NIH Public Access

NIH Public Access
Author Manuscript
J Adolesc. Author manuscript; available in PMC 2014 April 01.
NIH-PA Author Manuscript
Published in final edited form as:

J Adolesc. 2013 April ; 36(2): . doi:10.1016/j.adolescence.2012.12.007.
Identifying Gender-Specific Developmental Trajectories of

Nonviolent and Violent Delinquency from Adolescence to Young
Adulthood
Yao Zheng1 and H. Harrington Cleveland1
1Human Development and Family Studies, Pennsylvania State University, USA
Abstract
Most research examining gender differences in developmental trajectories of antisocial behavior

does not consider subtypes of antisocial behavior and is difficult to generalize due to small
nonrepresentative samples. The current study investigated gender difference in developmental
trajectories from adolescence to young adulthood while addressing those limitations. Analyses
were limited to respondents ages 15 and 16 in wave 1 (1617 in wave 2, and 2122 in wave 3) of
the National Longitudinal Study of Adolescent Health (n = 6244, 49.5% males). Self-report
nonviolent and violent delinquencies were simultaneously entered into latent class analysis. Four
latent classes were identified: low, desister, decliner, and chronic (male-only). In addition to
finding a male-specific chronic class, gender differences included differences in levels of
nonviolent and violent delinquency between synonymous classes of males and females, and
differences in prevalence of classes across genders. Neighborhood disadvantage and family
support predicted trajectories.
Keywords
nonviolent delinquency; violent delinquency; gender; latent class analysis; adolescence; young
adulthood
With the onset of developmental trajectory research over the last decade our understanding
of antisocial behavior across adolescence has become more complex. This research has
revealed distinct developmental patterns of antisocial behavior across this developmental
period (for comprehensive reviews, see Moffitt, 2006; Nagin & Tremblay, 2005a, 2005b).
However, these findings are predominantly based on male samples, perhaps due to males
greater involvement in antisocial behavior (Archer & Ct, 2005). However, extant research
on gender differences in antisocial behavior suggests that males and females may follow
different trajectories (see Fontaine, Carbonneau, Vitaro, Barker, & Tremblay [2009] for a
comprehensive review). Additional shortcomings of the current research include sample
quality, measurement, and model identification methods. Our goal was to address these
limitations with a study based on a large nationally representative sample with specific
measures of nonviolent and violent antisocial behaviors. Before setting out specific aims, we
2013 The Association for Professionals in Services for Adolescents. Published by Elsevier Ltd. All rights reserved.
Correspond to: Yao Zheng, Human Development and Family Studies, Pennsylvania State University, S-110 Henderson, University
Park, PA, 16802, USA; yzz122@psu.edu; Tel: 814-321-7667.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our
customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of
the resulting proof before it is published in its final citable form. Please note that during the production process errorsmaybe
discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Zheng and Cleveland
Page 2
first review literature on the developmental trajectories of antisocial behavior in general, and
then studies specifically examining gender differences in these developmental trajectories.
Developmental Trajectories of Antisocial Behavior

While most antisocial adults demonstrate a life history of behavior problems that can be
traced back to early childhood, not all antisocial children (in fact, only a relative few)
continue to show antisocial behavior later in life (Maughan & Rutter, 2008). The most wellknown model of antisocial behavior trajectories is Moffitts (1993) developmental
taxonomy, which provides that the general population of delinquents includes two distinct
antisocial subgroups that follow different pathways and have unique etiologies: a small
early-onset/life-course-persistent group (510%) and a larger adolescence-limited group
(1530%). Life-course-persistent individuals demonstrate early-onset symptoms of
antisocial behavior, partially linked to neuropsychological characteristics (e.g., difficult
temperament, neuropsychological deficits), which reciprocally interact with criminogenic
environments (e.g., disruptive family, poor parenting) (Moffitt, 1993, 2006). Adolescencelimited individuals, however, manifest antisocial behavior primarily during adolescence due
to temporary deviant peer influences and poor parental monitoring, but desist thereafter.
These adolescence-limited individuals are also less inclined towards aggressive and violent
behaviors compared to life-course-persistent individuals (Moffitt, 1993, 2006).
In addition to the life-course-persistent and adolescence-limited groups, which have been

found in multiple independent empirical studies with various samples across countries (e.g.,
Broidy et al., 2003; DeLisi, 2001; Piquero & Brezina, 2001), other subgroups have been
identified. These subgroups include a childhood-limited group, manifesting antisocial
behavior mostly during childhood but desisting in adolescence (e.g., Odgers et al., 2008), as
well as the never/low group, which is typically the largest, abstains from antisocial behavior,
and is highly prosocial and high-functioning (e.g., Piquero, Brezina, & Turner, 2005).
Additional identified groups, with labels varying to accommodate child- or adolescent-only
samples, include: Chronic, which resembles the life-course-persistent group, and the lowdesister/moderate-declining and high-desister/high-declining groups, which differ in
severity, onset, and timing of desistence, but both resemble the adolescence-limited group
(Nagin & Tremblay, 2005a, 2005b).
Gender Differences in the Developmental Trajectories of Antisocial

Behavior
Despite well-documented gender differences in antisocial behavior (e.g., Archer & Ct,
2005; Moffitt et al., 2001), gender differences in the developmental trajectories of antisocial
behavior haven been relatively understudied. Most research is based predominantly on
males, or on mixed samples without explicit examination of gender differences. Given that
males are more likely to be chronic offenders and involved in more violent behaviors than
females (Moffitt et al., 2001), it is reasonable to expect that the prevalence of life-coursepersistent groups would be higher among males, whereas the prevalence of the adolescencelimited group would be similar across gender groups. There is some evidence that trajectory
prevalence differs by genders. For example, males seem more likely to be early-onset/lifecourse-persistent than females, whereas they are only slightly more likely to belong to the
adolescent-limited group (Moffitt & Caspi, 2001; Moffitt et al., 2001). Conversely, the
prevalence of the low trajectory appears to be higher in females than in males (Lahey et al.,
2006; Odgers et al., 2008).
Among the few researchers to focus explicitly on gender, Silverthorn and Frick (1999)
proposed a female-specific adolescent-delayed-onset group that shares similar risk factors
Zheng and Cleveland
Page 3
(e.g., neuropsychological deficits, dysfunctional family environment) with male early-onset/

