Callous-unemotional traits and social information processing: Multiple risk-factor models for understanding aggressive behavior in antisocial youth

ORIGINAL ARTICLE

Callous-Unemotional Traits and Social Information Processing:Multiple Risk-Factor Models for Understanding AggressiveBehavior in Antisocial Youth

Timothy R. Stickle Æ Neil M. Kirkpatrick ÆLauren N. Brush

Published online: 9 January 2009! American Psychology-Law Society/Division 41 of the American Psychological Association 2008

Abstract This study examined multiple risk factor mod-els of links among callous-unemotional traits, aggression

beliefs, social information processing, impulsivity, and

aggressive behavior in a sample of 150 antisocial adoles-cents. Consistent with past research, results indicated that

beliefs legitimizing aggression predicted social information

processing biases and that social information processingbiases mediated the effect of beliefs on aggressive behavior.

Callous-unemotional traits accounted for unique variance in

aggression above and beyond effects of more establishedrisk factors of early onset of antisocial behavior, social

information processing, and impulsivity. These findings add

to recent research showing that callous-unemotional traitsare a unique risk factor associated with aggression and

criminal offending and suggest that targeting both affective

and cognitive vulnerabilities may enhance clinical inter-vention with antisocial youth.

Keywords Psychopathy ! Callous-unemotional traits !Aggression

Effectively understanding, preventing, and treating antiso-

cial behavior is hampered by lack of clear, integrated theory

and empirical findings. Indeed, Dodge and Pettit’s (2003)state-of-the-science review of the literature noted that

research on the origins of antisocial development over the

past 20 years has been characterized by numerous rigorous,focused studies that have proceeded without regard to each

other. Thus, although this body of research has yielded

enormous empirical support for individual aspects ofantisocial development, it largely lacks integration. Conse-

quently, as these authors noted, two decades of impressive

research has resulted in identification of dozens of looselyconnected risk factors and predictors of antisocial behavior,

but little understanding of how the identified genetic and

biological predispositions, life experiences, cognitive andemotional processes, and sociocultural contexts work toge-

ther to produce enacted antisocial behavior such as

aggression (Dodge & Pettit, 2003).One step toward improving on current knowledge is to

design and conduct studies that reach across and integrate

previously separate areas of study. This study makes amodest step in this direction by examining the direct and

mediating effects of two important, but mostly parallel,

lines of research on processes related to antisocial out-comes: (a) social cognition, and (b) affective deficits

represented by callous and unemotional traits. Specifically,this study examines the direct, indirect, and mediating

effects of social information processing and the direct,

indirect, and mediating effects of callous-unemotionaltraits, a key dimension of psychopathy, in a multiple risk

factor analytic model predicting aggressive behavior

among adjudicated, antisocial youth.The study of social information processing, those

inferences and judgments made about social stimuli in

social situations, and its association with aggression hasproven to be one of the most fruitful areas of study for

understanding processes implicated in persistent aggres-

sion. Biased social information processing has been relatedto both vulnerability for developing aggressive, antisocial

behavior and maintenance of such behavior once it has

T. R. Stickle (&)Department of Psychology, University of Vermont,Burlington, VT, USAe-mail: [email protected]

N. M. Kirkpatrick ! L. N. BrushUniversity of Vermont, Burlington, VT, USA

123

Law Hum Behav (2009) 33:515–529

DOI 10.1007/s10979-008-9171-7

onset (e.g., Crick & Dodge, 1994; Dodge & Pettit, 2003).

Information processing during social interaction involvessequential steps of cognitive and emotional processes

leading to enacted behavior (Dodge & Pettit, 2003). For

example, if a boy is hit by a ball on the playground, does hetoss it back to the ongoing game, ignore it, or retaliate

aggressively by kicking it away or throwing it at those

playing the game? Models derived from social informationprocessing theory describe behavioral responses as result-

ing from a sequence of steps consisting of (a) encoding and

interpreting social cues (e.g., ‘‘I was hit by accident,’’ or‘‘He tried to hurt me’’), (b) accessing and generating

aggressive or prosocial responses from a remembered

repertoire (e.g., kicking the ball, ignoring it and walkingaway), (c) evaluating these potential responses including

whether a response will yield a positive or negative out-

come (outcome expectancy), and (d) choosing and enactinga response behaviorally (Crick & Dodge, 1994; Dodge &

Pettit, 2003; Dubow & Reid, 1994; Erdley & Asher, 1998;

Huesmann, 1998).Contemporary models of social information processing

also note that life experiences lead to identifiable patterns of

thinking about life events (see e.g., Dodge & Pettit, 2003;Huesmann, 1988) called knowledge structures (sometimes

referred to as schemas or scripts). Social knowledge struc-

tures are cognitive processes or patterns of thinking aboutsocial relationships. For example, beliefs about social

norms, such as the appropriateness or legitimacy of behav-ing aggressively, constitute one such knowledge structure.

These normative beliefs about aggression formed through

life experiences appear to predict individual differences inaggressive behavior (Huesmann & Guerra, 1997).

Thus, contemporary social information processing

models of aggression and antisocial behavior generallyinclude some measurement of beliefs (e.g., beliefs about

aggression), as evidence indicates that social information

processing mediates the link between social knowledgestructures and aggressive behavior (e.g., Huesmann &

Guerra, 1997). Figure 1 illustrates relations among the

variables in such a model. Despite findings from numerousstudies supporting relations among deficits in social infor-

mation processing and aggression (see Crick & Dodge, 1994

for review), prediction of aggression using even these morecomplete models is typically modest (Dodge & Pettit, 2003).

Moreover, considering mediation by multiple stages of

information processing simultaneously is essential becauseprocesses identified in these stages function together to

affect behavior. Evidence suggests that considering multiple

mediating processes of social information processing toge-ther improves the magnitude of prediction of aggression

(e.g., Zelli, Dodge, Lochman, Laird, & Conduct Problems

Prevention Research Group, 1999). Nevertheless, predictionis still moderate, lacks consistency across studies in terms of

which social information processing biases are associated

with aggression, and the associations appear limited to

particular settings, such as school.

CONSIDERING MULTIPLE RISK PREDICTORSMAY IMPROVE PREDICTION OF ANTISOCIALOUTCOMES

It is well established that antisocial behavior is heteroge-

neous in its risk factors, course, and outcomes. Thus, it is not

surprising that focusing on a single level of risk processessuch as social cognition yields modest prediction. The the-

oretical underpinnings of social information processing

models emphasize emotional and behavioral undercontrol,such as reactivity to threat, leading to aggression (e.g.,

hostile attributional bias). Recent research, however, indi-

cates that temperamental vulnerability to either emotionalundercontrol or to emotional overcontrol confers vulnera-

bility for the development of conduct problems (see Frick &

Morris, 2004 for review).For example, past approaches that focus on identification

of high-risk youth through individual antisocial outcomes,

such as the type (e.g., proactive vs. reactive aggression) orseverity of aggressive and antisocial behavior, postulate that

emotional and behavioral undercontrol in the forms of

emotion dysregulation, deficits in cognitive skills, andimpulsivity are the primary causal vulnerabilities leading to

aggression and other conduct problems (e.g., Dodge & Pettit,2003; Eisenberg, Fabes, Guthrie, & Reiser, 2000; Giancola,

Mezzich, & Tarter, 1998). In contrast, an emerging body of

recent research has focused on vulnerability processes in theyouth’s affective (e.g., absence of guilt, constricted emotion)

and interpersonal (e.g., failure to show empathy, use of

others for gain) style. This affective and interpersonal style,designated as callous-unemotional traits, does indeed appear

to confer heightened risk for severe (Flight & Forth, 2007;

Aggression Beliefs

HAB

Aggressive response

bias

Prosocial response

bias

Outcome Expectancy

Aggression

Fig. 1 Theoretical model of aggression beliefs mediated by socialinformation processing associated with aggressive behavior

516 Law Hum Behav (2009) 33:515–529

123

Frick, Cornell, Barry, Bodin, & Dane, 2003; Salekin, 2006;

Salekin, Neumann, Leistico, DiCicco, & Duros, 2004), andpersistent antisocial and aggressive behavior (Frick, Stickle,

Dandreaux, Farrell, & Kimonis, 2005; Salekin, 2008; Sale-

kin, Rosenbaum, & Lee, 2008). Specifically, this recentresearch indicates that antisocial youth who show emotional

overcontrol as evidenced by high levels of callous-unemo-

tional traits (e.g., Caputo, Frick, & Brodsky, 1999; Christian,Frick, Hill, Tyler, & Frazer, 1997; Frick et al., 2003; Mul-

lins-Nelson, Salekin, & Leistico, 2006; Stickle & Frick,2002), viewed by many as the cornerstone of psychopathy

(Cleckley, 1976; Frick & White, 2008; Hare, 1998), are at

risk for particularly severe and persistent antisocial behavior(Barry, Barry, Deming, & Lochman, 2008; Burke, Loeber, &

Lahey, 2007; Dadds, Whiting, & Hawes, 2006; Frick et al.,

2005; Lynam, Loeber, & Stouthamer-Loeber, 2008; Salekinet al., 2008; Skeem & Cauffman, 2003; see Frick & Dickens,

2006 for a review).

Although persistently antisocial and aggressive youthare generally characterized by impaired emotional and

behavioral regulation, including high rates of impulsivity,

cognitive, and neuropsychological deficits (e.g., Moffitt,1993), the developmental and risk processes leading to

dysregulation appear to differ for youth characterized by

specific sets of individual differences. Persistently antiso-cial youth characterized primarily by impulsivity and

conduct problems (I/CP) tend to have an early onset (by

age 10) of antisocial behavior and are characterized bydysregulation resulting from the interaction of a vulnerable

temperament and an inadequate rearing environment (e.g.,

Moffitt, 1993). Callous-unemotional traits, however,appear to be a marker for an even more severe pattern of

antisocial behavior among early onset youth whose char-

acteristics are consistent with another temperamental style,one associated with low emotional reactivity to aversive

stimuli. This low reactivity is characterized physiologically

by underreactivity in the sympathetic nervous system, andbehaviorally by low fearfulness to novel or threatening

situations and poor responsiveness to cues of punishment

(Glenn, Raine, Venables, & Mednick, 2007; Kagan &Snidman, 1991; Stickle & Frick, 2002). This tempera-

mental style can impair development of the affective

components of conscience (Blair, 1999; Frick et al., 2003;Kochanska, 1993). The emotional deficits exhibited by

such youth are consistent with those observed in adults

with psychopathic traits, and appear to be a key factor inthe pattern of severe aggression and violence shown by

both antisocial adults and youth with callous-unemotional

traits (e.g., Flight & Forth, 2007; Hemphill, 2007; Porter &Woodworth, 2006; Salekin, Rogers, & Sewell, 1996).

Taken together, these findings suggest that better

understanding of characteristics associated with increasedaggression and other antisocial outcomes is likely to be

enhanced by considering processes indicative of both un-

dercontrol, such as I/CP and biases in social informationprocessing, and processes indicative of overcontrol, such as

callous-unemotional traits. Limited previous findings sug-

gest that callous-unemotional traits among antisocial youthmay indeed be associated with relations among information

processing and aggression (Pardini, Lochman, & Frick,

2003) that differ from that of antisocial youth low in cal-lous-unemotional traits. In that study, social cognition in

youth high in callous-unemotional traits was characterizednot by undercontrol and misinterpretation of social cues

such as hostile attributional bias, but by perceived rewards

in a higher expectation of positive outcomes from aggres-sive behavior. Flight and Forth (2007) also found

significant associations between a measure of psychopathy

and expectation of positive outcomes from aggression (i.e.,instrumental aggression) among adolescents, which the

authors suggested could be attributed to the affective def-

icits associated with callous-unemotional traits. Neither ofthese studies, however, examined multiple stages of social

information processing, or more directly examined the

specific links between social cognition (beliefs and infor-mation processing) and multiple informant reports of

aggression. Consequently, although this promising evi-

dence points to links among callous-unemotional traits,social cognition, and aggression, observed relationships

were limited to associations between higher levels of cal-

lous-unemotional traits and aggressive intentions andexpectations about outcomes of aggression under particular

circumstances.

Based on these considerations, this study examinedwhether callous-unemotional traits are uniquely associated

with multiple informant ratings of aggressive behavior

above and beyond the level of aggressive behavioraccounted for by deficits in social information processing

(e.g., intentions and interpretations of social provocations)

and impulsivity (I/CP). In order to simultaneously examinemultiple risk factors that incorporate processes that are

associated with emotional overcontrol and emotional un-

dercontrol (i.e., callous-unemotional traits, I/CP, and socialinformation processing), we tested multiple mediator

models of the associations among aggression beliefs,

multiple stages of social information processing, I/CP,callous-unemotional traits, and aggression in antisocial

adolescents. Specifically, it was predicted that callous-

unemotional traits would be positively associated withhigher scores on legitimacy of aggression beliefs, fre-

quency of aggressive responses to provocation (intent), and

expected positive outcomes of aggression. Callous-unemotional traits were also expected to account for unique

variance in multiple informant ratings of aggression above

and beyond the variance associated with beliefs, socialinformation processing, and I/CP.

Law Hum Behav (2009) 33:515–529 517

123

Much research has been conducted on social informa-

tion processing and aggression in school-age children up toabout age 10 (see Crick & Dodge, 1994). Little research

has been conducted on older youth or those involved in the

juvenile justice system. Yet, it is arguably these youth whoexperience the most difficulty with aggression. An adju-

dicated adolescent sample was chosen because (1) there is

inadequate understanding of the relationship betweensocial information processing and aggression for youth in

this age range and with an advanced developmental level ofantisocial behavior, (2) aggression can potentially be

altered in adolescents through established cognitive

behavioral interventions (Kazdin & Weisz, 1998), (3)oversampling in an antisocial population ensures partici-

pation by a sufficient number of youth with a low base rate

characteristic such as callous-unemotional traits, necessaryto achieve statistical power for tests of relations among

callous-unemotional traits and the other processes, and (4)

testing risk processes under a variety of circumstances andsampling at the extreme ends of severity have been iden-

tified as critical areas of needed study for advancing

knowledge of antisocial behavior (Dodge & Pettit, 2003). Itmight be argued that this sampling strategy limits gener-

alizability of the findings with regard to application of

callous-unemotional traits and social information process-ing with non-adjudicated youth. However, the advantage of

including enough youth with high levels of callous-

unemotional traits to meaningfully test these relationshipswas deemed to outweigh this disadvantage. Moreover,

persistently antisocial youth are understudied and these

processes are not well understood in this population.

METHOD

Participants

Participants were 150 youth from two juvenile detention

centers in the same small state. Because of the small

population of the state, the detention centers serve the samepopulation. One center is co-ed and the second center

serves only females. The centers have overlap of approx-

imately 90% of youth (i.e., most female youth in each havebeen in the other center). Because we oversampled both

females and eligible minority youth, the composition of the

sample was 40% female and 15% ethnic minority, withapproximately 4% African American, 5% Hispanic, 3%

Asian American or Pacific Islander, and 3% from other

minority groups. These distributions exceed the statepopulation of approximately 5% minority, and the deten-

tion population of approximately 30% females. Inclusion

criteria included youth: 1) between 11 and 17 years-of-age,2) functioning within the normal range of intelligence as

determined by cognitive assessments conducted by aca-

demic staff upon entry to detention including a WideRange Achievement Test (WRAT-3) basic reading ability

standard score greater than or equal to 80 (Wilkinson,

1993), and adequate academic functioning, and 3) resi-dence at the detention center for at least 2 weeks to ensure

completion of the protocol and sufficient time for adequate

observation by teachers and staff, who completed behav-ioral rating scales. Exclusion criteria included evidence of:

1) intellectual impairment that may limit ability to validlycomplete measures, 2) otherwise limited mental compe-

tency including any of the following conditions: Pervasive

Developmental Disorders, Mental Retardation, SelectiveMutism, Organic Mental Disorders, Schizophrenia, Other

Psychotic Disorders, or the inability to give informed,

written assent to participate in research. A total of 15 youthwere excluded: Five youth were excluded because of lim-

ited intellectual functioning; five youth were excluded

because they were unexpectedly transferred out of thefacility before completing all study measures; five youth

declined participation after caseworker consent had been

obtained.

Measures

Multiple Informant Measures

Aggressive Behavior The Youth Self-Report (YSR) and

Teacher Report Form (TRF) (Achenbach, 1991a, 1991b)are 118-item behavior checklists completed by youths 4–18

(YSR), and teachers (TRF) as an interview or filled out

directly. The checklists provide numerous scores, includingantisocial and aggressive behavior (Achenbach, 1991c), the

latter being the major scale of interest for this study.

Reliability is good for these instruments, with average test–retest reliability coefficients reported across age and gender

groups for the aggressive and delinquent behavior scales of

.91 and .86 (Achenbach, 1991a). The TRF is also appro-priate for use by residential counselors and other non-

teaching adults, given an adequate sample of a child’s

behavior in a classroom or other environment with peers(Achenbach, 1991b). A composite aggression score from

these scales was used as the primary dependent measure.

Cronbach’s alpha for the aggressive subscales on the TRFand YSR for this sample were adequate and similar to other

samples at .96 and .81.

Impulsive Conduct Problems (I/CP) The Antisocial Pro-cess Screening Device (APSD) (Frick & Hare, 2001) rating

scale (teacher, parent, and self-report versions are avail-

able) asks informants to rate how true 20 statements are ofa youth on a 3-point scale. Factor analyses have identi-

fied two correlated factors: a callous-unemotional factor

518 Law Hum Behav (2009) 33:515–529

123

(6 items) relating to a lack of guilt and remorse, low

empathy, and shallow constricted affect; and an impulsiveconduct problems factor (10 items) relating to antisocial

behavior, impulsivity and recklessness, and a tendency to

be easily bored (Frick, O’Brien, Wootton, & McBurnett,1994). The factors have good temporal stability and

internal consistency (Frick, Bodin, & Barry, 2000). The

self-report version reliably differentiates subgroups ofantisocial youth in adolescent offender samples (Caputo

et al., 1999). Cronbach’s alphas for this sample for staff,youth, and combined were .77, .72, and .73.

Callous-Unemotional Traits The Inventory of Callous

and Unemotional Traits (ICU) (Essau, Sasagawa, & Frick,2006) is a 24-item measure, also with child, parent, and

teacher versions. The ICU was based on the six-item cal-

lous-unemotional traits scale of the APSD, which has beenshown to identify a subgroup of antisocial youth with traits

similar to those of adult psychopaths in adjudicated, clin-

ical, and community samples (Essau et al., 2006; Fricket al., 2000). To create the ICU, the four APSD items that

consistently loaded on the callous-unemotional traits scale

in both community and clinical samples were expandedinto three positively worded and three negatively worded

items, to create the 24 items. Additional items appear to

have increased reliability over the APSD in measuringcallous-unemotional traits. In an initial study using a large

community sample of adolescents, Cronbach’s alpha was

reported as .77 (Essau et al., 2006). These authors alsoreported evidence supporting construct validity (conver-

gent and discriminant) of this measure. Results in the

current sample were consistent with those in Essau et al.(2006), with Cronbach’s alphas for staff, youth, and com-

bined of .88, .84, and .80, respectively.

A multi-informant composite score for each of the abovethree measures, the APSD (I/CP), ICU (callous-unemo-

tional traits), and the aggressive subscale of the TRF and

YSR (aggressive behavior), was computed. Following therecommendation of Piacentini, Cohen, and Cohen (1992)

for maximal sensitivity in assessing syndromes and disor-

ders, and consistent with past research and the publishedAPSD manual (Frick & Hare, 2001), these composites

were formed by considering endorsement of an item by any

informant to represent the presence of that trait, behavior,or frequency. Thus, the highest score among informants for

each item was used to index callous-unemotional traits, I/

CP, and aggressive behavior. As has been noted in previ-ously published research, motivation to underreport highly

socially undesirable traits is much more likely than that to

over report them (Frick et al., 2003). Thus, this method ofusing multiple informant data is considered optimal for

identifying youth along the range of severity of these

dimensions (Piacentini et al., 1992).

Youth Self-Report and Hypothetical Situation Measures

Demographics Data on youth current age (date of birth),age of onset of antisocial behavior, age of first arrest, and

other demographic characteristics were gathered as part of

the intake process.

Beliefs About Aggression The Normative Beliefs about

Aggression Scale (NOBAGS) is a 20-item scale assessing

beliefs about the acceptability of aggressive responses insocial situations (Huesmann & Guerra, 1997). Eight items

gauge general beliefs about aggression (e.g., ‘‘in general

it’s OK to use violence’’), and 12 items gauge beliefs aboutthe acceptability of aggression when provoked. Youth

respond on a 4 point scale ranging from ‘‘really wrong’’ to

‘‘perfectly OK.’’ The NOBAGS has good internal consis-tency (Guerra, Huesmann, Tolan, Van Acker, & Eron,

1995), test–retest reliability, and correlates with observa-

tional measures of aggressive behavior (Huesmann &Guerra, 1997). Scores on aggression beliefs in response to

provocation were used as the measure of aggression beliefs

in order to examine the effect of aggression beliefs undersuch circumstances. Reliability for this sample was ade-

quate with Cronbach’s alpha of .93.

Positive Outcome Expectancy The Outcome ExpectancyQuestionnaire (OEQ) (Perry, Perry, & Rasmussen, 1986)

consists of eight brief vignettes. These vignettes ask youth

to imagine themselves considering the use of aggressivebehavior against a same sex peer to obtain a particular

outcome such as a tangible reward (e.g., threatening a peer

for money). Participants evaluate the likelihood they wouldreceive the tangible reward (i.e., instrumental or social

goal) or decrease the aversive behavior, be punished, and

achieve a sense of dominance if they performed theaggressive act. The OEQ has several scales, each of which

has adequate reliability and reliably discriminates between

antisocial youth and controls (Hall & Skrowronski, 1998),and has shown similar psychometrics and association with

outcomes for adjudicated youth (Pardini et al., 2003).

Scores from this measure gauging expectation of tangibleor social reward for aggressive behavior were used for the

measure of the social information processing stage of

outcome expectancy. Reliability in this sample was fair togood, with Cronbach’s alphas for the individual scales

between .66 and .90.

Beliefs About Relational Aggression The RelationalAggression Measure (RAM) was created for this study and

was used to assess participants’ beliefs about the accept-

ability of relational aggression in social situations.Relevant to this study were four items identical in form to

those used on the NOBAGS (Guerra et al., 1995). This

Law Hum Behav (2009) 33:515–529 519

123

measure was used to include relational aggression in

indexing beliefs across different types of aggression andsocial situations. Thus, scores on beliefs about relational

aggression were combined with NOBAGS scores in order

to capture beliefs relevant to both genders by includingrelationally aggressive items (Crick, Grotpeter, & Bigbee,

2002). Reliability was adequate for the 4-item beliefs scale

with Cronbach’s alpha of .82. The combined aggressionbeliefs score from the RAM and NOBAGS had adequate

reliability with a Cronbach’s alpha of .93.

Social Information Processing The Attribution and

Response to Ambiguous Provocation Scale (ARAPS) is a

hypothetical situation instrument adapted from one ini-tially used by Dodge (1980), and variously expanded and

adapted by others (e.g., Crick, 1995). It consists of 12

vignettes describing a provocation by a peer, in whichthe intentions of the peer are ambiguous. One vignette

was written for this study while the remaining vignettes

were drawn from previous versions and adapted for ageappropriateness (lunchroom instead of playground), con-

tent (CD player instead of radio), and gender balance.

The 12 vignettes were counterbalanced for type ofprovocation (e.g., instrumental vs. relational) and gender

of the provoking youth. Participants are asked to respond

to four questions following each vignette, (1) to choosefrom four possible reasons for the provocation, with two

responses describing benign intent and two describing

hostile intent, (2) to answer an open-ended question ontheir response, ‘‘What would you do…?,’’ (3) to rate the

likelihood they would also engage in each of six ran-

domly ordered behavioral alternatives (physically,verbally, covertly, and relationally aggressive, prosocial,

avoidant), and (4) to rate their emotional response ‘‘Not

upset or mad at all’’ to ‘‘Very upset or mad.’’ Open-ended responses were recorded verbatim, and immedi-

ately coded as in previous studies (do nothing, ask why,

command the peer, threaten/seek adult punishment, ordirectly retaliate) (e.g., Zelli et al., 1999). Also consistent

with past studies, the last two responses were coded as

aggressive. We also coded relational aggression sepa-rately in order to examine possible gender differences. A

random subsample of responses was coded by a different

trained coder, and interrater agreement was assessed withkappa. Consistent with past studies, there was good

interrater agreement (k = .85). Scores from this measure

were used to index social information processing stagesof hostile attributions, aggressive response intentions

(including physical, relational, verbal, and covert

aggression), and prosocial response intentions (problemsolving) when provoked. Reliability ranged from fair to

good, with Cronbach’s alphas for these scales of .67, .95,

and .67.

Procedure

Youth in the detention facilities were in legal custody of thestate. Consent for youth was obtained from Division of

Children and Families caseworkers (legal guardians) prior to

notifying youth of their eligibility to participate. Eligibleyouth with signed consent were notified by the detention

director about periodic meetings held at the center by

research assistants (trained graduate students in clinicalpsychology) describing the project, and all eligible youth

received invitations to participate. Each participant had a

private meeting with project staff. Because incarceratedyouth are a vulnerable population, a representative from the

State Juvenile Defender’s Office was available on site or by

phone at all times to ensure that eligible youth had allquestions answered and could freely agree to or decline

participation. Youth participants, and classroom teachers and

detention staff were advised of the project’s voluntary nat-ure, goals, potential benefits, and compensation and invited

to participate. Participants reviewed and completed consent

or assent forms. Youth participants completed the protocoladministered at the detention facility by project staff.

The protocol consisted of a two-part same day assessment

with a 15-min break between. In order to eliminate biasrelated to differences in reading ability or understanding of

items, all measures were administered in a face-to-face

interview format using standardized electronic forms on alaptop computer. In addition, the interviewer provided the

participant with color-coded cue cards for each measure

indicating the response scale and word anchors for possibleresponses on that particular measure. The interviewer

entered participant responses on the computerized forms,

which generated a database for each measure. All adminis-trators, detention staff, and teachers met initially with the P.I.

and project staff, were provided with an overview of the

project procedures and timeframe before agreeing to par-ticipate. Participating teachers and detention staff met with

project staff at the detention center, received paper-and-

pencil measures, and project staff returned to collect com-pleted measures. Participant incentives of gift certificates

worth $10 to youth on completion of the protocol, and entries

in a drawing for 1 of 3 awards (worth $25–50) to teachers anddetention staff for each 25 completed youth protocols were

provided. The protocol was approved by both the University

and the State Human Services IRBs.

Data Analytic Approach

Few authors have devoted attention to statistical issues insimultaneous testing of multiple mediation (see e.g.,

MacKinnon, 2000; Preacher & Hayes, 2004). One barrier totesting multiple mediator models in psychological research

has been the lack of appropriate analytic tools outside of

520 Law Hum Behav (2009) 33:515–529

123

structural equation modeling (SEM) programs. Although

there are numerous advantages to SEM, its requirement oflarge samples to achieve valid estimates of statistical rela-

tionships has restricted all but the largest studies from

examining such hypotheses. Studies of adjudicated youth areparticularly sparse in the social cognition literature, and

samples of these youth are difficult to obtain. Recent

developments incorporating bootstrapping into path analyticapproaches has extended many of the advantages of SEM-

based approaches to path analysis with observed variables.The current study used available SAS programs to

examine multiple mediator path models with bootstrap

estimates of standard errors, providing unbiased signifi-cance tests in formal tests of mediation for each indirect

pathway (predictor ? mediator ? outcome) (see Preacher

& Hayes, 2004, 2008). Several advantages over simplemediation models using a causal steps approach (e.g., Baron

& Kenny, 1986) are incorporated into these contemporary

developments. The causal steps approach described byBaron and Kenny (1986), tests each path or bivariate rela-

tionship separately and then tests the predictor, mediator,

and dependent variable in a combined regression. Signifi-cant associations in these tests provide a basic indication of

potential mediation for a simple (single) mediator variable.

The causal steps approach, however, has been recognized asinadequate for complete understanding of complex media-

tion. It is limited to a single mediator and no significance test

can be computed for the complete indirect effect (product ofcoefficients) of a predictor variable operating through the

mediator to a dependent variable (e.g., Kenny, Kashy and

Bolger, 1998; Preacher & Hayes, 2004). In addition, there isa lack of agreement on the requirements for a direct rela-

tionship between each pair of variables (see e.g., