NIH Public Access

Author Manuscript
Aggress Behav. Author manuscript; available in PMC 2013 July 18.
Published in final edited form as:
NIH-PA Author Manuscript

Aggress Behav. 2011 ; 37(1): 98–106. doi:10.1002/ab.20370.

Risk and Protective Factors Associated with Patterns of

Antisocial Behavior among Nonmetropolitan Adolescents
Christian M. Connella,*, Emily C. Cooka, Will M. Aklinb,1, Jeffrey J. Vanderploega, and
Robert A. Brexc
aDepartment of Psychiatry, Yale University School of Medicine

bDepartment of Psychiatry and Behavioral Sciences, Johns Hopkins University School of

Medicine
cNortheast Communities Against Substance Abuse (NECASA) Dayville, Connecticut

Abstract
The current study used latent class analysis (LCA) to identify patterns of antisocial behavior
NIH-PA Author Manuscript

(ASB) in a sample of 1,820 adolescents in a nonmetropolitan region of the Northeast. Self-

reported antisocial behaviors including stealing, fighting, damaging property, and police contact
were assessed. Latent class analysis identified four classes of ASB including a non-ASB class, a
mild, a moderate, and a serious ASB class. Multinomial logistic regression indicated that parent-
child relationships served as a protective factor against engaging in ASB and peer, school, and
community risk and protective factors differentiated mild patterns of ASB from more intense
patterns of involvment. These findings suggest utility in using a latent class analysis to better
understand predictors of adolescent ASB to inform more effective prevention and intervention
efforts targeting youth who exhibit different patterns of behavior.

Keywords
problem behavior; aggression; adolescence; risk and protective factors; latent class analysis

Antisocial behavior patterns (e.g., involvement in aggressive or delinquent behavior) often

begin to emerge during adolescence and may contribute to adjustment problems during
adolescence and into adulthood (e.g., decreased educational attainment, increased mental
NIH-PA Author Manuscript

health problems; French & Conrad, 2005; Hair, Park, Ling, & Moore, 2009). Treatment of
these behaviors and resulting consequences represents an enormous cost to society (e.g.,
crime, incarceration, reduced job productivity, prevention, and treatment efforts; Cohen,
1998). Despite evidence that rates of adolescent problem behaviors have increased across
urban and non-urban environments, the majority of research studies on adolescent risk
behaviors (including ASB) focus on urban populations (Fahs et al., 1999; Osgood &
Chambers, 2003). Relatively few studies have examined rates and predictors of ASB
specifically among rural and nonmetropolitan adolescent populations. Such research may
provide insight into the prevention needs for youth in such communities. Given the potential
negative outcomes associated with ASB in adolescence and the cost to society, it is critical
that more effective programs aimed at preventing or reducing involvement in delinquent
activities be developed. These efforts would be strengthened if research identified particular

Correspondence concerning this article should be addressed to Christian M. Connell, Ph.D., Yale University School of Medicine, The
Consultation Center, New Haven, CT 06516. christian.connell@yale.edu.
1Will M. Aklin is now at Division of Clinical Neuroscience & Behavioral Research, National Institute on Drug Abuse, Bethesda,
Maryland
 Connell et al. Page 2

subgroups of problem behavior associated with specific contextual risk and protective
characteristics in a given population (Harris & Jones, 1999).
NIH-PA Author Manuscript

Research and theory suggest that ASB among adolescents may be better represented by a
person-centered perspective that identifies distinct subgroups of youth based upon specific
patterns of behavior, rather than treating ASB as a single continuum within the adolescent
population (Wiesner & Windle, 2004). Research using a person-centered approach has
generally followed one of two forms: (1) identification of subgroups based upon growth in
the level of involvement over time, or (2) identification of subgroups based upon types of
ASB behavior (e.g., aggression, property offense, etc.). More research has focused on
identification of different groups of youth based upon growth of involvement over time.
Moffitt and colleagues (2006, 2008) and others (e.g., Fergusson, Horwood, & Nagin, 2000;
Windle, 2000) have identified four patterns of ASB using this approach – life-course
persistent, adolescent-onset, childhood limited, and low-to-no antisocial behavior.

Fewer studies have examined differential patterns of ASB using the second approach in
which groups are defined based on the types of behaviors in which individuals engage. Of
those that have, the majority used official court records or clinical symptom-type behaviors
for classification. Studies using juvenile offense data and those relying on clinical samples
generally identified a group of adolescents engaging primarily in status and property
offenses, and an aggressive/violent and delinquent behavior group (Frick et al., 1993;
NIH-PA Author Manuscript

Odgers et al., 2007; Soothill, Francis, Ackerley, & Humphreys, 2008). Extension of this
research to school-based samples of adolescents would increase understanding of influences
on divergent patterns of ASB among normative groups of youth.

