when the course of aggressive behavior in childhood does ... · hanno petras,a cindy m. schaeffer,b...

When the course of aggressive behaviorin childhood does not predict antisocialoutcomes in adolescence and youngadulthood: An examination of potentialexplanatory variables

HANNO PETRAS,a CINDY M. SCHAEFFER,b NICHOLAS IALONGO,a

SCOTT HUBBARD,a BENGT MUTHÉN,c SHARON F. LAMBERT,a

JEANNE PODUSKA,d and SHEPPARD KELLAMd

aJohns Hopkins Bloomberg School of Public Health;bUniversity of Maryland, BaltimoreCounty; cUniversity of California, Los Angeles; anddAmerican Institutes of Research

AbstractTheoretical models and empirical studies suggest that there are a number of distinct pathways of aggressivebehavior development in childhood that place youth at risk for antisocial outcomes in adolescence and youngadulthood. The prediction of later antisocial behavior based on these early pathways, although substantial, is notperfect. The goal of the present study was to identify factors that explain why some boys on a high-riskdevelopmental trajectory in middle childhood do not experience an untoward outcome, and, conversely, why someboys progressing on a low-risk trajectory do become involved in later antisocial behavior. To that end, we exploreda set of theoretically derived predictors measured at entrance to elementary and middle school and examined theirutility in explaining discordant cases. First-grade reading achievement, race, and poverty status proved to besignificant early predictors of discordance, whereas the significant middle-school predictors were parent monitoring,deviant peer affiliation, and neighborhood level of deviant behavior.

Childhood aggressive behavior is widely rec-ognized as a precursor for antisocial behaviorin adolescence and adulthood. Numerous pro-spective studies have demonstrated that con-duct problems identified as early as preschoolpredict later delinquent behavior and drug use

~Ensminger, Kellam, & Rubin, 1983; Hawk-ins, Herrenkohl, Farrington, Brewer, Catal-ano, Harachi, & Cothern, 2000; Loeber & Hay,1997; Lynam, 1996; McCord & Ensminger,1997; Moffit, 1993; Yoshikawa, 1994!. Yet theavailable evidence also suggests that a sub-stantial proportion of those children who dis-play high levels of aggressive behavior inchildhood do not manifest antisocial behaviorin adolescence or adulthood~Maughan & Rut-ter, 1998!. Indeed, there appear to be “desist-ers” as well as “persisters”~McCord, 1983!.Moreover, a considerable number of childrenappear to be “late starters”~Moffitt, 1993!,engaging in average levels of aggressive be-havior in the early childhood years but pro-ceeding to engage in serious antisocial behaviorin adolescence and adulthood. It is important

This research was supported by grants from the NationalInstitutes of Mental Health~RO1 MH42968 to SheppardG. Kellam, PI, and T-32 MH18834 to Nick Ialongo, PI!and the Centers for Disease Control and Prevention~R490CCR318627-03!. We thank the Baltimore City PublicSchools for their continuing collaborative efforts and theparents, children, teachers, principals, and school psychol-ogists and social workers who participated.

Address correspondence and reprint requests to: HannoPetras, The Johns Hopkins Bloomberg School of PublicHealth, 624 N. Broadway, Baltimore, MD 21205; E-mail:[email protected].

Development and Psychopathology16 ~2004!, 919–941Copyright © 2004 Cambridge University PressPrinted in the United States of AmericaDOI: 10.10170S0954579404040076

919

to discriminate various pathways early in thedevelopmental course so that the limited re-sources available for preventive and interven-tion efforts may be more precisely targeted.

Models of Antisocial BehaviorDevelopment

Several influential theories of antisocial be-havior development have described qualita-tively different trajectories to different typesof delinquency and criminal involvement. Pat-terson, DeBaryshe, and Ramsey’s~1992! modelargues for two distinct pathways toward adultcriminality: early-starters~i.e., coercive par-enting, school failure, and antisocial behaviorproblems starting in childhood! and late-starters ~i.e., poor parental monitoring, op-positionality, and deviant peer involvementstarting in early adolescence!. Moffitt’s ~1993!model also proposes two mutually exclusivesubgroups of antisocial youth:life-course per-sistent offenderswho show high levels of ag-gression throughout development and continueto be violent as adults; andadolescence-limited offenderswho engage in nonviolentforms of antisocial behavior only during theteen years.

Loeber and Stouthamer–Loeber~1998!, inan extension of their earlier work~Loeber,Wung, Keenan, Giroux, Stouthamer–Loeber,Van Kammen, & Maughan, 1993!, proposedfive distinct subtypes to account for the ob-served heterogeneity among boys withinthese dual-pathway models. They proposetwo types of life-course persistent youth, onewith a preschool onset of aggression and co-morbid attention-deficit0hyperactivity disor-der~ADHD! and one with a middle childhoodonset of aggression without ADHD. They alsopropose two limited-duration groups: onewhose initially high level of aggression desistsin elementary school and another whose ag-gression desists in late adolescence or earlyadulthood. The final group, late-onset offend-ers, is thought to comprise those youth whoshow no antecedent problems in aggressionbut who develop antisocial behavior problemsin late adolescence or early adulthood. Thesemodels have helped to shift the study of youth

antisocial behavior away from a variable-centered focus on describing broad predictorsof behavior toward a more person-centeredfocus emphasizing individual differences in de-velopment~Magnusson, 1998!.

Modeling Studies of IndividualDifferences in AntisocialBehavior Development

Various methodological techniques have beenused to determine individual differences in lon-gitudinal patterns in antisocial behavior. Onetechnique is to make post hoc classificationsof individuals based on their progression onvariables of interest over time, using empiri-cally or theoretically defined cutoff scores~e.g.,see Loeber, Wei, Stouthamer–Loeber, Huiz-inga, & Thornberry, 1999; Moffitt & Caspi,2001; Moffitt, Caspi, Harrington, & Milne,2002; Patterson, 1996; Patterson, Forgatch, Yo-erger, & Stoolmiller, 1998; Tolan & Gorman–Smith, 1998!. Although post hoc classificationof youth to various developmental trajectorieshas heuristic value, this method is fraught withproblems as well. Post hoc classification doesnot allow for empirical testing, and it may over-or underclassify youth to various trajectorieswhile failing to identify other trajectories en-tirely. An alternative method for identifyingdevelopmental patterns in longitudinal data isto use growth modeling techniques to empiri-cally define distinct subgroups within a sam-ple~Muthén, 2000, 2001; Nagin, 1999!. Thesetechniques, which treat group membership asan unobserved variable, have been used to de-scribe individual differences in developmentfor a range of behaviors, including readingachievement~e.g., Crijnen, Feehan, & Kellam,1998!, coping strategies~e.g., Sandler, Tein,Mehta, Wolchik, & Ayers, 2000!, substanceuse~e.g., Curran, Muthén, & Harford, 1998;Oxford, Gilchrist, Morrison, Gillmore, Lohr& Lewis, 2003; White, Xie, Thompson, Loe-ber, & Stouthamer–Loeber, 2001!, and aggres-sion ~e.g., Muthén, 2001; Schaeffer, Petras,Ialongo, Poduska, & Kellam, 2003!.

Several groups of researchers have used la-tent growth modeling to classify antisocial boyson the basis of their longitudinal behavioralpatterns~Broidy et al., 2003; Maughan, Pick-

920 H. Petras et al.

les, Rowe, Costello, & Angold, 2000; Nagin& Tremblay, 1999; Shaw, Gilliom, Ingoldsby,& Nagin, 2003!. Taken together, the resultsfrom these longitudinal studies suggest thatthere are empirically identifiable subgroups ofyouth with distinct developmental trajectoriesof antisocial behavior. Those studies that mod-eled aggression over time identified differentnumbers of trajectories, but the trajectorieswere otherwise similar. Each study identifiedone to two normative subgroups~about 60%of boys! that never showed serious problemswith aggression and were not at increased riskfor later criminal behavior. These samples alsoidentified two groups of boys with sustainedproblems in antisocial behavior over time: achronic group~4–12% of boys! whose aggres-sive behavior was consistently high through-out development, and a high but desistinggroup ~20–28% of boys! whose aggressionstarted at a high level but decreased over time.Across samples, aggressive trajectories wereassociated with later antisocial and criminalbehavior in adolescence~Maughan et al., 2000;Nagin & Tremblay, 1999!.

Most recently, Schaeffer et al.~2003! usedgeneral growth mixture modeling~GGMM!to find empirical evidence for pathways de-scribed in theoretical models within an epide-miological sample of urban, primarily AfricanAmerican boys. Teacher-rated aggression mea-sured longitudinally through elementary schoolwas used to define growth trajectories. Con-sistent with prior studies, evidence was foundfor a chronic high aggression and a low-riskaggression trajectory. The Schaeffer et al. studyalso empirically identified a trajectory of in-creasing aggression, which corresponds closelyto Loeber and Stouthamer–Loeber’s~1998!hypothesized life-course persistent, childhood-onset group. Relative to the low-risk trajec-tory, boys with chronic high and increasingtrajectories were at increased risk for juve-nile arrest in adolescence and antisocial per-sonality disorder~ASPD! and adult arrest inyoung adulthood. Concentration problemswere highest among boys with a chronic highaggression trajectory and also differentiatedboys with increasing aggression from boyswith stable low aggression. Peer rejection alsowas higher among boys with chronic high

aggression relative to the low aggressiongroup.

Present Study Rationale and Goals

Among the questions left unanswered bySchaeffer et al.~2003! is how to explain thefact that although boys in the chronic high andincreasing aggression trajectories were at in-creased risk for a number of untoward out-comes in adolescence and early adulthood, atleast 26% of these boys did not manifest anyof the antisocial outcomes assessed. Con-versely, although boys in the stable low-aggression trajectory had the lowest risk of anuntoward outcome, a small percentage~9–16%! of these boys did manifest an antisocialoutcome in adolescence and0or young adult-hood. The primary purpose of this article wasto determine whether we could predict whichboys will have outcomes that are inconsistentwith their early aggressive behavior growthtrajectory. To that end, we examined a set ofpredictors measured at entrance to elementaryand middle school within the growth-mixturemodeling framework. We drew upon two ma-jor theoretical models in selecting our predic-tors. First, consistent with Bronfenbrenner’s~1979! social ecological model of child devel-opment, we selected first- and sixth-grade pre-dictors that represent a range of systemicinfluences~i.e., individual, family, peer, school,and neighborhood factors!. Second, Patter-son, Reid, and Dishion’s~1992! early starterand late starter model of the development ofantisocial behavior provided the rationale forspecific hypotheses.

