International Journal of Behavioral

Development http://jbd.sagepub.com

A cohort-sequential multivariate latent growth curve analysis of normative CBCL aggressive and
delinquent problem behavior: Associations with harsh discipline and gender
P. Prinzie, P. Onghena and W. Hellinckx
International Journal of Behavioral Development 2006; 30; 444
DOI: 10.1177/0165025406071901

The online version of this article can be found at:

http://jbd.sagepub.com/cgi/content/abstract/30/5/444

Published by:

http://www.sagepublications.com

On behalf of:

International Society for the Study of Behavioral Development

Additional services and information for International Journal of Behavioral Development can be found at:

Email Alerts: http://jbd.sagepub.com/cgi/alerts

Subscriptions: http://jbd.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations (this article cites 55 articles hosted on the

SAGE Journals Online and HighWire Press platforms):
http://jbd.sagepub.com/cgi/content/abstract/30/5/444#BIBL

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
International Journal of Behavioral Development © 2006 The International Society for the
2006, 30 (5), 444–459 Study of Behavioural Development
http://www.sagepublications.com DOI: 10.1177/0165025406071901

A cohort-sequential multivariate latent growth curve analysis of

normative CBCL aggressive and delinquent problem behavior:
Associations with harsh discipline and gender

P. Prinzie P. Onghena and W. Hellinckx

Utrecht University, The Netherlands Katholieke Universiteit Leuven, Belgium

The aim of this study was to examine the normative developmental trajectories of aggressive and
delinquent behavior in young children. Cohort-sequential univariate latent growth modeling (LGM)
analyses were employed to conceptualize and analyze intraindividual changes in children’s aggressive
and delinquent behavior and interindividual differences in these changes. A multivariate model was
tested that related the two developmental trajectories to each other and to harsh discipline. The
longitudinal data included mother and father ratings on the Child Behavior Checklist (CBCL)
(Achenbach, 1991), the “Leuvens Instrument voor Coërcief Opvoedingsgedrag” (Leuvens Instru-
ment of Coercive Parenting Behavior, LICO) (Hellinckx et al., 2000) and the Parenting Scale
(Arnold, O’Leary, Wolff, & Acker, 1993) of 674 school-aged boys and girls of a proportional strat-
ified general population sample, assessed annually for three years. A significant nonlinear decline in
aggressive and a significant linear decline in delinquent problem behavior were found both in the
mother and in the father ratings. A multivariate latent growth analysis indicated that trajectories in
aggressive and delinquent problem behavior were positively associated. The association was stronger
for boys than for girls. Parenting behaviors were differentially related to children’s aggressive and
delinquent problem behavior. Coercion was significantly related to aggressive behavior but not to
delinquent behavior. Higher scores on coercion were related to higher initial levels and a slower
decrease of aggressive behavior. High scores on overreactivity were associated with higher initial levels
of aggressive and delinquent problem behavior but not with the growth rates. Boys were higher than
girls in initial status. Conversely, the rate of change was not related to gender. The results were repli-
cated in the father reports.

Keywords: aggressive behavior; CBCL; cohort-sequential latent growth analysis; delinquent

behavior; development; externalizing behavior; harsh discipline; sex differences

Introduction behavior. Theoretically, the detection of factors accounting for

individual variability about the population’s mean trajectory
Externalizing problem behaviors are the most common and increases our knowledge of the development of psycho-
persistent forms of childhood maladjustment (Campbell, pathology in children. The knowledge of mechanisms under-
1995) and change not only in frequency but also in expression lying changes in problem behavior among non-referred
over the course of development. Therefore, there is a growing children has promise for both effective prevention and
consensus that externalizing behavior must be studied from a broadened approaches to later intervention (Sroufe, 1997).
developmental perspective (Loeber & Farrington, 1997; Several studies investigated the development of externaliz-
Sroufe, 1997). Longitudinal research on representative general ing behavior in the general population (e.g., Bongers, Koot,
population samples is necessary to determine the typical Van der Ende, & Verhulst, 2003; Hofstra, Van der Ende, &
course of childhood problems. As is emphasized by Sroufe Verhulst, 2000; Loeber, Green, Lahey, Frick, & McBurnett,
(1997, p. 258) “the starting point for a developmental 2000; Moffitt, Caspi, Harrington, & Milne, 2002). Most of
approach to psychopathology is always a consideration of these studies focused on the development of externalizing
normal development”. The same laws that govern normal behavior problems in boys or were based on mother ratings.
development govern the pathological as well (Cicchetti, 1984). Although the significance of developmental trajectories is
If externalizing problem behavior in children fails to show emphasized, there has been little attention to relations between
persistence across time, this would cast doubt on the necessity trajectories and relevant time-varying covariates.
of treatment. However, if persistence of problem behavior Within cross-sectional studies, empirically-based multi-
confirms to be substantial, this would speak against a wait- variate studies have identified two general types of externaliz-
and-see policy. Finally, the variability in pathways of external- ing behavior problems: overtly aggressive behavior and more
izing problem behavior argues for a need to understand the covertly delinquent behavior (Achenbach, Conners, Quay,
specific factors that have an influence on changes in problem Verhulst, & Howell, 1989; Frick et al., 1993). The aggressive

Correspondence should be sent to P. Prinzie, Department of Child

and Adolescent Studies, Utrecht University, P.O. Box 80 140, 3508
TC Utrecht, The Netherlands; e-mail: [email protected]

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 445

behavior syndrome includes behaviors such as teasing, provided information. Because the aim of this study was to
fighting, or hitting, but also incorporates aspects of aggressive describe the normative co-development of aggressive and
personality such as arguing, bragging and hot temper. The delinquent behavior, in the next sections we only review those
delinquent behavior syndrome includes behaviors such as studies that included samples from the general population.
stealing, lying and cheating. Because relatively few develop- Further, we restrict our review to studies that used the CBCL
mental studies have made a distinction between different forms (Achenbach, 1991; Verhulst, Van der Ende, & Koot, 1996), to
of externalizing behavior (see e.g., Bongers, Koot, Van der enable a direct comparison with our own perspective and data.
Ende, & Verhulst, 2004; Stanger, Achenbach, & Verhulst,
1997), little is known about the normative co-development of
aggressive and delinquent behavior in the general population.
Average development of aggressive and delinquent
The differentiation of externalizing behavior in these two types
behavior problems
enables the investigation of differential developmental patterns With regard to age effects, findings are not unambiguous. To
for different manifestations of externalizing behavior. our knowledge, the study by Stanger et al. (1997) is one of the
Moreover, distinguishing different types of externalizing first studies that investigated the aggressive and delinquent
behavior allows the study of comorbidity of behavior clusters syndromes separately. The aggressive syndrome was found to
and developmental pathways both within and across domains. be more stable over time than the delinquent syndrome. Using
For example, studies by Loeber and colleagues (e.g., Loeber repeated measures analysis of variance (ANOVA) of data from
et al., 1993) suggest three different pathways of externalizing seven cohorts of Dutch children, they reported that aggression
behavior problems in males that predict different outcomes. declined steadily from about age 4 on, whereas delinquency
Recently, in their cross-sectional study Barnow, Lucht, and declined slightly from age 4 until age 10 and then started to
Freyberger (2005) found different correlates for aggressive and increase again, reaching a peak at about age 17. However,
delinquent behaviors in adolescents. ANOVA-analyses focus only on the factor means. A fully
During the last decade, important advances in the field of expanded latent growth curve analysis on the other hand, takes
measurement and analysis of normative change enabled into account both factor means and variances. This combi-
scientists to incorporate both invariant characteristics and nation of individual and group levels of analysis is unique to
continuous change in developmental models (Collins & Sayer, the latent growth modeling procedure (LGM). Therefore,
2001; Singer & Willett, 2003). Examples of these quantitative growth curve modeling is well suited to answer questions
methods are hierarchical models (Bryk & Raudenbush, 1992), concerning systematic differences in individual developmental
group-based trajectory modeling (Nagin, 1999, 2005), and processes.
latent growth models (Duncan, Duncan, Strycker, Li, & Bongers et al. (2003) showed a declining trajectory of
Alpert, 1999). Although these procedures differ in important mother-reported aggressive behavior over time for both boys
ways (Collins & Sayer, 2001; Willett & Sayer, 1994), all and girls. Children with a lower level of aggressive behavior
estimate trajectory parameters (e.g., initial level and rate of changed at a faster rate. Delinquent behavior showed a
change) for each person and specify how these parameters vary curvilinear developmental trajectory peaking at age 11 years.
throughout the population. The group-based method focuses Children with a lower initial score had a greater quadratic
on identification of different trajectory shapes and on examin- change than children with a higher initial value on delinquent
ing how the prevalence of the shape and the shape itself relate behavior. In a review of Loeber, Lahey, and Thomas (1991) it
to predictors. By contrast, standard growth curve modeling appears that most symptoms of oppositional defiant disorder
focuses on the population mean trajectory and how individual are common by at least age 4–5 years and then decline in
variation about that mean relates to predictors (Nagin & prevalence with increasing age in most children (see also,
Tremblay, 2005). The group-based approach is ideally suited Achenbach & Edelbrock, 1981). In a longitudinal study from
for testing whether distinctive developmental trends are preschool to third grade, Munson, McMahon, and Spieker
present in the outcome variable of interest. By contrast, (2001) found an increasing mean growth trajectory for exter-
standard growth curve modeling is more appropriate to test nalizing behavior problems of children of adolescent mothers.
whether some identifiable characteristic (e.g. sex) is associated The greater the mother’s depression, the faster externalizing
with individual differences in trajectories of development problems tended to increase. Other studies reported also that
(Nagin, 2005). The group-based approach lends itself to the trajectories of boys’ externalizing behaviors tend to
analyzing questions that are framed in terms of the shape of increase or remain fairly stable through the period from kinder-
the developmental course of the outcome of interest, whereas garten to seventh grade (see e.g., Loeber et al., 1993).
standard growth curve modeling lends itself to analyzing ques-
tions framed in terms of predictors of the outcome’s develop-
mental course (Nagin, 2005, p. 9).
Gender differences in the development of aggressive
Until recently, technical limitations restricted analyses of
and delinquent behavior problems
individual change to one measurement series at a time. Previous studies of gender differences for externalizing behav-
However, advances in latent growth modeling now allow the iors have reported conflicting results (Bongers et al., 2003).
investigation of developmental trajectories in several domains Moreover, findings are not ubiquitous or clear because of vari-
simultaneously and to investigate how these trajectories relate ations in sampling, informants, measurement strategies, and
to one another (Blozis, 2004; Willett & Sayer, 1996). The statistical analysis (Lahey, Miller, Gordon, & Riley, 1999). In
present study examined the co-development of aggressive and the study by Stanger et al. (1997) boys were scored higher than
delinquent behavior in a general population-based sample of girls on both aggressive and delinquent behaviors. The study
children aged 4 to 9 years including boys as well as girls. In by Keiley, Bates, Dodge, and Pettit (2000) used a single-cohort
addition, the relations between these trajectories and harsh longitudinal design and reported that mothers and teachers
parenting were investigated. Fathers as well as mothers differed in their reports on externalizing behavior. Teachers