life-course-persistent groups, but does not manifest antisocial behavior until adolescence.
This group was also found in a small at-risk sample (Silverthorn, Frick, & Reynolds, 2001).
Using data from a large longitudinal cohort study and official records, DUnger, Land, and
McCall (2002) identified three female trajectories: Non-offenders, low-rate, and high-rate
adolescent peaked, which were similar to never/low, low-desisting, and high-desisting,
respectively. Two other trajectories were identified in males: low-rate chronic and high-rate
chronic. Among both lowrate and high-rate adolescent groups, males reported more police
contacts than did females; and the female high-rate adolescent group showed more similarity
to the male low-rate chronic group than to the male high-rate adolescent group.
Previous Limitations and the Scope of the Current Study
As demonstrated in the above studies, gender differences in trajectories do exist. The goal of
the current study was to add to the understanding of such gender differences in antisocial
behavior trajectories by addressing several limitations in the current literature. First, and
perhaps most important in relation to potential gender differences, most studies use
aggregated measures of antisocial behavior without differentiating subtypes of antisocial
behavior (e.g., Moffitt & Caspi, 2001; Odgers et al., 2008). This measurement issue could
potentially mask differences in developmental pathways for antisocial behavior subtypes of
nonviolent and violent delinquency. These different aspects of antisocial behavior have been
shown to have different etiologies and may represent qualitatively different types of
antisocial behavioral trajectories (e.g., Burt & Neiderhiser, 2009; Lacourse et al., 2002). The
importance of these differences is suggested in findings from Fergusson and Horwood
(2002), which revealed similar latent class trajectories but different prevalences when using
property offenses vs. violent offenses to identify trajectories.
Second, many previous studies used a priori criteria or single threshold criteria (e.g., age of
onset) to identify trajectories (e.g., Mazerolle, Brame, Paternoster, Piquero, & Dean, 2000;
Silverthorn et al., 2001) that could be subjective and lead to rather different findings
(Fontaine et al., 2009). Recent advances in group-based methods such as growth mixture
modeling (GMM; Muthn & Shedden, 1999), latent class growth analysis (LCGA; Nagin,
1999, 2005) and latent class analysis (LCA; Collins & Lanza, 2009) could provide more
robust estimation. These mixture approaches assume that there are mutually exclusive and
exhaustive latent groups in the population and are especially useful in capturing
heterogeneity in developmental trajectories.
Third, many studies have used official crime records (e.g., DeLisi, 2001; DUnger et al.,
2002; Mazerolle et al., 2000). These records have been shown to primarily capture more
chronic and severe antisocial behaviors and underestimate overall rates of antisocial
behavior (Thornberry & Krohn, 2000), especially during adolescence and for minor
nonviolent delinquent acts. The use of such records could affect the particular groups
identifiedin particular, making it more likely to identify groups made up of more chronic
and violent life-course-persistent individuals. For example, using conviction records across
20 years (from age 10 to 30), Nagin (2005) found two persistent groups: low chronic and
high chronic. In contrast, using teacher-, parent- and self-report conduct problems, Odgers et
al. (2008) only identified one persistent group: the early-onset-persistent trajectory.
Fourth, research in this area is challenged by the use of samples that are relatively small, atrisk, or clinically-based (e.g., Silverthorn et al., 2001) or restricted in age (e.g., limited to
adolescence; Nagin & Tremblay, 1999; Odgers et al., 2008). These samples may make it
more difficult to identify subgroups that may be low in prevalence (e.g., 5% life-coursepersistent) or distinguished from others by their childhood or adult behaviors.
Zheng and Cleveland
Page 4
Attempting to address these limitations, using latent class analysis the current study drew on
a large nationally representative sample from the National Longitudinal Adolescent Health
data set (Add Health) to examine gender differences in developmental trajectories of
selfreported antisocial behavior from adolescence to young adulthood. We expected that, in
addition to differences in the prevalence of analogous trajectories across genders, genderspecific trajectories also might arise. This study had two secondary and related goals: to 1)
determine the impact of distinguishing between nonviolent and violent delinquency on
identifying latent trajectories, and 2) determine the risk factors associated with identified
trajectories. Because different types of antisocial behavior have different influences and
prevalence across genders, distinguishing between subtypes of antisocial behavior is
especially important for investigating potential gender differences.
Method
Data and Sample
The current study used the first three waves of In-Home Interview data of the National
Longitudinal Study of Adolescent Health (Add Health; Harris et al., 2009), which included
over 20,000 respondents (mean age = 16.7 years, 49.5% males, 50.6% non-Hispanic Whites)
in the first wave. The analysis sample consisted of 6,244 respondents (48.8%% males,
53.4% non-Hispanic Whites) interviewed in 1995 (1st wave, 1516 years old) and
approximately one year later in 1996 (2nd wave, 1617 years old). The third wave was
conducted between July 2001 and April 2002 when respondents were in young adulthood
(2122 years old). The age group of 15 16 years old (mid-adolescence, 45.6% 15 years old)
was chosen in part because it represented the largest age group in the first wave. We
analyzed data from a specific age group rather than all available respondents with three
waves of data because including early adolescents (1213 years old) and late adolescents
(1819 years old) simultaneously in the analysis could confound developmental patterns in
the same age with developmental differences across different ages. For example, a decline in
nonviolent delinquency for a 16-year-old respondent from wave 1 to wave 2 does not have
the same developmental meaning as the same decline for a 12 or 18 year old.
Measures
Nonviolent and violent delinquencyThe nonviolent and violent delinquency