Understanding contextual influences on ASB is an important step in developing more

targeted prevention efforts for community samples. The social development model
(Catalano & Hawkins, 1996) provides a useful framework for organizing contextual
influences on problem behavior. The model draws upon social learning, social control, and
differential association theories to identify risk and protective factors across family, peer,
school, and community contexts that are associated with adolescent problem behavior.
Important influences within the model include attachment to prosocial or antisocial
influences (e.g., parents, peers), bonding with normative institutions (e.g., school), and
opportunities to engage in positive or negative behavior. Within this framework parent-child
relationships, affiliation with deviant peers, attachment to school, and neighborhood disorder
are important risk and protective factors that have been associated with adolescent ASB
(Ardelt & Day, 2002; Fergusson & Horwood, 2002; Najaka, Gottfredson, & Wilson, 2001;
Windle, 2000).
NIH-PA Author Manuscript

Family, peer, school, and neighborhood risk and protective factors may be differentially
associated with divergent patterns of ASB. The social development model hypothesizes that
increases in severity and continued offending throughout adolescence are a result of
exposure to influences outside the family that reinforce ASB (Catalano & Hawkins, 1996).
Thus, during adolescence peer, school, and neighborhood factors might have a stronger
relationship with more severe patterns of ASB (i.e., number of behaviors and/or severity of
the type of behavior) than parenting. Few studies have examined the relationship between
risk and protective factors and patterns of ASB using person-centered approaches and those
that have focused on the effect of risk and protective factors on various trajectories of ASB
(Odgers et al., 2008; Wiesner & Silbereisen, 2003). Furthermore, studies using a latent class
approach generally have ignored the effects of school and neighborhood on various types of
ASB. However, research relying on a variable-centered approach has found that after taking
into account parenting, problems at school and in the neighborhood are associated with an
increase in ASB (Ayers, Williams, Hawkins, Peterson, & Abbott, 1999, Beyers, Bates,

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 3

Pettit, & Dodge, 2003). Using LCA to extend previous research can inform if prevention
efforts should target specific factors that are salient to different patterns of ASB.
NIH-PA Author Manuscript

The purpose of the current study was to address these gaps in the existing research and
examine patterns of ASB and predictors of those patterns in a school-based sample of
adolescents living in a nonmetropolitan area. We hypothesized that adolescents would
cluster into a minimum of three distinct classes of ASB including a group engaged in
minimal levels of problem behavior and additional groups differentiated both by degree and
type of problem behaviors. Further, based upon previous research, we hypothesized that
proximal social influences (e.g., family and peer relationships) would play a critical role in
differentiating risk for both initiation and degree of involvement in ASB, while more distal
social influences (e.g., bonding with school, exposure to neighborhood disorder) would
likely play a more central role in risk for escalated patterns of ASB.

Method
Site
Northeast Communities Against Substance Abuse (NECASA) is a community-based
coalition representing the 21 towns and municipalities in northeast Connecticut. The region
comprises approximately 171,000 people and covers an area of approximately 740 square
miles. With an average population density of approximately 231 individuals per square mile,
NIH-PA Author Manuscript

the region is among the least populated areas of the state, though it includes both small town
and rural locales. Many towns in the region have fewer than 5,000 inhabitants. The median
household income in the region is $58,199, with 6.4% of the population below the poverty
line.

Participants
Participants were 1,820 students from 5 area public/non-vocational high schools and 9
middle schools serving the region (a total of 14 schools). Approximately 42% of
respondents were in the 8th grade, 32% were in the 9th grade, and 26% were in the 10th
grade, and the average age of respondents was 14.1 years (SD = 1.0 years). Fifty-four
percent of respondents were female, and the racial and ethnic composition was as follows:
83% Caucasian, 9% Hispanic/Latino, 2.5% African American, 1% Asian/Pacific Islander,
1% Native American, and 2.5% Other.

Measures
The survey instrument in the present study was based on a self-report survey developed for
statewide epidemiological assessment of adolescent substance use patterns (Ungemack,
NIH-PA Author Manuscript

Cook, & Damon, 2001) and draws on methods employed in a number of national surveys
(e.g., Monitoring the Future). The survey assesses a range of adolescent risk behaviors (e.g.,
substance use, antisocial behavior) as well as a wide range of risk and protective factors
associated with these behaviors. In addition to the specific subscales used in the present
analyses (described below), youth provided basic demographic information including age,
gender, and race/ethnicity.