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
446 PRINZIE ET AL. / A COHORT-SEQUENTIAL MULTIVARIATE LATENT GROWTH CURVE ANALYSIS

reported significantly more externalizing behaviors in boys relate to one another. Until now, most longitudinal studies that
than in girls and a decline over time that was faster for boys investigated growth curves of antisocial behaviors have focused
than for girls. A similar declining trajectory was found for on change in only one domain at a time (e.g., Bongers et al.,
mother-reported externalizing behavior, but no effect for 2003). To our knowledge, no studies have used an integrative
gender was found. Recently, Bongers et al. (2003) showed that analytical model to longitudinal data on problem behavior in
for aggressive behavior boys at age 4 scored higher than girls a gender and time balanced sample of young children to
and that boys decreased at a faster rate than girls. For delin- describe the intraindividual changes of aggressive and delin-
quent behavior at age 4, boys had higher scores than girls. No quent behavior over time and the interindividual differences in
gender differences on the slope were reported. In a cross- these changes. Adequately describing these changes is a necess-
sectional study, Lahey et al. (2000) reported no gender differ- ary condition to link more closely the conceptualization of
ences in opposition but aggression and property violations children’s development of externalizing problem behavior and
were more common among boys. However, in the study by the associated longitudinal data. These analyses provide initial
Munson et al. (2001), from preschool through third grade no insight into the characteristics of developmental trajectories in
significant gender difference was observed for externalizing problem behavior prior to introducing additional explanatory
scores of children of adolescent mothers. Recently, Bongers measures of interest.
et al. (2004) reported that the shape of the developmental This study used mother and father reports on children’s
trajectories of aggressive and oppositional behavior hardly problems on the Child Behavior Checklist (CBCL) (Achen-
differed between males and females. Moffitt and Caspi (2001) bach, 1991; Verhulst et al., 1996) obtained at three time
found that with regard to adolescent delinquency, males and points with 1-year intervals. The focus was on five questions
females in the life-course persistent antisocial pathway shared related to the development of aggressive and delinquent
the same childhood risk factors. It is important to note that behavior problems in young children: First, is there evidence
most of these studies did not explicitly model individual for systematic change and individual variability in change in
change but rather statistically controlled for previous levels of aggressive and delinquent behavior during childhood? Based
problems. on the literature (see e.g., Bongers et al., 2003), we expected
a declining normative developmental trajectory on the aggres-
sive and delinquent behavior scale. Second, are the develop-
The role of harsh parenting in the development of mental trajectories of aggressive and delinquent behavior
externalizing behavior related to each other? Because we were not aware of any other
In contrast to the divergent findings with regard to the impact study that has investigated the particular connections between
of age and gender, extensive empirical work exists linking the normative development of aggressive and delinquent
parenting practices to childhood externalizing behavior behaviors, we did not formulate specific hypotheses on the
problems (Patterson, 1982; Patterson, Reid, & Dishion, 1992; relation between the initial levels and growth. Third, are
Rothbaum & Weisz, 1994; Snyder, Cramer, Afrank, & developmental pathways the same for boys and girls? Based on
Patterson, 2005). From a social interactional perspective previous research (see e.g., Bongers et al., 2003), we expected
Patterson and his colleagues (e.g., Patterson et al., 1992) that boys scored higher on aggressive and delinquent behavior
stated that family members inadvertently provide reinforce- compared to girls. With respect to the growth, we had no
ments for antisocial behavior and fail to provide effective expectation on differences in growth. Fourth, to what extend
punishment for misbehaviors.While some of the reinforcement is harsh discipline related to developmental processes in
for coercive behaviors is positive (attend, laugh, or approve), aggressive and delinquent problem behavior? Based on the
the most important set of contingencies for coercive behavior literature (see e.g., Patterson et al., 1992), we expected signifi-
consists of escape-conditioning contingencies (Patterson et al., cant associations between harsh discipline and the initial levels
1992). In the latter, the child uses aversive behaviors to stop and growth of aggressive and delinquent problem behavior.
aversive intrusions by other family members. These reinforc- Fifth, are the developmental pathways and the associations
ing contingencies embedded in social interactions are actually with harsh parenting the same for mother as for father ratings?
the direct determinants for children’s aggression. As this We expected that fathers would also make significant contri-
process continues, the child and other family members butions to children’s normal development.
progressively escalate in the intensity of their coercive behav-
iors, often leading to high-amplitude behaviors such as hitting
and physical attacks. In this training, the child eventually learns
to control other family members through coercive means
Methods
(Patterson, 1982; Patterson et al., 1992).
Participants
A sample of regular elementary schools was randomly
Research questions selected. Within the schools that agreed to participate, a
The purpose of this study was to extend prior research on proportional stratified sample of school-aged children was
young children’s development of externalizing problem randomly selected. Strata were constructed according to
behavior by modeling an appropriate growth curve that parsi- geographical location (province), sex and age. Parents
moniously and accurately depicts intraindividual changes in received an invitation letter to participate in “a study of
aggressive and delinquent problem behavior and interindivid- parenting and child development”. If these parents agreed to
ual differences in these changes from age 4 to 9 years, using be contacted by the investigators, the first author called the
cohort-sequential latent growth analyses. We estimated the parents, explained the study and obtained written permission.
initial status and rate of change in both domains simul- All subjects took part voluntarily, and anonymity and confi-
taneously and modeled how the changes in these domains dentiality were guaranteed.

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 447

The initial sample consisted of 800 families from which 674 Measures
(84.3%) responded to the mailed questionnaires with two
Aggressive and delinquent problem behavior. Parent’s global
postal reminders and two telephone calls. At the first assess-
perceptions of their child’s problem behaviors were assessed
ment target children in these families ranged in age from 4 to
using the Dutch translation of the Child Behavior Checklist
7 years (M = 5 years 10 months, SD = 1.12). There were 337
(CBCL) (Achenbach, 1991; Verhulst et al., 1996).This widely-
boys (M = 5 years 10 months, range: 4 years–7 years 11
used instrument has two parts, one part measuring children’s
months, SD = 1.12) and 337 girls (M = 5 years 10 months,
competencies and a second part consisting of 113 items describ-
range: 4 years–7 years 11 months, SD = 1.12). Data for both
ing a broad range of problems. Only the findings from the latter
parents were available for 627 children. From 42 children
part of the CBCL were used for the purpose of this study. Each
only the mother, and from 5 children only the father, rated
item on the CBCL is rated as 0 (not true), 1 (somewhat or some-
the questionnaires. All parents had Belgian nationality. The
times true), or 2 (very true or often true). The Child Behavior
mean age of the mothers was 34 years 11 months (range: 25
Checklist is an extensively-validated instrument that has
years 1 month–50 years, SD = 3. 62) and of the fathers 37
adequate reliability and validity when describing child behavior
years (range: 25 years 11 months–59 years 10 months, SD =
(Achenbach, 1991; Vignoe, Bérubé, & Achenbach, 2000). The
4.26). Percentages of mothers (M) and fathers (F) with
CBCL broadband externalizing scale encompasses the
various educational levels: elementary school (M: 0.9; F: 3.0),
syndromes aggressive behavior and delinquent behavior.
secondary education (M: 41.1, F: 43.3), non-university
The aggression subscale contains 20 items, including overt
higher education (M: 45.2; F: 34.4), university (M: 12.8; F:
aggressive behaviors such as arguing a lot, destroying one’s
19.2). The mean occupational level of mothers on a six-point
own and other’s belongings, being disobedient at home and at
scale (6 = highest; Van Westerlaak, Kropman, & Collaris,
school, fighting with other children, attacking others, and
1975) was 3.49 (SD = 1.38, range 1–6), and of fathers 3.59
threatening others. The delinquency subscale is made up of 13
(SD = 1.58, range 1–6) which is representative for the Belgian
items including more covert behaviors such as lying, cheating,
population.
being truant, having no guilt, stealing at home and elsewhere,
For the accelerated longitudinal design, data were used from
using drugs and alcohol. These scales are traditionally used
four different age cohorts (4, 5, 6, and 7 years of age at the
in raw score form by summing the score across all items
initial assessment). The 4- through 7-year-old cohorts were
(Achenbach, 1991). Cronbach’s alphas for the aggression
each measured three times at approximately 1-year intervals
subscale were in mother reports .89 and in the father reports
(Time 1 to Time 3), thus representing an overall age range of
.88 at each wave. Cronbach’s alphas for the delinquency
4 to 9 years. These age cohorts were “approximately staggered”
subscale ranged in the mother ratings from .48 to .58 and in
(McArdle & Anderson, 1990), meaning that the average age of
the father ratings from .44 to .49.
the first cohort at the second measurement period was about
the same as the average age of the second cohort at the initial
measurement, and so forth. The sample sizes for the four Harsh discipline. At the second assessment, both parents rated
cohorts were 162, 175, 167, and 170, respectively (N = 674). the “Leuvens Instrument voor Coërcief Opvoedingsgedrag”
At time 2, parents of 609 children (90%) and on time 3, (LICO) (Leuvens Instrument of Coercive Parenting Behavior)
parents of 588 children (87%) participated in the study. (Hellinckx et al., 2000). This questionnaire intends to measure
Attrition and other forms of missing data (e.g., omissions) coercion as described by Patterson et al. (1992). When parents
across the 3 years approximated 17% for the mother data and are inconsistent and capitulate to the child, Patterson (1982)
19% for the fathers, resulting in a sample of 555 and 511 hypothesized that they enter a “reinforcement trap” where
respectively, with complete data. Percentages of children in short-term gains (e.g., peace and quiet) are obtained at the cost
cohort 1, cohort 2, cohort 3, and cohort 4 with less than three of strengthening the child’s difficult behavior. This instrument
time points amount to: for the mother data (17.4; 16.3; 15.7; is novel in that it is based on the outcome of whole sequences
18.8) and for the father data (19.3; 18.1; 19.6; 19.5). of conflict rather than on immediate reactions to particular
In the mother and father reports, subjects with missing data individual behaviors. The LICO contains 10 descriptions of a
had higher Time1 externalizing problem scores than subjects situation in which the child is confronted with an aversive
with no missing data, but the difference was not statistically intrusion of the parents (e.g., clean up toys, go to bed, take a
significant. Also with respect to sex, age, parental occupational bath). For each situation, parents rated at maximum six items,
and educational level, no significant differences were found i.e., three sequences of actions of the child (e.g., when you ask
between the two groups. By estimating the maximum likeli- your child to go to bed, how will your child usually act?) and
hood function at the individual level (Arbuckle, 1996; Schafer reactions of the parent (e.g., given that your child acts like that
& Graham, 2002), data from all fathers (N = 632) and all . . . how do you usually react?). The response categories of the
mothers (N = 669), regardless of their pattern of missingness, child behavior range on a continuum from 1 (obey) to 4 (get
could be included in these analyses. Expected Maximization angry, hit). Parent behaviors range from 1 (give in) to 5 (punish
(EM) is a maximum likelihood procedure that uses iterations severely). If the child complies during the first or second
to impute missing values based on all available data. Schafer sequence, parents go on with the next situation. If the parents
and Graham (2002) recommend this procedure as being a capitulate to the child, a coercion score is calculated taking into
highly efficient way to use the available data under the assump- account the duration of the conflict (i.e., the longer the child
tion that data are missing at random. resists the request, the higher the coercion score) and the
Mothers and fathers rated children’s aggressive and delin- intensity of the aversive child behavior (i.e., the more aversive
quent behavior problems at three time points with an approx- the child reacts, the higher the coercion score). The total score
imately 1-year interval. At the second and third assessment for coercion is summated over the 10 situations. Cronbach’s
wave, parents completed a parenting questionnaire. Both alphas for the LICO were in the mother and father ratings .88
parents completed the questionnaires independently. and .91, respectively.