Predictors of discordant casesin the chronic high class

We hypothesized that concentration problemsand peer rejection in the presence of high lev-els of aggressive0disruptive behavior at theentrance to first grade would promote the per-sistence of aggressive0disruptive behavior overtime and heighten the risk of later antisocialbehavior in adolescence and adulthood. Thus,we expected that those boys with a chronichigh-aggression trajectory who went on to anti-social outcomes in adolescence and young adult-

When childhood aggressive behavior does not predict antisocial outcomes 921

hood would have significantly higher levels ofconcentration problems and peer rejection infirst grade than boys with chronic high aggres-sion who did not manifest later serious anti-social behavior.There is considerable empirical~Jensen, Martin, & Cantwell, 1997; Lahey,McBurnett, & Loeber, 2000; Satterfield &Schell, 1997! and theoretical~e.g., Loeber et al.,1993; Moffitt, 1993; Patterson et al., 1989! ev-idence linking ADHD and concentration prob-lems to chronic antisocial behavior patterns. Interms of potential mechanisms, consistent withPatterson et al.’s~1992! early starter model,concentration problems may make demand0compliance bouts more likely, which in turnlead to more frequent coercive child behaviormanagement on the part of parents and teach-ers and more sustained levels of aggressive0disruptive behavior over development.

Peer rejection is another common correlateof chronic aggressive behavior problems~Ha-selager, Van Lieshout, Riksen–Walraven, Cil-lessen, & Hartup, 2002; Hektner, August, &Realmuto, 2000; Schwartz, 2000! and, in ac-cord with Patterson et al.~1992!, may serve tohasten the aggressive0disruptive child’s driftinto a deviant peer group in late childhoodand early adolescence. In these deviant peergroups, aggressive0disruptive behavior is re-inforced, thereby canalizing the pathway tolater antisocial behavior~Deater–Deckard,2001; French, Conrad, & Turner, 1995; Hekt-ner et al., 2000!. Thus, we expected that boyswho displayed a chronic high aggression tra-jectory in childhood and who went on to seri-ous antisocial behavior in adolescence andyoung adulthood would have higher levels ofpeer rejection in the early elementary schoolyears than their chronic high aggressive be-havior trajectory counterparts who did not man-ifest serious antisocial behavior in youngadulthood.

Poor academic achievement is another cor-relate of chronic aggressive behavior, and con-sistently predicts later delinquency~Denno,1990; Maguin & Loeber, 1996!. Furthermore,academic failure during the elementary gradesalso has been related to increased risk for laterviolent behavior~Farrington, 1989; Maguin& Loeber, 1996!. Accordingly, we hypoth-esized that those boys with a chronic high ag-

gressive growth trajectory in childhood but whodo not go onto antisocial behavior in youngadulthood would have higher levels of aca-demic achievement than their chronic high-aggressive behavior counterparts who doengage in serious antisocial behavior in youngadulthood.

Predictors of discordant casesin the stable low class

Patterson et al.’s~1992! late starter model ofthe development of antisocial behavior pro-vided the conceptual framework for predict-ing which boys who demonstrated a lowaggression trajectory in childhood would goon to serious antisocial behavior in adoles-cence and young adulthood. Patterson et al.argue that late starters typically exhibit mar-ginal levels of social adaptation in the elemen-tary school years~e.g., poor academic andsocial skills!, making them more vulnerableto perturbations in parental monitoring andsupervision. More specifically, Patterson et al.hypothesize that the escalation in antisocialbehavior during early adolescence among latestarters is due to disruptions in parental mon-itoring and supervision, which are broughton by serious family adversities that first sur-face in the middle-school years and tend tobe chronic in nature. The disruptors may in-clude a divorce, serious financial distress as-sociated with the loss of a job, and0or theonset of parental psychiatric distress or sub-stance abuse. As a result of their coerciveand antisocial behavior, late-onset children arealso rejected by their mainstream natural rat-ers. Their limited social survival skills andthe rejection by their mainstream natural rat-ers precipitate drift into a deviant peer group,where antisocial behavior, substance use, andrejection of mainstream social values, mores,and institutions are reinforced. Based on thelate starter model, we hypothesized that lowerparental monitoring, higher deviant peer af-filiation, and lower academic skills would dis-criminate those boys in the low-aggressiontrajectory who went on to antisocial out-comes in young adulthood from those whoevaded an untoward outcome.


Predictors of discordant casesin the increasing class

We also predicted that boys with increasingaggression~childhood onset subtype! who wenton to have an antisocial outcome would showhigher levels of concentration problems andpeer rejection relative to boys with increasingaggression who did not have an untoward out-come. In formulating this hypothesis, onceagain we drew upon Patterson et al.’s~1992!conceptualization of the development of anti-social behavior among late starters. Specifi-cally, attention0concentration problems athome and in the classroom may increase thelikelihood of demand0compliance bouts withparents and teachers. Such bouts may then leadto more coercive and inconsistent forms of dis-cipline on the part of parents and teachers,which in turn, leads to the development ofantisocial behavior. Attention0concentrationproblems also may result in rejection frommainstream peers and a subsequent drift intoantisocial peer groups, which further pro-motes antisocial behavior. Thus, we hypoth-esized that among those children who displaylow levels of aggressive0disruptive behaviorearly on in their elementary-school careers,the higher the level of attention0concentrationproblems and peer rejection, the more likelythat they will manifest antisocial behavior laterin their development.

Poverty, race, and neighborhoodcharacteristics as predictors ofdiscordant cases across allaggressive behavior trajectories

Family poverty is associated with a wide ar-ray of negative health, cognitive and socio-emotional outcomes in children~Bradley &Corwyn, 2002; Loeber et al., 1993!. Similarly,children growing up in economically deprivedneighborhoods are more likely to get involvedin crime and violence~Farrington, 1989;Yoshikawa, 1994!. One potential mechanismfor the effects of poverty on aggressive behav-ior is the role that poverty plays in disruptingparent monitoring and discipline, which in turn,increases the likelihood of youth exposureto deviant peers~Patterson et al., 1992!.

Ingoldsby and Shaw~2002! hypothesize thatneighborhood poverty and neighborhood crim-inality operate as risk indicators early in youthdevelopment and exert a more direct influ-ence as youth gain increased exposure to de-viant neighborhood influences in adolescence.In urban environments, African Americanyouth are more likely than Caucasian youth toexperience poverty on the family and neigh-borhood level and to experience neighbor-hood criminality, potentially placing them atgreater risk for later antisocial behavior. More-over, given the same level of antisocial behav-ior, African American youth are more likely tobe arrested and detained than are Caucasianyouth ~Beck & Karberg 2001; Sickmund,2000!.

These realities led us to a set of trajectory-specific hypotheses. Within the chronic highand the increasing trajectories of aggression,we hypothesized that boys who did not mani-fest antisocial behavior in young adulthoodwere more likely to be Caucasian, come fromhigher income families, and live in neighbor-hoods with less deviance. On the other hand,within the low aggression trajectory, we hy-pothesized that African American youth fromfamilies with lower income levels and higherlevels of neighborhood deviance would be ata higher risk for later antisocial behavior.

Strengths of the Present Study

Consistent with Schaeffer et al.~2003!, thepresent study improves upon previous longi-tudinal work in several important ways. First,in contrast to previous studies comprised pri-marily of Caucasian, urban~e.g., Nagin &Tremblay, 1999; Shaw et al., 2003! or Cau-casian, rural~e.g., Maughan et al., 2000! youth,the longitudinal sample used in the presentstudy is comprised of primarily African Amer-ican urban youth, consistent with the US Sur-geon General call for more mental healthresearch among ethnic minority populations~US Public Health Service, 2001!. Second, thepresent study utilizes an epidemiologically de-fined population, representative of all youthentering first grade in 19 public schools withinfive urban areas defined by census tract dataand vital statistics. Accordingly, the present


study allows for generalization to similar pop-ulations of students entering school from ur-ban neighborhoods with multiple problems~i.e., high rates of poverty, unemployment, andcrime!. Third, the present study includes botha psychiatric and a criminological young adultoutcome~diagnosis of ASPD and arrest! mea-sured at a much later point in development~young adulthood!, thereby providing strongexternal validity for the identified trajectoriesand spanning several important periods foryouth antisocial behavior development. Fi-nally, the present study employs the newestgeneration of latent growth modeling tech-niques, GGMM~Muthén & Muthén, 2000!, toaddress methodological limitations~i.e., as-sumption of invariance between classes andtime points, uncorrected estimates of covari-ates and young adult outcomes! of previouslongitudinal modeling studies.

Method

Participants

The participants included 675 boys who werefirst assessed at age 6 as part of an evaluationof two school-based, universal preventive in-terventions targeting early learning and aggres-sion in first and second grade~Dolan, Kellam,Brown, Werthamer–Larsson, Rebok, Mayer,Laudolff, Turkkan, Ford, & Wheeler, 1993;Kellam, Werthamer–Larsson, Dolan, Brown,Mayer, Rebok, Anthony, Laudolff, Edelsohn,& Wheeler, 1991! in 19 Baltimore City publicschools. These 675 boys were members of thecontrol group within the evaluation design. The19 schools were drawn from five geographicareas within the eastern half of the city, whichwere defined by census tract data and vitalstatistics obtained from the Baltimore CityPlanning Office. The five areas varied by eth-nicity, type of housing, family structure, in-come, unemployment, violent crime, suicide,and school drop out rates. However, each areawas defined so that the population within itsborders was relatively homogenous with re-spect to each of the above characteristics.

Special education and gifted classroomswere excluded from the pool of potential class-rooms in light of the fact the preventive inter-

ventions targeted regular or mainstreamclassrooms. In schools with three or fewer reg-ular first-grade classrooms, all classrooms par-ticipated, whereas in larger schools, three first-grade classrooms were randomly selected forinclusion in the study. Children had been ran-domly assigned to classrooms prior to assign-ment of classrooms to intervention conditions.Schools were randomly assigned to either anintervention or control condition within a geo-graphic area. In all analyses, standard errorsare corrected to reflect the fact that individualparticipants are clustered within classes andwithin schools~Jo, Muthén, Ialongo, & Brown,2004!.

A total of 675 male control participants wereoriginally available within the 19 participat-ing Baltimore City public schools in first grade.Seventy-eight of these 675 control boys didnot have a fall of first-grade teacher rating,and, consequently, they were not included inthe analyses for this article. The resulting sam-ple of 597 boys was 60.8% African American,38.0% Caucasian American heritage, and 1.2%other ethnicity~American Indian or Hispanic!.At entrance into first grade, the boys had amean age of 6.3 years~SD6 0.47!. Fifty-twopercent of the children received free or re-duced school lunch, a proxy for family in-come. There were no differences in terms ofethnicity, age, or standardized achievement testscores between the 78 boys with missing dataand the 597 boys with baseline teacher andcovariate data.