Antisocial behavior—Past year involvement in antisocial behaviors was assessed using

an 11-item scale in which youth indicated the frequency of involvement in a range of
behaviors in the following domains: aggression-related, property-related, and drug-related
behaviors and consequences related to problem behavior. Items were rated on the following
scale: 0 (no history), 1 (one to two times), and 2 (three or more times). Because prevalence
of frequent involvement in behaviors was relatively low, these items were dichotomized to
represent: 0 (no history of the specified behavior) and 1 (one or more times engaged in the

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 4

specified behavior). A total score also was created by summing affirmative responses (α
=0.82). This score was used to compare identified classes on overall rates of antisocial
behavior.
NIH-PA Author Manuscript

Risk and protective factors—Risk and protective factors associated with problem
behavior classes were selected to represent individual, family, peer, school, and community
domains.

Individual-level characteristics included in the model as covariates were age and gender.
Age was centered on the mean and gender was dummy coded 0 (female) and 1 (male).

Family-level risk was measured by a 10-item scale that assessed youth ratings of parental
supervision and parent-adolescent relationship quality. Sample items include, “when I am
away from home, my parent/guardian(s) knows where I am and who I am with,” and “my
parent/guardian(s) give me help and support when I need it.” Items were rated on a 4-point
Likert scale from 1(definitely not true) to 4 (definitely true). Responses were averaged and
higher scores indicated more positive parent-adolescent relationships (α =0.84).

Peer-level risk was assessed by youth ratings of substance use among their close friends.
This scale was created by taking the mean for five items in which respondents indicated the
number of close friends that used alcohol, tobacco, and other drugs. Items were scored on a
NIH-PA Author Manuscript

four-point scale (none to most) and had good internal consistency (α = 0.84).

School-level risk was assessed by a 4-item scale reflecting their attitudes towards school and
education. Sample items on this scale included, “it is important for me to get good grades,”
and “I try hard to do good work at school.” Items were rated on a 4-point scale 1 (definitely
not true) to 4 (definitely true) and had good internal consistency (α =0.78).

Finally, neighborhood-level risk was assessed by a 13-item scale reflecting perceived

neighborhood safety and collective efficacy. Sample items included “you’ve got to be tough
to get along in my neighborhood,” and “people in my neighborhood are trying to keep kids
off drugs.” Items were rated on a 4-point scale from ‘definitely not true’ to ‘definitely true.’
The thirteen items were averaged and higher scores indicated more neighborhood disorder
(α =0.81).

Procedure
Surveys were conducted at 14 area middle and high schools from the spring of 2006 to the
winter of 2008. Passive consent procedures were approved by school administrations, with a
letter sent home to parents describing the survey and providing parents and youth with the
NIH-PA Author Manuscript

opportunity to opt-out of participation. Surveys were administered in a group format, with

youth completing surveys on their own and marking responses on forms that included no
identifying information to ensure confidentiality and anonymity of responses. Surveys were
checked for consistency of responding, and a small number of surveys (fewer than 3%) were
discarded because of failure to complete the survey, evidence of over-reporting of problem
behavior (e.g., indicating use of a fictitious drug included in the survey), or inconsistent
response patterns (e.g., lack of consistency among use variables within substances).

Data Analysis
Latent class analysis (LCA; Lubke & Muthén, 2005; McCutcheon, 1987) with Mplus 5.1
(Muthén & Muthén, 2006) was used to identify sub-groups (i.e., classes) based upon
similarity of responses to measured variables, and also to assess the effect of covariates on
class membership (Nylund, Asparouhov, & Muthén, 2007). LCA has advantages over
alternative approaches to classification (e.g., cluster analysis, a priori assignment based upon

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 5

observed data) including use of model-based methods for estimating population

characteristics derived from sample data and formal statistical procedures for determining
the number of classes, (Magidson & Vermunt, 2002; Muthén & Muthén, 2006; Nylund et
NIH-PA Author Manuscript

al., 2007). LCA provides estimates of class membership probabilities (e.g., problem
behavior classes), and behavioral probability estimates within class (Lanza, Flaherty, &
Collins, 2003). Covariates are incorporated into models with more than one latent class
using multinomial logistic regression models based upon posterior-probability estimates of
latent class membership. Missing data are addressed via full information maximum
likelihood methods.

We conducted a series of models, beginning with a one class model and subsequently
incorporating additional classes, and compared fit indices to determine the optimal solution.
For models with more than one class, we compared fit for unrestricted models, in which
class problem behavior probabilities were freely estimated, as well as fit for restricted
models that imposed a zero-behavior class (i.e., a “non-ASB” class). Latent class structure
was chosen based upon existing research recommendations of statistical model fit, as well as
interpretability of the classes. Optimal fit was indicated by a low BIC and significant LMR-
LRT (Muthén & Muthén, 2006; Nylund et al., 2007).

Risk and protective factors were incorporated into the final model using multinomial logistic
regression based upon posterior-probability estimates of latent class membership. A similar
NIH-PA Author Manuscript

procedure was used to compare total scores on the antisocial behavior scale across groups
and assess the extent to which class structure was a function of differences in overall level of
antisocial activity. Finally, because data were clustered within 14 schools we conducted an
additional set of analyses in which school was treated as a fixed effect in the model. These
analyses provided nearly identical results to the model without school effects, and none of
the school-level identifier effects were statistically significant. Thus, the school-level fixed
effects model was not included in the reported results.