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
448 PRINZIE ET AL. / A COHORT-SEQUENTIAL MULTIVARIATE LATENT GROWTH CURVE ANALYSIS

At the third assessment, participants rated a translated specific age cohort can be linked together with similar
version of the Parenting Scale (Arnold, O’Leary, Wolff, & segments from other temporally related age cohorts to deter-
Acker, 1993). The Parenting Scale was originally developed as mine the existence of a common developmental trend or
a parent-report measure of dysfunctional parenting practices. growth curve. In that way, a long-term longitudinal study is
The scale consists of 30 items presenting discipline encoun- approximated by conducting several short-term longitudinal
ters (e.g., “When my child misbehaves . . .”) followed by two studies of different age cohorts at the same time.
options that act as opposite anchor points for a 7-point scale A noticeable advantage of the cohort-sequential design over
(e.g., “I do something about it right away” versus “I do some- the single-cohort longitudinal design is the shorter follow-up
thing about it later”). Based on factor analytic findings, three period. This reduces the problems of cumulative testing effects
subscales were developed: overreactivity, laxness and verbosity. and attrition, and produces quicker results. Another advantage
The overreactivity and laxness factor have adequate test–retest is that tracking several cohorts, rather than one, allows the
reliability, distinguish clinical from nonclinical samples, and researcher to determine whether those trends observed in the
have been validated against behavioral observations of parent- repeated observations are corroborated within short time
ing (Arnold et al., 1993). An exploratory factor analysis of the periods for each age cohort.
translated version revealed two interpretable factors corre- In this study, statistical analyses were performed in five steps.
sponding with the overreactivity and laxness factors identified In the first step, univariate latent growth models were fitted to
in previous studies of the parenting scale (Prinzie, Onghena, the mother ratings of children’s aggressive and delinquent
& Hellinckx, in press). This study focused on the overreactivity problem behavior to determine the form of the growth trajec-
(OVR) scale. The OVR scale contains nine items and measures tory that most adequately described intraindividual change
a tendency exhibited by parents to respond with anger, frus- and interindividual differences in each setting. These analyses
tration, meanness and irritation, impatiently and aversively to provided initial insight into the characteristics of develop-
problematic behavior of their children. Arnold et al. (1993) mental trajectories in aggressive and delinquent problem
equated the overreactivity factor with an authoritarian parent- behavior. In the second step, multivariate latent growth curves
ing style, which included threats and physical punishment. were tested to investigate the covariation between the develop-
Cronbach’s alphas for the overreactivity scale were in the mental processes of aggressive and delinquent behavior. In the
mother and father sample .78 and .77, respectively. third step, we performed a multigroup analysis to investigate
the effect of child’s gender. In the fourth step, the effect of
harsh discipline on the intercept and the slope of aggressive
Statistical analysis and delinquent behavior was explored. Finally, to address
The techniques of latent growth curve modeling (LGM) were possible problems with post hoc model fitting, we tested
used to examine changes in aggressive and delinquent behavior whether the final multivariate model with predictors could be
problems across time. LGM is a powerful and flexible tech- replicated across the father sample. This invariance testing
nique that can be used to model longitudinal change in strategy gives an indication for the stability of the model. As
repeated observations of one variable (e.g., Duncan et al., Bollen (1989) has indicated, this excessively rigid test of cross-
1999; Meredith & Tisak, 1990). LGM gives an indication of validation is appropriate when a multigroup focus is directed
the persistence measured by comparing the absolute level of more toward the equality of structural – rather than measure-
problem behavior an individual has at the three time points. ment – parameters. We opted for this approach and not for a
The focus in latent growth analysis is not so much on the full latent variable model approach because we had only two
observed repeated measures of a theoretical construct over indicators (mother and father ratings) for the major latent
time but on the unobserved latent factors that are thought to constructs. Bollen (1989), among others suggests that three
have given rise to the set of observed measures. The strongest indicators are a minimum and others point out that four
advantage of latent growth curve analyses for time-ordered indicators per latent variable are necessary to avoid a just-
approaches is that LGM has the ability to incorporate infor- identified model.
mation concerning the group or population and also contains The model depicted in Figure 1 represents a cohort-
specific information concerning changes in the individual. In sequential univariate latent growth model in which the basic
other words, the subjects differ in their level at the first parameters describe a systematic pattern of individual differ-
measurement occasion and develop at different rates. Meredith ences in change over time. The same developmental model is
and Tisak (1990) noted that repeated measures polynomial assumed in each age cohort (4-, 5-, 6-, and 7-year-olds), each
analysis of variance (ANOVA) models are actually special cases measured at three approximately equal time intervals over a
of LGMs in which only the factor means are of interest. In 3-year period, allowing for tests of hypotheses concerning
contrast, a fully-expanded growth analysis takes into account convergence across separate groups and the feasibility of
both factor means and variances. This combination of the specifying a common growth trajectory over the six years repre-
individual and group levels of analysis is unique to this sented by the latent variable cohort sequential design. The εs
procedure. represent error variances in the repeated measures. The first
In this study, the cohort-sequential (or accelerated) latent factor (F1) represents the intercept and corresponds to
approach was used to model the development of children’s the initial status of children’s aggressive problem behavior, that
externalizing problem behavior from 4 to 9 years of age over a is, the aggressive score at Time 1. The intercept is a constant
3-year period. The cohort-sequential design (Nesselroade & for any individual across time and represents information
Baltes, 1979) consisted of limited repeated measurements of concerning the mean, represented by µI , and variance, repre-
independent age cohorts, resulting in temporally-overlapping sented by σ2I , of the collection of individual intercepts that
measurements of the various groups. The technique, which characterize each individual’s growth curve. Hence, in our
was first introduced by Bell (1953, 1954) provides a way by models all measured variables from the three assessments have
which adjacent segments of limited longitudinal data of a loadings with the common factor representing the intercept or

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 449

µI µS
F1 σIS F2 COH 1
σ2I σ2S
0
1 1 1 0.2
0.4

4 5 6 7 8 9

ε1 ε2 ε3
µI σIS µS
F1 F2
σ2 I σ2S
0.2
1 1 1 0.4 COH 2
0.6

4 5 6 7 8 9

ε2 ε3 ε4
µI σIS µS
F1 F2 COH 3
σ2I σ2S
0.4
1 1 1 0.6
0.8

4 5 6 7 8 9

ε3 ε4 ε5

COH 4 µI σIS µS
F1 F2
σ2 I σ2S
0.6
1 1 1 0.8
1.0

4 5 6 7 8 9

ε4 ε5 ε6

Figure 1. Representation of the cohort-sequential linear growth model for change in aggressive behavior from ages 4 to 9 years. Cohort 1
represents the 4-year-old cohort with data for ages 4, 5, and 6 years across years 1–3 respectively. Cohort 2 represents the 5-year-old cohort
with data for ages 5, 6, and 7 years. Cohort 3 represents the 6-year-old cohort with data for ages 6, 7, and 8 years. Cohort 4 represents the
7-year-old cohort with data for ages 7, 8, and 9 years. F1 represents Factor 1, the intercept, and F2 represents Factor 2, the slope. The mean
of the intercept, µI, and the slope, µS, and variances of the intercept, σ2I, and slope, σ2S, as well as the covariance between the intercept and
the slope, σIS, are estimated. A linear growth trajectory is specified by fixing the factor loadings on the slope from 0 to 1 across ages 4–9 years.

initial status, constrained to 1.0. The second factor (F2), increase less/more rapidly than people who start at a high level
represents the slope, i.e. the rate of change (increase or of aggressive behavior. The factor loadings on the slope factor
decrease) in aggressive problem behavior over the period of carry the information of the underlying time metric. The
study (i.e., from Time 1 to Time 3). The slope factor has a interpretation of the factors is determined by the fixed values
mean, σS, and variance, σ2S, across the whole sample. The of the factor loadings. Meredith and Tisak (1990) have shown
mean and the variance can also be estimated from the repeated that at least two slope factor loadings must be fixed to two
measures. The variances of the slope and the intercept indicate different values to identify the model. Because in this study the
how much individuals in the group vary. A large value means time intervals are equal, a trajectory corresponding to a linear
that the initial level, or rate of change, varies widely from change (as in Figure 1) can be evaluated by fitting a model that
person to person. Small variances mean that the group is specified the slope factor loadings to be 0, 0.2, 0.4, 0.6, 0.8
homogeneous, starting at much the same level and changing and 1 for ages 4, 5, 6, 7, 8, and 9, respectively. Fixing the first
at much the same rate. The double-headed curved arrow and last estimate to 0 and 1, respectively, and freely estimat-
between the two factors indicates that both latent factors are ing the other four parameters makes it possible to model an
allowed to covary (estimated as σSI ). This happens if people unspecified trajectory where the shape of the trajectory is
who start at a lower level of aggressive behavior decrease/ determined by the data (i.e., an unspecified two-factor model).

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
450 PRINZIE ET AL. / A COHORT-SEQUENTIAL MULTIVARIATE LATENT GROWTH CURVE ANALYSIS

Within this approach, the freely estimated parameters reflect convergent evidence of model fit. The chi-square difference
the developmental function with maximal fit to the data (Rao, test was used to assess statistical significance of incremental fit
1958). In the unspecified two-factor model, the “slope” factor in nested model comparisons. Multiple imputation with EM
is more accurately a “slope/shape” factor in which information (Schafer, 1997) is a new feature implemented in LISREL for
about the slope (i.e., the general linear trend) and the shape of imputing incomplete data that are missing at random under
the growth curve are confounded. The difference in fit between the assumption of an underlying multivariate normal
the unspecified two-factor model and the linear trajectory distribution.
model provides an index of the incremental gain in model fit
due to the inclusion of information about the shape (i.e., curvi-
linearity). The chi-square difference test performed in the
nested model comparison is used to determine the statistical
Results
significance of incremental fit due to the inclusion of curvi-
linearity in model specification.
Descriptive statistics
For all latent growth models reported in this study, the Means and standard deviations for the cohort-sequential
LISREL 8.54 structural equation modeling software (Jöreskog samples are presented in Table 1. Because the aggressive
& Sörbom, 2003) was used to fit the growth model to the syndrome has more items, it has a wider range of possible
observed mean vector and covariance structure. In addition to scores (0–40) and a higher mean than the delinquent syndrome
the chi-square, three model fit indices were used to assess (range = 0–26). It can be seen that for all informants mean
model fit. They were Jöreskog and Sörbom’s (1989) goodness- levels of aggressive and delinquent problem behavior showed
of-fit index (GFI), Bentler’s (1990) comparative fit index a decreasing trajectory over time and that boys scored higher
(CFI), and Steiger’s (1990) root mean square of error of than girls. Because these means are computed from observed
approximation (RMSEA). For values of GFI and CFI, the scores, measurement error may obscure the nature of the true
convention of above .90 was adopted here as an indication of growth trajectory. Latent growth analysis models the true
good fit. Browne and Cudeck (1993) indicated that an growth trajectory by separating observed scores into a
RMSEA-value of about .05 or less suggests a close fit of the component describing true growth and a component repre-
model in relation to the degrees of freedom. Values up to .08 senting the stochastic effect of the measurement error.
represent reasonable errors of approximation in the popu- Additionally, Table 2 shows that pairwise correlations of
lation. MacCallum, Browne, and Sugawara (1996) have measures of aggressive problem behavior were higher than the
elaborated on these cut points and noted that RSMEA values correlations for delinquent problem behavior. Pairwise corre-
ranging from .08 to .10 indicate mediocre fit. All three model lations of measures of aggressive behavior were also high to
fit indices were used in evaluating each model to provide moderate within waves of measurements (mean rMF ranging

Table 1
Descriptive statistics for childhood aggressive and delinquent behavior problems according to mothers (N = 669) and fathers (N = 632)