Of the 597 control males with a fall of first-grade teacher rating, 40 refused to participatein the age 19–20 follow-up, 16 had died priorto the follow-up as confirmed by a search ofthe National Death Index and0or an immedi-ate family member or friend, and the remain-ing 126 young adults either failed to respondto repeated requests for an interview or wereunable to be located during the fieldingperiod. Thus, 415 boys~66.0% African Amer-ican, 33.0% European Americans, 1.0% Amer-ican Indian or Hispanic! contributed data forvariables from the age 19–20 follow-up. As todifferences between those 415 males with afall of first-grade teacher data and who com-pleted ages 19–20 versus those missing ages19–20 follow-up data, no differences were


found in terms of a fall of first-grade free lunchstatus, math achievement, child’s rating ofanxiety and depression, and teacher ratings ofaggressive behavior and concentration prob-lems. Those who were missing the age 19–20follow-up had slightly lower standard readingscores than those who were interviewed~.20SD!, but the magnitude of this difference wasquite small. In addition, African Americanswere more likely to complete the age 19–20follow-up ~75.5% African American vs. 60.3%European Americans!.

Assessment design

Data for this report were gathered in the falland spring of first grade, the spring of secondto fifth grades, the spring of sixth grade, andat the age 19 or 20 follow-up assessment. Thedata gathered in the first-grade assessmentsincluded teacher reports of child aggressive0disruptive behavior, attention0concentrationproblems, and peer rejection. Data on freelunch eligibility and race also were collectedin first grade. Teacher reports of child aggres-sive0disruptive behavior were collected annu-ally or semiannually in Grades 1–5. Teacherratings of academic performance and youthratings of parental monitoring, neighborhooddeviance, and deviant peer affiliation were col-lected in the spring of sixth grade. At the ageof 19–20 years, a follow-up structured clinicalinterview was used to ascertain whether theparticipant met criteria for ASPD, and juve-nile and adult adjudication records wereobtained.

Measures for elementary school(Grades 1–5)

Teacher Observation of Classroom Adap-tation—Revised~TOCA-R!. Teacher ratingsof aggressive0disruptive behavior, attention0concentration problems, and peer rejectionwere obtained in the fall and spring of firstgrades using the TOCA-R~Werthamer–Larsson, Kellam, & Wheeler, 1991!. There-after, teacher ratings of aggressive0disruptiveusing the TOCA-R were collected annually inthe spring of Grades 2–5. Thus, although the

same teacher rated the youth in first grade,each subsequent year a different teacher pro-vided ratings of aggression for the child.

The TOCA-R is a structured interview withthe teacher, which is administered by a trainedassessor. Teachers respond to 36 items pertain-ing to the child’s adaptation to classroom taskdemands over the last 3 weeks. Adaptation israted by teachers on a 6-point frequency scale~15 almost neverto 65 almost always!. Theaggressive0disruptive behaviorsubscale in-cludes the following items:~a! breaks rules,~b! harms others and property,~c! breaksthings,~d! takes others property,~e! fights,~f !lies, ~g! trouble accepting authority,~h! yellsat others,~i! stubborn, and~ j! teases class-mates. The coefficient alphas for the aggres-sive0disruptive behaviors subscale ranged from.92 to .94 over Grades 1–7 or ages 8–13. The1-year test–retest intraclass reliability coeffi-cients ranged from .65 to .79 over Grades 2–3,3– 4, and 4–5. Scores on the aggressive0disruptive behavior subscale were signifi-cantly related to the incidence of school sus-pensions within each year from Grades 1–5~i.e., the higher the score on aggressivebehavior, the greater the likelihood of beingsuspended from school that year!.

The TOCA-R’s attention0concentration sub-scale consists of the following items:~a! com-pletes assignments,~b! concentrates,~c! pooreffort, ~d! works well alone,~e! pays atten-tion, ~f ! learns up to ability,~g! eager tolearn, ~h! works hard, and~i! stays on task.Werthamer–Larsson et al.~1991! report alphavalues of .91 and .83 in first grade. In terms ofconcurrent validity; each single unit of in-crease in teacher rated attention0concentrationproblems was associated with a twofold in-crease in risk of teacher perception for the needfor medication for such problems.

Teacher ratings ofpeer rejectionwere basedon a single item “rejected by classmates,” with1 indicatingtotal acceptanceand 6 represent-ing total rejection. The 4-month intraclass cor-relation coefficient for this item was .74, andit correlated significantly with peer nomina-tions ~not reported in this study! for the ques-tions “which kids don’t you like?”~r 5 .43!and “which kids are your best friends?”~r 5 2.58!.


California Achievement Test (CAT, Forms Eand F). The CAT was administered in the fallof the grade and represents one of the mostfrequently used standardized achievement bat-teries~Wardrop, 1989!. Subtests in CAT-E andF cover both verbal~reading, spelling, and lan-guage! and quantitative topics~computation,concepts, and applications!. The CAT was stan-dardized on a nationally representative sam-ple of 300,000 children. Internal consistencycoefficients for virtually all of the subscalesexceed .90. Alternate form reliability coeffi-cients are in the .80 range~Wardrop, 1989!.

Eligibility for free lunch.Eligibility for a freeschool lunch upon entry into first grade waschosen as a proxy for family income in thepresent study. Conceptually, eligibility for freelunch is likely to be a proxy variable for arange of economic and other stressors operat-ing at the family and neighborhood levels. Pre-vious research has demonstrated that free luncheligibility correlates highly with family in-come and other traditional measures of socio-economic status~Ensminger, Forrest, Riley,Kang, Green, Starfield, & Ryan, 2000!. Eligi-bility was treated as a binary variable~i.e.,eligible or not eligible!. In the present study,free lunch status had a strong negative corre-lation with parent education status measuredin fourth grade.

Race0ethnicity.Race is an important factor toconsider in studies of antisocial behavior, giventhat African American youth are disproportion-ately represented in the juvenile justice sys-tem ~Snyder & Sickmund, 1999!, and may berated higher by teachers on externalizing be-havior problems~Zimmerman, Khoury, Vega,& Gil, 1995!. Race was treated as a binaryvariable ~African American vs. CaucasianAmerican!.

Measures for middle school

Academic performance.As part of theTOCA-R, sixth-grade teachers were asked torate participating students’ academic progresson a scale of 15 failing to 6 5 excellent. Inthis study, academic progress was treated as a

binary variable, using the mean as the cut point~1 $ mean, 0, mean!.

Deviant peer affiliation.As elaborated above,Patterson et al.~1992! and colleagues havetheorized that drift into a deviant peer groupincreases the risk for antisocial behavior. Theyargue that antisocial behavior is not only mod-eled but also reinforced by deviant peers. InGrade 6, we used a subset of items from Ca-paldi and Patterson’s~1989! self-report scaleto measure deviant peer affiliation. Youths areasked in forced choice format to indicate howmany of their friends~1 5 noneto 5 5 all ofthem! have engaged in antisocial behavior, suchas hitting or threatening someone, stealing, anddamaging others’ property. Coefficient alphain sixth grade was .78. This variable was alsotreated as a binary variable, using the mean ascut point~1 $ mean, 0, mean!.

Neighborhood level of deviant behavior.Toassess the level of the youth exposure to devi-ant behavior in the neighborhood, four itemsfrom the Neighborhood Environment Scale~NES; Elliott, Huizinga, & Ageton, 1985! wereused:~a! kids get beat up in neighborhood,~b! adults get beat up in neighborhood,~c!neighbors often steal, and~d! drug dealers havemost money. The four items are coded so thatlow scores indicate high levels of exposureand high scores indicate low levels of expo-sure. The reliability coefficient in sixth gradewas .70. This variable was dichotomized usingthe mean as the cut point.

Structured Interview of Parent ManagementSkills and Practices—Youth Report (SIPMSP).The Monitoring subscale of the SIPMSP~Capaldi & Patterson, 1989! was used to as-sess parent monitoring in the sixth grade. Youthare asked to respond to questions regardingtheir parent monitoring practices in a forcedchoice response format~1 5 neverto 55 al-ways!. The questions making up the Monitor-ing subscale include, “When you get homefrom school, how often is there an adult therewithin one hour?” and “If you are at homewhen your parents are not, how often do youknow how to get in touch with them?” Thisvariable also was treated as a binary variable,


using the mean as cut point~1 $ mean, 0,mean!.

Measures for young adults

ASPD diagnosis.As part of a larger telephoneinterview at age 19–20, a scale was developedand administered to determine whether theparticipant metDiagnostic and Statistical Man-ual Fourth Edition (DSM-IV; American Psy-chiatric Association, 1994! criteria for ASPD.The questions comprising the scale were keyedto DSM-IVcriteria and the diagnoses derivedin accord with those criteria. To reduce thelikelihood of socially desirable responses, par-ticipants were asked to maintain their owncount of yes responses as opposed to respond-ing yes or no to the interviewer’s questions.To ensure against the respondents losing trackof the count, they were asked to have a penciland sheet of paper available to mark down thenumber of yes responses. In addition, the ques-tions were divided into three sections and atthe end of each section of the interview theinterviewer obtained a count of yes responses.In terms of concurrent validity, relative to thosewho did not meet criteria for ASPD, partici-pants with an ASPD diagnosis were four timesmore likely than those who did not to have ajuvenile or adult adjudication record~odds ra-tio @OR# 5 4.91, 95% confidence interval@CI#:2.77–8.71!.

Juvenile and adult adjudication records.Ju-venile police and court records also were ob-tained throughout the follow-up period todetermine the frequency and nature of policecontacts and criminal convictions during ado-lescence. Juvenile records were updated afterall participants had aged out of the juvenilecourt system~i.e., after everyone in the sam-ple had reached their 18th birthday! and thusrepresent complete juvenile court data for thissample. Adult court records were obtained atthe time of the young adult follow-up inter-view when participants were on average 20years of age. For the present study, both juve-nile and adult court records were treated asbinary variables~i.e., presence or absence of arecord!.

Analytic plan

The statistical methods applied in this studywere consistent with a person-centered ap-proach to data analysis, which emphasizesindividual differences in development~Mag-nusson, 1998!. GGMM ~Muthén, 2001; Muthén& Muthén, 2000; Muthén & Shedden, 1999!,implemented with the Mplus Version 2.14 sta-tistical software package~Muthén & Muthén,1998!, was used to identify distinct patterns ofgrowth in aggression over time. The observedtime variant indicators consisted of teacher-rated classroom aggression measured at six timepoints: fall and spring of first grade, and springof second through fifth grades.

Like traditional growth modeling tech-niques, GGMM estimates latent variables basedon multiple indicators. The multiple indica-tors of latent growth parameters correspond torepeated univariate outcomes at different timepoints. However, rather than assuming that thepopulation is constructed of a single continu-ous distribution, GGMM tests whether the pop-ulation is constructed of two or more discreteclasses~pathways! of individuals, with the goalof determining optimal class membershipfor each individual. As suggested in Muthén~2003!, we incorporated first-grade covariatesas antecedents for class membership andgrowth factors to correctly specify the model,to find the proper number of classes, and tocorrectly specify class proportions and classmembership.

Evidence for these different pathways inaggressive behavior development exists whenmodels involving two or more latent classesof growth provide a better fit than a traditional~single-class! growth model. Nested modelsare compared based on their specific log like-lihood; nonnested models were compared usingseveral test statistics available in the Mplussoftware package. The Bayesian InformationCriterion~BIC!, the sample size adjusted BIC~SSA BIC!, and the Akaike Information Cri-terion ~AIC ! were used for comparing non-nested models; lower scores represent betterfitting models~Akaike, 1987; Schwartz, 1978;Sclove, 1987!. In addition, the Lo-Mendell–Rubin~LMR! likelihood ratio test of model fitand an adjusted version were used to compare


the estimated and alternative models~Lo, Men-dell, & Rubin, 2001!. The obtainedp valuerepresents the probability that the null hypoth-esis~i.e., there is no difference in how the twomodels fit the data! is true. A lowp value in-dicates that the estimated model is preferableto a model with one fewer class. Finally, asummary measure of the overall classificationquality was given by the entropy measure~Ramswamy, DeSarbo, Reibstein, & Robin-son, 1993!. Entropy values range from zero to1, with values closer to 1 indicating better clas-sifications of individuals to specific classes.The estimation for a model with an increasingnumber of classes was stopped when none ofthe fit indices showed further improvement.

The analytical model used in this study isshown in Figure 1. Three pathways~P0, P1,P2! are of particular interest in this model.Pathway P0 stands for the base model, wherethe prevalence of the outcome for each of theaggression trajectory classes is presented. Inpathway P1, it is explored to what extent earlycovariates are related to the occurrence of theyoung adult outcome, once the model controlsfor class membership. Finally, in pathway P2,the association of middle-school covariateswith the two outcomes is tested after control-ling for class membership and the impact ofearly covariates.

Missing data

The estimates of parameters in the models wereadjusted for attrition. All longitudinal studiesexperience attrition when following partici-pants over time~Hanson, Tobler, & Graham,1990!. The Mplus software program used fullinformation maximum likelihood estimationunder the assumption that the data were miss-ing at random~MAR!. MAR assumes that thereason for the missing data is either random~i.e., not related to the outcome of interest! orrandom after incorporating other variablesmeasured in the study~Arbuckle, 1996; Little,1995!. Full information maximum likelihood,used in the present study, is widely acceptedas an appropriate way of handling missing data~Muthén & Shedden, 1999; Schafer & Graham,2002!.

Overall, the percentage of boys in the sam-ple missing data at a given time point was asfollows: missing 0–1 time points, 59.1%; 2–3time points, 34.3%; 4–6, 6.6%. The Mplussoftware bases its estimates on all availabletime points for a given case. Only cases withmissing on all of the repeated aggression mea-sures or with missing on the first-grade covari-ates are excluded from the analysis. To assessthe extent of missing data in the dataset, theMplus software provides a covariance “cover-age” matrix that gives the proportion of avail-able observations for each indicator variableand pairs of variables, respectively. The min-imum coverage necessary for models to con-verge is .10~Muthén & Muthén, 1998!. In thepresent study, coverage ranged from .47–.89,more than adequate for acceptable estimation.

Results

The relationship between patterns of growthin aggressive0disruptive behavior~hereafter re-ferred to as “aggression”!, covariate informa-tion in the fall of first grade and spring ofsixth grade, and the young adult outcomes wasestimated using GGMM. The results are pre-sented in three parts. First, we describe thebasic growth mixture model that includes onlythe set of a fall of first-grade covariates andthe young adult outcome. Second, for eachof the two outcomes~ASPD diagnosis andarrest record! we investigate if any of the first-grade covariates contribute to the predictionof the young adult outcomes over and abovewhat is explained by the membership in oneof the aggression trajectories. Third, we de-scribe the association between the sixth-gradecovariates~level of neighborhood deviance andparental monitoring! and the two young adultoutcomes in the final section.

Growth model, trajectory classes,and young adult outcomes

In this section, the model building procedureis discussed, and the resulting model is de-scribed in terms of growth trajectories, char-acteristics of class membership0growth, andprevalence of the two young adult outcomes


Figure 1. The analytical model, using a growth mixture framework.

92

9

~ juvenile or adult arrest record, ASPDdiagnosis!.

Model building procedure of growth mixturemodel.Prior model testing was used to deter-mine the growth shape of aggression, the num-ber of different developmental trajectoriesneeded, and the impact of the early covariateson growth and class membership~Muthén,2003!. Further improvements were added tothis base model. First, to account for the factthat the same teacher provided ratings in thefall and spring of first grade, the residual cor-relation between aggression ratings at thesetime points was added to the model. Second,due to low and nonsignificant estimates, thevariance of the quadratic slope as well as co-variation between the three growth param-eters~intercept, linear slope, quadratic slope!were set to zero. In all models tested, the fallof first-grade covariates were allowed to in-fluence the overall growth parameters~inter-cept, slope, quadratic slope!, as well as classmembership~Muthén, 2003!.

Assuming equal residual and growth vari-ance, we compared models with increasingnumbers of classes~see Table 1!. Althoughthe BIC value indicated almost equal fit for amodel with two or three classes, the samplesize adjusted BIC and the AIC indicated a four-class solution as the best model. The LMRtest indicated a two-class model was better fit-ting than a single-class model, whereas no bet-ter fit was found for the unmodified three- andfour-class models.

To further explore the impact on model fitonce the equal variance assumption was re-laxed, class-specific differences in variationwere integrated into the model. Specifically,to account for the fact that low-aggressive boysappeared to be a more homogenous group, in-tercept and residual variances were allowed tobe different in the low-aggression class, andslope means~linear and quadratic! and the vari-ance of the linear slope were set to zero in thatclass. With these modifications, two-, three-,and four-class models were compared. Thethree-class model showed the lowest BIC valueand thep value of the LMR test was signifi-cant. On the other hand, the sample size ad-justed BIC and the AIC were lowest for a four-class solution. Given the small differences infit between a three- and four-class solution andthe nontrivial difference of nine additional pa-rameters in the four-class model, we decidedon the modified three-class model as the mostparsimonious solution. Parameter estimates forthe model are presented in Table 2.

In a second step, the two young adult out-comes~crime record, ASPD diagnosis! wereadded to the model~BIC 5 7188.48, SSABIC 5 7026.58, entropy5 0.789!. A graphicaldepiction of the resulting solution with youngadult outcomes is presented in Figure 2 . Threedistinct trajectories were identified: achronichigh aggression~CHA! trajectory, consistingof those boys~16.1%! whose aggression startedhigh in first grade and remained high through-out elementary school; anincreasing aggres-sion~IA ! trajectory, consisting of boys~52.5%!

Table 1. Comparison of model fit among models with different numbers of classes

No. ofClasses

No. ofParameters BIC SSA BIC AIC

LMR,Adj. LRT Entropy

1 22 6552.442 6482.601 6456.494 — —2 31 6465.373 6366.961 6330.173 0.0021 0.8863 40 6465.607 6338.623 6291.155 0.5652 0.8174 49 6487.820 6332.265 6274.117 0.1821 0.769

Modified 2 36 6100.178 5985.892 5943.171 0.0007 0.772Modified 3 45 6030.977 5888.119 5834.718 0.0114 0.779Modified 4 54 6045.758 5874.329 5810.248 0.8127 0.772

Note: BIC, Baysian Information Criterion; SSA BIC, sample size adjusted BIC; AIC, Akaike information criterion;LMR Adj. LRT, Lo–Mendell–Rubin adjusted likelihood ratio test; Entropy, classification accuracy.


whose first-grade aggression was low but whobecame increasingly more aggressive throughfifth grade; and astable low aggression~SLA!trajectory, consisting of boys~31.4%! with con-sistently low levels of aggression over time.Consistent with Schaeffer et al.~2003!, wewere unable to identify a group of boys whoseaggression decreased over time~i.e., desisters!.

First-grade covariates and class membership.As noted, the first-grade predictors selectedfor this study were included in model building

as a means of improving model fit and increas-ing the accuracy of assignments of individualsto trajectory classes~Muthén, 2003!. For thispurpose, class membership and growth param-eters~intercept, slope! were simultaneously re-gressed on the fall of first-grade covariates.

The effects of first-grade covariates on classmembership are presented in Table 3. The riskfor class membership is given in reference tothe SLA class. Because this part of the analyt-ical work builds the foundation for the lateranalyses, we chose to express these findings

Table 2. Parameter estimates for the three-class GGMM model, including two distaloutcomes

CHA IA SLA

Parameter Estimate SE Estimate SE Estimate SE

a0 2.824 0.432 1.143 0.327 0.790 0.271a1 20.594 0.179 0.491 0.131 0.000 Fixeda2 0.103 0.041 20.064 0.022 0.000 FixedV~z0! 0.199 0.068 0.199 0.068 0.023 0.012V~z1! 0.019 0.008 0.019 0.008 0.000 FixedV~z2! 0.000 Fixed 0.000 Fixed 0.000 Fixedg06concentration 0.106 0.037 0.106 0.037 0.106 0.037g06peer rejection 0.175 0.082 0.175 0.082 0.175 0.082g06reading 20.001 0.008 20.001 0.008 20.001 0.008g06lunch status 20.029 0.093 20.029 0.093 20.029 0.093g06race 0.172 0.063 0.172 0.063 0.172 0.063g16concentration 20.012 0.010 20.012 0.010 20.012 0.010g16peer rejection 20.034 0.021 20.034 0.021 20.034 0.021g16reading 0.002 0.001 0.002 0.001 0.002 0.001g16lunch status 0.038 0.028 0.038 0.028 0.038 0.028g16race 20.001 0.021 20.001 0.021 20.001 0.021V~«1F! 0.212 0.071 0.212 0.071 0.096 0.040V~«1S! 0.438 0.059 0.438 0.059 0.097 0.037V~«2S! 1.082 0.134 1.082 0.134 0.170 0.035V~«3S! 1.003 0.123 1.003 0.123 0.119 0.030V~«4S! 0.745 0.087 0.745 0.087 0.092 0.029V~«5S! 0.728 0.085 0.728 0.085 0.116 0.093C~«1f, «1S! 0.064 0.028 0.064 0.028 0.064 0.028C~a0, a1! 0.000 Fixed 0.000 Fixed 0.000 FixedC~a0, a2! 0.000 Fixed 0.000 Fixed 0.000 FixedC~a1, a2! 0.000 Fixed 0.000 Fixed 0.000 Fixedac1 210.203 2.557ac2 22.130 1.358ac3 0.000 Fixed

Note: Estimates of covariates on latent class membership are depicted in Table 3. Estimates of the class-specificprevalence of the young adult outcomes are displayed in Figure 2.Log likelihood5 23432.028,df 5 51, BIC5 7188.482, entropy5 0.789Model: yit 5 h0i 1 h1i at 1 h2i at

2 1 «it at 5 0, 0.5, 1.5, 2.5, 3.5, 4.5h0i 5 a0k 1 g0kCi 1 z0i h1i 5 a1k 1 g1kCi 1 z1i h2i 5 a2k 1 g2kCi 1 z2i

Ci 5 concentration problems, peer rejection, reading, lunch status, raceV~z6classk! 5 ck V~« 6classk! 5 Qk


only in reference to the group~SLA! least atrisk for later young adult outcomes. Relativeto the SLA class, boys with higher levels ofreading achievement, concentration problems,

and peer rejection were more likely to be inthe CHA class. Similarly, high levels of con-centration problems and receiving a free schoollunch were associated with membership in theIA class. Across all classes, African Americanboys ~est. 5 0.172, SE 5 0.063! with highlevels of concentration problems~est.5 0.106,SE5 0.037! and peer rejection~est.5 0.175,SE5 0.082! had higher intercepts of aggres-sion. Furthermore, the course~slope! of ag-gression was accelerated by boys’ reading level~est.5 0.002,SE5 0.001!.

Class membership and young adult outcomes.Of the boys with a CHA trajectory, 27.2% hadan ASPD diagnosis and 44.8% were arrestedas a juvenile or adult~percentages are basedon threshold estimates from the regression ofdistal outcomes on class membership!. Of theboys with IA, 25.6% had an ASPD diagnosisand 52.6% had a juvenile or adult court record.In comparison, of the boys who were in theSLA class, 11.1% were diagnosed with ASPDand 12.9% had an arrest record. Significancetesting revealed that boys in the CHA trajec-tory were three times more likely to be diag-nosed with ASPD~OR52.99, 95% CI50.74–11.94! and more than five times more likely~OR5 5.48, 95% CI5 2.59–11.559! to have

Figure 2. The prevalence of antisocial personality disorder and arrest as a function of growth in aggression.

Table 3. Association between classmembership and first grade predictors

CovariateTrajectory

Class OR 95% CI

Reading CHA 1.20* 1.03, 1.40achievement IA 1.02 0.94, 1.09

SLA 1 Fixed

Concentration CHA 3.22* 2.22, 4.67problems IA 1.65* 1.25, 2.20

SLA 1 Fixed

Peer rejection CHA 2.35* 1.36, 4.08IA 1.44 0.93, 2.23SLA 1 Fixed

Eligible for CHA 2.27 0.61, 8.43free lunch IA 1.86* 1.05, 3.30

SLA 1 Fixed

Race CHA 0.78 0.26, 2.34IA 1.29 0.66, 2.52SLA 1 Fixed

Note: OR, odds ratio; CI, confidence interval; CHA,chronic high aggression; IA, increasing aggression; SLA,stable low aggression. SLA is the reference group.*p , .05.


an arrest record when compared to boys in theSLA class. Boys with an IA trajectory werealmost three times more likely to be diag-nosed with ASPD~OR52.77, 95% CI51.26–6.11! and more than seven times more likelyto have an arrest record~OR5 7.46, 95% CI52.96–11.47! when compared to boys in theSLA class.

Contribution of first-grade covariates tothe prediction of young adult outcomes

In this section, we present the class specificimpact of the fall of first-grade covariates onthe likelihood to have an arrest record or anASPD diagnosis~see Table 4!. We hypoth-esized that concentration problems, peer rejec-tion, and low levels of reading would beassociated with an increase in risk for boys inthe CHA and the IA trajectories. Poverty andbeing of African American heritage were hy-pothesized to increase risk across all classes,but would be particularly instrumental in ex-plaining discordant cases in the SLA class.

For boys in the CHA trajectory, none of thecovariates contributed significantly to the like-lihood of an arrest over and above the riskassociated with class membership.Among boysin the IA class, race~African American!, lunchstatus~eligible for free lunch!, and readingachievement~lower! in the fall of first gradewere significantly related to risk of an arrestrecord. Among boys in the SLA class, beingeligible for a free school lunch and being Af-rican American significantly increased the riskfor later arrest. With respect to an ASPD diag-nosis, none of the covariates in any of the threeclasses contributed to the prediction over andabove class membership.

Middle-school covariates andyoung adult outcomes

It was hypothesized that low levels of parentalmonitoring and poor academic progress andhigh levels of deviant peer affiliation andneighborhood exposure to deviant behaviorwould be associated with an increase in risk

Table 4. Association between first-grade predictors and young adult outcomesby class membership

Arrest ASPD

CovariateTrajectory

Class OR 95% CI OR 95% CI

Reading achievement CHA 1.02 0.89, 1.19 0.91 0.75, 1.11IA 0.87* 0.79, 0.95 0.94 0.83, 1.08SLA 0.82 0.65, 1.02 0.94 0.81, 1.11

Concentration problems CHA 1.04 0.67, 1.62 1.09 0.65, 1.82IA 0.91 0.68, 1.22 0.87 0.57, 1.32SLA 0.69 0.31, 1.52 0.71 0.38, 1.33

Peer rejection CHA 1.20 0.71, 2.00 1.21 0.71, 2.06IA 0.81 0.56, 1.18 0.87 0.46, 1.63SLA 0.81 0.32, 2.04 1.23 0.46, 3.27

Eligible for free lunch CHA 2.60 0.86, 7.86 2.47 0.40, 15.30IA 0.95 0.47, 1.96 1.71 0.67, 4.36SLA 4.14* 1.37, 12.52 1.11 0.26, 4.83

Race CHA 1.88 0.79, 4.43 2.73 0.40, 18.80IA 2.40* 1.29, 4.45 4.09 0.98, 16.99SLA 6.53* 1.28, 33.25 1.96 0.47, 3.27

Note:OR, odds ratio; CI, confidence interval; CHA, chronic high aggression; IA, increasing aggression;SLA, stable low aggression.*p , .05.


for the two outcomes, especially in the IA andthe SLA class~see Table 5!.

Academic performance in sixth grade.Regard-ing a diagnosis of ASPD, higher academicachievement was associated with a nonsignif-icant decrease in risk in the CHA and SLAtrajectory classes, and with a nonsignificantincrease in risk for boys in the IA class. Re-garding later criminal record, higher aca-demic achievement was associated with anonsignificant increase in risk for boys in theIA and SLA classes and a decrease in risk forboys in the CHA class~see Table 5!.

Deviant peer affiliation.Boys in the IA classwho were above the mean on deviant peer af-filiation were more than three times more likelyto be diagnosed with ASPD, whereas no sig-nificant associations were found in the CHAor the CLA classes. Regarding the likelihoodof a criminal record, no significant associa-tions with deviant peer affiliation were found.However, estimates indicated an increase inrisk in the increasing and in the low aggres-sive classes.

Exposure to neighborhood level deviantbehavior. Boys’ perception of neighborhood

level deviant behavior was not significantlyrelated to the probability of an ASPD diagno-sis. However, estimates of ORs indicated anincrease in risk for ASPD ranging from 11%in the CHA to 232% in the SLA class. Forarrest, a significant increase in risk was foundfor two of the classes. Boys in the CHA classwhose perception of neighborhood deviancewas above the mean were more than eight timesmore likely to have an arrest record by age 21,and boys in the IA class were more than twotimes more likely. Neighborhood deviant be-havior increased risk in the SLA class but notsignificantly.

Parental monitoring.With respect to an ASPDdiagnosis, parental monitoring reduced the riskin all three classes but did not reach signifi-cance. With respect to an arrest record, sixth-grade level of parental monitoring significantlyreduced the risk by 55% for boys in the IAclass. For boys in the CHA class and in theSLA class, parental monitoring also was asso-ciated with a reduction in risk but did not reachsignificance.

Discussion

Prior research~e.g., Broidy et al., 2003; Nagin& Tremblay, 1999; Schaeffer et al., 2003! and

Table 5. Association between middle school predictors and young adultoutcomes by class membership

Arrest ASPD

CovariateTrajectory

Class OR 95% CI OR 95% CI

Teacher-rated academic CHA 0.52 0.05, 5.66 0.55 0.02, 19.20performance IA 1.13 0.61, 2.08 3.32 0.73, 15.13

SLA 1.44 0.31, 6.82 0.88 0.27, 2.60

Deviant peer affiliation CHA 0.48 0.16, 1.47 0.86 0.16, 4.68IA 1.11 0.67, 1.83 3.35* 1.95, 5.76SLA 1.94 0.52, 7.35 0.82 0.16, 4.35

Neighborhood exposure CHA 0.12* 0.02, 0.60 0.90 .069, 3.46to deviant behavior IA 0.44* 0.23, 0.86 0.67 0.39, 1.14

SLA 0.44 0.11, 1.76 0.43 0.08, 2.45

Parent monitoring CHA 0.28 0.08, 1.01 0.59 0.18, 1.99IA 0.45* 0.22, 0.92 0.65 0.35, 1.21SLA 0.76 0.22, 2.60 0.76 0.19, 3.03

Note:OR, odds ratio; CI, confidence interval; CHA, chronic high aggression; IA, increasing aggression;SLA, stable low aggression. Covariates were dichotomized~mean split!.*p , .05.


theory ~e.g., Loeber & Stouthamer–Loeber,1998; Moffitt, 1993; Patterson et al., 1989!point to a number of distinct pathways fromaggressive behavior in childhood to later anti-social outcomes in young adulthood. Yet theprediction of later antisocial behavior basedon these early pathways, though substantial,is not perfect~Schaeffer et al., 2003!. Indeed,Schaeffer et al. found a significant level ofdiscordance between the early course of ag-gression and later antisocial outcomes. The goalof the present study was to identify factorsthat explain why some boys on high-risk de-velopmental trajectories in middle childhooddo not experience an untoward outcome, and,conversely, why some boys progressing on alow-risk trajectory do become involved in laterantisocial behavior. To that end, we exploreda set of predictors measured at entrance to ele-mentary and middle school and examined theirutility in explaining discordant cases. Bron-fenbrenner’s~1979! social ecological modeland Patterson et al.’s~1992! early and latestarter model of the development of antisocialbehavior served as the theoretical basis for ourchoice of these predictors.

Trajectory class membership, first-gradecovariates, and young adult outcomes

Before discussing findings regarding predic-tors of discordance, it is important to point outthat, consistent with Schaeffer et al.~2003!,aggressive behavior trajectory membership inelementary school was a significant predictorof criminal arrest and ASPD in young adult-hood. Moreover, consistent with Schaeffer et al.and in line with Patterson et al.’s~1992! model,teacher-rated concentration problems and peerrejection in first grade significantly predictedaggressive behavior trajectory membership.The higher the level of teacher-rated concen-tration problems and peer rejection in firstgrade, the greater the likelihood of being inthe CHA and IA trajectory classes in elemen-tary school.

It is also important to point out that ourexamination of the early predictors of discor-dance centers on their role in explaining vari-ation within trajectory class in terms of theyoung adult outcomes. This is also the case

for the middle-school predictors. However, inregard to middle-school predictors, analysesalso control for the effects of trajectory classmembership and the early covariates on themiddle-school predictors. Thus, the results forthe middle-school predictors in terms of dis-cordance answer the question of what factorswithin trajectory classes explain discordancebetween early course and the young adult out-comes beyond first-grade covariates and classmembership itself.

Early predictors of discordant cases

To understand early childhood factors thatmight predict later discordance, predictors fromthe individual ~i.e., attention0concentrationproblems, race!, family ~i.e., poverty status!,peer~i.e., peer rejection!, and school~i.e., first-grade reading achievement! were examined.Of these, reading achievement proved predic-tive of discordant cases for boys with an IAtrajectory, such that those with higher readingachievement were less likely to have a crimi-nal arrest record. Poverty and race also werepredictive of discordant cases amongst boyswith a SLA trajectory in elementary school.More specifically, boys with SLA who wereeligible for free school lunch in first gradewere four times more likely to have a criminalarrest record than their SLA counterparts whowere not eligible for free lunch. In addition,African American boys with SLA were sixtimes more likely to have an arrest record thanwere their Caucasian counterparts. Of note,for boys with a CHA trajectory, none of theearly predictors explained the discordant casesfor either antisocial outcome~i.e., arrest orASPD diagnosis!.

There are at least two possible explana-tions why reading achievement predicted dis-cordance in terms of later arrest among boyswith IA. First, and consistent with Pattersonet al. ~1992!, early reading achievement maybuffer at-risk youth from engagement in laterantisocial activities through mechanisms suchas mainstream peer acceptance, greater attach-ment to school, enhanced job prospects inyoung adulthood, or better cognitive resources~e.g., foresight and planning! for anticipatingthe negative consequences of engaging in


crime. Alternatively, it may be that IA boyswith higher reading achievement are simplybetter at avoiding arrest when compared toyouth with lower cognitive abilities.

Regarding the role of race and poverty sta-tus among boys in the SLA trajectory, our find-ings may be further evidence for the tendencyof police to arrest African American youth forminor misbehaviors that are overlooked or han-dled informally when committed by Cauca-sian youth~Sickmund, 2000!. For the boys inthe present study, race and poverty were closelyrelated, such that African American boys weremuch more likely than Caucasian boys to alsobe poor. Thus, it seems likely that broader so-cial policies related to poverty and commu-nity violence or crime, such as increased policepresence in low-income innercity neighbor-hoods, may be responsible for these discrep-ancies in arrest rates rather than~or in additionto! racial profiling per se.

Middle-school predictors ofdiscordant cases

Middle-school predictors from the family~i.e.,parental monitoring!, peer~i.e., deviant peeraffiliation!, school ~i.e., academic perfor-mance!, and neighborhood~i.e., level of devi-ant behavior! domains were examined tounderstand later factors affecting discor-dance. We found that the lower the level ofneighborhood-level deviant behavior, the lowerthe risk of criminal arrest among boys withCHA and IA trajectories. In addition, a highlevel of parental monitoring was associatedwith a significant decrease in the risk of crim-inal arrest among boys with an IA trajectory.Teacher-rated academic performance in sixthgrade was not predictive of discordant casesin any of the three trajectory classes. More-over, none of the middle-school predictorsexplained discordant cases for an ASPD diag-nosis in any of the aggressive behavior tra-jectories. However, with few exceptions, andregardless of significance level, the effects ofthe predictors on ASPD diagnosis and crimi-nal arrest were in the hypothesized direc-tion within each of the aggressive behaviortrajectories.

As noted, parental monitoring in sixth gradewas associated with a reduction in risk for bothoutcomes for all three trajectories; however,its protective effect was found to be signifi-cant only for boys with an IA trajectory. It ispossible that monitoring has somewhat of areduced effect on boys with a pattern of CHA~i.e., early starters! because of problems in theparent–child affective relationship. Recent ev-idence suggests that monitoring is most effec-tive in the context of a strong parent–childemotional bond~Stattin & Kerr, 2000! that bymiddle school may have eroded significantlyfor boys with CHA who by then have a longhistory of compliance bouts and coercive cy-cles with caregivers. In addition, CHA boysmay have earned antisocial reputations amongteachers, peers, and police that are not af-fected by parental monitoring, and that keepthese boys at risk for arrest. Alternatively, theremay be a dose–response relationship betweenmonitoring and later antisocial outcomes, suchthat CHA boys require a higher level of mon-itoring and supervision to protect them fromthose outcomes. Regarding the failure to finda significant relationship between parent mon-itoring and the young adult outcomes in theSLA trajectory, one possible explanation wasthe relatively small number of boys who fellinto the low level of parent monitoring. Theremay have been too little variation in parentmonitoring among these SLA boys for it toserve as a predictor of discordance.

Contrary to predictions, a protective effectfor sixth-grade academic performance was notfound for either of the high-risk classes in termsof arrest or ASPD diagnosis. Once again, apossible explanation for this finding is the highnumber of boys in the CHA and IA trajectoryclasses with below the mean levels of aca-demic performance~.80%!. It is not clearwhy sixth-grade academic performance did notoperate in the same way as reading achieve-ment in first grade for boys in the IA class.One explanation for this discrepancy mightbe method variance, given that a standardizedachievement test was used to measure readingin first grade, while academic performance insixth grade was assessed by teacher ratings.

Neighborhood level of deviant behavioroperated in the hypothesized direction for an


arrest record and showed a sizable, but nonsig-nificant increase in risk for anASPD diagnosisin the IAand SLAtrajectory classes.These find-ingssuggest thatexposure toneighborhood leveldeviant behavior may be related to opportunis-tic outcomes, such as arrest, but not to endur-ing patterns of antisocial conduct, such asASPDdiagnosis. This differential effect of neighbor-hood on antisocial outcomes may be evidencethat ASPD has more microsystemic underpin-nings~e.g., family and peer influences! whereasbeingarrestedhasagreatermacrosystemiccom-ponent~e.g., differential police practices!. Al-ternatively, this finding may reflect a confoundin the present study design between the ASPDand conduct disorder~CD! diagnoses; becausea diagnosis of CD is a prerequisite for ASPD,it is possible that boys with ASPD already metcriteria for CD at the time that neighborhoodeffects were assessed.

To summarize, consistent with Pattersonet al. ~1992!, we found evidence that middle-school factors may serve as a developmentalbridge between early aggressive behavior andantisocial outcomes in adolescence and youngadulthood, particularly with respect to crimi-nal arrest. Recall that our middle-school pre-dictor findings emerged despite the fact thatwe controlled for the effects of aggressiongrowth trajectory and first-grade predictorson both the young adult outcomes and themiddle-school predictors themselves. Thus,these findings suggest that antisocial behaviorin adolescence and young adulthood is notsolely a function of early-risk behaviors andrisk factors. The results of the present studyare consistent with the organizational theoryof development~Cicchetti & Schneider–Rosen,1984!, which suggests that although compe-tence at one developmental period is likely toexert a positive influence toward achievingcompetence at the next period, factors operat-ing at the level of the child, family, peer group,school, neighborhood, and greater society maymediate between early and later adaptation andpermit alternative outcomes.

Implications for prevention

Our findings suggest preventive interventionsshould be targeted at the entrance to both ele-

mentary and middle school. Universal inter-ventions targeting parenting and classroommanagement were implemented successfullyin the Reid, Eddy, and Fetrow~1999! trial infirst and fifth grades. In line with Ialongo, Wer-thamer, Kellam, Brown, Wang, and Lin~1999!,we recommend a nested approach to the pre-vention of antisocial behavior, wherein youthwho do not respond to universal interventionsare routed to either a selected or indicated in-tervention.As we argue in Ialongo et al.~1999!,a universal preventive intervention can serveas a diagnostic function in terms of the needfor more intensive intervention. That is, theneed for a selected or indicated interventioncould be determined on the basis of the re-sponse to the universal intervention as op-posed to some crudely measured risk factorsassessed at one point in time. Thus, the uni-versal intervention potentially provides us witha more accurate and therefore more cost-effective strategy for identifying individualsin need of more intensive intervention.

As noted, among boys with IA across ele-mentary school, the higher the early readingachievement in first grade, the lower the riskof a criminal arrest in adolescence or adult-hood. This finding is consistent with Kellam,Mayer, Rebok, and Hawkins~1998!, who foundthat a first-grade intervention targeting read-ing improvement had a small but beneficialeffect on concurrent aggressive0disruptive be-havior. Accordingly, we conclude that elemen-tary and middle-school interventions aimed atpreventing later antisocial behavior should in-clude a focus on academic achievement, whichis in line with the crucial role “social sur-vival skills” are afforded in Patterson et al.’s~1992! model of the development of antisocialbehavior.

Of concern is the fact that race and povertystatus were the only significant predictors ofdiscordant cases among boys with SLA withrespect to criminal arrest in adolescence andyoung adulthood. As noted in Greenberg,Domitrovich, and Bumbarger’s~2001! reviewof programs for the prevention of mental dis-orders in children and adolescents, inter-ventions found to be effective in preventingantisocial behavior focus almost exclusivelyon improving child social cognitions, peer re-


lations, or parent and teacher behavior man-agement. Thus, when it comes to antisocialbehavior, the field’s most effective preventiveinterventions are likely to prove of little ben-efit if you are well behaved in elementaryschool, but are African American and poor andlive in an urban environment. Clearly, our re-sults with respect to the role of race in explain-ing discordant cases among boys with SLAneed to be replicated. In addition, there needsto be further study of the potential mecha-nisms by which race and poverty influencecourse and outcome among boys who exhibitlittle in the way of aggressive0disruptive be-havior in elementary school.

Strengths of the study

Consistent with Schaeffer et al.~2003!, thepresent study improves upon previous longitu-dinal research in several important ways. First,the longitudinal sample used in the present studyis comprised of primarily African Americanurban youth, consistent with the US SurgeonGeneral’s call for more mental health researchamong ethnic minority populations~US PublicHealth Service, 2001!. Second, the study’s epi-demiologically defined population allows forgeneralization to similar populations of stu-dents entering school from urban neighbor-hoods with multiple problems~i.e., high ratesof poverty, unemployment, and crime!. Third,the study spans several important periods foryouth antisocial behavior development~ele-mentary school through young adulthood! anduses young adult outcomes that provide strongexternal validity for identified aggression tra-jectories. Fourth, the study examines predic-tors of discordance across important domainsin the child’s social ecology~i.e., individual,family, peer, school, and neighborhood fac-tors!. Fifth, the present study employs the new-est generation of latent growth modelingtechniques, GGMM~Muthén & Muthén,2000!, to address methodological limitationsof previous longitudinal modeling studies.

Limitations and Future Directions

A significant limitation of the present study isthe limited number of predictors examined with

respect to discordance. In addition, whereaswe have repeated assessments of aggressive0disruptive behavior throughout the elementary-school years, the assessments of the predictorsof discordance were limited to two time points:the entrance to first grade, and middle school.Repeated assessments of a wider array of po-tential predictors result in better prediction ofdiscordant cases and provide further evidencethat developmental pathways to antisocial be-havior remain malleable even for those boyswith a prolonged course of aggressive behavior.

Consistent with Schaeffer et al.~2003!, therewas no evidence in the present study for ahigh-but-desisting group as has been found inprevious studies modeling aggression withhigh-risk samples~e.g., Broidy et al., 2003;Maughan et al., 2000; Nagin & Tremblay,1999!. Although boys in the CHA group whodid not go on to have an antisocial outcomemight be considered to be a distinctive de-sister group, the latent class approach used inthe present study failed to empirically identifysuch a group. First, the relatively small sam-ple size in this study as well as the low prev-alence of boys with a CHA pattern~16%! inthis community sample may have resulted in asubsample size too small for further empiricalsubdivision~i.e., not enough meaningful vari-ation within this class to justify the extractionof an additional group!. Second, it is possiblethat the probability of desisting from aggres-sion will be much smaller in urban areas char-acterized by high rates of antisocial behaviorand violence. Another possibility is that thefinal time point used to identify growth trajec-tories in this study~i.e., spring of fifth grade!may have been too early in youth develop-ment for desistence from aggression yet to haveoccurred.

The fact that youth and teacher reports werethe primary source of data is an additional lim-itation. Multiple reporters and assessmentmethods would be preferable from the stand-point of reliability and validity. Furthermore,as is the case in most longitudinal studies, theend of the study is not the end of developmentfor study participants. Indeed, it is possiblethat a number of the participants engaged inantisocial behavior for the first time followingthe study’s age 19–20 end point. It is also pos-


sible that participants who had engaged in anti-social behavior prior to the study end pointmay have “desisted” from such behavior asadults. A final limitation of the present studyis the exclusive focus on boys. Future re-

search should include gender comparisons re-garding the strength of the association betweenthe course of aggression and later antisocialbehaviors as well as the impact of first-gradeand middle-school covariates.

References

Akaike, H. ~1987!. Factor analysis and AIC.Psycho-metrika, 52, 317–332.

American Psychiatric Association.~1994!. Diagnostic andstatistical manual of mental disorders~4th ed.!. Wash-ington, DC: Author.

Arbuckle, J. L.~1996!. Full information estimation in thepresence of incomplete data. In G. A. Marcoulides &R. E. Schumacker~Eds.!, Advanced structural equa-tion modeling: Issues and techniques. Mahwah, NJ:Erlbaum.

Beck, A. J., & Karberg, J.~2001!. Prison and Jail Inmatesat Midyear 2000. Washington, DC: Bureau of JusticeStatistics.

Bradley, R. H., & Corwyn, R. F.~2002!. Socioeconomicstatus and child development.Annual Review of Psy-chology, 53, 371–399.

Broidy, L. M., Nagin, D. S., Tremblay, R. E., Bates, J. E.,Brame, B., Dodge, K. A., Fergusson, D., Horwood,J. L., Loeber, R., Laird, R., Lyman, D. R., Moffitt,T. E., Pettit, G. S., & Vitaro, F.~2003!. Developmen-tal trajectories of childhood disruptive behaviors andadolescent delinquency:Asix-site, cross-national study.Developmental Psychology, 39, 222–245.

Bronfenbrenner, U.~1979!. The ecology of human devel-opment: Experiments by nature and design. Cam-bridge, MA: Harvard University Press.

Capaldi, D. M., & Patterson, G. R.~1989!. Psychometricproperties of fourteen latent constructs from theOregon Youth Study. New York: Springer–Verlag.

Cicchetti, D., & Schneider–Rosen, K.~1984!. Toward atransactional model of childhood depression. In D.Cicchetti & K. Schneider–Rosen~Eds.!, Childhooddepression a developmental perspective~pp. 5–28!.San Francisco, CA: Jossey–Bass.

Crijnen, A. A. M., Feehan, M., & Kellam, S. G.~1998!.The course and malleability of reading achievementin elementary school: The application of growth curvemodeling in the evaluation of a mastery learning in-tervention.Learning and Individual Differences, 10,137–157.

Curran, P. J., Muthén, B. O., & Harford, T. C.~1998!. Theinfluence of changes in marital status on developmen-tal trajectories of alcohol use in young adults.Journalof Studies on Alcohol, 59, 647–658.

Deater–Deckard, K.~2001!. Annotation: Recent researchexamining the role of peer relationships in the devel-opment of psychopathology.Journal of Child Psy-chology and Psychiatry and Allied Disciplines, 42,565–579.

Denno, D. W.~1990!. Biology and violence: From birthto adulthood. Cambridge: Cambridge University Press.

Dolan, L. J., Kellam, S. G., Brown, C. H., Werthamer–Larsson, L., Rebok, G. W., Mayer, L. S., Laudolff, J.,Turkkan, J., Ford, C., & Wheeler, L.~1993!. The short-term impact of two classroom-based preventive inter-ventions on aggressive and shy behaviors and poor

achievement.Journal of Applied Developmental Psy-chology, 14, 317–345.

Elliot, D. S., Huizinga, D., & Ageton, S. S.~1985!. Ex-plaining delinquency and drug use. Beverly Hills, CA:Sage.

Ensminger, M. E., Forrest, C. B., Riley, A. W., Kang, M.,Green, B. F., Starfield, B., & Ryan, S. A.~2000!. Thevalidity of measures of socioeconomic status of ado-lescents.Journal of Adolescent Research, 15, 392–419.

Ensminger, M. E., Kellam, S. G., & Rubin, B. R.~1983!.School and family origins of delinquency: Compari-sons by sex. In K. T. Van Dusen & S. A. Mednick~Eds.!, Prospective studies of crime and delinquency~pp. 73–97!. Boston: Kluwer–Niejhoff.

Farrington, D.P.~1989!. Early predictors of adolescentaggression and violence.Violence and Victims, 4,79–100.

French, D. C., Conrad, J., & Turner, T. M.~1995!. Adjust-ment of antisocial and nonantisocial rejected adoles-cents.Development and Psychopathology, 7, 857–874.

Greenberg, M., Domitrovich, C., & Bumbarger, B.~2001!.The prevention of mental disorders in school-aged chil-dren: current state of the field.Prevention & Treat-ment, 4.

Hanson, W. B., Tobler, N. S., & Graham, J. W.~1990!.Attrition in substance abuse prevention research: Ameta-analysis of 85 longitudinally followed cohorts.Evaluation Review, 14, 677–685.

Haselager, G. J. T., Van Lieshout, C. F. M., Riksen–Walraven, J. M. A., Cillessen, A. H. N., & Hartup,W. W. ~2002!. Heterogeneity among peer-rejectedboys across middle childhood: Developmental path-ways of social behavior.Developmental Psychology,38, 446–456.

Hawkins, J. D., Herrenkohl, T. I., Farrington, D. P., Brewer,D., Catalano, R. F., Harachi, T. W., & Cothern, L.~2000!, Predictors of youth violence.Office of Juve-nile Justice and Delinquency Prevention Bulletin, 4,1–11.

Hektner, J. M., August, G. J., & Realmuto, G. M.~2000!.Patterns and temporal changes in peer affiliation amongaggressive and nonaggressive children participatingin a summer school program.Journal of Clinical ChildPsychology, 29, 603–614.

Ialongo, N. S., Werthamer, L., Kellam, S. G., Brown, C. H.,Wang, S., & Lin, Y.~1999!. Proximal impact of twofirst-grade preventive interventions on the early riskbehaviors for later substance abuse, depression, andantisocial behavior.American Journal of CommunityPsychology, 27, 599–641.

Ingoldsby, E. M., & Shaw, D. S.~2002!. Neighborhoodcontextual factors and early-starting antisocial path-ways.Clinical Child and Family Psychology Review,5, 21–55.

Jensen, P. S., Martin, D., & Cantwell, D. P.~1997!. Co-morbidity in ADHD: Implications for research, prac-


tice, andDSM-IV. Journal of the American Academyof Child and Adolescent Psychiatry, 36, 1065–1079.

Jo, B., Muthén, B., Ialongo, N. S., & Brown, C. H.~2004!.Cluster randomized trials with nonadherence. Manu-script submitted for publication.

Kellam, S. G., Werthamer–Larsson, L., Dolan, L. J., Brown,C. H., Mayer, L. S., Rebok, G. W., Anthony, J. C.,Laudolff, J., Edelsohn, G., & Wheeler, L.~1991!. De-velopmental epidemiologically-based preventive tri-als: Baseline modeling of early target behaviors anddepressive symptoms.American Journal of Commu-nity Psychology, 19, 563–584.

Kellam, S. G., Mayer, L. S., Rebok, G. W., & Hawkins,W. E. ~1998!. The effects of improving achievementon aggressive behavior and of improving aggressivebehavior on achievement through two prevention in-terventions: An investigation of causal paths. In B.Dohrenwend~Ed.!, Adversity, stress and psychopa-thology ~Vol. 27, pp. 486–505!. New York: OxfordUniversity Press.

Lahey, B. B., McBurnett, K., & Loeber, R.~2000!. Areattention-deficit0hyperactivity disorder and opposi-tional defiant disorder developmental precursors toconduct disorder? In A. J. Sameroff & M. Lewis~Eds.!,Handbook of developmental psychopathology~2nd ed.,pp. 431–446!. New York: Plenum Press.

Little, R. J. ~1995!. Modeling the dropout mechanism inrepeated-measures studies.Journal of the AmericanStatistical Association, 90, 1112–1121.

Lo, Y., Mendell, N. R., & Rubin, D. B.~2001!. Testing thenumber of components in a normal mixture.Bio-metrika, 88, 767–778.

Loeber, R., & Hay, D.~1997!. Key issues in the develop-ment of aggression and violence from childhood toearly adulthood.Annual Review of Psychology, 48,371–410.

Loeber, R., & Stouthamer–Loeber, M.~1998!. Develop-ment of juvenile aggression and violence: Some com-mon misconceptions and controversies.AmericanPsychologist, 53, 242–259.

Loeber, R., Wei, E., Stouthamer–Loeber, M., Huizinga,D., & Thornberry, T. P.~1999!. Behavioral anteced-ents to serious and violent offending: Joint analysesfrom the Denver Youth Survey, Pittsburgh Youth Study,and the Rochester Youth Development Study.Studieson Crime and Crime Prevention, 8, 245–263.

Loeber, R., Wung, P., Keenan, K., Giroux, B., Stouthamer–Loeber, M., Van Kammen, W. B., & Maughan, B.~1993!. Developmental pathways in disruptive childbehavior. Development and Psychopathology, 5,103–133.

Lynam, D. R.~1996!. The early identification of chronicoffenders: Who is the fledging psychopath?Psycho-logical Bulletin, 120, 209–234.

Magnusson, D.~1998!. The logic and implications of aperson-oriented approach. In R. B. Cairns, L. R. Berg-man, & J. Kagan~Eds.!, Methods and models for study-ing the individual~pp. 33–64!. Thousand Oaks, CA:Sage.

Maguin, E., & Loeber, R.~1996!. Academic performanceand delinquency. In M. Tomry~Ed.!, Crime and Jus-tice: A review of research~Vol. 20, pp. 145–264!. Chi-cago: University of Chicago Press.

Maughan, B., Pickles, A., Rowe, R., Costello, E. J., &Angold, A. ~2000!. Developmental trajectories of ag-gressive and nonaggressive conduct problems.Jour-nal of Quantitative Criminology, 16, 199–221.

Maughan, B., & Rutter, M.~1998!. Continuities and dis-continuities in antisocial behavior from childhood toadult life. In T. H. Ollendick & R. J. Prinz~Eds.!,Advances in clinical child psychology~Vol. 20, pp.1–47!. New York: Plenum Press.

McCord, J. ~1983!. A longitudinal study of aggressionand antisocial behavior. In K. T. Van Dusen & S. A.Mednick~Eds.!, Studies of crime and delinquency~pp.269–275!. Boston: Kluwer–Nijhoff.

McCord, J., & Ensminger, M. E.~1997!, Multiple risksand co-morbidity in an African-American population.Criminal Behavior and Mental Health, 7, 339–352.

Moffitt, T. E. ~1993!. Adolescent-limited and life-coursepersistent antisocial behavior: A developmental tax-onomy.Psychological Review, 100, 674–701.

Moffitt, T. E., & Caspi, A. ~2001!. Childhood predictorsdifferentiate life-course persistent and adolescence-limited antisocial pathways among males and fe-males. Development and Psychopathology, 13,355–375.

Moffitt, T. E., Caspi, A., Harrington, H., & Milne, B. J.~2002!. Males on the life-course-persistent andadolescence-limited antisocial pathways: Follow-upat age 26 years.Development and Psychopathology,14, 179–207.

Muthén, B.~2000!. Methodological issues in random co-efficient growth modeling using a latent variable frame-work: Applications to the development of heavydrinking in ages 18–37. In J. S. Rose, L. Chassin, C.Presson, & J. Sherman~Eds.!, Multivariate applica-tions in substance use research: New methods for newquestions~pp. 113–140!. Mahwah, NJ: Erlbaum.

Muthén, B.~2001!. Latent variable mixture modeling. InG. A. Marcoulides & R. E. Schumacker~Eds.!, Newdevelopments and techniques in structural equationmodeling~pp. 1–33!. Washington, DC: APA.

Muthén, B.~2003!. Statistical and substantive checkingin growth mixture modeling: Comment on Bauer andCurran~2003!. Psychological Methods, 3, 369–377.

Muthén, B., & Muthén, L. K.~2000!. Integrating person-centered and variable-centered analyses: Growth mix-ture modeling with latent trajectory classes.Alcoholism:Clinical and Experimental Research, 24, 1–10.

Muthén, B., & Shedden, K.~1999!. Finite mixture mod-eling with mixture outcomes using the EM algorithm.Biometrics, 6, 463–469.

Muthén, L. K., & Muthén, B. O.~1998!. Mplus usersguide. Los Angeles, CA: Muthén and Muthén.

Nagin, D. S.~1999!. Analyzing developmental trajecto-ries: A semiparametric, group-based approach.Psy-chological Methods, 4, 139–157.

Nagin, D., & Tremblay, R. E.~1999!. Trajectories of boys’physical aggression, opposition, and hyperactivity onthe path to physically violent and nonviolent juveniledelinquency.Child Development, 70, 1181–1196.

Oxford, M. L., Gilchrist, L. D., Morrison, D. M., Gill-more, M. R., Lohr M. J., & Lewis, S. M.~2003!. Al-cohol use among adolescent mothers: Heterogeneityin growth curves.Prevention Science, 4, 15–26.

Patterson, G. R.~1996!. Some characteristics of a devel-opmental theory for early-onset delinquency. In M. F.Lenzenweger & J. J. Haugaard~Eds.!, Frontiers ofdevelopmental psychopathology~pp. 81–124!. NewYork: Oxford University Press.

Patterson, G. R., DeBaryshe, B. D., & Ramsey, E.~1989!.A developmental perspective on antisocial behavior.American Psychologist, 44, 329–335.


Patterson, G. R., Forgatch, M. S., Yoerger, K. L., &Stoolmiller, M.~1998!. Variables that initiate and main-tain an early-onset trajectory for juvenile offending.Development and Psychopathology, 10, 531–547.

Patterson, G. R., Reid, J. B., & Dishion, T. J.~1992!. Asocial learning approach: Vol. 4. Antisocial boys.Eugene, OR: Castilia.

Ramaswamy, V., DeSarbo, W., Reibstein, D., & Robin-son, W. ~1993!. An empirical pooling approach forestimating marketing mix elasticities with PIMS data.Marketing Science, 12, 103–124.

Reid, J., Eddy, M., & Fetrow, R.~1999!. Description andimmediate impacts of a preventive intervention forconduct problems.American Journal of CommunityPsychology, 27, 483–517.

Sandler, I. N., Tein, J. Y., Mehta, P., Wolchik, S., & Ayers,T. ~2000!. Coping efficacy and psychological prob-lems of children of divorce.Child Development, 71,1099–1118.

Satterfield, J. H., & Schell, A.~1997!. A prospective studyof hyperactive boys with conduct problems and nor-mal boys: Adolescent and adult criminality.Journalof the American Academy of Child and AdolescentPsychiatry, 36, 1726–1735.

Schaeffer, C., Petras, H., Ialongo, N., Poduska, J., &Kellam, S.~2003!. Modeling growth in boys’ aggres-sive behavior across elementary school: Links to latercriminal involvement, conduct disorder, and anti-social personality disorder.Developmental Psychol-ogy, 39, 1020–1035.

Schwartz, G.~1978!. Estimating the dimensions of amodel.The Annals of Statistics, 6, 461–464.

Schafer, J. L., & Graham, J. W.~2002!. Missing data: Ourview of the state of the art.Psychological Methods, 7,147–177.

Schwartz, G.~1978!. Estimating the dimension of a model.The Annals of Statistics, 6, 461–464.

Schwartz, D.~2000!. Subtypes of victims and aggressorsin children’s peer groups.Journal of Abnormal ChildPsychology, 28, 181–192.

Sclove, L. S.~1987!. Application of model-selection cri-teria to some problems in multivariate analysis~Tech.Rept.!. Los Angeles: University of California.

Shaw, D. S., Gilliom, M., Ingoldsby, E. M., & Nagin,D. S. ~2003!. Trajectories leading to school-ageconduct problems.Developmental Psychology, 39,189–200.

Sickmund, M.~2000!. Offenders in Juvenile Court, 1997.Juvenile Justice Bulletin, 7.

Snyder, H. N., & Sickmund, M.~1999!. Juvenile offend-ers and victims: 1999 national report. Washington,DC: National Center for Juvenile Justice.

Snyder, H. N., & Sickmund, M.~2000!. Challenging themyths.Juvenile Justice Bulletin, May 25th.

Stattin, H., & Kerr, M.~2000!. Parental monitoring.ChildDevelopment, 71, 1072–1085.

Tolan, P. H., & Gorman–Smith, D.~1998!. Developmentof serious, violent, and chronic offenders. In R. Loe-ber & D. Farrington~Eds.!, Never too early, never toolate: Serious, violent, and chronic juvenile offenders~pp. 68–85!. Beverly Hills, CA: Sage.

US Public Health Service.~2001!. Mental health: Cul-ture, race, and ethnicity. A supplement to mentalhealth: A report of the Surgeon General. Rockville,MD: US Department of Health and Human Services,Substance Abuse and Mental Health Services Admin-istration, Center for Mental Health Services, Na-tional Institutes of Health, National Institute of MentalHealth.

Wardrop, J. L.~1989!. Review of the California Achieve-ment Tests, Forms E and F. In J. Close Conoley & J.Kramer~Eds.! The tenth mental measurements year-book ~pp. 128–133!, Lincoln, NE: University of Ne-braska Press.

Werthamer–Larsson, L., Kellam, S. G., & Wheeler, L.~1991!. Effect of first-grade classroom environmenton child shy behavior, aggressive behavior, and con-centration problems.American Journal of CommunityPsychology, 19, 585–602.

White, H. R., Xie, M., Thompson, W., Loeber, R., &Stouthamer–Loeber, M.~2001!. Psychopathology asa predictor of adolescent drug use trajectories.Psy-chology of Addictive Behaviors, 15, 210–218.

Yoshikawa, H.~1994!. Prevention as cumulative protec-tion: Effects of early family support and education onchronic delinquency and its risks.Psychological Bul-letin, 115, 28–54.

Zimmerman, R. S., Khoury, E. L., Vega, W. A., & Gil,A. G. ~1995!. Teacher and parent perceptions of be-havior problems among a sample of African-American,Hispanic, and non-Hispanic White students.Ameri-can Journal of Community Psychology, 23, 181–197


when the course of aggressive behavior in childhood does ... · hanno petras,a cindy m. schaeffer,b...

Documents