Results
Prevalence rates for adolescents’ report of each of the antisocial behaviors during the past
year are presented in Table 1. On average youth indicated having been involved in 1.3 (s.d.
= 2.0) behaviors during the past 12 months. Involvement in a physical fight was the most
frequently endorsed item (24.4%), followed by threatening someone (24.2%) and stealing
something worth less than $50 (21.2%).

Latent class models were fit to the data reflecting self-reported ASB involvement. As
indicated in Table 2, the BIC and LMR-LRT suggested a four-class model with a forced
NIH-PA Author Manuscript

non-ASB class (i.e., a four-class, restricted model) provided a significantly better fit to the
data than other models.

Conditional probabilities of reporting any history of ASB for the four-class, restricted LCA
are depicted in Figure 1. Class 1 (‘non-ASB’, 37% of youth) was restricted to adolescents
with no reported ASB involvement. Class 2 (‘mild ASB’, 45% of youth) reported a
relatively low probability of aggressive behaviors (e.g., threatening, fighting) and stealing,
with very low probability of engaging in any other behaviors. Class 3 (‘moderate ASB’,
12% of youth) reported higher probabilities of engaging in a range of aggressive behaviors
(e.g., threatening, fighting, attacking someone with intent to harm) and property-related
offenses (e.g., stealing, damaging property), but reported little-to-no involvement in drug-
related activities or trouble resulting from antisocial behavior (e.g., suspension, police
contact, arrest). Finally, Class 4 (‘serious ASB’, 6% of youth) reported comparable
probabilities of engaging in aggressive behaviors to those of class three (as well as increased

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 6

probability of bringing a weapon on school grounds). Property-related offenses were also

elevated among this group. This class also was more likely to indicate drug involvement
(i.e., buying or selling drugs) and consequences of ASB (e.g., suspension, police contact,
NIH-PA Author Manuscript

arrest).

Post-hoc comparisons of total antisocial behavior scores revealed significant differences in

the number of affirmative responses across all groups in both overall and pairwise
comparisons. The non-ASB class evidenced the lowest rate (M = 0.04, s.d. = 0.26), followed
by the mild ASB class (M = 1.04, s.d. = 1.16), moderate ASB class (M = 3.94, s.d. = 1.95),
and serious ASB class (M = 7.39, s.d. = 2.51).

Multinomial logistic regression analyses were conducted to estimate the association between
hypothesized individual, family, peer, school, and community domains and class
membership. The odds ratios resulting from these analyses are presented in Table 3.

Individual-level influences had a somewhat inconsistent relationship with ASB class

membership. As anticipated, males were at greater risk than females of being classified in
the mild, moderate, or serious ASB classes than the non-ASB class, and were also at greater
risk of being classified in the serious ASB class than either the mild or moderate ASB
classes. Child age was not significantly related to ASB class membership. The only
significant finding for these indicators was that older youth were more likely to be classified
NIH-PA Author Manuscript

in the serious ASB class relative to the mild and moderate ASB classes.

Family and peer influences also were significant predictors of ASB class membership.
Youth reporting more positive parent-child relationships were significantly less likely to be
classified in the mild, moderate, or serious ASB, although parent-child relationships did not
differentiate among classes engaged in antisocial behavior. Peer substance use had a more
consistent negative effect on youth problem behavior, differentiating non-problem involved
youth from each problem behavior class and also differentiating among classes engaging in
ASB.

Broader contextual influences also played a role in differentiating among ASB classes. Both
school attachment and neighborhood disorder played a significant and substantial role in
differentiating moderate and serious ASB classes from non- and mild ASB classes.
Attachment to school served as a potential protective factor – reducing risk of identification
in the moderate and serious ASB classes relative to the non- and mild ASB classes.
Conversely, neighborhood disorder was associated with increased risk of membership in the
moderate and serious problem behavior classes than both the non- and mild ASB classes.

A final set of models were conducted to examine whether gender differences might be found
NIH-PA Author Manuscript

with respect to either the expression of ASB or the relation of risk factors to specific patterns
of behavior. The first set of analyses treated gender as a grouping variable to see whether the
class structure was consistent for males and females. Analyses revealed that a four-class
model fit well for both girls and boys and that the item-response probabilities did not differ
significantly by gender. A second set of analyses included gender-by-covariate interaction
effects in the multinomial logistic models, but no significant interactions effects were found
suggesting that the effects of parents, peers, school attachment, and neighborhood disorder
did not differ as a function of adolescents’ gender.

Discussion
The aim of this study was to examine heterogeneity in the expression of antisocial behavior
patterns among adolescents in a nonmetropolitan region and investigate the influence of
contextual risk and protective factors to identified patterns of behavior. Four patterns of

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 7

behavior were identified representing variation in the probability of engaging in a range of

antisocial activities including aggressive-, property-, drug-, and consequence-related
behavior. Individual, family, peer, school, and community risk and protective factors
NIH-PA Author Manuscript

differentiated membership in identified classes. Results are discussed in terms of the

implications for prevention research

Findings reveal a potential progression in terms of the range and type of ASB among
adolescents. A minimal level of involvement in problem behavior is the overwhelming norm
among this population of nonmetropolitan youth in the transition from middle to high
school. These results are consistent with previous research conducted with state and national
data. Our observed rate of involvement in a physical fight (24.4%), for example, is slightly
lower than the state and national estimates (28.3% and 31.5%, respectively; CDC, 2010),
though these estimates include older adolescents than our sample population. Of youth
involved in higher rates of ASB, approximately 12% were identified in a moderate antisocial
behavior class and 6% were classified in a serious antisocial behavior class. These groups
were marked by significantly higher probabilities of involvement in aggressive behaviors,
property-related offenses, and probability of consequence-related behaviors; though the
serious ASB class evidenced significantly higher rates of both drug-related behavior and
negative consequences including policy contact and arrest.

It is interesting to note that this pattern of findings is also reflected in the significant
NIH-PA Author Manuscript

differences in total antisocial behavior scores across groups. Although this relationship may
raise concern that the identified groups simply reflect differences in overall level of
antisocial behavior, one would expect that if this were the case individual item response
probabilities within classes would be more evenly distributed across behaviors (e.g., 0.1 for
the mild ASB class, 0.4 for the moderate ASB class). Instead, the observed patterns suggest
that as youth increase their level of antisocial activity they do so in a consistent manner,
beginning with mild aggression or theft of non-valuable items, followed by escalation of
other antisocial activities. These might include an expansion to a broader range of
aggressive behaviors and property-related offenses, but eventually expand to a broader range
of risky behaviors (e.g., drug buying or selling), and result in negative consequences from
such activities. This pattern is consistent with research demonstrating that distinct types of
problem behaviors follow different trajectories over the middle school years, with the early
emergence of aggressive behaviors predicting growth in other delinquent behaviors or drug
use (Farrell, Sullivan, Esposito, Meyer, & Valois, 2005). A stage-sequential pattern of
escalation, such as the one suggested by these results, could guide development of
prevention and intervention efforts with more universal strategies focused on general
reduction of aggressive behavior to reduce escalation of antisocial behavior, and more
indicated interventions focused on youth engaged in a broader range of serious patterns of
NIH-PA Author Manuscript

ASB. Further research using Latent Transition Analysis, a longitudinal variant of LCA
(Lanza et al., 2003), would provide a means of better understanding youths’ progression
through these identified stages of problem behavior.

In terms of particular risk and protective factors that may be targeted within an prevention
and intervention framework, these results highlight the influence of social and contextual
factors on risk of engaging in more serious patterns of ASB. Closer parent-child relations
serve a protective function against engaging in ASB, differentiating members of the non-
ASB class from those in the mild, moderate, and serious ASB classes. However, ratings of
parent-child relations did not differentiate among ASB-involved classes. Peer behavior, on
the other hand, was strongly associated with both initiation and intensity of ASB
involvement. Furthermore, school and neighborhood influences differentiated normative
patterns of ASB from more intense patterns of involvement. These findings are consistent
with previous theory and research which suggest socialization influences outside the family

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 8

may have a stronger association with more intense patterns of ASB involvement during
adolescence (Ayers et al., 1999; Catalano & Hawkins, 1996).
NIH-PA Author Manuscript

The use of person-centered approaches has direct implications for development of effective
prevention/intervention programs. Our findings, for example, suggest that focusing on
parent-child relationships and supervision may be an effective universal prevention strategy
that reduces risk for initiating ASB, but is not likely to have a significant direct effect on
reducing intensity of involvement once youth begin to engage in such behaviors. Prevention
and intervention programs that target association with deviant peer groups likely are critical
to reduce risk for escalation to more serious patterns of ASB. An approach targeting both
parents and peers is also likely to be maximally effective, since there is evidence that
parenting practices and parent-child relationships have an indirect effect on antisocial
activity through effects on opportunities to socialize with deviant peer groups (Laird, Criss,
Pettit, Dodge, & Bates, 2008). Programs that rely on such an approach (e.g., the Adolescent
Transitions Program, Guiding Good Choices) have been shown to be particularly effective
in this regard. Furthermore, providing youth with a sense of attachment to school, as well as
intervening at the community level are important intervention strategies to prevent youth
from escalating to higher levels of ASB.

This study contributes to our understanding of patterns and predictors of ASB in a

nonmetropolitan sample of adolescents, however, some limitations should be noted. Most
NIH-PA Author Manuscript

notably, the current study relied solely on adolescent report to assess both risk and protective
factors and ASB. The validity of adolescent self-report of potentially sensitive information
(e.g., ASB) has been criticized (Brown & Zimmerman, 2004). However, evidence indicates
that adolescents generally provide accurate reports of antisocial behaviors (Harrison, Haaga,
& Richards, 1993; Needle, Jou, & Su, 1989) and similar youth surveys have been used to
provide valid estimates of ASB. Furthermore, relying on self-report for both independent
and dependent variables may result in shared method variance that arbitrarily inflates the
relationship among risk/protective factors and ASB (Bank, Dishion, Skinner, & Patterson,
1990). Future research should employ multiple methods when assessing key study
constructs.

A second limitation is that data were cross-sectional. This may be a particular problem given
our finding that peer substance use had the strongest association with probability of
classification into the mild, moderate, and serious ASB groups. There is an extensive
literature that indicates selection effects (i.e., ASB-involved youth selecting into
relationships with ASB-involved peers) as well as socialization are important factors that
explain why peer ASB and youths’ own ASB are so strongly related (Dishion, Patterson, &
Griesler, 1994; Monahan, Steinberg, & Cauffman, 2009). Thus, additional research utilizing
NIH-PA Author Manuscript

longitudinal data is paramount to assess the causal relation of risk and protective processes
to class membership.

Finally, the generalizability of the findings may be influenced by the characteristics of the
sample. The extent to which these findings would generalize to youth in more urban settings
is not clear. Studies have not explored whether patterns of ASB and correlates of those
patterns vary as a function of geographic location and ethnic/racial group. Given that our
findings are similar to research conducted with urban samples (Fergusson, Horwood, Nagin,
2000; Odgers et al., 2007), future research should utilize more representative samples to
examine if patterns of ASB differ as a function of geographic locale and ethnic/racial group.

Despite limitations, this study contributes to our understanding of patterns of ASB and risk
and protective factors associated with those patterns in an understudied sample of
nonmetropolitan youth. Future research should build on this study and examine changes in

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 9

patterns of ASB in adolescence and early adulthood and individual and contextual risk and
protective factors associated with those changes.
NIH-PA Author Manuscript

Acknowledgments
This research study was supported by a grant from the Substance Abuse and Mental Health Services Administration
(SAMHSA SP13960-01; PI: R. Brex) to the Northeast Communities Against Substance Abuse (NECASA). Dr.
Cook’s involvement was supported by funding from the National Institute on Drug Abuse (NIDA T32-DA019426;
PI: D. Snow); Dr. Aklin’s involvement was supported by funding from the National Institute on Drug Abuse
(NIDA R25 DA020515; PI: S. Ball). The authors’ wish to acknowledge NECASA staff involved in the survey
administration including: Mary Ann Murphy-Patton, Susan Harrod, Michelle Rawcliffe, and Bonnie Wolters; as
well as the school administrators, teachers, and students who participated in the survey administration. We also
wish to acknowledge helpful comments by members of the Division of Prevention and Community Research at
Yale University School of Medicine on an earlier draft of this manuscript.

References
Ardelt M, Day L. Parents, siblings, and peers: Close social relationships and adolescent deviance.
Journal of Early Adolescence. 2002; 22:310–349.
Ayers CD, Williams JH, Hawkins JD, Peterson PL, Abbott RD. Assessing correlates of onset,
escalation, deescalation, and desistance of delinquent behavior. Journal of Quantitative
Criminology. 1999; 15:277–306.
Bank, L.; Dishion, TJ.; Skinner, M.; Patterson, G. Method variance in structural equation modeling. In:
NIH-PA Author Manuscript

Patterson, GR., editor. Depression and aggression in family interaction. Hillsdale, NJ: Lawrence
Erlbaum Associates; 1990. p. 247-279.
Beyers JM, Bates JE, Pettit GS, Dodge KA. Neighborhood structure, parenting processes and the
development of youths’ externalizing behaviors: A multilevel analysis. Journal of Community
Psychology. 2003; 31:35–54.
Brown TL, Zimmerman RS. Are adolescents accurate reporters of their alcohol use? Individual
Differences Research. 2004; 2:17–25.
Catalano, RF.; Hawkins, JD. The social development model: A theory of anti-social behavior. In:
Hawkins, JD., editor. Delinquency and crime: Current theories. New York: Cambridge University
Press; 1996. p. 149-197.
Center for Disease Control and Prevention. Youth Risk Behavior Surveillance – United States.
Morbidity and Mortality Weekly Report. 2010; 59:1–148. [PubMed: 20075837]
Cohen MA. The monetary value of saving a high-risk youth. Journal of Quantitative Criminology.
1998; 26:343–353.
Dishion, TJ.; Patterson, GR.; Griesler, PC. Peer adaptations in the development of antisocial behavior:
A confluence model. In: Huessmann, LR., editor. Aggressive behavior: Current perspectives.
Plenum series in social/clinical psychology. New York: Plenum Press; 1994. p. 61-95.
Fahs PSS, Smith BE, Atav AS, et al. Integrative research review of risk behaviors among adolescents
NIH-PA Author Manuscript

in rural, suburban, and urban areas. Journal of Adolescent Health. 1999; 24:230–243. [PubMed:
10227342]
Farrell AD, Sullivan TN, Esposito LE, Meyer AL, Valois RF. A latent growth curve analysis of the
structure of aggression, drug use, and delinquent behaviors and their interrelations over time in
urban and rural adolescents. Journal of Research on Adolescence. 2005; 15:170–204.
Fergusson DM, Horwood LJ. Male and female offending trajectories. Development and
Psychopathology. 2002; 14:159–177. [PubMed: 11893091]
Fergusson DM, Horwood LJ, Nagin DS. Offending trajectories in a New Zealand birth cohort.
Criminology. 2000; 38:525–551.
French DC, Conrad J. School dropout as predicted by rejection and antisocial behavior. Journal of
Research on Adolescence. 2005; 11:225–244.
Frick PJ, Lahey BB, Loeber R, Tannenbaum L, Vanhorn Y, Christ M, Hart EA, Hanson K.
Oppositional defiant disorder and conduct disorder: A meta-analytic review of factor analyses and
cross-validation in a clinical sample. Clinical Psychology Review. 1993; 13:319–340.

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 10

Hair EC, Park MJ, Ling TJ, Moore KA. Risky behaviors in late adolescence: co-occurrence,
predictors, and consequences. Journal of Adolescent Health. 2009; 45:253–261. [PubMed:
19699421]
NIH-PA Author Manuscript

Harris PW, Jones PR. Differentiating delinquent youths for program planning and evaluation. Criminal
Justice and Behavior. 1999; 26:403–434.
Harrison ER, Haaga J, Richards T. Self-reported drug use data: What do they reveal? American
Journal of Drug and Alcohol Abuse. 1993; 19:423–441. [PubMed: 8273764]
Laird RD, Criss MM, Pettit GS. Parents’ monitoring knowledge attenuates the link between antisocial
friends and adolescent delinquent behavior. Journal of Abnormal Child Psychology. 2008;
36:299–310. [PubMed: 17874291]
Lanza, ST.; Flaherty, BP.; Collins, LM. Latent class and latent transition analysis. In: Schinka, JA.;
Velicer, WF., editors. Handbook of psychology: Vol. 2, research methods in psychology.
Hoboken, NJ, US: John Wiley & Sons Inc; 2003. p. 663-685.
Lubke GH, Muthén B. Investigating population heterogeneity with factor mixture models.
Psychological Methods. 2005; 10:21–39. [PubMed: 15810867]
Magidson J, Vermunt JK. Latent class models for clustering: A comparison with K-means. Canadian
Journal of Marketing Research. 2002; 20:37–44.
McCutcheon, A. Latent class analysis. Beverly Hills, CA: Sage Publications; 1987.
Moffitt, TE. Life-course-persistent versus adolescence-limited antisocial behavior. In: Cicchetti, D.;
Cohen, DJ., editors. Developmental psychopathology, Vol 3: Risk, disorder, and adaptation. 2.
Hoboken, NJ, US: John Wiley & Sons, Inc; 2006. p. 570-598.
NIH-PA Author Manuscript

Monahan KC, Steinberg L, Cauffman E. Affiliation with antisocial peers, susceptibility to peer
influence, and antisocial behavior during the transition to adulthood. Developmental Psychology.
2009; 45:1520–1520. [PubMed: 19899911]
Muthén, LK.; Muthén, BO. Mplus user's guide. 4. Los Angeles, CA: Muthén & Muthén; 2006.
Najaka SS, Gottfredson DC, Wilson DB. A meta-analytic inquiry into the relationship between
selected risk factors and problem behavior. Prevention Science. 2001; 2:257–271. [PubMed:
11833928]
Needle R, Jou SC, Su S. The impact of changing methods of data collection on the reliability of self-
reported drug use of adolescents. American Journal of Drug and Alcohol Abuse. 1989; 15:275–
289. [PubMed: 2788363]
Nylund KL, Asparouhov T, Muthén BO. Deciding on the number of classes in latent class analysis and
growth mixture modeling: A Monte Carol Simultation Study. Structural Equation Modeling. 2007;
14:535–569.
Odgers CL, Moffitt TE, Broadbent JM, Dickson N, Hancox RJ, Harrington H, et al. Female and male
antisocial trajectories: From childhood origins to adult outcomes. Development and
Psychopathology. 2008; 20:673–716. [PubMed: 18423100]
Odgers CL, Moretti MM, Burnette ML, Chauhan P, Waite D, Reppucci ND. A latent variable
modeling approach to identifying subtypes of serious and violent female juvenile offenders.
NIH-PA Author Manuscript

Aggressive Behavior. 2007; 33:339–352. [PubMed: 17593559]

Osgood, DW.; Chambers, JM. Community correlates of rural youth violence (Juvenile Justice
Bulletin). Washington, D.C: U.S. Department of Justice, Office of Juvenile Justice and
Delinquency Prevention; 2003.
Soothill K, Francis B, Ackerley E, Humphrey L. Changing patterns of offending behaviour among
young adults. British Journal of Criminology. 2008; 48:75–95.
Ungemack, JA.; Cook, MJ.; Damon, D. The Governor's Prevention Initiative for Youth 2000 Student
Survey: State of Connecticut. Farmington, CT: University of Connecticut Health Center; 2001.
Wiesner M, Silbereisen RK. Trajectories of delinquent behavior in adolescence and their covariates:
Relations with initial and time-averaged factors. Journal of Adolescence. 2003; 26:753–771.
[PubMed: 14643745]
Wiesner M, Windle M. Assessing covariates of adolescent delinquency trajectories: A latent growth
mixture modeling approach. Journal of Youth and Adolescence. 2004; 5:431–442.
Windle M. A latent growth curve model of delinquent activity among adolescents. Applied
Developmental Science. 2000; 4:193–207.

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 11
NIH-PA Author Manuscript

Figure 1.
Conditional probabilities of reporting any history of ASB for a four class, restricted LCA
NIH-PA Author Manuscript

model.
NIH-PA Author Manuscript

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 Connell et al. Page 12

Table 1
Percentage of youth indicating affirmative engagement in ASB or related consequences in the past year (n =
NIH-PA Author Manuscript

1820).

Domain In the past year, did you: % Yes

Pick a fight with someone (not including a family member)? 24.4%
Threaten to hurt someone? 24.2%
Aggression-related Behavior
Attack someone with the idea of seriously hurting them? 8.7%
Bring a knife, gun, or other weapon to school? 3.3%

Steal something worth less than $50? 21.2%

Property-related Behavior
Purposely damage or destroy property that did not belong to you? 15.6%

Buy illegal drugs? 7.0%

Drug-related Behavior
Sell illegal drugs? 3.1%

Get into trouble with the police? 10.7%

Consequence-related Behavior Get suspended or expelled from school? 7.2%
Get arrested? 3.2%
NIH-PA Author Manuscript
NIH-PA Author Manuscript

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

 NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Table 2
Fit statistic comparisons of exploratory LCA models (n = 1820).

Model Description BIC LMR LRT

p-value
Connell et al.

1 One-class 13717.29 —
2 Two-class, unrestricted 11294.06 <0.001
3 Two-class, restricted a 12033.80 <0.001

4 Three-class, unrestricted 11099.19 <0.001

5 Three-class, restricted a 11128.61 <0.001

6 Four-class, unrestricted 11100.65 0.55

7 Four-class, restricted a 11088.23 <0.001

8 Five-class, unrestricted 11127.92 0.001

9 Five-class, restricted a 11095.32 0.24

Notes: BIC – Bayesian Information Criterion. LMR LRT – Lo-Mendell-Rubin Likelihood Ratio Test p-value for (K-1)-classes.
a
The responses to antisocial behavior involvement variables were set to 0 for one class.

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

Page 13
 NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Table 3
Odds ratio results from latent multinomial logistic regression models: Covariate effects on latent class membership with non-antisocial class as reference
(n = 1,820).

Mild ASB a Moderate. ASB a Serious ASB a Moderate ASB b Serious ASB b Serious ASB c
Connell et al.

Age 0.97 0.84 1.26 0.87 1.30 1.51**

Gender (Male) 1.33* 1.53* 2.91** 1.15 2.19** 1.91*
Parent-Child Relationship 0.68* 0.51** 0.46** 0.76 0.69 0.90

Peer Substance Use 2.01** 4.68** 8.46** 2.33** 4.22** 1.81**

School Attachment 0.78 0.37** 0.26** 0.48** 0.34** 0.71

Neighborhood Disorder 1.18 2.04** 2.88** 1.74** 2.45** 1.41

*
p < 0.05,
**
p < 0.01

Note:
a
Referent Class is Non-ASB
b
Referent Class is Mild ASB
c
Referent Class is Moderate ASB

Aggress Behav. Author manuscript; available in PMC 2013 July 18.

Page 14