, & White, 1984.) tion, three proximal measures of neuropsychological functioning were
For the present study, we included only male participants because, in assessed at T3 in the areas of verbal abilities and executive function-
general population samples, life-course-persistent delinquency is limited ing. Unlike the personality and environmental measures (see below), these
almost exclusively to men (Moffitt, 1993a) and because, in our sample, the measures were not available until T3. Although this is not ideal, studies
number of female participants with serious delinquency was very small. All have shown that levels of neuropsychological functioning are relatively
698 male participants in our sample were included in the analysis. All
available observations on delinquency were used, with an average of 3.75
observations for each participant over the four waves. For 75% of the 1
The data set did not contain independent measures of delinquency (such as
participants (n = 523), complete covariate information was available; for
arrest records); therefore, the analyses were based on self-report data for both
the remaining 25%, partial covariate information was available. To take
the delinquency and the risk factors. Individuals' response styles and person-
advantage of the full available data, we used multiple imputation. Specif-
ality factors can affect their responses to self-report items, which in turn can
ically, we used PROC MI in SAS, with options set to use multivariate
create measurement error. This problem may be most serious for variables that
normal Markov Chain Monte Carlo, to impute 10 data sets with full
rely on a participant's perception, such as the measure of parental hostility. For
covariate information (SAS, 2001). All analyses described below were first
example, delinquents may be more negative in their attitudes about everything,
done separately on each imputed data set and then combined by standard
including their parents' treatment of them. Therefore, the results should be
statistical methods designed to incorporate the additional uncertainty due to
evaluated in light of possible measurement limitations.
the imputed data. This multiple imputation takes full advantage of all 2
available data and avoids the bias that may result from using only the Several operationalizations of life-course-persistent delinquency have in-
complete cases (Schafer, 1997). cluded early onset of delinquency as a defining characteristic (e.g., Tibbetts &
Piquero, 1999; J. White et al., 1990). Because the HHDP began as a study of
the development of drug use behaviors, no data were collected on age of onset
Measures for delinquency, and thus early onset could not be included in the definition
scale used in the present analysis. Only eight delinquency items were included
The data came from self-report questionnaires. Self-reports are generally in the delinquency scale in order to create a scale that was exactly the same
accepted as reliable and valid indicators of delinquent behavior (Hindelang, across all four test occasions. This number is fewer than the number of items
Hirschi, & Weis, 1981) if conducted under nonthreatening circumstances typically used in general delinquency scales (e.g., Elliott et al., 1989). Never-
(Dembo, Williams, Wish, & Schmeidler, 1990). Elliott, Huizinga, and theless, the scale covered a broad range of acts in terms of type (i.e., property
Menard (1989) suggested that self-reports provide a more direct, sensitive, offenses and offenses against persons) and severity of delinquency.
3
and complete measure of various forms of deviant behavior than do At T4, using a weapon in a fight was replaced by gang fighting in the
measures based on official law enforcement and institutional records. aggression scale due to changes in the questionnaire items. However, given
There are also limitations involved in using self-report data, such as that the aggression scale was examined separately at each age and not used
respondents' inability to remember past events, misunderstanding the ques- as a repeated measure to develop trajectories over time, it is unlikely that
tions, and efforts to conceal or exaggerate (Chaiken & Chaiken, 1990). this change was problematic.
4
When both the dependent and independent measures are assessed with These birth risk measures were collected from mothers retrospectively
self-reports, there is also a potential influence of shared method variance.1 when the child was between 12 and 18 years old. It could be argued that
Nevertheless, this tradition of using self-reports is fairly common in the retrospective recall may have made these measures unreliable. However,
delinquency literature (e.g., Elliott et al., 1989). (For further discussion of these two questions tap unusual circumstances that should be remembered
the advantages and disadvantages of self-report data, see Elliott et al., by most informants. Other studies have reported that retrospective recall of
1989, and Hindelang et al., 1981.) To maximize reliability and validity of such circumstances (e.g., birth weight) is reliable (see Tibbetts & Piquero, 1999).
ADOLESCENCE-LIMITED VERSUS PERSISTENT DELINQUENCY 603

stable from childhood to adolescence (Moffitt, 1990) and from adolescence Table 1
into young adulthood (Bates & Labouvie, 1998). Confirmatory factor Measures
analyses of the structure and stability of our neuropsychological battery test
scores over 7 years from T3 to T4 demonstrated very high stability of latent Measure Tl T2 T3 T4
factors (>.90; Bates & Labouvie, 1998), and the specific manifest scores
used in the present study also showed high stability from T3 to T4 (>.70). Delinquency scale X X X X
Verbal ability (VERB) was measured by the Shipley Institute of Living Aggression scale X X X X
Birth risk X
Scale (SILS) Vocabulary Test (Zachary, 1986). It consists of a 40-item
Verbal ability X
multiple-choice test and provides a time-saving (10 min) yet valid alter- BCT X
native to more extensive verbal IQ assessments (Parker, Birnbaum, Boyd, Trail Making Test X
& Noble, 1980; Paulson & Lin, 1970). Executive function was measured Impulsivity X
by two separate tests: the Trail Making Test—Part B (TRAIL) and the Harm avoidance X
Booklet Category Test (BCT). TRAIL (Reitan & Wolfson, 1985) involves Disinhibition" X X
drawing a continuous line to connect circled numbers and letters in an SES X
alternating sequence. Response time is the dependent variable. Longer Family structure X
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

response times signal impairment in complex conceptual tracking skills, Parental hostility X
This document is copyrighted by the American Psychological Association or one of its allied publishers.

reduced ability to follow more than one mental sequence at a time, and
Note. An X means that the assessment was made at that time. Tl = Time
inflexibility in shifting cognitive-behavioral activities (Escon, Yen, &
1; T2 = Time 2; T3 = Time 3; T4 = Time 4. BCT = Booklet Category
Bourke, 1978; Gaudino, Geisler, & Squires, 1995). The BCT (DeFillippis Test; SES = socioeconomic status.
& McCampbell, 1991) requires developing and elaborating abstract con- a
Disinhibition was measured at Tl for the middle and oldest cohorts and
cepts that relate subsets of multidimensional visual stimuli. Increasing at T2 for the youngest cohort.
errors are indicative of a deficit in the ability to develop (with feedback)
conceptual referents or rules to use in solving novel, unfamiliar problems,
that is, shirting cognitive sets and adapting. A shortened version (Russell &
Levy, 1987) was used, which correlates highly with the full BCT and has Nagin, 1999). This growth mixture model method allows for cross-group
been found to be as accurate as full scores in discriminating brain-damaged differences in the shape of developmental trajectories and is, therefore,
from normal participants (Russell & Barren, 1989). Higher scores indicate especially suited for identifying, rather than assuming, heterogeneity in
higher neuropsychological risk. types of developmental trajectories (Nagin & Tremblay, 1999). The ap-
Two of the personality measures were the Impulsivity and Harm Avoid- proach is based on the assumption that the population is composed of a
ance subscales from a shortened version of Jackson's (1968) Personality mixture of distinct groups defined by their developmental trajectories.
Research Form E (PRF E). Higher scores indicate higher impulsivity (IMP; Thus, this approach can test whether theoretically predicted developmental
M = 5.35, SD = 2.32) and lower harm avoidance (HARM; M = -5.65, trajectories are actually present in the population.
SD = 2.79). The PRF E provides a broadly defined measure of personality To take advantage of the cohort sequential design, age scores for
structure, and high levels of internal consistency (.94 for HARM; .92 for delinquency were used to model the trajectories. There were seven ages at
IMP) and test-retest reliability (.90 for HARM; .81 for IMP) have been which delinquency was measured (i.e., ages 12, 15, 18, 21, 24.5, 28, and
reported (Jackson, 1968). Sensation seeking was measured by the Disin- 31). The number of participants included for each delinquency measure-
hibition subscale (M = 4.37, SD = 2.0) from Zuckerman's (1979) Sensa- ment varied depending on age. For example, delinquency at age 12 was
tion Seeking Scale. Higher scores indicate higher disinhibition. This scale assessed in the youngest cohort (at Tl), whereas two cohorts supplied
has been demonstrated to have high test-retest reliability (.91) and internal information for delinquency at age 15 (the middle cohort at Tl and the
consistency (>.70). All of the personality variables were assessed at Tl youngest cohort at T2). The oldest cohort was 24 at T3 and the youngest
with the exception of disinhibition. Disinhibition was assessed at Tl for the cohort was 25 at T4; their scores were combined to provide a measure of
middle and oldest cohorts; however, it was assessed at T2 for the youngest delinquency at age 24.5.
cohort because of the lack of demonstrated reliability prior to age 14 A custom-written Splus library named mmlcr (mixed-mode latent class
(Zuckerman, 1979). In this sample, the alpha reliability coefficient was .64 regression), run on the Splus statistical programming environment (Math-
at Tl and .66 at T2. soft, 1999), was used to test the models (available at StatLib, http://
Three measures of family adversity were included. Family socioeco- lib.stat.cmu.edu). An important step in the analysis is to determine the
nomic status (SES) at Tl was the product of highest parental educational optimal number of groups that best describes the data. This is accomplished
level (1 = less than high school, 1 = postgraduate) and highest parental by repeating the analyses and adding an additional group each time.
occupational level (1 = unemployed, 10 = professional; alpha reliability = Although there is no definitive statistic for determining the optimal number
.66). The final SES scale was divided into quintiles representing five levels of groups, most experts suggest that the Bayes information criterion (BIC)
of SES, with a higher score indicating lower SES (M = 3.26, SD = 1.46). should be used to determine the optimal number of trajectory groups when
A Tl family structure (FAMST) variable was coded 1 for a single-parent testing mixture models (D'Unger, Land, McCall, & Nagin, 1998; Keribin,
family and 0 for a two-parent family. We also included a measure of 2000). The model with the smallest absolute BIC is chosen. The BIC
parental hostility (HOST; M = 47.30, SD = 12.67) at Tl from the Streit criterion rewards parsimony and so tends to favor fewer groups, but it is
(1978) Youth Perception Inventory. This was a 17-item (e.g., how often known to be consistent (Keribin, 2000). A difference of more than 10 in the
parents forget to help you when you need it, act as though you are in the BIC is considered very strong evidence against the model with the higher
way, make you feel you are not loved) scale with an alpha reliability BIC (Raftery, 1995). For each possible trajectory group, the observed
coefficient of .87. Table 1 presents a list of all of the measures included in delinquency score was modeled with a Poisson model quadratic in time.
the analyses and when they were assessed. For each class, this model was fit using quasi-likelihood, ensuring the
consistency of the estimates within each class even if the model is mis-
specified (Wedderburn, 1974). Using the methods recommended in
Analyses Bandeen-Roche, Miglioretti, Zeger, and Rathouz (1997) for checking
model adequacy in latent class models, we found this model to give a
A semiparametric, group-based modeling technique was used to develop satisfactory fit to the data. (The model-checking analyses are available on
trajectories of delinquency (Muthen & Shedden, 1999; Roeder, Lynch, & request from Helene Raskin White.)
604 WHITE, BATES, AND BUYSKE

The EM algorithm was used to simultaneously find the trajectories and to the these three predicted groups, a fourth group was also
the membership probabilities. Earlier applications of the growth mixture identified. This group, labeled E in Figure 1, comprises the esca-
modeling approach assigned each individual to the class to which he or she lating delinquents, who increased their levels of delinquency into
had the highest probability of belonging (e.g., Nagin et al., 1995). In this adulthood (13%).
study, each individual's probability of membership in each trajectory group
Chi-square analyses indicated that there were no significant
was used as a weight in the analyses. This technique reduces bias and
measurement error, especially for individuals whose observed behavior differences in the representativeness of the three age cohorts
does not clearly match any one group's trajectory (see Nagin & Tremblay, among the trajectory groups, ^(6, N = 698) = 1.16, p = .98. In
1999; Roeder et al., 1999). After determining the number of groups, we addition, when separate trajectories for each age cohort were
examined the association between group membership and risk factors by plotted (not shown, but available from Helene Raskin White), they
using polytomous logistic regression, with the group membership proba- matched the overall trajectory well, providing convergent support
bilities as case weights. for the chi-square analyses and indicating that age cohort effects
were negligible.
Results
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Delinquency Trajectories Levels of Delinquency

In all, two-group, three-group, four-group, five-group, and six- Table 2 shows the weighted mean levels and standard deviations
group models were tested. There was improvement in the BIC (i.e., for the delinquency and aggression scales at each age for the
it decreased) as the number of latent classes increased from two trajectory groups. Tukey post hoc comparisons were conducted to
(BIC = 10,405) to three (BIC = 10,140) and from three to four determine significant group differences. Persistent delinquents
(BIC = 10,009). However, the BIC began to increase as the scored significantly (p < .05) higher on the delinquency scale than
number of classes increased to five (BIC = 10,082) and six adolescence-limited delinquents at all ages except age 12. Al-
(BIC = 10,111). Thus, on the basis of the BIC criteria, we selected though persistent compared with adolescence-limited delinquents
a four-group model as the best-fitting model. were more aggressive at all ages except age 31, the differences
Figure 1 shows the four trajectory groups. Group N represents were only statistically significant at ages 18 and 24.5.
nondelinquents (47% of the total sample). Group A comprises the The escalating group was significantly lower in their delin-
adolescence-limited delinquents (33%). Group P represents the quency than the adolescence-limited delinquent group at ages 12
adolescence-to-adulthood-life-course-persistent delinquents (7%, and 15 and the persistent group at ages 12 through 18. The
which is consistent with estimates by Moffitt, 1993a). In addition escalating group became significantly more delinquent than the

8
w
£

'33
Q

12 15 18 21 24.5 28 31

Age

Figure 1. Delinquency trajectories for the total sample. N = nondelinquents; A = adolescence-limited


delinquents; E = escalating delinquents; P = persistent delinquents.
ADOLESCENCE-LIMITED VERSUS PERSISTENT DELINQUENCY 605

Table 2
Means (and Standard Deviations) of Delinquency and Aggression

Adolescence-limited Escalating Persistent


Variable Nondelinquents delinquents delinquents delinquents

Delinquency
Age 12 0.35 (0.70)a 2.02(1.92)b 1.05(1.44)c 2.99 (4.26)b,c
Age 15 0.61 (0.91)a 3.76 (2.52)b 2.02(1.91)c 6.93 (4.97)d
Age 18 0.76(1.03)a 3.54 (2.27)b 3.74 (2.69)b 10.76 (4.85)c
Age 21 0.61 (0.95)a 1.97(1.92)b 3.73 (2.70)c 7.55 (4.61)d
Age 24.5 0.40 (0.77)a 0.82(1.13)b 3.87 (2.33)c 5.51 (4.09)c
Age 28 0.16 (0.52)a 0.33 (0.64)a 3.99(2.71)b 2.46 (2.64)b
Age 31 0.16(0.48)a 0.12 (0.35)a 4.30 (3.04)b 1.56(1.96)c
Aggression
Age 12 0.13(0.39)a 0.49 (0.88)b 0.33(0.99)ab 1.41(2.31) ab
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Age 15 0.14(0.46)a 0.62(1.28)bc 0.27 (0.78)a,b 1.13(1.61)c


This document is copyrighted by the American Psychological Association or one of its allied publishers.

Age 18 0.14 (0.47)a 0.60(1.15)b 0.41 (0.86)b 1.80(2.25)c


Age 21 0.11 (0.48)a 0.38 (0.97)b 0.63(1.35)b 0.93 (1.57)b
Age 24.5 0.06 (0.35)a 0.14(0.47)a 0.30(0.90)ab 0.65(1.14)b
Age 28 0.04 (0.28)a 0.04 (0.27)a 0.36 (1.05)a 0.11 (0.46)a
Age 31 0.02(0.14)a 0.03 (0.24)a 0.13 (0.43)a 0.07 (0.25)a

Note. Means with different subscripts differ significantly (p < .05) by Tukey post hoc tests.

adolescence-limited group at age 21 and older, and was signifi- interactions, F(36, 49045) = 1.21, p = .18, and the personality-
cantly more delinquent than the persistent group at age 31.s environment interactions, F(27, 26043) = 1.42, p = .07, did not
significantly improve the model fit. Within these models, we
Trajectory Profiles examined odds ratios to determine which risk factors differentiated
among trajectory groups. Table 3 shows the odds ratios (OR) for
Figure 2 presents the age-adjusted, weighted means and 95%
the significant comparisons.
confidence intervals for the nondelinquent, adolescence-limited,
First, nondelinquents were compared with the other three tra-
escalating, and persistent groups on each of the risk factors to
provide an overall description of the trajectory groups (see Moffitt jectory groups. The main effects model indicated that those ado-
et al., 1996). For this presentation, the means were standardized to lescents with higher impulsivity (OR = 1.15 for every unit in-
a mean of 0 and a standard deviation of 1 (higher scores indicate crease), lower harm avoidance (OR = 1.11 for every unit increase
higher risk). Tukey post hoc comparisons (available from Helene on our reverse-coded scale), higher disinhibition (OR =1.41 for
Raskin White on request) revealed that all three delinquent groups every unit increase), higher parental hostility (OR = 1.03 for every
were significantly (p < .05) higher than nondelinquents in impul- unit increase), and one versus two parents (OR = 2.16) were
sivity, harm avoidance, and disinhibition. No other significant significantly more likely to be delinquents than nondelinquents.
group differences were found. Second, in the primary comparison of the adolescence-limited
delinquent trajectory with the persistent delinquent trajectory, only
Trajectory Group Differences one of the risk factors differed significantly between groups.
Persistent delinquents were significantly higher in disinhibition
To test the hypotheses regarding whether putative neuropsycho- (OR = 1.19 for every unit increase).
logical, personality, and environmental risk factors and their in- Third, escalating delinquents were compared with persistent
teractions could distinguish among delinquency trajectories, we delinquents. Although this comparison was not planned, it was
conducted hierarchical polytomous logistic regression analyses. conducted to help elucidate the risk factors associated with in-
Four sets of models were examined. The first set included the main
effects of the predictors and age cohort (as two dummy variables).6
The second set included the main effects plus the neuro- 5
Trajectory groups were also compared on the individual delinquency
psychological-personality variable interactions. The third set items. For the most part, the persistent group had the highest 3-year
included the main effects plus the neuropsychological- prevalence rates for serious and nonserious types of delinquency until their
environmental variable interactions. The final set included the late 20s and the escalating group exceeded them on every offense by either
main effects plus the personality-environmental variable interac- age 28 or age 31. The adolescence-limited group engaged in both serious
tions. For comparing nested models, we used the multiple impu- and nonserious offenses during adolescence; however, their prevalence
rates of nonserious delinquency were closer to those of persistent delin-
tation version of the likelihood ratio test, which uses an F statistic
quents than were their rates of serious delinquency.
instead of a chi-square difference (Meng & Rubin, 1992). 6
The youngest cohort and oldest cohort were entered as two dummy
The main effects model was significant, F(30, 352268) = 5.19,
variables compared with the middle cohort as the reference group. We
p < .0001. The addition of the neuropsychological-environment chose to use cohort rather than age as the control in case the risk factors
interactions significantly, F(36, 37665) = 5.19,p = .03, improved were not linearly related to age. Interactions of cohort with each of the risk
the model fit. The addition of the neuropsychological-personality factor variables were also tested, but none were statistically significant.
606 WHITE, BATES, AND BUYSKE

oq
0 Nondelinquent
Adolescent Limited
Escalating
CD Persistent
d

t\j
d
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

i i i
BIRISK VERB TRAIL BCT IMP HARM DIS SES FAMST PARHOS

Figure 2. Standardized means on the risk factors for the four trajectory groups. All risk factors are coded so
that a higher score means greater risk. BIRISK = birth risk; VERB = verbal ability; TRAIL = Trail Making
Test; BCT = Booklet Category Test; IMP = impulsivity; HARM = harm avoidance; DIS = disinhibition;
SES = socioeconomic status; FAMST = family structure; PARHOS = parental hostility.

creases in delinquency over time. Higher disinhibition (OR = 1.20 Finally, we examined the significant set of neuropsychological-
for every unit increase), lower verbal ability (OR = 1.06 for every environment interactions. Because of the large number of specific
unit increase), lower harm avoidance (OR = 1.15 for every unit interactions terms examined, we used the Bonferroni formula to
increase), higher parental hostility (OR = 1.03 for every unit establish a significance level of .0014. One interaction in each
increase), and coming from a one- compared with a two-parent group comparison was significant at this level; birth risk interacted
family (OR ~ 3.49) differentiated persistent from escalating with family structure to distinguish nondelinquents from delin-
delinquents. quents, adolescence-limited from persistent delinquents, and esca-
lating from persistent delinquents.7 However, the odds ratio for
birth risk in each of the three group comparisons was not signif-
Table 3 icant (p > .05) when family structure was set at either level (i.e.,
one-parent vs. two-parent family; Aiken & West, 1991). This
Odds Ratios for Significant Group Comparisons
failure to achieve significance is not surprising given that birth risk
Predictor N vs. A, E, & P A vs. P E vs. P was not a significant main effect in any of the models.

Higher birth risk


Lower verbal ability 1.06 Discussion
More BCT errors
Slower on Trail Making Test This study examined trajectories of delinquency from adoles-
Higher impulsivity 1.15 cence into adulthood. Four different trajectory groups were iden-
Lower harm avoidance 1.11 1.15 tified, which partially confirms the hypotheses. In accord with
Higher disinhibition 1.41 1.19 1.20 Moffitt's (1993a) model, an adolescence-limited and a persistent
Lower SES
delinquent trajectory were both clearly differentiated from a non-
One-parent family structure 2.17 3.49
Higher parental hostility 1.03 1.03 delinquent trajectory. In addition, a fourth trajectory group com-

Note. N = nondelinquents; A = adolescence-limited delinquents; E =


7
escalating delinquents; P = persistent delinquents. BCT = Booklet Cate- To give better numerical stability to the interaction models, the birth
gory Test; SES = socioeconomic status. risk scale was dichotomized to no risk versus one or two risks.
ADOLESCENCE-LIMITED VERSUS PERSISTENT DELINQUENCY 607

prising escalating delinquents, who engaged in relatively little accumulation of negative consequences. We did not have data to
delinquency in early adolescence but whose delinquency increased test for this; we recommend that future research should examine
from late adolescence into adulthood, was also identified (dis- this possibility.
cussed below). Other limitations of this study could be related to our failure to
A major aim of this study was to determine the ability of differentiate adolescence-limited from persistent delinquents, ex-
putative risk factors to differentiate adolescence-limited from cept in terms of disinhibition. First, the neuropsychological tests
adolescence-to-adulthood-persistent delinquents. Only one of the were first administered when participants were 18 to 24 years of
risk factors could distinguish membership in the persistent versus age (T3). Given the stability coefficients discussed in the Method
the adolescence-limited trajectory group. Persistent delinquents section, it seems unlikely that the timing of these measurements
were higher in disinhibition in adolescence (see also H. White et seriously affected the validity of results. Second, and more impor-
al., 1985). tant, a limitation may be the selection of the specific executive
When persistent, escalating, and adolescence-limited delin- functions assessed. Although problem-solving ability and cogni-
quents as a group were contrasted with nondelinquents, five main tive flexibility are important executive functions (Cummings,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

effects were significant: higher disinhibition, impulsivity, parental 1995; Pennington & Ozonoff, 1996), it is not known whether these
This document is copyrighted by the American Psychological Association or one of its allied publishers.

hostility, lower harm avoidance, and less intact family structure skills represent the dimensions of executive control that are most
were related to delinquent behavior. Although personality differ- relevant for identifying life-course-persistent delinquents. For ex-
ences between adolescence-limited delinquents and nondelin- ample, potentially important deficits in planning and response
quents were not predicted by Moffitt's theory (1993a), she and her inhibition are not well tapped by these measures. In addition, the
colleagues found similar differences when they compared nonde- reliance on self-report data may have affected the results inasmuch
linquents (abstainers) with adolescence-limited and persistent de- as some delinquents may have under- or overreported their delin-
linquents (Moffitt et al., 1996). The ability of the risk factors tested quency involvement. Finally, some individuals may have been
here to more robustly distinguish membership in any delinquency institutionalized for an extended period of time and therefore
trajectory versus the nondelinquent trajectory (rather than distin- unable to engage in continued delinquent activities. Although we
guish the persistent from the adolescence-limited trajectory) raises did not have the data to specifically address this question, it is
the possibility that some of the unique characteristics of childhood- unlikely that this would have seriously biased the results, given the
to-adolescence-persistent (or early-onset) delinquents that have relatively lengthy intervals between assessments. Despite these
been reported in the literature (e.g., Moffitt et al., 1994; J. White limitations, our results for neuropsychological risk factors and
et al., 1990) may primarily reflect differences between early-onset impulsivity are consistent with those of Nagin and colleagues
delinquents and those who never as opposed to later become (1995), who used alternative measures of risk factors, a lower SES
delinquent. sample, and convictions rather than self-report data for criminal
These results may suggest that many of the risk factors that behavior.
relate to onset and stability of deviant behaviors from childhood to The escalating trajectory was not anticipated on the basis of
adolescence simply are not related to persistence from adolescence Moffitt's (1993a) two-trajectory taxonomy of delinquency, al-
to adulthood and, therefore, are not related to life-course persis- though others studies have identified a similar trajectory (e.g.,
tence of antisocial behavior. It is possible that social and environ- D'Unger et al., 1998). This small group (13% of the total) dis-
mental factors beyond adolescence, rather than childhood individ- played higher levels of delinquency in adulthood (statistically
ual and environmental characteristics, may be more important for significant at age 31 but also higher at age 28) than the
the persistence of delinquency from adolescence to adulthood. For adolescence-to-adulthood-persistent group. These individuals
example, Laub and Sampson (1993) suggested that social bonds in were adult criminals who potentially would persist but who did not
adulthood and life events (e.g., a stable job, and a cohesive engage in delinquency heavily during adolescence. It is possible
marriage or cohabiting relationship) can lead to desistance from that some of these individuals engaged in antisocial behavior
delinquency. during childhood (which we did not measure), decreased this
An alternative explanation is that more aggressive and "patho- behavior throughout adolescence, and then increased it again in
logical" life-course-persistent delinquents were underrepresented adulthood. However, Moffitt (1993a) did not identify a hiatus for
in this working-class and middle-class sample8 or more were lost persistent delinquents during adolescence but rather an escalation.
to attrition. Moreover, the primarily middle-class nature of this In addition, some members of this group may have lied about their
sample may have limited the variation in SES and may have
contributed to lower levels of neuropsychological deficit and lower
8
birth risk, thereby providing less power to detect interactions than Although levels of delinquency were lower in this sample than nation-
in lower SES samples (see Tibbetts & Piquero, 1999). In addition, ally representative samples (e.g., Elliott et al., 1989), the persistent delin-
some of the adolescence-limited delinquents may have been pro- quents still engaged in serious acts in adulthood (e.g., at age 28, 18% had
pelled into a persistent trajectory although they were not high on stolen things worth more than $50 and 3% had committed a breaking and
purported risk factors. In other words, some may have experienced entering offense, and at age 31, 10% had stolen things worth more than
$50). The escalating delinquents were also frequent and serious offenders
negative consequences or "snares" due to their early delinquency
in adulthood (e.g., 31% and 23% had stolen things worth more than $50 at
(e.g., acquiring a police record, becoming embedded in a deviant
ages 28 and 31, respectively, and 11% and 8% had committed a breaking
peer group), which steered them into a delinquent career (see and entering offense at ages 28 and 31, respectively). Thus, while these
Moffitt, 1993a). Their inclusion in the persistent trajectory group adult offenders may not represent the worst felons nationwide, they are still
could dilute observed differences between pathologically driven a group of individuals who are committing illegal acts in adulthood that
persistent delinquents and those who persisted because of the should be of concern to criminology researchers.
608 WHITE, BATES, AND BUYSKE

delinquent activity in adolescence. Yet, others, such as Farrington Research and clinical form manual. Odessa, FL: Psychological Assess-
(1991), have also found that only one half of antisocial adults were ment Resources.
antisocial as children (see also Loeber & Hay, 1997). The esca- Dembo, R., Williams, L., Wish, E., & Schmeidler, J. (1990). Urine testing
of detained juveniles to identify high-risk youth. Washington, DC: U.S.
lating delinquents differed from the persistent delinquents in terms
Department of Justice.
of having higher verbal ability and harm avoidance, lower parental
Donnellan, M. B., Ge, X., & Wenk, E. (2000). Cognitive abilities in
hostility and disinhibition, and coming from more intact families. adolescence-limited and life-course-persistent criminal offenders. Jour-
Thus, the stronger family support, better neuropsychological func- nal of Abnormal Psychology, 109, 396-402.
tioning, and more protective personality characteristics of the Duffy, J., & Campbell, J. (1994). The regional prefrontal syndromes: A
escalating delinquents compared with the persistent delinquents theoretical and clinical overview. Journal of Neuropsychiatry and Clin-
may have buffered the expression of delinquency in adolescence. ical Neurosciences, 6, 379-387.
Perhaps these individuals escalated their delinquency only after D'Unger, A. V., Land, K. C., McCall, P. L., & Nagin, D. S. (1998). How
leaving the supervision of their family of origin. many latent classes of delinquent/criminal careers? Results from mixed
In sum, the data suggest that many of the neuropsychological Poisson regression analyses. American Journal of Sociology, 103, 1593-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

risk factors that distinguish early-onset persistent delinquency 1630.


This document is copyrighted by the American Psychological Association or one of its allied publishers.

from adolescence-limited delinquency are not capable of distin- Elliott, D. S., Huizinga, D., & Menard, S. (1989). Multiple problem youth:
Delinquency, substance use, and mental health problems. New York:
guishing adolescence-limited from adolescence-to-adulthood-
Springer-Verlag.
persistent delinquency. As predicted by Moffitt (1993a, 1993b), Escon, M. E., Yen, J. K., & Bourke, A. L. (1978). Assessment of recovery
the present results provide support for noteworthy individual dif- from serious head injury. Journal of Neurology, Neurosurgery and
ferences in the stability of antisocial behavior across time and Psychiatry, 41, 1036-1042.
extend this to a later portion of the life course than has typically Farrington, D. P. (1991). Antisocial personality from childhood to adult-
been studied in longitudinal, self-report studies. Furthermore, the hood. The Psychologist: Bulletin of the British Psychological Society, 4,
results suggest that researchers may need to go beyond childhood 389-394.
to determine the risk factors for life-course-persistence and exam- Farrington, D. P., & Hawkins, J. D. (1991). Predicting participation, early
ine transitions and turning points in development (see H. White, onset and later persistence in officially recorded offending. Criminal
Bates, & Labouvie, 1998). Thus, our findings support the validity Behavior and Mental Health, I, 1-33.
of applying a dynamic, life-course perspective to the criminal Gaudino, E. A., Geisler, M. W., & Squires, N. W. (1995). Construct
validity in the Trail Making Test: What makes Part B harder? Journal of
career paradigm.
Clinical and Experimental Neuropsychology, 17, 529-535.
Giancola, P. R. (1995). Evidence for dorsolateral and orbital prefrontal
References cortical involvement in the expression of aggressive behavior. Aggres-
sive Behavior, 21, 431-450.
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and Giancola, P. R., Moss, H. B., Martin, C. S., Kirisci, L., & Tarter, R. E.
interpreting interactions. Newbury Park, CA: Sage. (1996). Executive cognitive functioning predicts reactive aggression in
Bandeen-Roche, K., Miglioretti, D., Zeger, S. L., & Rathouz, P. J. (1997). boys at high risk for substance abuse: A prospective study. Alcoholism:
Latent variable regression for multiple discrete outcomes. Journal of the Clinical and Experimental Research, 20, 740-744.
American Statistical Association, 92, 1375-1386. Gottfredson, M. R., & Hirschi, T. (1990). A general theory of crime.
Bartusch, D., Jeglum, R., Lynam, D. R., Moffitt, T. E., & Silva, P. A. Stanford, CA: Stanford University Press.
(1997). Is age important? Testing a general versus a developmental Hindelang, M. J., Hirschi, T., & Weis, J. G. (1981). Measuring delin-
theory of antisocial behavior. Criminology, 35, 13-48. quency. Beverly Hills, CA: Sage Publications.
Bates, M. E., & Labouvie, E. W. (1998, June). Do neurocognitive abilities Hoaken, P. N. S., Giancola, P. R., & Pihl, R. O. (1998). Executive
vary with alcohol and other drug use from adolescence to adulthood? cognitive functions as mediators of alcohol-related aggression. Alcohol
Answers from variable-centered and person-centered approaches. Paper & Alcoholism, 33, 47-54.
presented at the annual meeting of the Research Society on Alcoholism, Horwitz, A. V., & White, H. R. (1987). Gender role orientations and styles
Hilton Head, SC. of pathology among adolescents. Journal of Health and Social Behav-
Benton, A. L. (1994). Neuropsychological assessment. Annual Review of ior, 28, 158-170.
Psychology, 45, 1-23. Jackson, D. N. (1968). Personality Research Form. Goshen, NY: Research
Block, J. (1995). On the relation between IQ, impulsivity, and delinquency: Psychologists Press.
Remarks on the Lynam, Moffitt, and Stouthamer-Loeber (1993) inter- Keribin, C. (2000). Consistent estimation of the order of mixture models.
pretation. Journal of Abnormal Psychology, 104, 395-398. Sanka Series A, 62, 49-66.
Blumstein, A., Cohen, J., & Farrington, D. (1988). Criminal careers re- Laub, J. H., & Sampson, R. J. (1993). Turning points in the life course:
search: Its value for criminology. Criminology, 26, 1-35. Why change matters to the study of crime. Criminology, 31, 301-325.
Caspi, C., Moffitt, T. E., Silva, P. A., Stouthamer-Loeber, M., Krueger, Linden, R. (1978). Myths of middle-class delinquency: A test of the
R. F., & Schmutte, P. S. (1994). Are some people crime-prone? Repli- generalizability of social control theory. Youth and Society, 9, 407-431.
cations of the personality-crime relationship across countries, genders, Loeber, R. (1988). Natural histories of conduct problems, delinquency, and
races, and methods. Criminology, 32, 163-195. associated substance use. In B. B. Lahey & A. E. Kazdin (Eds.),
Chaiken, J., & Chaiken, M. (1990). Drugs and predatory crime. In M. Advances in clinical child psychology (Vol. 11, pp. 73-124). New York:
Tonry & J. Q. Wilson (Eds.), Drugs and crime (Vol. 13, pp. 203-239). Plenum.
Chicago: University of Chicago Press. Loeber, R., & Hay, D. (1997). Key issues in the development of aggression
Cummings, J. L. (1995). Anatomic and behavioral aspects of frontal- and violence from childhood to early adulthood. Annual Review of
subcortical circuits. Annals of the New York Academy of Sciences, 769, Psychology, 48, 371-410.
1-13. Luria, A. R. (1973). The working brain: An introduction to neuropsychol-
DeFillippis, N. A., & McCampbell, E. (1991). The Booklet Category Test: ogy. New York: Basic Books.
ADOLESCENCE-LIMITED VERSUS PERSISTENT DELINQUENCY 609

Lynam, D. R., Moffitt, T. E., & Stouthamer-Loeber, M. (1993). Explaining opmental psychopathology. Journal of Child Psychology and Psychia-
the relation between IQ and delinquency: Class, race, test motivation, try, 37, 51-87.
school failure, or self-control? Journal of Abnormal Psychology, 102, Raftery, A. E. (1995). Bayesian model selection in social research. Socio-
187-196. logical Methodology, 25, 111-164.
Mathsoft. (1999). S-Plus 2000programmer's guide. Seattle: Data Analysis Reitan, R. M., & Wolfson, D. (1985). The Halstead-Reitan Neuropsycho-
Products Division, Mathsoft Institute. logical Test Battery: Theory and clinical interpretation. Tucson, AZ:
Meng, X. L., & Rubin, D. B. (1992). Performing likelihood ratio tests with Neuropsychology Press.
multiply-imputed data sets. Biometrika, 79, 103-111. Roeder, K., Lynch, K. G., & Nagin, D. S. (1999). Modeling uncertainty in
Moffitt, T. E. (1990). Juvenile delinquency and attention deficit disorder: latent class membership: A case study in criminology. Journal of the
Developmental trajectories from age three to fifteen. Child Develop- American Statistical Association, 94, 766-776.
ment, 61, 893-910. Russell, E. W., & Barren, J. H. (1989). A difference that is not a difference:
Moffitt, T. E. (1993a). Adolescent-limited and life-course-persistent anti- Reply to Vanderploeg and Logan. Journal of Consulting and Clinical
social behavior: A developmental taxonomy. Psychological Review, Psychology, 57, 317-318.
100, 674-701. Russell, E. W., & Levy, M. (1987). Revision of the Halstead Category
Test. Journal of Consulting and Clinical Psychology, 55, 898-901.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Moffitt, T. E. (1993b). The neurobiology of conduct disorder. Development


SAS. (2001). SAS/STAT software: Changes and enhancements (Ver-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

and Psychopathology, 5, 135-151.


Moffitt, T. E., Caspi, A., Dickson, N., Silva, P., & Stanton, W. (1996). sion 8.2). Cary, NC: Author.
Childhood-onset versus adolescent-onset antisocial conduct problems in Schafer, J. L. (1997). Analysis of incomplete multivariate data. New York:
males: Natural history from ages 3 to 18 years. Development and Chapman & HalVCRC.
Psychopathology, 8, 399-424. Seguin, J. R., Pihl, R. O., Harden, P. W., Tremblay, R. E., & Boulerice, B.
Moffitt, T. E., Lynam, D. R., & Silva, P. A. (1994). Neuropsychological (1995). Cognitive and neuropsychological characteristics of physically
aggressive boys. Journal of Abnormal Psychology, 104, 614-624.
tests predicting persistent male delinquency. Criminology, 32, 277-300.
Simons, R. L., Wu, C., Conger, R. D., & Lorenz, F. O. (1994). Two routes
Moffitt, T. E., & Silva, P. A. (1988). Self-reported delinquency, neuropsy-
to delinquency: Differences between early and late starters in the impact
chological deficit, and history of attention deficit disorder. Journal of
of parenting and deviant peers. Criminology, 32, 247-275.
Abnormal Child Psychology, 16, 533-569.
Smith, D. A., & Brame, R. (1994). On the initiation and continuation of
Muthen, B., & Shedden, K. (1999). Finite mixture modeling with mixture
delinquency. Criminology, 32, 607-628.
outcomes using the EM algorithm. Biometrics, 55, 463-469.
Streit, F. (1978). Technical manual: Youth Perception Inventory. Highland
Nagin, D. S., & Farrington, D. P. (1992). The onset and persistence of
Park, NJ: Essence Publications.
offending. Criminology, 30, 501-523.
Tibbetts, S. G., & Piquero, A. (1999). The influence of gender, low birth
Nagin, D. S., Farrington, D. P., & Moffitt, T. E. (1995). Life-course
weight, and disadvantaged environment in predicting early onset of
trajectories of different types of offenders. Criminology, 33, 111-139.
offending: A test of Moffitt's interactional hypothesis. Criminology, 37,
Nagin, D. S., & Land, K. C. (1993). Age, criminal career, and population
843-878.
heterogeneity: Specification and estimation of a nonparametric mixed
Tremblay, R. E., Pihl, R. O., Vitaro, F., & Dobkin, P. L. (1994). Predicting
Poisson model. Criminology, 31, 327-362.
early onset of male antisocial behavior from preschool behavior. Ar-
Nagin, D. S., & Tremblay, R. E. (1999). Trajectories of boys' physical
chives of General Psychiatry, 51, 732-739.
aggression, opposition, and hyperactivity on the path to physically
U.S. Bureau of the Census. (1981). Current population survey: Money,
violent and nonviolent juvenile delinquency. Child Development, 70,
income and poverty status of families and persons in the United States,
1181-1196.
1980, (Series P-60, No. 127). Washington, DC: U.S. Government Print-
Pandina, R. J., Labouvie, E. W., & White, H. R. (1984). Potential contri- ing Office.
butions of the life span developmental approach to the study of adoles- Wedderburn, R. W. M. (1974). Quasi-likelihood functions, generalized
cent alcohol and drug use: The Rutgers Health and Human Development linear models, and the Gauss-Newton method. Biometrika, 61, 439-447.
Project, a working model. Journal of Drug Issues, 14, 253-268. White, H. R., Bates, M. E., & Labouvie, E. W. (1998). Adult outcomes of
Pandina, R. J., & White, H. R. (1984, April). The relationship between adolescent drug use: A comparison of process-oriented and incremental
alcohol and marijuana use and crime: Implications for intervention and analyses. In R. Jessor (Ed.), New perspectives on adolescent risk behav-
prevention. Paper presented at the State-of-the-Art Conference on Juve- ior (pp. 150-181). Cambridge, England: Cambridge University Press.
nile Offenders With Serious Drug, Alcohol, and Mental Health Prob- White, H. R., Labouvie, E. W., & Bates, M. E. (1985). The relationship
lems, Bethesda, MD. between sensation seeking and delinquency: A longitudinal analysis.
Parker, E. S., Birnbaum, I. M., Boyd, R. A., & Noble, E. P. (1980). Journal of Research in Crime and Delinquency, 22, 197-211.
Neuropsychological decrements as a function of alcohol intake in male White, J. L., Moffitt, T. E., Caspi, A., Bartusch, D. J., Needles, D. J., &
students. Alcoholism: Clinical and Experimental Research, 4, 330-334. Stouthamer-Loeber, M. (1994). Measuring impulsivity and examining
Pasternoster, R., & Brame, R. (1997). Multiple routes to delinquency? A its relationship to delinquency. Journal of Abnormal Psychology, 103,
test of developmental and general theories of crime. Criminology, 35, 192-205.
49-84. White, J. L., Moffitt, T. E., Earls, F., Robins, L., & Silva, P. A. (1990).
Patterson, G. R., & Yoerger, K. (1993). Developmental models for delin- How early can we tell? Predictors of childhood conduct disorder and
quent behavior. In S. Hodgins (Ed.), Crime and mental disorder. New- adolescent delinquency. Criminology, 28, 507-533.
bury Park, CA: Sage. Zachary, R. A. (1986). Shipley Institute for Living Scale: Manual. Los
Paulson, M. J., & Lin, T. (1970). Predicting WAIS IQ from the Shipley- Angeles, CA: Western Psychological Services.
Hartford scores. Journal of Clinical Psychology, 26, 453-461. Zuckerman, M. (1979). Sensation seeking: Beyond the optimal level of
Pennington, B. F., & Bennetto, L. (1993). Main effects or transactions in arousal. Hillsdale, NJ: Erlbaum.
the neuropsychology of conduct disorder? Commentary on "The neuro-
psychology of conduct disorder." Development and Psychopathology, 5, Received March 27, 2000
153-164. Revision received May 15, 2001
Pennington, B. F., & Ozonoff, S. (1996). Executive functions and devel- Accepted May 17, 2001 •

You might also like