Boys Girls

T1 T2 T3 T1 T2 T3

mean SD mean SD mean SD mean SD mean SD mean SD

Mother ratings:
Aggressive behavior (Coh 1) 9.05 6.48 9.40 6.64 8.48 6.04 7.24 5.36 6.22 4.85 6.20 4.54
Aggressive behavior (Coh 2) 10.12 6.70 7.92 5.79 7.47 5.67 5.93 4.42 5.45 4.97 5.20 4.49
Aggressive behavior (Coh 3) 7.90 5.98 7.06 5.30 7.20 5.96 5.84 5.49 5.42 5.26 4.78 5.05
Aggressive behavior (Coh 4) 7.82 6.20 7.04 5.98 6.76 6.15 7.02 5.57 4.99 4.49 5.38 5.48
Delinquent behavior (Coh 1) 1.52 1.62 1.61 1.59 1.46 1.60 1.18 1.27 1.16 1.13 1.08 1.52
Delinquent behavior (Coh 2) 1.74 1.90 1.22 1.25 1.27 1.44 0.88 1.20 0.87 1.06 0.76 1.02
Delinquent behavior (Coh 3) 1.41 1.59 1.19 1.38 1.25 1.59 1.08 1.23 0.96 1.29 0.76 1.32
Delinquent behavior (Coh 4) 1.50 1.81 1.43 1.79 1.38 1.90 1.26 1.50 0.92 1.22 0.94 1.62

Father ratings:
Aggressive behavior (Coh 1) 9.21 6.10 8.69 6.15 7.29 5.45 6.73 4.71 5.81 4.34 5.49 4.15
Aggressive behavior (Coh 2) 8.50 6.16 7.56 6.12 6.87 5.78 6.32 5.05 5.12 4.33 4.81 4.39
Aggressive behavior (Coh 3) 6.40 5.91 6.17 5.43 5.95 5.73 5.91 5.48 4.99 4.38 4.39 4.62
Aggressive behavior (Coh 4) 6.42 5.05 6.21 5.25 5.81 5.51 6.11 4.93 4.43 3.67 4.51 3.86
Delinquent behavior (Coh 1) 1.50 1.47 1.21 1.20 1.25 1.33 1.26 1.51 1.01 1.00 0.99 1.03
Delinquent behavior (Coh 2) 1.17 1.45 1.08 1.59 1.00 1.28 0.75 1.05 0.78 0.93 0.81 1.08
Delinquent behavior (Coh 3) 1.21 1.73 1.03 1.30 1.03 1.32 1.09 1.29 0.73 1.14 0.69 1.27
Delinquent behavior (Coh 4) 1.29 1.48 1.05 1.45 1.35 1.48 1.13 1.31 0.68 0.90 0.77 0.96

Note. Mother ratings: Coh 1 Boys n = 82, Girls n = 79; Coh 2 Boys n = 89, Girls n = 83; Coh 3 Boys n = 81, Girls n = 85; Coh 3 Boys n = 82,
Girls n = 88.
Father ratings: Coh 1 Boys n = 80, Girls n = 74; Coh 2 Boys n = 84, Girls n = 73; Coh 3 Boys n = 81, Girls n = 85; Coh 3 Boys n = 77, Girls
n = 82.

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 451

Table 2
Intercorrelations between childhood aggressive and delinquent behavior problems and harsh discipline according to mother (N = 669) and
father ratings (N = 632)a

Mothers 1 2 3 4 5 6 7 8
Coh 1
1. Agg 1 . – .78*** .66*** .58*** .47*** .37** .06 .15
2. Agg 2 .75*** . – .72*** .44*** .58*** .51*** .05 .18
3. Agg 3 .65*** .79*** . – .45*** .53*** .62*** –.11 .18
4. Del 1 .71*** .59*** .55*** . – .51*** .46*** –.12 –.02
5. Del 2 .59*** .77*** .73*** .67*** . – .63*** –.14 .14
6. Del 3 .54*** .70*** .77*** .64*** .79*** . – –.07 .13
7. Coercion .25† .16 .11 .01 .06 –.02 .– .11
8. Overreactivity .21 .37** .38** .24* .26* .33* .16 . –
Coh 2
1. Agg 1 . – .75*** .69*** .40** .40** .41** .12 .34*
2. Agg 2 .75*** . – .76*** .33* .52*** .45*** .09 .33*
3. Agg 3 .72*** .79*** . – .17 .45*** .59*** .20 .42***
4. Del 1 .73*** .50*** .51*** . – .40** .34* .01 –.01
5. Del 2 .51*** .65*** .49*** .55*** . – .49*** .15 .02
6. Del 3 .54*** .61*** .68*** .57*** .64*** . – .26† .32*
7. Coercion .32* .31* .31* .27† .14 .07 .– .13
8. Overreactivity .38** .39** .43*** .29* .31* .35** .37** . –
Coh 3
1. Agg 1 . – .87*** .80*** .65*** .66*** .59*** .30* .44***
2. Agg 2 .68*** . – .82*** .58*** .69*** .56*** .27† .46***
3. Agg 3 .65*** .75*** . – .46*** .59*** .65*** .31* .52***
4. Del 1 .61*** .39*** .50*** . – .68*** .61*** .08 .21
5. Del 2 .42*** .57*** .44*** .43*** . – .71*** .15 .30*
6. Del 3 .50*** .57*** .74*** .56*** .59*** . – .15 .23†
7. Coercion .29* .21 .22 .15 –.12 .12 .– .19
8. Overreactivity .29* .32* .41** .18 .15 .39** .09 . –
Coh 4
1. Agg 1 . – .79*** .78*** .64*** .61*** .54*** .15 .33*
2. Agg 2 .79*** . – .80*** .61*** .66*** .61*** .20 .21†
3. Agg 3 .77*** .84*** . – .50*** .59*** .77*** .11 .28*
4. Del 1 .75*** .70*** .64*** . – .57*** .55*** .05 .29*
5. Del 2 .72*** .75*** .68*** .64*** . – .73*** –.01 .11
6. Del 3 .63*** .70*** .84*** .64*** .73*** . – –.06 .09
7. Coercion .27† .27† .13 .15 .22 .18 .– .19
8. Overreactivity .33* .26† .32* .26† .21 .24† .13 . –

Fathers 1 2 3 4 5 6 7 8
Coh 1
1. Agg 1 . – .65*** .55*** .54*** .38** .18 .01 .20
2. Agg 2 .71*** . – .58*** .41** .61*** .25† .12 .29†
3. Agg 3 .80*** .73*** . – .40** .40** .48*** –.10 .42**
4. Del 1 .56*** .51*** .60*** . – .37* .44** .09 .25†
5. Del 2 .44*** .62*** .58*** .41** . – .37* .03 .24†
6. Del 3 .49*** .63*** .69*** .49*** .62*** – –.04 .10
7. Coercion .12 .09 .02 –.08 .00 –.09 .– –.17
8. Overreactivity .33* .35* .35* .31* .33* .31* .17 . –
Coh 2
1. Agg 1 . – .78*** .67*** .71*** .44*** .19 .13 .17
2. Agg 2 .68*** . – .75*** .48*** .51*** .23† .28† .31*
3. Agg 3 .62*** .78*** . – .39** .39*** .49*** .29† .36*
4. Del 1 .72*** .56*** .49*** . – .38** .21 .14 .07
5. Del 2 .55*** .79*** .51*** .57*** . – .35** .15 .16
6. Del 3 .50*** .59*** .71*** .61*** .57*** – .26† .15
7. Coercion .20 .22† .26† .23† .25† .30* .– .20
8. Overreactivity .17 .17 .32* .25† .09 .40** .30* . –

Continued

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
452 PRINZIE ET AL. / A COHORT-SEQUENTIAL MULTIVARIATE LATENT GROWTH CURVE ANALYSIS

Table 2
Continued

Fathers 1 2 3 4 5 6 7 8

Coh 3
1. Agg 1 . – .70*** .73*** .56*** .40** .38** .26† .10
2. Agg 2 .69*** . – .78*** .49*** .64*** .54*** .35* .17
3. Agg 3 .71*** .85*** . – .47*** .51*** .55*** .15 .28†
4. Del 1 .81*** .51*** .50*** . – .42** .42** .00 .10
5. Del 2 .50*** .66*** .63*** .41** . – .47*** .37** .02
6. Del 3 .43*** .56*** .67*** .40** .68*** . – .22† .21
7. Coercion .13 .01 .07 .18 .04 .04 .– –.04
8. Overreactivity .33* .32* .35* .28† .25† .30* .19 . –
Coh 4
1. Agg 1 . – .72*** .69*** .68*** .46*** .52*** –.01 .44***
2. Agg 2 .78*** . – .80*** .54*** .55*** .49*** .04 .38**
3. Agg 3 .71*** .81*** . – .53*** .50*** .53*** –.02 .41**
4. Del 1 .58*** .57*** .46*** . – .62*** .58*** –.03 .34*
5. Del 2 .43*** .63*** .53*** .60*** . – .47*** .08 .25†
6. Del 3 .55*** .70*** .78*** .64*** .61*** . – .12 .42***
7. Coercion .26† .19 .23† .31* .36* .27† .– .05
8. Overreactivity .22† .32* .46*** .27† .22 .39** .21 . –

a Correlationsfor boys are below the diagonal; correlations for girls are above.
Note. †p < .05; *p < .01; **p < .001; ***p < .0001. Agg = aggressive problem behavior, Del = delinquent problem behavior.

from .71 for Cohort 3 and Cohort 4 to .72 for Cohort 1 and behavior. In a third step, a multigroup analysis was performed
Cohort 2) and highly correlated across waves (mother mean r to examine a possible gender effect (Model 3.2). The associ-
ranging from .74 for Cohort 1 and Cohort 2 to .80 for Cohort ations of harsh discipline with the intercept and slope of
4; father mean r ranging from .70 for Cohort 2 to .76 for aggressive and delinquent behavior were explored in Model 4.
Cohort 4). Pairwise correlations of measures of delinquent Finally, we investigated whether the final model for the mother
behavior were moderate within waves of measurements (mean reports could be replicated in the father reports (Model 5).
rMF ranging from .49 for Cohort 1 to .56 for Cohort 4) and
highly correlated across waves (mother mean r from .52 for Univariate latent growth analysis of aggressive problem behavior.
Cohort 2 to .65 for Cohort 1; father mean r ranging from .46 As shown in Table 3, the no-growth model (Model 1.1)
for Cohort 1 to .60 for Cohort 4). Moderate positive correla- provided a poor fit to the aggressive problem behavior
tions were found between harsh discipline and aggressive and problems rated by the mothers, χ2(28, N = 669) = 130.39, p <
delinquent problem behavior. .001, GFI = .975, CFI = .913, RMSEA = .148 indicating that
some form of growth was present over the three time points.
The linear trajectory model (Model 1.2), provided a better fit,
Univariate latent growth models χ2(25, N = 669) = 43.97, p = .01, GFI = .983, CFI = .984,
From the hypotheses, developmental processes can be RMSEA = .068. The nested model comparison using the chi-
deduced for aggressive and delinquent problem behavior. square difference test showed that the linear trajectory model
However, specific hypotheses about the functional form, or provided significant incremental fit over the no-growth model,
shape, of the developmental trajectories have not been formu- ∆χ2 = 86.42, ∆df = 3, p < .001, indicating that a linear growth
lated. Therefore, in a first step, univariate latent growth models model represented the nature of intraindividual change over
for children’s aggressive and delinquent problem behavior were the three time points better than a no-growth model. To para-
tested separately for mother ratings. Table 3 contains the fit meterize an unspecified model, the first and the last slope
indices for the different models. The first model (Model 1.1 factor loadings were fixed to 0 and 1 to identify the model. This
and 2.1) fitted to the data was a “no-growth” or strict-stability model (Model 1.3) provided an excellent fit to the data, χ2(21,
model (Stoolmiller, 1994), which specified that no growth N = 669) = 29.53, p = .10, GFI = .988, CFI = .993, RMSEA
occurred at all over the three time points (i.e., a model with = .049. The nested model comparison showed that this
only an intercept factor). This model is the most restricted unspecified two-factor model provided significant incremental
model and is nested under the other models in the table. The fit over the linear trajectory model, ∆χ2 = 14.44, ∆df = 4, p <
second model (Model 1.2 and 2.2) fitted to the data was a .01, indicating that a curvilinear growth trajectory represented
linear trajectory model, which specified a straight-line growth the nature of intraindividual change over the three time points
over the three time points. The third model (Model 1.3 and better than a linear growth trajectory. Therefore, this unspeci-
2.3) fitted to the data was the unspecified two-factor model fied two-factor model was selected as the final growth model
discussed above. The second model is nested under the third. for children’s aggressive behavioral problems rated by the
In a second step, a multivariate latent growth model (Model mothers. To test whether the growth trajectories for the four
3.1) was specified to investigate the relationship between the cohorts converged, we tested several models in which one cross
intercept and growth of aggressive and delinquent problem cohort constraint at a time was released. The results of these

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 453

Table 3
Model fit indices and nested model comparisons of children’s aggressive and delinquent problem behavior in mother and father ratings

Model χ2 df p Model comparison ∆χ2 ∆df GFI CFI RMSEA RMSEA90

Mother ratings:
1. Aggresssion (univariate latent growth)
Model 1.1 (no growth) 130.39 28 .000 .975 .913 .148 .123–.174
Model 1.2 (linear trajectory) 43.97 25 .010 M1.1 vs M1.2 86.42 3 .983 .984 .068 .032–.099
Model 1.3 (unspecified trajectory) 29.53 21 .100 M1.2 vs M1.3 14.44 4 .988 .993 .049 .000–.088
2. Delinquency (univariate latent growth)
Model 2.1 (no growth) 51.94 22 .000 .975 .956 .090 .059–.123
Model 2.2 (linear trajectory) 26.73 19 .110 M2.1 vs M2.2 25.21 3 .996 .987 .049 .000–.090
Model 2.3 (unspecified trajectory 24.97 15 .050 M2.2 vs M2.3 1.76 4 .997 .984 .063 .000–.106
3. Multivariate latent growth
Model 3.1 86.73 60 .010 .977 .992 .052 .024–.075
Model 3.2 (boys-girls) 165.63 124 .010 .953 .986 .064 .034–.088
Model 3.3 (EQUAL) 220.08 138 .000 M3.3 vs M3.2 64.45 14 .956 .974 .085 .063–.105
Model 3.4 179.35 134 .010 M3.4 vs M3.2 13.72 10 .951 .985 .064 .036–.087
4. Multivariate latent growth + Harsh parenting
Model 4.1 270.39 215 .010 .933 .981 .056 .031–.076
Model 4.2 289.90 223 .000 M4.1 vs M4.2 19.51 8 .930 .979 .060 .038–.079
Model 4.3 278.50 222 .010 M4.1 vs M4.3 8.11 7 .933 .981 .056 .032–.075
Model 4.4 282.55 226 .010 M4.3 vs M4.4 4.05 4 .937 .981 .055 .031–.075
5. Mother + Father ratings
Model 5.1 596.12 451 .000 .949 .971 .063 .049–.077
Model 5.2 648.80 478 .000 M5.1 vs M5.2 52.68 27 .935 .965 .067 .053–.080
Model 5.3 628.23 474 .000 M5.1 vs M5.3 32.11 23 .942 .968 .064 .050–.077

model comparisons, in addition with the good fit indices for obtained in a multivariate model is strongly dependent on the
the fully constrained model, indicated that there were no adequate specification of growth in the associate univariate
cohort differences. models (Stoolmiller, 1994). In this multivariate model, the
parameters of interest are the between variables growth factor
covariances. Under the assumption that mothers’ measure-
Univariate latent growth analysis of delinquent problem behavior.
ment errors for aggressive problem behaviors were related to
Analysis of the univariate latent growth curves for delinquent
those for delinquent behavior problems, we allowed the error
behavior problems rated by the mothers indicated that the no
variances to covary within time points. This multivariate model
growth curve (Model 2.1) provided a poor fit to the data,
(Model 3) provides an adequate fit to the data, χ2(60, N = 669)
χ2(22, N = 669) = 51.94, p < .001, GFI = .975, CFI = .956,
= 86.73, p = .01, GFI = .977, CFI = .992, RMSEA = .052.
RMSEA = .090. The linear trajectory model (Model 2.2)
The parameter estimates of the multivariate model are
provided a better fit, χ2(19, N = 669) = 26.73, p = .11, GFI
presented in Table 4. The significant mean estimate for the
= .996, CFI = .987, RMSEA = .049. The nested model
intercept (µI(Agg) = 8.25, t = 24.06, p < .05) indicates that on
comparison using the chi-square difference test showed that
average, children had a significant initial score on aggressive
the linear trajectory model provided significant incremental fit
behavior problems of 8.25. Also the slope factor means depart
over the no-growth model, ∆χ2 = 25.21, ∆df = 3, p < .001. The
significantly from 0, µS(Agg) = –2.54, t = –6.32, p < .05, indi-
fit indices for the unspecified model (Model 2.3) were χ2(15,
cating that in the mother ratings there was a decline in aggres-
N = 669) = 24.97, p = .05, GFI = .997, CFI = .984, RMSEA
sive problem behavior over time. The slope factor loadings
= .063. The nested model comparison showed that this
converged at [0, .07, .45, .62, .93, 1.00]. Table 4 also shows
unspecified two-factor model provided no significant incre-
that the variance for the intercept factor departs significantly
mental fit over the linear trajectory model, ∆χ2 = 1.76, ∆df =
from zero, σ2I(Agg) = 19.49, t = 3.65, p < .05, indicating that
4, p = .78, indicating that for delinquent behavior problems a
there were systematic individual differences in aggressive
linear growth trajectory represented the nature of intraindivid-
behavior problems in the mother ratings at initial status (T1).
ual change over the three time points better than a nonlinear
The slope factor variance was also significantly different from
growth trajectory. Additional model comparisons, in which
zero, σ2S(Agg) = 19.49, t = 3.65, p < .05, indicating systematic
cohort constraints were released, revealed that the convergence
individual differences in rate of change. The factor covariance
assumption was tenable.
between intercept and slope was also significant, σS(Agg)I(Agg) =
–13.32, t = –3.58, p < .05, indicating that mothers of children
Multivariate latent growth curves. In the second step, the focal with higher initial levels of aggressive behavior problems
variables of the univariate models were entered in one multi- tended to report lower rates of change in behavior problems
variate model to investigate the relationships between growth over time. The obtained R2 ranged from .74 to .83 indicating
processes in aggressive and delinquent behavior. Step 1 was that most of the variance in the observed measures was
prerequisite to step 2 because the validity of the parameters accounted for by growth factors.

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
454 PRINZIE ET AL. / A COHORT-SEQUENTIAL MULTIVARIATE LATENT GROWTH CURVE ANALYSIS

Table 4 positive correlation between the intercept (r = .77) of aggres-

Parameter estimates of the multivariate latent growth models for sive and delinquent problem behavior. A high start on aggres-
aggressive and delinquent problem behavior (t-values between sive behavior corresponded to a high start on delinquent
parentheses) behavior. The slopes of aggressive and delinquent behavior
were not significantly correlated.
Multivariate
Gender differences. To investigate a possible gender effect, we
Agg Del
performed a multivariate multigroup analysis on the boys and
Mean initial level 8.25 (24.06) 1.38 (19.97) girls data. In Model 3.2, model specifications describing the
Variance of initial level 32.84 (10.01) 1.32 (7.30) multivariate model (Model 3.1) were similarly specified for
Mean rate of growth –2.54 (–6.32) –0.39 (–3.72) boys and girls. As presented in Table 3 the goodness-of-fit
Variance of growth rate 19.49 (3.65) 1.25 (2.27) statistics indicate an adequate fit to the data representing both
Covariance between level and –13.32 (–3.58) –0.34 (–1.17) boys and girls (GFI = .953, CFI = .986, RMSEA = .064). In
growth Model 3.3, equality constraints were placed on the parameters
Covariance between level Agg 5.09 (8.39) across the two groups. Results from the estimation of this
and level Del highly restrictive multigroup model yielded a χ2 value of
Covariance between growth Agg 2.56 (1.94)
220.08 with 138 degrees of freedom. To assess the tenability
and growth Del
of these equality constraints, this model was compared with
Covariance between growth Agg –1.61 (–2.10)
and level Del Model 3.2 in which no constraints were imposed. Accordingly,
Covariance between level Agg –0.68 (–0.75) this comparison yielded a ∆χ2(14) = 64.45, which is statisti-
and growth Del cally significant (p < .001). In a next step, to pinpoint the
non-invariant parameters, we inspected modification indices of
the parameters for which equality constraints were imposed.
The equality constraint for the mean of the aggression and
For delinquent behavior, the significant mean estimate for delinquent intercept factor, the variance of the latent delin-
the intercept (µI(Del) = 1.38, t = 19.97, p < .05) indicates that quency intercept and the covariance between the aggression
on average, children had a significant initial score on delin- and delinquency intercept were released. The goodness-of-fit
quent behavior problems of 1.38. Also the slope factor means statistics of this model (Model 3.4) indicated an acceptable fit,
depart significantly from 0, µS(Del) = –0.39, t = –3.72, p < .05, χ2(134, N = 669) = 179.35, p = .01, GFI = .951, CFI = .985,
indicating that in the mother ratings there was a decline in RMSEA = .064. A comparison with Model 3.2 yielded a
delinquent problem behavior over time. The estimated ∆χ2(10) = 13.72, which is not statistically significant (p = .19).
variance for the intercept was 1.32 (t = 7.30) indicating that The mean of the intercept for aggressive (µI(Gagg) = 7.32, t =
there were systematic individual differences in the mother 18.81) and delinquent behavior (µI(Gdel) = 1.16, t = 15.30) was
ratings at initial status. The estimated variance of the slope was lower for girls compared to boys (µI(Bagg) = 9.38, t = 23.91;
significantly different from 0 (1.25, t = 2.27). The factor µI(Bdel) = 1.56, t = 17.82). The variance of the intercept for
covariance between intercept and slope was not significant, delinquency (σ2I(Bdel) = 1.46, t = 7.57) was higher for boys
σS(Del)I(Del) = –0.34, t = –1.17, indicating that there was no compared to girls (σ2I(Gdel) = 0.95, t = 5.62) indicating that
relation between the initial level of delinquent behavior and the there were for boys more systematic individual differences in
rated changes in delinquent behavior over time. The obtained the initial level of delinquent problem behavior. Finally, the
R2 ranged from .50 to .73. relation between the initial levels of aggressive and delinquent
The focus of this analysis is on the interrelationships behavior was stronger for boys (σI(Agg)I(Del) = 5.16, t = 8.26)
between the level and growth of aggressive and delinquent compared to girls (σI(Agg)I(Del) = 4.04, t = 6.92). The repro-
problem behavior. The significant positive intercept–intercept duced means for the model testing a common growth trajec-
covariance (σI(Agg)I(Del) = 5.09, t = 8.39) indicated a positive tory for aggressive and delinquent behavior across the four
association between initial scores on aggressive and delinquent cohorts from age 4 to age 9 are represented in Figure 2.
behavior. The slope of aggressive behavior was negatively
correlated with the initial level of delinquent behavior Harsh discipline as correlate. Because significant individual
(σS(Agg)I(Del) = –1.61, t = –2.10) indicating that slower variability was found in the intercept and slope factors, it is
decreases in aggressive behavior were related to higher initial appropriate to try to explain this variability by explanatory
levels of delinquent behavior. Table 5 represents the concur- variables. In Model 4.1, the effect of harsh discipline was
rent correlations between the intercept and the growth rates added to Model 3.4. This model (Model 4.1) showed an
for aggressive and delinquent behavior. We found a significant adequate fit to the data (χ2(215, N = 669) = 270.39, p = .01,

Table 5
Correlations between growth parameters in the multivariate latent growth curve model

Mean initial level Agg Mean slope Agg Mean initial level Del Mean slope Del

Mean initial level Agg 1.00

Mean slope Agg –.53 1.00
Mean initial level Del .77 –.32 1.00
Mean slope Del –.11 .52 –.26 1.00

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 455

10
9
8

MEAN CBCL SCORE

7
6
5
4
3
2
1
0
4 5 6 7 8 9
AGE

AGG BOYS AGG GIRLS DEL BOYS DEL GIRLS

Figure 2. Reproduced means for aggressive and delinquent problem behavior in boys and girls.

GFI = .933, CFI = .981, RMSEA = .056). In a next step, the delinquent behaviors. In the data for girls, these variables
effects were set equal for boys and girls. The fit of this model accounted for 16% of the variance in the initial level of
(Model 4.2) was χ2(223, N = 669) = 289.90, p = .001, GFI = aggression, for 2% of the variance in the growth of aggression,
.930, CFI = .979, RMSEA = .060. A comparison with Model and for 6% of the variance in the initial level of delinquent
4.1 yielded a ∆χ2(8) = 19.51, which is statistically significant (p behaviors.
< .05). Based on the modification indices, the equality
constraint for the path from overreactivity to the initial level of Father ratings. To investigate the stability of this final model,
delinquency was released. Fit indices of this model (Model we tested whether the final model could replicate across the
4.3) were χ2(222, N = 669) = 278.50, p = .01, GFI = .933, father ratings. First, a multigroup baseline model was estab-
CFI = .981, RMSEA = .056. A comparison with Model 4.1 lished against which subsequent models that include equality
yielded a ∆χ2(7) = 8.11, which is not statistically significant (p constraints were compared. In Model 5.1, model specifications
= .32). Finally, in a trimming process (Model 4.4), nonsignif- describing the final model for the mother ratings (Model 4.4)
icant paths were removed from the model, one at a time, begin- were similarly specified for the father ratings. The goodness-
ning with the path with the smallest t value (Kline, 1998). of-fit statistics reflect the simultaneous estimation of the final
Model 4.4 provided a good fit to the data, χ2(226, N = 669) model for both the mother and the father sample and are
= 282.55, p = .01, GFI = .937, CFI = .981, RMSEA = .055. presented in Table 3. The GFI value of .949 and the CFI values
The comparison with model 4.3 yielded a ∆χ2(4) = 4.05, which of .971 indicate an adequate fit to the data representing both
is not statistically significant (p = .40). Table 6 presents the mothers and fathers. This model was used as the yardstick
estimated coefficients of the regression of the growth against which to determine the tenability of the imposed
parameters on overreactivity and coercion. Higher scores on equality constraints. In Model 5.2, equality constraints were
coercion were related to higher scores on the initial level of placed on the parameters across the mother and the father
aggressive behavior (β = 0.59, t = 4.75, p < .01) and to slower sample. Results from the estimation of this highly restrictive
decreases (β = –0.44, t = –2.55, p < .01) of aggressive behavior. multigroup model yielded a χ2 value of 648.80 with 478
The significantly positive effect of overreactivity on the degrees of freedom. To assess the tenability of these equality
aggression intercept indicated that high scores on overreactiv- constraints, this model was compared with Model 5.1 in which
ity were related to higher initial levels of aggressive (β = 2.18, no constraints were imposed. Accordingly, this comparison
t = 9.78, p < .01) behavior problems. The significantly positive yielded a ∆χ2(27) = 52.68, which is statistically significant (p <
effect of overreactivity on the delinquency intercept indicated .01). Based on the modification indices, the equality constraint
that high scores on overreactivity were related to higher initial for the mean of the aggression and delinquent intercept factor
levels of delinquent behavior; however, this effect was higher for boys was released. The goodness-of-fit statistics of this
for boys (β = 0.50, t = 6.84, p < .01) compared to girls (β = model (Model 5.3) indicated an acceptable fit, χ2(474, N =
0.27, t = 4.24, p < .01). In the data for boys, the two explana- 669) = 628.23, p < .001, GFI = .942, CFI = .968, RMSEA =
tory variables accounted for 14% of the variance in the initial .064. A comparison with Model 5.1 yielded a ∆χ2(23) = 32.11
level of aggression, for 2% of the variance in the growth of which is not statistically significant (p = .10). These results
aggression and for 12% of the variance in the initial level of indicated that the mean initial score on aggressive and

Table 6
Regression coefficients of the growth parameters on harsh parenting (t-values between parentheses)

Mean initial level Agg Mean slope Agg Mean initial level Del Mean slope Del

Growth parameters on coercion 0.59 (4.75) –0.44 (–2.55) – –

Growth parameters on overreactivity 2.18 (9.78) – 0.50 (6.84)/0.27 (4.24)* –

Note. * Boys/Girls.

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
456 PRINZIE ET AL. / A COHORT-SEQUENTIAL MULTIVARIATE LATENT GROWTH CURVE ANALYSIS

delinquent behavior problems for boys was significantly lower Coie and Dodge (1998), infancy and toddlerhood can be
in the father data (µIAF = 0.93, t = 1.51; µIDF = –0.03, t = considered as the period of highest frequency in aggression.
–0.17) than in the mother data (µIAM = 3.34, t = 4.24; µIDM Loeber et al. (1991) found that temper tantrums, irritable and
= –0.26, t = –1.49). This means that the effects of the explana- annoying behaviors are common among preschool children but
tory variables could be replicated in the father ratings. usually have largely disappeared by 8 or 9 years. Several studies
have shown that parental socialization attempts are associated
with a decline in child externalizing problems (e.g., Denham
Discussion et al., 2000). From a social interactional perspective, the
decline may reflect age-related changes in reactions from
The analyses in this study suggest that significant develop- adults and peers to aggressive behaviors. Snyder et al. (2003)
mental decreases existed in the repeated measures for aggres- hypothesize that the efficient use of discipline strategies “drive
sive and delinquent behavior problems. The multivariate latent overt forms of antisocial behavior underground”. As the child
growth analysis indicated that trajectories in aggressive and ages, aggressive behaviors are less reinforced by adults. In this
delinquent problem behavior were positively associated. way, the child learns to inhibit behaviors that are annoying and
Moreover, the association between the initial scores for aggres- irritating for others. Further, parental induction, warm respon-
sive and delinquent behaviors was stronger for boys than for siveness, and less frequent use of physical punishment
girls. Furthermore, boys were higher than girls in initial status. generally were linked to the development of self-regulation and
Conversely, the rate of change was not related to gender. The internalized standards of conduct (e.g., Hoffman, 1994),
most important finding was that parenting behaviors were which in turn are thought to underlie children’s capacity to
differentially related to children’s aggressive and delinquent refrain from aggressive behavior (e.g., Kuczynski &
problem behavior. The results of the mother reports were Hildebrandt, 1997).
replicated in the father reports. This suggests that fathers make The declining trend of externalizing problem behavior
significant contributions to children’s development in much seems to continue later on in the development. Loeber (1982)
the same way that mothers do and that paternal factors are pointed out that most longitudinal studies of children show a
substantially related to children’s adjustment. negative trend in ratings of aggressive behavior as they enter
adolescence. This evidence suggests that, although some new
forms of antisocial behavior (e.g., delinquent acts such as theft
Normative development of aggressive and delinquent and vandalism) are emerging at later ages, other forms are
behavior problems decreasing (e.g., interpersonal aggression) (Patterson, 1993).
A decreasing nonlinear growth trajectory most adequately
described the development of aggressive behaviors in young Parenting effects. An important finding of this study is that
children. The decline in delinquent problem behavior was best dysfunctional parenting behaviors were differentially related to
described by a linear trajectory. Caution is needed in the children’s aggressive and delinquent problem behavior.
interpretation of the results for delinquent behavior problems. Coercion was significantly related to aggressive behavior but
Due to the age range of the children included and the normal not to delinquent behavior. Higher scores on coercion were
population sample, several items had a very low prevalence. related to higher initial levels and a slower decrease of aggres-
However, the lower Pearson correlations indicating the lower sive behavior. High scores on overreactivity were associated
stability of delinquent behavior over time have also been with higher initial levels of aggressive and delinquent problem
reported in both clinical and nonclinical samples (e.g., behavior. The stronger association between harsh discipline
McConaughy, Stanger, & Achenbach, 1992, Stanger et al., and aggressive problem behavior is consistent with Patterson’s
1997). Possible explanations for the higher stability of aggres- coercion model (Patterson, 1982; Patterson et al., 1992).
sive behavior are the higher heritability estimates (Edelbrock, Overreactive and coercive parenting behavior might lead to
Rende, Plomin, & Thompson, 1995; Tuvblad, Eley, & inconsistent behavioral contingencies, a capricious and un-
Lichtenstein, 2005) and the link with biochemical variables for predictable environment, and a decreased parent’s sense of
both boys (Gabel, Stadler, Bjorn, Shindledecker, & Bowden, control. This in turn might increase the likelihood of external-
1993) and girls (Paikoff, Brooks-Gunn, & Warren, 1991). izing problem behaviors. As described in the coercion theory
The results of the multivariate analyses illuminate the strong of Patterson (Patterson et al., 1992), the negative reinforce-
linkages between individual-level aggressive and delinquent ment of externalizing behavior may increase the frequency, the
behavior trajectories in young children. The positive associ- intensity and the duration of this problem behavior.
ation between the initial scores on aggressive and delinquent Dysfunctional parenting practices were related to the initial
behavior indicated the comorbidity between the aggressive and level of delinquent behaviors but not to changes in delinquent
delinquent syndromes. This association corroborates and behaviors. This suggests that harsh discipline is partly respon-
extends previous cross-sectional findings on the comorbidity of sible for the onset of delinquent problem behaviors, but not
aggressive and delinquent behaviors (Barnow et al., 2005; their persistence. Previous studies reported that delinquent
Snyder, Reid, & Patterson, 2003). The negative association problem behaviors are more environmentally influenced. They
between the initial level of aggressive and delinquent behavior may reflect circumstances, such as a context of delinquent
on the one hand and the decline in aggressive and delinquent peer affiliation, and the use of social mimicry or attempts to
behavior on the other hand pointed to the persistence of this gain peer-group respect (Eley, Lichtenstein, & Moffitt, 2003).
problem behavior. This persistence is in accordance with Richman, Stevenson and Graham (1982) observed that
findings for the developmental trajectories of other forms of family factors, measured when children were three years old
externalizing behavior (Tremblay, 2000). had an influence on the development of problem behaviors
The decreasing trend of these problem behaviors is in five years later, whereas they had no effect on the outcome of
accordance with previous studies. According to the review of problem behavior once established. The time-limited effect of

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 457

parenting indicates that changes in growth rates are sensitive No gender differences were found in the relation between
to ongoing changes in the family environment or to the intro- coercive parenting and the developmental trajectory of aggres-
duction to and changes in other environments (e.g., school sive behavior. Eddy, Leve and Fagot (2001) reported that
settings). Also in the Dunedin Study, other variables than similar coercion processes apply to both boys and girls.
parenting practices (such as neurocognitive deficits, under- Recently, Prinzie et al. (2003) found that significant effects of
controlled temperament, and hyperactivity) were found to have children’s personality characteristics and negative parenting
a stronger predictive effect (Moffitt & Caspi, 2001). practices on children’s externalizing behavior were not differ-
entially patterned for boys and girls.
Gender differences. Significant gender differences were found Overreactivity on the other hand was stronger related to
with respect to the mean aggressive and delinquent scores. initial levels of delinquent behavior in boys compared to girls.
Boys had higher initial levels than girls. In addition, the In a meta-analysis Rothbaum and Weisz (1994) found that in
positive relation between the initial level of aggressive and studies of preadolescents, associations between parenting and
delinquent problem behavior was stronger for boys compared externalizing behavior were stronger for boys than for girls. The
to girls. A strong association between childhood problem differentiation of externalizing behavior in aggressive and
behavior and later delinquency for boys is also reported in delinquent behavior problems revealed that this is the case for
other studies (see e.g., Loeber, Farrington, Stouthamer- delinquent behavior but not for aggressive behavior. Lytton
Loeber, Moffitt, & Caspi, 1998). From a longitudinal and Romney’s (1991) meta-analysis on differential parental
perspective, Broidy et al. (2003) found that the linkage socialization revealed that physical punishment was more often
between childhood physical aggression and later delinquent used for boys than for girls in Western countries. Physical
behavior was more consistent among boys than among girls. punishment, in turn, was found to be associated with higher
That the association between aggressive and delinquent levels of delinquent behavior (Gershoff, 2002).
behavior is already stronger for boys at the age of 4 years is an
important finding of this study. Informant effects. Compared to fathers, mothers reported
The consistent gender difference in mean aggressive and significant higher initial levels of aggressive and delinquent
delinquent scores is in line with other empirical studies behavior problems for boys but not for girls. This is in accord-
(Bongers et al., 2003; Broidy et al., 2003; Loeber & Hay, ance with the study of Seiffge-Krenke and Kollmar (1998)
1994). There is some evidence that boys are at greater risk who reported that the agreement between parental reports on
for developing externalizing behavior than girls (Sanson, problem behavior was greater with respect to daughters as
Oberklaid, Pedlow, & Prior, 1991). It is likely that biological compared to sons and that mothers reported more problems
differences have an important contribution to the development than fathers.
of sex differences in externalizing behavior problems (Keenan
& Shaw, 1997). In his extensive review on causes of aggressive Limitations and future directions. The findings of this study
and disruptive behaviors in childhood and adolescence, Hill contribute to the developmental study of aggressive and delin-
(2002) suggested that testosterone is a strong candidate to quent problem behavior in young children. However, some
explain gender differences. Mazur and Booth (1998) have limitations should be noted.
linked high testosterone levels to dominant behavior. Children A first limitation lies in the reliance on questionnaire
with a high need to dominate their environment are more likely measures. This increases the likelihood of method bias among
to use aggressive behavior. the measures. Therefore, a multimethod measurement strategy
Gender differences in externalizing problem behavior can (by the inclusion of observational measures) may more accu-
also be explained by differences in social interactions. In rately assess children’s individual differences and hence further
contacts with peers, boys are more likely than girls to be more strengthen the results.
physical in their interactions. Charlesworth and Dzur (1987) The design of this study accounted for the requirements that
and Leaper (1991) found that supremacy exchanges and three waves of data is the minimum number to evaluate trends
forceful styles are more common in the social interactions of in the change of problem behavior. Additional time points
boys. Boys develop styles of relating to one another that would allow consideration of more complex forms of growth
encourage and maintain forceful and even aggressive ways of (e.g., a quadratic growth factor) as well as increased power to
interacting. Girls on the other hand develop more cooperative detect effects.
styles of group interaction and use more verbal protest and The focus in this study was on harsh discipline. However,
negotiation during conflicts, which may prevent the escalation other unmeasured parenting behaviors may also have an
of conflicts into aggressive behaviors. important influence (Locke & Prinz, 2002). Future research is
The rate of change was not related to gender. This finding necessary to test the possible effects of positive parenting
suggests that the developmental trajectory of aggressive and (gentle, nonpower) on the development of children’s behavior.
delinquent problem behavior ran parallel in boys and girls. This In the background of this study, additional future research
result is congruent with the findings of Broidy et al. (2003) may contrast the development of externalizing problem
who reported that the developmental trajectory of aggressive behaviors in specific clinical groups with the present norma-
behavior hardly differed between males and females. Moffitt tive data. In addition, it will be interesting to examine the
and Caspi (2001) reported that in the Dunedin Study a single distribution of aggressive and delinquent behavior into distinct
model described the antisocial development of both males and developmental pathways and the degree to which these
females. This conclusion was based on childhood predictors at pathways tend to increase, decrease or remain stable. Several
age 3 years and peer delinquency in adolescence. The finding studies used semi-parametric mixture models to estimate
that boys and girls follow a similar trajectory during the latent trajectory classes of physical aggression in boys (e.g.,
elementary school age years is an important contribution of Nagin & Tremblay, 1999) and girls (e.g., Côté, Zoccolillo,
this study. Tremblay, Nagin, & Vitaro, 2001).

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
458 PRINZIE ET AL. / A COHORT-SEQUENTIAL MULTIVARIATE LATENT GROWTH CURVE ANALYSIS

Conclusion recent research. Journal of Child Psychology and Psychiatry and Allied Disci-
plines, 36, 113–149.
Charlesworth, W.R., & Dzur, C. (1987). Gender comparisons of preschoolers’
The trajectories as described in this study reflect the norma-
behavior and resource utilization in group problem solving. Child Development,
tive development of aggressive and delinquent behavior 58, 191–200.
problems in young children when problem behaviors are Cicchetti, D. (1984). The emergence of developmental psychopathology. Child
repeatedly assessed using parent reports. In addition, the Development, 55, 1–7.
multivariate latent growth analyses illuminate the linkages Coie, J.D., & Dodge, K.A. (1998). Aggression and antisocial behavior. In
W. Damon (Series Ed.) & N. Eisenberg (Vol. Ed.), Handbook of child psychol-
between individual-level aggressive and delinquent behavior ogy: Vol. 3. Social, emotional, and personality development (5th edn, pp. 779–862).
trajectories in young children. Boys had higher initial levels New York: Wiley.
than girls on both aggressive and delinquent problem behavior Collins, L.M., & Sayer, A.G. (2001). New methods for the analysis of change.
but the rate of change was not related to gender. An import- Washington, DC: American Psychological Association.
Côté, S., Zoccolillo, M., Tremblay, R.E., Nagin, S., & Vitaro, F. (2001). Predict-
ant finding of this study is that dysfunctional parenting behav-
ing girls’ conduct disorder in adolescence from childhood trajectories of
iors were differentially related to children’s aggressive and disruptive behaviors. Journal of the American Academy of Child & Adolescent
delinquent problem behavior. A statistically-significant associ- Psychiatry, 40, 678–684.
ation was found between harsh discipline and aggressive Denham, S., Workman, E., Cole, P., Weissbrod, C., Kendziora, K., & Zahn-
problem behavior. Harsh discipline was not related to changes Waxler, C. (2000). Prediction of externalizing behavior problems from early
to middle childhood: The role of parental socialization and emotion
in delinquent behaviors. Our findings with regard to the expression. Development and Psychopathology, 12, 23–45.
persistence indicate that a wait-and-see policy cannot be justi- Duncan, T.E., Duncan, S.C., Strycker L.A., Li, F., & Alpert, A. (1999). An intro-
fied. This highlights the importance of early prevention and duction to latent variable growth curve modeling: Concepts, issues and applications.
intervention and may contribute to the development of effec- Mahwah, NJ: Erlbaum.
Eddy, J.M., Leve, L.D., & Fagot, B.I. (2001). Coercive family processes: A
tive programs. This study demonstrates the feasibility and
replication and extension of Patterson’s coercion model. Aggressive Behavior,
validity of LGM in combination with an accelerated longitu- 27, 14–25.
dinal design to describe problem behaviors over a longer study Edelbrock, C., Rende, R., Plomin, R., & Thompson, L.A. (1995). A twin study
period. of competenceand problem behavior in childhood and early adolescence.
Journal of Child Psychology and Psychiatry, 36, 775–785.
Eley, T.C., Lichtenstein, P., & Moffitt, T.E. (2003). A longitudinal behavioral
genetic analysis of the etiology of aggressive and nonaggressive antisocial
References behavior. Development and Psychopathology, 15, 383–402.
Frick, P.J., Lahey, B.B., Loeber, R., Tannenbaum, L.E., Van Horn, Y., Christ,
Achenbach, T.M. (1991). Manual for the Child Behavior Checklist/4–18 and 1991 M.A.G., et al. (1993). Oppositional defiant disorder and conduct disorder: A
Profile. Burlington: University of Vermont, Department of Psychiatry. meta-analytic review of factor analyses and cross-validation in a clinic sample.
Achenbach, T.M., Conners, C.K., Quay, H.C., Verhulst, F.C., & Howell, C.T. Clinical Psychology Review, 13, 319–340.
(1989). Replication of empirically derived syndromes as a basis for taxonomy Gabel, S., Stadler, J., Bjorn, J., Shindledecker, R., & Bowden, C. (1993).
of child/adolescent psychopathology. Journal of Abnormal Child Psychology, 17, Dopamine-beta-hydroxylase in behaviorally disturbed youth. Relationship
299–323. between teacher and parent ratings. Biological Psychiatry, 34, 434–442.
Achenbach, T.M., & Edelbrock, C.S. (1981). Behavioral problems and compe- Gershoff, E.T. (2002). Corporal punishment by parents and associated child
tencies reported by parents of normal and disturbed children aged four behaviors and experiences: A meta-analytic and theoretical review. Psycho-
through sixteen. Monographs of the Society for Research in Child Development, logical Bulletin, 128, 539–579.
46(1, Serial No. 188). Hellinckx, W., Prinzie, P., Onghena, P., Ghesquière, P., Colpin, H., Grietens,
Arbuckle, J.L. (1996). Full information estimation in the presence of incomplete H., et al. (2000). Het Leuvens Instrument voor Coërcief Opvoedingsgedrag
data. In G.A. Marcoulides & R.E. Schumacker (Eds.), Advanced structural (LICO). Leuven: Katholieke Universiteit Leuven, Section orthopedagogics.
equation modeling: Issues and techniques (pp. 243–277). Mahwah, NJ: Erlbaum. Hill, J. (2002). Biological, psychological and social processes in the conduct
Arnold, D.S., O’Leary, S.G., Wolff, L.S., & Acker, M.M. (1993). The Parent- disorders. Journal of Child Psychology and Psychiatry, 43, 133–164.
ing Scale: A measure of dysfunctional parenting and discipline situations. Hoffman, M.L. (1994). Discipline and internalization. Developmental Psychology,
Psychological Assessment, 5, 137–144. 30, 26–28.
Barnow, S., Lucht, M., & Freyberger, H.J. (2005). Correlates of aggressive and Hofstra, M.B., Van der Ende, J., & Verhulst, F.C. (2000). Continuity and change
delinquent conduct problems in adolescence. Aggressive Behavior, 31, 24–39. of psychopathology from childhood into adulthood: a 14-year follow-up
Bell, R.Q. (1953). Convergence: An accelerated longitudinal approach. Child study. Journal of the American Academy of Child and Adolescent Psychiatry, 39,
Development, 24, 145–152. 850–858.
Bell, R.Q. (1954). An experimental test of the accelerated longitudinal approach. Jöreskog, K.G., & Sörbom, D. (1989). LISREL 7: A guide to the program and
Child Development, 25, 281–286. applications (2nd edn). Chicago: SPSS.
Bentler, P.M. (1990). Comparative fit indices in structural models. Psychological Jöreskog, K.G., & Sörbom, D. (2003). LISREL 8.54 [Computer software].
Bulletin, 107, 238–246. Chicago: Scientific Software International.
Blozis, S.A. (2004). Structured latent curve models for the study of change in Keenan, K., & Shaw, D.S. (1997). Developmental influences on young girls’
multivariate repeated measures. Psychological Methods, 9, 334–353. behavioral and emotional problems. Psychological Bulletin, 121, 95–113.
Bollen, K.A. (1989). Structural Equations with Latent Variables. New York: Wiley. Keiley, M.K., Bates, J.E., Dodge, K.A., & Pettit, G.S. (2000). A cross-domain
Bongers, I.L., Koot, H.M., Van der Ende, J., & Verhulst, F.C. (2003). The growth analysis: Externalizing and internalizing behavior during 8 years of
normative development of child and adolescent problem behavior. Journal of childhood. Journal of Abnormal Child Psychology, 28, 161–179.
Abnormal Psychology, 112, 179–192. Kline, R. (1998). Principles and practice of structural equation modeling. New York:
Bongers, I.L., Koot, H.M., Van der Ende, J., & Verhulst, F.C. (2004). Develop- Guilford Press.
mental trajectories of externalizing behaviors in childhood and adolescence. Kuczynski, L., & Hildebrandt, N. (1997). Models of conformity and resistance
Child Development, 75, 1523–1537. in socialization theory. In J.E. Grusec & L. Kuczynski (Eds.), Parenting and
Broidy, L.M., Nagin, D.S., Tremblay, R.E., Bates, J.E., Brame, B., Dodge, K.A., the internalization of values: A handbook of contemporary theory (pp. 227–256).
et al. (2003). Developmental trajectories of childhood disruptive behaviors New York: Wiley.
and adolescent delinquency: A six site, cross-national study. Developmental Lahey, B.B., Miller, T.L., Gordon, R.A., & Riley, A.W. (1999). Developmental
Psychology, 39, 222–245. epidemiology of the disruptive behavior disorders. In H.C. Quay & A.E.
Browne, M.W., & Cudeck, R. (1993). Alternative ways of assessing model fit. Hogan (Eds.), Handbook of disruptive behavior disorders (pp. 23–48). New York:
In K.A. Bollen & J.S. Long (Eds.), Testing Structural Equation Models Plenum.
(pp. 132–162). Beverley Hills, CA: Sage. Lahey, B.B., Schwab-Stone, M., Goodman, S.H., Waldman, I.D., Canino, G.,
Bryk, A.S., & Raudenbush, S.W. (1992). Hierarchical linear models: Applications Rathouz, P.J., et al. (2000). Age and gender differences in oppositional
and data analysis methods. Newbury Park: Sage. behavior and conduct problems: A cross-sectional household study of middle
Campbell, S.B. (1995). Behavior problems in preschool children: A review of childhood and adolescence. Journal of Abnormal Psychology, 109, 488–503.

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.
 INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2006, 30 (5), 444–459 459

Leaper, C. (1991). Influence and involvement in children’s discourse: Age, Patterson, G.R., Reid, J.B., & Dishion, T.J. (1992). Antisocial boys. Eugene, OR:
gender and partner effects. Child Development, 62, 797–811. Castalia.
Locke, L.M., & Prinz, R.J. (2002). Measurement of parental discipline and Prinzie, P., Onghena, P., & Hellinckx, W. (in press). The validity of the Parent-
nurturance. Clinical Psychology Review, 22, 895–929. ing Scale for assessing the discipline practices of parents of elementary
Loeber, R. (1982). The stability of antisocial and delinquent child behavior: A school-aged children. European Journal of Psychological Assessment.
review. Child Development, 53, 1431–1446. Prinzie, P., Onghena, P., Hellinckx, W., Grietens, H., Ghesquière, P., & Colpin,
Loeber, R., & Farrington, D.P. (1997). Strategies and yields of longitudinal H. (2003). The additive and interactive effects of parenting and children’s
studies on antisocial behavior. In D.M. Stoff, J. Breiling, & J.D. Maser (Eds.), personality on externalising behaviour. European Journal of Personality, 17,
Handbook of antisocial behavior (pp. 125–139). New York: Wiley. 95–117.
Loeber, R., Farrington, D.P., Stouthamer-Loeber, M., Moffitt, T.E., & Caspi, Rao, C.R. (1958). Some statistical models for comparison of growth curves.
A. (1998). The development of male offending: Key findings from the first Biometrics, 14, 1–17.
decade of the Pittsburgh Youth Study. Studies on Crime and Prevention, 7, Richman, M., Stevenson, J., & Graham, P.J. (1982). Preschool to school: A behav-
1–31. ioral study. London: Academic Press.
Loeber, R., Green, S.M., Lahey, B.B., Frick, P.J., & McBurnett, K. (2000). Rothbaum, F., & Weisz, J.R. (1994). Parental caregiving and child externalizing
Findings on disruptive behavior disorders from the first decade of the behavior in nonclinical samples. A meta-analysis. Psychological Bulletin, 116,
Developmental Trends Studies. Clinical Child and Family Psychology Review, 55–74.
3, 37–60. Sanson, A., Oberklaid, F., Pedlow, R., & Prior, M. (1991). Risk indicators:
Loeber, R., & Hay, D.F. (1994). Developmental approaches to aggression and Assessment of infancy predictors of pre-school behavioural maladjustment.
conduct problems. In M. Rutter & D.F. Day (Eds.), Development through life: Journal of Child Psychology and Psychiatry, 32, 609–626.
A handbook for clinicians (pp. 488–516). Oxford: Blackwell. Schafer, J.L. (1997). Analysis of incomplete multivariate data. London: Chapman
Loeber, R., Lahey, B., & Thomas, C. (1991). Diagnostic conundrum of & Hall.
oppositional defiant disorder and conduct disorder. Journal of Abnormal Schafer, J.L., & Graham, J.W. (2002). Missing data: Our view of the state of the
Psychology, 100, 379–390. art. Psychological methods, 7, 147–177.
Loeber, R., Wung, P., Keenan, K., Giroux, B., Stouthamer-Loeber, M., Seiffge-Krenke, I., & Kollmar, F. (1998). Discrepancies between mothers’ and
VanKammen, W.B., et al. (1993). Developmental pathways in disruptive child fathers’ perceptions of sons’ and daughters’ problem behaviour: A longitudi-
behavior. Development and Psychopathology, 5, 103–133. nal analysis of parent–adolescent agreement on internalising and externalis-
Lytton, H., & Romney, D.M. (1991). Parents’ differential socialization of boys ing problem behavior. Journal of Child Psychology and Psychiatry, 39, 687–697.
and girls: A meta-analysis. Psychological Bulletin, 109, 267–296 Singer, J.D., & Willett, J.B. (2003). Applied Longitudinal Data Analysis. New York:
MacCallum, R.C., Browne, M.W., & Sugawara, H.M. (1996). Power analysis Oxford University Press.
and determination of sample size for covariance structure modeling. Psycho- Snyder, J., Cramer, A., Afrank, J., & Patterson, G.R. (2005). The contributions
logical Methods, 1, 130–149. of ineffective discipline and parental hostile attributions of child misbehavior
Mazur, A., & Booth, A. (1998). Testosterone and dominance in men. Behavioral to the development of conduct problems at home and school. Developmental
& Brain Sciences, 21, 353–397. Psychology, 41, 30–41.
McArdle, J.J., & Anderson, E. (1990). Latent variable growth models for Snyder, J., Reid, J., & Patterson, G. (2003). A social learning model of child and
research on aging. In J.E. Birren & K.W. Schaie (Eds.), Handbook of the adolescent antisocial behavior. In B.B. Lahey, T.E. Moffit, & A. Caspi (Eds.),
psychology of aging (pp. 21–43). New York: Academic Press. Causes of conduct disorder and juvenile delinquency (pp. 27–47). New York:
McConaughy, S.H., Stanger, C., & Achenbach, T.M. (1992). Three-year course Guilford Press.
of behavioral/emotional problems in a national sample of 4- to 16-year-olds: Sroufe, L.A. (1997). Psychopathology as an outcome of development. Develop-
I. Agreement among informants. Journal of the American Academy of Child and ment and Psychopathology, 9, 251–268.
Adolescent Psychiatry, 31, 932–940. Stanger, C., Achenbach, T.M., & Verhulst, F.C. (1997). Accelerated longitudi-
Meredith, W., & Tisak, J. (1990). Latent curve analysis. Psychometrika, 55, nal comparisons of aggressive versus delinquent syndromes. Developmental
107–122. Psychopathology, 9, 43–58.
Moffitt, T.E., & Caspi, A. (2001). Childhood predictors differentiate life-course Steiger, J.H. (1990). Structural model evaluation and modification: An interval
persistent and adolescence-limited antisocial pathways among males and estimation approach. Multivariate Behavioral Research, 25, 173–180.
females. Development and Psychopathology, 13, 355–375. Stoolmiller, M. (1994). Antisocial behavior, delinquent peer association, and
Moffitt, T.E., Caspi, A., Harrington, H., & Milne, B. (2002). Males on the life- unsupervised wandering for boys: Growth and change from childhood to early
course persistent and adolescence-limited antisocial pathways: Follow-up at adolescence. Multivariate Behavioral Research, 29, 263–288.
age 26. Development and Psychopathology, 14, 179–206. Tremblay, R.E. (2000). The development of aggressive behavior during child-
Munson, J.A., McMahon, R.J., & Spieker, S.J. (2001). Structure and variability hood: What have we learned in the past century? International Journal of
in the developmental trajectory of children’s externalizing problems: Impact Behavioral Development, 24, 129–141.
of infant attachment, maternal depressive symptomatology, and child sex. Tuvblad, C., Eley, T.C., & Lichtenstein, P. (2005). The development of anti-
Development and Psychopathology, 13, 277–296. social behaviour from childhood to adolescence. A longitudinal twin study.
Nagin, D.S. (1999). Analyzing developmental trajectories: A semi-parametric, European Child & Adolescent Psychiatry, 14, 216–225.
group-based approach. Psychological Methods, 4, 139–157. Van Westerlaak, J.M., Kropman, J.A., & Collaris, J.W.M. (1975). Beroepenklapper
Nagin, D.S. (2005). Group-based modeling of development. Cambridge, MA: [Manual for occupational level]. Nijmegen, The Netherlands: Instituut voor
Harvard University Press. Toegepaste Sociologie.
Nagin, D.S., & Tremblay, R.E. (1999). Trajectories of boys’ physical aggression, Verhulst, F.C., Van der Ende, J., & Koot, H.M. (1996). Handleiding voor de
opposition, and hyperactivity on the path to physically violent and nonviolent CBCL/4–18 [Manual for the CBCL/4–18]. Rotterdam, the Netherlands:
juvenile delinquency. Child Development, 70, 1181–1196. Erasmus University and Department of Child and Adolescent Psychiatry,
Nagin, D.S., & Tremblay, R.E. (2005). What has been learned form the group- Sophia Children’s Hospital.
based trajectory modeling? Examples from physical aggression and other Vignoe, D., Bérubé, R.L., & Achenbach, T.M. (2000). Bibliography of published
problem behaviors. The Annals of the American Academy of Political and Social studies using the Child Behaviour Checklist and related materials (1999 edn).
Science, 602, 82–117. Burlington: University of Vermont, Department of Psychiatry.
Nesselroade, J.R., & Baltes, P.B. (Eds.) (1979). Longitudinal research in the study Willett, J.B., & Sayer, A.G. (1994). Using covariance structure analysis to detect
of behavior and development. New York: Academic Press. correlates and predictors of change. Psychological Bulletin, 116, 363–381.
Paikoff, R.L., Brooks-Gunn, J., & Warren, M.P. (1991). Effects of girls’ Willett, J.B., & Sayer, A.G. (1996). Cross-domain analyses of change over time:
hormonal status on depressive and aggressive symptoms over the course of Combining growth modeling and covariance structure analysis. In R.
one year. Journal of Youth and Adolescence, 2, 191–215. Schumacher & G. Marcoulides (Eds.), Advanced structural equation modeling
Patterson, G.R. (1982). Coercive family process. Eugene, OR: Castalia. techniques. Hillsdale, NJ: Erlbaum.
Patterson, G.R. (1993). Orderly change in a stable world: The antisocial trait as
a chimera. Journal of Consulting and Clinical Psychology, 61, 911–919.

Downloaded from http://jbd.sagepub.com at CAPES on April 16, 2007

© 2006 International Society for the Study of Behavioral Development. All rights reserved. Not for commercial use or unauthorized distribution.