measures covered behaviors with a type of scale widely used in delinquency and violence
research (Thornberry & Krohn, 2000), and were closely related to scales used in other Add
Health studies (e.g., Guo, Roettger, & Cai, 2008; Hagan & Foster, 2003). To make the
means of nonviolent and violent delinquency comparable across waves, only items included
in all three waves were used.
Nonviolent delinquency: A 5-item 4-point scale was used to measure nonviolent
delinquency: In the past 12 months, how often did you deliberately damage property that
didnt belong to you; steal something worth more than 50 dollars; go into a house or
building to steal something; sell marijuana or other drugs; steal something worth less than
50 dollars? All items response ranged from 0 never , 1 once or twice, 2 three or four
times to 3 five times or more. Cronbach were between 0.66 and 0.74 across three
waves.
Violent delinquency: A 3-item 4-point scale was used to measure violent delinquency: In
the past 12 months, how often did you use or threaten to use a weapon to get something
from someone; take part in a fight where a group of your friends was against another group;
hurt someone badly enough to need bandages or care from a doctor or nurse? All items
Zheng and Cleveland
Page 5
response ranged from 0 never , 1 once or twice, 2 three or four times to 3 five times
or more. Cronbach were between 0.58 and 0.63.
Scale transformation: Prior to entry in Latent Class Analysis (LCA), continuous values for
nonviolent and violent delinquency were categorized into three levels: Never, Mild, and
Moderate/Serious. This transformation was necessary for fitting latent mixture models.
More detail on this transformation is provided in analytic strategy. The non-count nature of
the scales also prevents them from being modeled in Latent Profile Analysis with Poisson
distribution. Scales were trichotomized because three levels maintains more of the full
meaning of the original scales than simply dichotomizing, which would combine mild
delinquents with moderate delinquent respondents, or combine mild delinquents with nondelinquents. To keep the meaning of each level comparable across waves, the same
numerical cut-offs were used rather than maintaining the same proportion in each level
across waves.
The specific cut-offs used to trichotomize the scales were as follows: All values of 0 in each
original scale were coded as Never. However, due to different numbers of items in
nonviolent and violent delinquency, different cut-offs were used between Mild and
Moderate/Serious. For nonviolent delinquency, values greater than 0 and less than 0.6
were coded into Mild, and values equal to or larger than 0.6 were coded into Moderate/
Serious. The cut-off of 0.6 was chosen because there were five items on the nonviolent
delinquency scale. Respondents would have a value of 0.4 if they responded once or twice
(1) for two of the five items but never (0) to the three other items, or if they responded
three or four times (2) to one of the 5 items but responded never to the other four items.
These respondents were regarded as mildly delinquent. Any respondent reporting more
nonviolent delinquency would receive a value of 0.6 or higher and were regarded as
moderately to seriously delinquent. For the 3-item violent delinquency scale, respondents
with mean values greater than 0.33 who had responded to two or more of the three items
with once or twice (1) or at least one item with three or four times (2) or greater were
categorized into Moderate/Serious. Respondents who responded only once or twice (1)
to a single item had a score of 0.33 on the 3-item scale and were categorized as Mild.
To ensure that the cut-offs used correctly ranked respondents into never, mild, and
moderate/serious levels across two scales, the means for nonviolent and violent delinquency
on the original scales were calculated by the three levels and compared to the overall sample
means (see Table 1). For example, the mean for wave 1 nonviolent delinquency for the full
sample was 0.18 (SD = 0.38). The mean for the Never level was 0. The mean for the
Mild level was 0.26, which was only 0.21 SD higher than the full sample mean, indicating
that the Mild level of nonviolent delinquency was mildly higher than the overall sample
mean. In contrast, the mean for the Moderate/Serious level was 1.01, more than 2 SD
higher than the full sample mean. Distributions were similar for wave1 violent delinquency
and for the other two waves.
Covariates
Neighborhood disadvantage: A composite of three proportion scores was used: proportion
of non-intact family households, proportion of low-income families (less than $15,000) and
proportion of unemployment rate measured in 1st wave. Higher scores indicated more
disadvantage. Data were collected from the Census of Population and Housing 1990
measured at the Block Group level. Cronbach was 0.85.
Family support: This construct was measured with a 10-item scale constructed from the
average of five items about mother and five about father. Items for mother were: How close
Zheng and Cleveland
Page 6
do you feel to your mother; how much do you think she cares about you; you are satisfied
with the way your mother and you communicate with each other; overall you are satisfied
with your relationship with your mother; and, most of the time your mother is warm and
loving towards you. All item responses ranged from 1 to 5 but with different anchors. The
first two items ranged from not at all to very much, and the others from strongly disagree to
strongly agree. Higher scores indicated more family support. Cronbach was 0.87.
Analysis Framework
Latent Class Analysis was fitted in SAS 9.2 using PROC LCA (Lanza, Collins, Lemmon, &
Schafer, 2007) to trichotomized nonviolent and violent delinquency scores over three waves.
This procedure provides Maximum Likelihood (ML) estimates using the ExpectationMaximization (EM) algorithm, which also handles missing data on latent class items (Lanza
et al., 2007). LCA can be considered a special case of GMM without modeling slope (e.g.,
Collins & Lanza, 2009, p. 186). GMM uses a latent mixture approach to model a single
outcomes development (intercept and slope). However, GMM typically requires welldistributed continuous variables and more observations with generally equal time intervals
(e.g., 4 or 5 times with 1 or 2 years between two points), thereby obtaining more accurate
slope estimates. The current available data from Add Health (only 3 times, one year apart in
the 2nd wave and 5 years later in the 3rd wave, ordinal response items) are not well-suited for
this approach.
Another rationale for choosing LCA over GMM is that currently available GMM software is
better suited for modeling univariate outcomes (e.g., delinquency) than simultaneously
estimating groups from multivariate outcomes (i.e., estimating the intercepts and slopes of
nonviolent and violent delinquency in the same model). A common solution is to use a twostep procedure where GMMs are separately fit for each outcome first; then, crosstabs are
calculated post hoc to describe patterns of classification based on each model (e.g.,
Fergusson & Horwood, 2002). By incorporating trichotomized nonviolent and violent
delinquency into the same model, however, LCA can investigate whether simultaneously
modeling these two subtypes of antisocial behavior can help identify otherwise hidden
patterns of antisocial behavioral trajectories, a main goal of the current study. As a result,
the LCA modeling of trichotomized data applied here traded the modeling of continuous
univariate outcomes and estimating slope possible in GMM for the advantages of
simultaneously modeling across the wave patterns of two aspects of antisocial behavior in
the same model.
Starting from a 2-class model with parameters constrained to be equal across genders, a
hundred iterations were run for each model using randomly generated starting values to
avoid local ML solutions. The models with the most frequent solutions were chosen.
Bayesian Information Criterion (BIC, the smaller the better) was primarily used to choose
the best trajectory number because it emphasizes parsimony, especially with large sample
sizes, together with Akaikes Information Criterion (AIC, the smaller the better) and entropy
(similar to R2). Using gender as a grouping variable, parameters were then allowed to vary
across gender after the model of best number of class was identified. Gender invariance was
examined by comparing the fit of gender in constrained vs. unconstrained models. In the
case of gender non-invariance, LCA was run in each gender group separately to identify
gender-specific trajectories. A final step used covariates to conduct multinomial logistic
regressions to predict different developmental trajectories in each gender separately.
Zheng and Cleveland
Page 7
Results
Gender-specific Latent Trajectory of Nonviolent and Violent Delinquency
As shown in the first column of each gender group in Table 3, females consistently reported
less nonviolent delinquency (e.g., 0.14 vs. 0.24 in the 1st wave) and violent delinquency
(e.g., 0.03 vs. 0.15 in the 3rd wave) than males throughout three waves. However, both
gender groups demonstrated similar patterns of desistence in nonviolent delinquency (0.14
to 0.05 for females; 0.24 to 0.16 for males) and violent delinquency (0.15 to 0.03 for
females; 0.26 to 0.15 for males) from adolescence to young adulthood. While there were no
obvious gender differences in neighborhood disadvantage, females reported slightly less
family support (M = 4.29, SD = 0.68) than did males (M = 4.46, SD = 0.53).
Based on BIC, a 5-class solution with parameters constrained to be equal across genders
(AIC = 1624.8, BIC = 2083.1, entropy = 0.69, 2 (1389) = 1488.8) was the best fit compared
to a 4-class model (AIC = 1811.9, BIC = 2175.8, entropy = 0.60) and a 6-class model (AIC
= 1545.1, BIC = 2097.7, entropy = 0.67). A nested model allowing parameters to be freely
estimated across genders provided a 5-class model with a 2(1329) of 1299.4, indicating
significant gender noninvariance (2diff(60) = 189.4, p < .001). This gender non-invariance
could have two sources: 1) the five classes have different characteristics and resulting
meanings across genders, and/or 2) there are gender-specific classes. To investigate the
sources of gender non-invariance, models were run separately by gender.
Female classesBased on BIC, females were best described by a 3-class solution (AIC
= 759.1, BIC = 989.8, entropy = 0.64) compared to a 2-class model (AIC = 875.0, BIC =
1026.7, entropy = 0.65) and a 4-class model (AIC = 724.9, BIC = 1034.5, entropy = 0.66).
Shown in the right half of Table 2, about 60% of females in the sample belonged to the first
classlow, defined by a high probability (p > .92) of reporting that they had never engaged
in antisocial behavior from adolescence to young adulthood. Means for nonviolent and
violent delinquency (see Table 3) suggested that low females reported the least antisocial
behavior across all three classes, substantially lower than females grand average.
Approximately one-third of females belonged to the second classdesister. In wave 1

desister had somewhat higher chances of reporting Never levels of nonviolent and violent
delinquency (p = .53 and .42, respectively) than other levels. Probabilities of Never in
wave 2 increased to .69 and .63, respectively, and approached total desistence by young
adulthood, with .85 and .91 probabilities of Never by wave 3. As a group, desister females
reported levels of antisocial behavior that were above the grand average of all females, but
gradually desisted to close to the overall average in young adulthood, although still higher
than the low group.
The third class, decliner, had a prevalence of 11%. Female decliners tended to report
Moderate/Serious levels of nonviolent delinquency during adolescence (p = .52 and .61,
respectively), but were more likely to report Never in young adulthood (p = .73). With
regard to violent delinquency, although females tended to report Never through three
waves, their probabilities of Never were much higher in young adulthood (p = .88) than in
adolescence (p = .46 and .51, respectively), and much lower than that of low. Means
indicated that decliner females actually reported the most antisocial behavior among all
classes through three waves, and consistently reported approximately twice the level of
nonviolent delinquency as violent delinquency. Despite a clear reduction in antisocial
behaviors by young adulthood, they failed to reach the same low levels of antisocial
behavior as desister females.
Zheng and Cleveland
Page 8
Male classesBased on BIC, males were best described by a 4-class solution (AIC =
914.3, BIC = 1281.4, entropy = 0.69) compared to a 3-class model (AIC = 1079.7, BIC =
1308.5, entropy = 0.72) and a 5-class model (AIC = 911.2, BIC = 1296.6, entropy = 0.71).
As shown in the left half of Table 2, the first class, low, appeared similar to its female
counterpart in reporting no engagement in antisocial behavior from adolescence to young
adulthood, with high probabilities (p > .82). Their means were the least among all male
classes, substantially lower than males average and comparable to females.
About a quarter of males belonged to the second classdesister. These respondents tended
to report Mild (e.g., p = .48 for nonviolent delinquency) in wave 1, but declined to the
point that Never was their most likely response in subsequent waves. The one exception to
the pattern of declining across waves was the lower than expected wave 2 violent
delinquency score. This score may have been due to uncertainty in class membership (see
the note on Table 3 for a more detailed explanation). Although their means were higher than
for the corresponding female class, the overall pattern of male desister is otherwise similar
to the pattern for female desister. Class means indicated a moderate level of antisocial
behavior compared to other male classes.
About 13% of the males belonged to the third classchronic, which showed a quite distinct
pattern from all other classes. Members of this group tended to respond Never for both
nonviolent and violent delinquency in young adulthood (p = .76 and .63, respectively),
Never for nonviolent delinquency (p = .69 and .57, respectively) in adolescence, but
Mild or Moderate/Serious for violent delinquency (p = .52 and .47 in 2nd wave). An
examination of their means showed that they consistently reported mild to moderate levels
of antisocial behavior from adolescence to young adulthood (the exception being a low
mean for 1st-wave nonviolent delinquency, which again may be due to class membership
uncertainty) without any clear pattern of desistence. A distinguishing aspect of their
antisocial behavior was a much higher rate of violent delinquency than the overall male
averagetwice as high as their nonviolent delinquency rate, which was similar to the full
sample mean across waves.
The last male class, decliner, had a prevalence rate of 12%. Male decliners tended to report
a Moderate/Serious level of nonviolent delinquency (p = .70 and .58, respectively) and
violent delinquency (p = .69 and .62, respectively) in adolescence but Never in young
adulthood (p = .50 and .54, respectively). They showed a pattern of desistence similar to that
for desister males but failed to reach the same low level of antisocial behavior as desister
males in young adulthood. Because they began with such high levels of antisocial behavior
and did not entirely desist in young adulthood, this group reported the most total antisocial
behavior across the three waves. Unlike decliner females, decliner males reported similar
levels of nonviolent and violent delinquency within each wave.
Predicting Trajectory Membership
Among males, those from more disadvantaged neighborhoods were less likely, compared to
low, to be in the desister class (.01 OR) but more likely to be in the chronic class (12.49 OR)
(see Table 4). Compared to being in the desister class, they were much more likely to be in
the chronic or decliner class (2090.99 and 384.29 OR). They were less likely to be in the
decliner class relative to chronic (.18 OR). Males with more family support were less likely
to be in the desister or decliner class (.63 and .35 OR) relative to being in the low class.
Those with more family support were less likely to be in the decliner class relative to being
in the desister (.56 OR) or chronic class (.38 OR).
Females from more disadvantaged neighborhoods were more likely to be in the desister
class but less likely to be in the decliner class (40.20 and .04 OR) relative to low (see Table
Zheng and Cleveland
Page 9
5). However, a greater level of neighborhood disadvantage predicted a lower likelihood of

being in the decliner compared to the desister class (.00 OR). Females with more family
support were less likely to be in the desister or decliner class relative to the low class (.65
and .31 OR). Similar to males, females with more family support were also less likely to be
in the decliner class compared to being in the desister class (.50 OR).
Discussion
The current study applied latent class analysis to data for 15 and 16 year olds during the
Wave I In-Home survey of the National Longitudinal Study of Adolescent Health to
determine developmental patterns of self-reported nonviolent and violent delinquency from
adolescence to young adulthood. Males and females demonstrated similarities and
differences in their respective antisocial behavior trajectories. Like some previous studies
(e.g., DUnger et al., 2002), several synonymous classes were found across genders (e.g.,
male desister and female desister). The low class appears to be the most similar class across
genders. Individuals in this class self-reported abstinence from antisocial behavior from
adolescence to young adulthood, and represented the majority of the population (e.g.,
Piquero et al., 2005). For this class the means for nonviolent and violent delinquency were
comparable between males and females across waves. Consistent with males greater
general involvement in antisocial behavior (Moffitt et al., 2001), however, the low class was
somewhat more prevalent among females than males (59% vs. 50%) (e.g., DUnger et al.,
2002; Odgers et al., 2008).
Two other classes, desister and decliner, were also fairly similar across genders. Both were
similarly prevalent across genders (25% and 29% for desister; 12% vs. 11% for decliner),
comparable with previous studies (e.g., Lahey et al., 2006; Moffitt & Caspi, 2001; Nagin,
1999). Both desister males and females reported antisocial behavior substantially above
each genders average in adolescence and gradually desisted to slightly above their genders
average in young adulthood. However, desister males consistently reported more antisocial
behavior than desister females across waves. In addition, desister females reported more
nonviolent delinquency than violent delinquency in adolescence, the opposite of the pattern
exhibited by desister males. Both decliner males and females reported the most antisocial
behavior among all classes, with means considerably higher than the averages of other
groups, including chronic. Their antisocial behaviors declined in young adulthood but did
not desist. An important gender difference involved decliner males, who reported
comparable means for nonviolent and violent delinquency within each wave. In contrast,
decliner females primarily reported engaging in nonviolent delinquency. Patterns exhibited
by both the decliner and desister classes are consistent with Moffitts adolescence-limited
pattern and resemble the high-desister/high-declining and lowdesister/moderate-declining
classes, respectively, identified by Nagin and colleagues (Nagin, 1999; Nagin & Tremblay,
2005b). They also resemble the high-rate adolescence-peaked and low-rate adolescencepeaked groups identified by DUnger and colleagues (2002).
The gender differences in the three groups (low, desister, and decliner) were relatively
subtle. In contrast, the chronic group was gender-specific, existing among males but not
females. DUnger et al. (2002) also identified chronic groups among males only. The allmale membership of the chronic group confirms that a gender may exhibit a distinct
developmental pattern. These males reported mild to moderate levels of antisocial behavior
without showing any clear pattern of desistence. An additional distinguishing characteristic
of this group is that their antisocial behavior was more characterized by violent than
nonviolent delinquency, with nearly the highest rates for the former but only moderate rates
for the latter. This class bears some resemblance to Moffitts life-course-persistent group
(1993), with about the same prevalence. However, unlike DUnger et al. (2002), who
Zheng and Cleveland
Page 10
identified low-rate and high-rate chronic groups, we only found one chronic group, perhaps
because we used self-reported antisocial behavior rather than crime records indicating more
serious antisocial behaviors.
Nonviolent and Violent Delinquency in Latent Trajectories
Specific consideration of nonviolent and violent delinquency is intrinsically linked to this
studys identification of the novel male class, chronic. Separately entering nonviolent and
violent delinquency also allowed differences between male and female decliners and
desisters to emerge. Decliner males reported similar levels of nonviolent and violent
delinquency, whereas decliner females reported more nonviolent delinquency than violent
delinquency. Desister males generally reported more nonviolent delinquency than violent
delinquency, whereas desister females were the opposite, reporting more adolescent violent
delinquency than nonviolent delinquency. It is also interesting to note that female desisters
may be more like male chronics than male desisters, as shown in DUnger et al. (2002).
These nuanced gender differences would not have been found had the study used an
aggregated antisocial behavior score, combining nonviolent and violent delinquency. Doing
so would have removed the opportunity to consider how these two aspects of antisocial
behaviorwhich have been found to have somewhat different, yet not unique, etiological
and developmental paths (Burt & Neiderhiser, 2009; Lacourse et al., 2002)characterize
individuals differently across time and genders.
Developmental Meanings and Prediction of Latent Trajectories

What is the developmental meaning of these groups? Desister seemed to correspond to
adolescent-limited delinquents. Chronic may be the best match for life-course-persistent
delinquents, at least with regard to their consistency. However, the decliner group, while
having reduced their antisocial behavior the most overtime, continued to report the highest
young adult levels of antisocial behavior. Thus, both the male chronics and male and female
decliners represent problematic developmental patterns.
Class prediction and identification are equally important because both can offer new
understandings of different trajectories. Key predictions included the following: being in a
more disadvantaged neighborhood made it more likely for a female or male to be in the
decliner than desister class, and more likely for a male to be a chronic delinquent; and
neighborhood quality distinguished low from chronic and decliner class for males and
decliner class for females. Both findings are consistent with social disorganization theory,
which argues that disadvantaged community and neighborhoods ineffective informal social
control increases adolescents antisocial behavior (Sampson, 1997; Shaw & McKay, 1969).
However, being in/from a disadvantaged neighborhood does not always increase the risk of
belonging to an antisocial group. Males from more disadvantaged neighborhoods were less
likely to be in the desister than low class. Females in disadvantaged neighborhoods were less
likely to be in the decliner compared to the low or desister class. These apparent protections
against being in the normative delinquent desister group may be linked to the meanings and
implications of delinquency in disadvantaged vs. adequate neighborhoods. It may be that,
among males for example, desisting delinquency is a privilege of living in an adequate
neighborhooda middle-class indulgence similar to emerging adulthood (see Osgood, Ruth,
Eccles, Jacobs, & Barber, 2005). In contrast, antisocial behavior in disadvantaged
neighborhoods may have greater long-term consequences. Similar behaviors may have very
different meanings and implications depending on the neighborhood. Thus, individuals from
more disadvantaged neighborhoods might actively and carefully avoid antisocial behavior
because their few chances for a bright future could be undercut by such behavior.
Zheng and Cleveland
Page 11
Family support generally acted as a protective factor against initiation or escalation, which
has been well documented in the literature (Moffitt, 1993, 2006; Patterson & StouthamerLoeber, 1984). In both gender groups, individuals with more family support were less likely
to belong to any antisocial class, desister or decliner, relative to low; and less likely to
escalate into more serious class, decliner, relative to desister. Thus, family support may
protect adolescents from deviant peers negative influences (Patterson & StouthamerLoeber, 1984). An interesting finding is that males with more family support were more
likely to be in the chronic than decliner class. The finding that decliners had the least family
supporta variable consistently related to risk behaviorsupports the argument that
decliner members were more problematic than chronic members. This finding underscores
the severe nature of the behaviors (more violence) exhibited by male decliners. However,
results could be driven in part by the large sample size, especially given the modest amount
of pseudo-R2 that they explained, so should be viewed cautiously.
Limitations
Several study limitations are worthy of attention and future study. One concerns
measurement. First, the Add Health data do not have information on antisocial behavior
between 1996 and 2002. It would be preferable to identify developmental patterns of
antisocial behavior with data from this time period. Sparse data coupled with the noncountable nature of the original scale items also made it difficult to adopt GMM estimation.
In addition, the current data did not measure early childhood antisocial behavior, preventing
identification of the childhoodlimited trajectory (Moffitt, 2006). Further, self-reports of
antisocial behavior, which have been shown to capture different forms from those found in
crime records data (Thornberry & Krohn, 2000), should be reviewed with caution for several
reasons, including the difficulty in making straightforward comparisons with some previous
studies (e.g., DUnger et al., 2002). Second, in order to use the same items across all three
waves, only a limited set of the total antisocial behavior items available in the Add Health
data were used. Another approach would be to use all items available across all waves
regardless of their inclusion across waves (e.g., Guo et al., 2008; Hagan & Foster, 2003).
Finally, it would have been possible to use different cut-offs to trichotomize between Mild
and Moderate/Serious levels. However, the current cut-offs successfully classified
respondents similarly into three levels for both nonviolent and violent delinquency, and also
distinguished between chronic and low classes. Similar results with previous studies add
validity to the current transformation. Another analysis using a stricter cutoff between mild
and moderate levels produced the same patterns of developmental trajectories.
Implications
The major contributions of this study are linked to the use of separate measures of
nonviolent and violent delinquency rather than a composite antisocial behavior score to
identify trajectories and to separate estimation of trajectories across genders. Findings
include differences in the prevalence of trajectory groups across genders, a gender-specific
class, and different patterns of the two subtypes of antisocial behavior across genders. These
results highlight the necessity of considering gender and subtypes of antisocial behavior in
developmental trajectory research (Fontaine et al., 2009). Predictions of trajectory
membership by adolescent neighborhood disadvantage and perceived family support reveal
expected and unexpected links between adolescents experiences and their developmental
patterns of antisocial behavior. Associations between group membership and family support
are straightforward: less support leads to more risk. Links between neighborhood
disadvantage and relative risk of membership in different trajectory groups are more subtle
and may reflect differences in both the potential risk inherent in neighborhood settings for
engaging in antisocial behavior as well as the implications of such behaviors across
Zheng and Cleveland
Page 12
neighborhood contexts. These findings send prevention research a clear message: prevention
efforts must be sensitive to different developmental needs and special attention must be paid
to developmental context and social settings (Tremblay, 2006).
Acknowledgments
This research uses data from Add Health, a program project directed by Kathleen Mullan Harris and designed by J.
Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill,
and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human
Development, with cooperative funding from 23 other federal agencies and foundations. Special acknowledgment
is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Information on how to obtain
the Add Health data files is available on the Add Health website (http://www.cpc.unc.edu/addhealth). No direct
support was received from grant P01-HD31921 for this analysis.
References

Archer, J.; Ct, S. Sex differences in aggressive behavior: A developmental and evolutionary
perspective. In: Tremblay, RE.; Hartup, WW.; Archer, J., editors. Developmental origins of
aggression. New York: Guilford Press; 2005. p. 425-443.
Broidy LM, Nagin DS, Tremblay RE, Bates JE, Brame B, Dodge KA, Vitaro F. Developmental
trajectories of childhood disruptive behaviors and adolescent delinquency: A six-site, cross-national
study. Developmental Psychology. 2003; 39:222245. [PubMed: 12661883]
Burt SA, Neiderhiser JM. Aggressive versus non aggressive antisocial behavior: Distinctive etiological
moderation by age. Developmental Psychology. 2009; 45:11641176. [PubMed: 19586186]
Collins, LM.; Lanza, ST. Latent class and latent transition analysis for the social, behavioral, and
health sciences. New York: Wiley; 2009.
DeLisi M. Scaling archetypal criminals. American Journal of Criminal Justice. 2001; 26:7792.
DUnger AV, Land KC, McCall PL. Sex differences in age patterns of delinquent/criminal career:
Results from Poisson latent class analyses of the Philadelphia cohort study. Journal of Quantitative
Criminology. 2002; 18:349375.
Fergusson DM, Horwood LJ. Male and female offending trajectories. Development and
Psychopathology. 2002; 14:159177. [PubMed: 11893091]
Fontaine N, Carbonneau R, Vitaro F, Barker ED, Tremblay RE. Research review: A critical review of
studies on the developmental trajectories of antisocial behavior in females. Journal of Child
Psychology and Psychiatry. 2009; 50:363385. [PubMed: 19236525]
Guo G, Roettger M, Cai T. The integration of genetic propensities into socialcontrol models of
delinquency and violence among male youths. American Sociological Review. 2008; 73:543568.
Hagan J, Foster H. S/hes a rebel: Toward a sequential stress theory of delinquency and gendered
pathways to disadvantage in emerging adulthood. Social Forces. 2003; 82:5386.
Harris, KM.; Halpern, CT.; Whitsel, E.; Hussey, J.; Tabor, J.; Entzel, P.; Udry, JR. The National
Longitudinal Study of Adolescent Health: Research design [WWW document]. 2009. URL: http://
www.cpc.unc.edu/projects/addhealth/design
Lacourse E, Ct S, Nagin DS, Vitaro F, Brendgen M, Tremblay RE. A longitudinal-experimental
approach to testing theories of antisocial behavior development. Development and
Lahey BB, Van Hulle CA, Waldman ID, Rodgers JL, DOnofrio BM, Pedlow S, Keenan K. Testing
descriptive hypotheses regarding sex differences in the development of conduct problems and
delinquency. Journal of Abnormal Child Psychology. 2006; 34:737755. [PubMed: 17033935]
Lanza ST, Collins LM, Lemmon D, Schafer JL. PROC LCA: A SAS procedure for latent class
analysis. Structural Equation Modeling. 2007; 14:671694. [PubMed: 19953201]
Maughan, B.; Rutter, M. Development and psychopathology: A life course perspective. In: Rutter, M.;
Bishop, DVM.; Pine, DS.; Scott, S.; Stevenson, J.; Taylor, E.; Thapar, A., editors. Rutters child
and adolescent psychiatry. (5th ed.). New York: Blackwell Publishing; 2008. p. 160-181.
Mazerolle P, Brame R, Paternoster R, Piquero A, Dean C. Onset age, persistence, and offending
versatility: Comparisons across gender. Criminology. 2000; 38:11431172.
Zheng and Cleveland
Page 13

Moffitt TE. Adolescence-limited and life-course-persistent antisocial behavior: A developmental

taxonomy. Psychological Review. 1993; 100:674701. [PubMed: 8255953]
Moffitt, TE. Life-course-persistent versus adolescence-limited antisocial behavior. In: Cicchetti, D.;
Cohen, DJ., editors. Developmental psychopathology, Vol. 3: Risk, disorder, and adaptation. (2nd
ed.). New York: John Wiley and Sons; 2006. p. 570-598.
Moffitt TE, Caspi A. Childhood predictors differentiate life-course persistent and adolescence-limited
antisocial pathways among males and females. Development and Psychopathology. 2001; 13:355
375. [PubMed: 11393651]
Moffitt, TE.; Caspi, A.; Rutter, M.; Silva, PA. Sex differences in antisocial behavior: Conduct
disorder, delinquency, and violence in the Dunedin longitudinal study. Cambridge: Cambridge
University Press; 2001.
Muthn B, Shedden K. Finite mixture modeling with mixture outcomes using the EM algorithm.
Biometrics. 1999; 55:463469. [PubMed: 11318201]
Nagin DS. Analyzing developmental trajectories: A semi-parametric group-based approach.
Psychological Methods. 1999; 4:139157.
Nagin, DS. Group-based modeling of development. Cambridge, MA: Harvard University Press; 2005.
Nagin DS, Tremblay R. Trajectories of boys physical aggression, opposition, and hyperactivity on the
path to physically violent and nonviolent juvenile delinquency. Child Development. 1999;
70:11811196. [PubMed: 10546339]
Nagin DS, Tremblay R. Developmental trajectory groups: Fact or a useful statistical fiction?
Criminology. 2005a; 43:873904.
Nagin DS, Tremblay R. What has been learned from group-based trajectory modeling? Examples from
physical aggression and other problem behaviors. ANNALS, AAPSS. 2005b; 602:82116.
Odgers C, Moffitt TE, Broadbent JM, Dickson N, Hancox RJ, Harrington H, Caspi A. Female and
male antisocial trajectories: From childhood origins to adult outcomes. Development and
Osgood, DW.; Ruth, G.; Eccles, JS.; Jacobs, JE.; Barber, BL. Six paths to adulthood: Fast starters,
parents without careers, educated partners, educated singles, working singles, and slow starters. In:
Settersten, RA., Jr; Furstenberg, FF., Jr; Rumbaut, RG., editors. On the frontier of adulthood:
Theory, research, and public policy. Chicago, IL: University of Chicago Press; 2005. p. 320-355.
Patterson GR, Stouthamer-Loeber M. The correlation of family management practices and
delinquency. Child Development. 1984; 55:12991307. [PubMed: 6488958]
Piquero AR, Brezina T. Testing Moffitts account of adolescent-limited delinquency. Criminology.
2001; 39:353370.
Piquero AR, Brezina T, Turner MG. Testing Moffitts account of delinquency abstention. Journal of
Research in Crime and Delinquency. 2005; 42:2754.
Sampson RJ. Collective regulation of adolescent behavior: validation results from eighty Chicago
neighborhoods. Journal of Adolescent Research. 1997; 12:227244.
Shaw, C.; McKay, H. Juvenile delinquency and urban areas. Chicago: University of Chicago Press;
1969. (Rev. ed.)
Silverthorn P, Frick PJ. Developmental pathways to antisocial behavior: The delayed-onset pathway in
girls. Development and Psychopathology. 1999; 11:101126. [PubMed: 10208358]
Silverthorn P, Frick PJ, Reynolds R. Timing of onset and correlates of severe conduct problems in
adjudicated girls and boys. Journal of Psychopathology and Behavioral Assessment. 2001;
23:171181.
Thornberry TP, Krohn MD. The self-report method for measuring delinquency and crime. Criminal
Justice. 2000; 4:3383.
Tremblay RE. Prevention of youth violence: Why not start at the beginning? Journal of Abnormal
Child Psychology. 2006; 34:481487. [PubMed: 16865544]
Moderate/Serious ( 0.6)
Delinquency
Delinquency
Violent
0.26 (0.09)
Mild (0 0.6)
1.00 (0.50)
0.21 (0.40)
0.33 (0.01)
Mild (0 0.33)
Overall
Never (0)
Moderate/Serious ( 0.33)
0.18 (0.38)
Overall
1.01 (0.50)
Never (0)
Nonviolent
M (SD)
Trichotomized Levels (Cut-offs)
Wave 1
6210
915
1094
4201
6214
777
1371
4066
0.14 (0.33)
0.98 (0.48)
0.33 (0.01)
0.15 (0.34)
0.99 (0.49)
0.27 (0.09)
M (SD)
Wave 2
5114
495
680
3939
5095
519
884
3692
0.08 (0.25)
0.94 (0.44)
0.34 (0.02)
0.10 (0.26)
0.87 (0.39)
0.26 (0.09)
M (SD)
Wave 3
4624
263
386
3975
4616
352
592
3672
Means (standard deviations) and group size of nonviolent and violent delinquency by trichotomized levels from adolescence to young adulthood
Table 1
Zheng and Cleveland
Page 14

0.12
0.03
0.92
0.95
0.05
0.00
0.82
0.89
0.06
Mild
Moderate/Serious
Never
Mild
Moderate/Serious
Never
Mild
Moderate/Serious
Never
Mild
Moderate/Serious
Never
Mild
Moderate/Serious
Violent
Delinquency
Wave 2
Nonviolent
delinquency
Wave 2
Violent
delinquency
Wave 3
Nonviolent
delinquency
Wave 3
Violent
delinquency
0.05
0.05
0.13
0.00
0.08
0.00
0.12
0.20
0.68
0.23
0.25
0.52
0.00
0.08
0.92
0.18
0.35
0.47
0.22
0.24
0.54
0.32
0.15
0.22
0.63
0.09
0.15
0.76
0.47
0.52
0.01
0.14
0.29
0.57
0.33
0.30
0.37
0.00
0.31
0.15
0.54
0.30
0.20
0.50
0.62
0.25
0.13
0.58
0.23
0.19
0.69
0.20
0.11
0.70
0.00
0.03
0.97
0.01
0.07
0.92
0.00
0.04
0.96
0.01
0.05
0.94
0.00
0.06
0.94
0.00
0.02
0.07
0.91
0.03
0.12
0.85
0.11
0.26
0.63
0.00
0.31
0.69
0.22
0.36
0.42
0.08
0.39
0.53
807
0.05
0.07
0.88
0.12
0.15
0.73
0.25
0.24
0.51
0.61
0.30
0.09
0.31
0.23
0.46
0.52
0.33
0.15
304
11.42%
Class 3
decliner
Estimated proportion of a specific class in the population. For example, about 25% of males in the population belong to desister.
Calculated by assigning each individual to the class with the highest class membership probability. Thirteen males and 9 females could not be classified into any class because they have equal probabilities
of belonging to more than one specific class.
Note.
0.84
Never
0.08
0.92
Moderate/Serious
0.26
Wave 1
0.31
0.69
Delinquency
0.48
0.04
0.11
Mild
2075
29.46%
Class 2
desister
Never
314
59.12%
Class 1
Low
Wave 1
410
11.73%
Class 4
decliner
Nonviolent
0.20
720
13.28%
Class 3
chronic
Conditional Item Response Probabilityc

0.89
1592
Group sizeb
24.93%
Class 2
desister
Females (n = 3,195)
Response
Item
50.06%
Prevalencea
Class 1
low
Males (n = 3049)
Gender-specific latent class model using nonviolent and violent delinquency and from adolescence to young adulthood
Table 2
Zheng and Cleveland
Page 15
The probability of an average hypothetical individual who belongs to a specific class endorsing different levels of each item. For example, on average, males in low had a probability of .84 responding
Never for wave 1 violent delinquency. To facilitate interpretation, item response probabilities > .65 were regarded as clearly differentiating between levels (bolded). Item response probabilities > .45 and
< .65 were regarded as moderately differentiating (italicized and bolded).
Zheng and Cleveland

Page 16

0.14
(0.10)
4.54
(0.47)
0.19
(0.39)
0.18
(0.39)
0.16
(0.32)
0.15
(0.34)
0.15
(0.11)
4.46
(0.53)
Wave 2 nonviolent
delinquency
Wave 2 violent
delinquency
Wave 3 nonviolent
delinquency
Wave 3 violent
delinquency
Neighborhood
disadvantage
Family support
4.40
(0.51)
0.10
(0.07)
0.18
(0.33)
0.27
(0.40)
0.01
(0.06)
0.27
(0.36)
0.31
(0.46)
0.49
(0.45)
Class 2
desister
4.56
(0.44)
0.19
(0.12)
0.24
(0.44)
0.12
(0.33)
0.59
(0.38)
0.19
(0.34)
0.45
(0.46)
0.08
(0.13)
Class 3
chronic
4.21
(0.67)
0.15
(0.11)
0.39
(0.55)
0.36
(0.47)
0.84
(0.61)
0.83
(0.69)
0.95
(0.68)
0.97
(0.62)
Class 4
decliner
4.29
(0.68)
0.15
(0.11)
0.03
(0.11)
0.05
(0.18)
0.10
(0.26)
0.11
(0.28)
0.15
(0.32)
0.14
(0.30)
All
females
4.41
(0.59)
0.15
(0.11)
0.01
(0.06)
0.03
(0.11)
0.01
(0.06)
0.02
(0.10)
0.02
(0.08)
0.02
(0.07)
Class 1
low
4.25
(0.64)
0.21
(0.14)
0.05
(0.15)
0.07
(0.20)
0.23
(0.32)
0.11
(0.14)
0.38
(0.37)
0.20
(0.26)
Class 2
desister
Females
3.80
(0.81)
0.11
(0.08)
0.07
(0.22)
0.17
(0.38)
0.35
(0.52)
0.77
(0.47)
0.43
(0.59)
0.75
(0.53)
Class 3
decliner
Note. To further clarify the meaning of different classes, respondents were assigned to classes based on their highest class membership probability. However, cautions have to be made because class
membership uncertainty disappeared when respondents were assigned to one specific class (rather than having a membership probability for each class) (see Collins & Lanza, 2009). Therefore, means in
each class merely served for the ease of class labeling and interpretation.
0.06
(0.21)
0.07
(0.18)
0.01
(0.07)
0.02
(0.06)
0.06
(0.17)
0.26
(0.46)
Wave 1 violent
delinquency
0.02
(0.07)
0.24
(0.43)
Class 1
low
Wave 1 nonviolent
delinquency
All
males
Males
Means (standard deviations) of nonviolent and violent delinquency and covariates by gender and class
Table 3
Zheng and Cleveland
Page 17

0.63
(0.45, 0.87)
0.01
(0.00, 0.07)
OR
95% CI
0.09
2.53
Ba
0.92
(0.63,1.32)
12.49
(3.13, 49.91)
OR
95% CI
0.38
7.65
Bb
Chronic
1.47
(0.93, 2.31)
2090.99
(129.64, 33726.00)
OR
95% CI
1.04
0.83
Ba
Males (n = 2,982)
0.35
(0.28, 0.44)
2.30
(0.66, 7.98)
OR
95% CI
0.58
5.95
Bb
0.56
(0.40, 0.80)
384.29
(24.99, 5909.80)
OR
95% CI
Decliner
0.96
1.69
Bc
0.38
(0.26, 0.55)
0.18
(0.04, 0.85)
OR
95% CI
The chronic class served as the reference group.
The desister class served as the reference group.
The low class served as the reference group.
helped to explain 2.4% more error in the male model identification in Table 2. McFaddens pseudo-R2 compared to an intercept-only model is 0.6%.
Note. OR = odds ratio; CI = confidence interval. All covariate tests were significant at p< .001. Sixtyseven males were excluded because of missing values in covariates. Inclusion of the two covariates
0.47
5.14
Neighborhood
disadvantage
Family support
Ba
Variables
Desister
Multinomial logistic regression predicting male class membership with adolescent neighborhood disadvantage and family support
Table 4
Zheng and Cleveland
Page 18

0.65
(0.51, 0.82)
40.20
(12.40, 130.33)
OR
95% CI
1.16
3.33
Ba
0.31
(0.26, 0.38)
0.04
(0.00, 0.33)
OR
95% CI
Females (n = 3,134)
0.69
6.89
Bb
Decliner
0.50
(0.38, 0.67)
0.00
(0.00, 0.01)
OR
95% CI
The desister class served as the reference group.
The low class served as the reference group.
helped to explain 3.1% more error in the female model identification in Table 2. McFaddens pseudo-R2 compared to an intercept-only model is 1.3%.
Note. OR = odds ratio; CI = confidence interval. All covariate tests were significant at p< .001. Sixty-one females were excluded because of missing values in covariates. Inclusion of the two covariates
0.44
3.69
Neighborhood
disadvantage
Family support
Ba
Variables
Desister
Multinomial logistic regression predicting female class membership with adolescent neighborhood disadvantage and family support
Table 5
Zheng and Cleveland
Page 19

NIH Public Access

Uploaded by

Copyright:

Available Formats

NIH Public Access

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

NIH Public Access

Uploaded by

Copyright:

Available Formats

NIH Public Access

NIH-PA Author Manuscript

Published in final edited form as:

Identifying Gender-Specific Developmental Trajectories of

NIH-PA Author Manuscript

Most research examining gender differences in developmental trajectories of antisocial behavior

NIH-PA Author Manuscript

Zheng and Cleveland

NIH-PA Author Manuscript

Developmental Trajectories of Antisocial Behavior

NIH-PA Author Manuscript

In addition to the life-course-persistent and adolescence-limited groups, which have been

Gender Differences in the Developmental Trajectories of Antisocial

J Adolesc. Author manuscript; available in PMC 2014 April 01.

Zheng and Cleveland

NIH-PA Author Manuscript

(e.g., neuropsychological deficits, dysfunctional family environment) with male early-onset/

Previous Limitations and the Scope of the Current Study

NIH-PA Author Manuscript

NIH-PA Author Manuscript

J Adolesc. Author manuscript; available in PMC 2014 April 01.

Zheng and Cleveland

NIH-PA Author Manuscript

NIH-PA Author Manuscript

NIH-PA Author Manuscript

Nonviolent and violent delinquencyThe nonviolent and violent delinquency

J Adolesc. Author manuscript; available in PMC 2014 April 01.

Zheng and Cleveland

NIH-PA Author Manuscript

NIH-PA Author Manuscript

NIH-PA Author Manuscript

J Adolesc. Author manuscript; available in PMC 2014 April 01.

Zheng and Cleveland

NIH-PA Author Manuscript

NIH-PA Author Manuscript

NIH-PA Author Manuscript

J Adolesc. Author manuscript; available in PMC 2014 April 01.

Zheng and Cleveland

NIH-PA Author Manuscript

NIH-PA Author Manuscript

NIH-PA Author Manuscript

Approximately one-third of females belonged to the second classdesister. In wave 1

J Adolesc. Author manuscript; available in PMC 2014 April 01.

Zheng and Cleveland

NIH-PA Author Manuscript

NIH-PA Author Manuscript

NIH-PA Author Manuscript

Zheng and Cleveland

NIH-PA Author Manuscript

5). However, a greater level of neighborhood disadvantage predicted a lower likelihood of

NIH-PA Author Manuscript

NIH-PA Author Manuscript

J Adolesc. Author manuscript; available in PMC 2014 April 01.

Zheng and Cleveland

NIH-PA Author Manuscript

NIH-PA Author Manuscript

Developmental Meanings and Prediction of Latent Trajectories

NIH-PA Author Manuscript

J Adolesc. Author manuscript; available in PMC 2014 April 01.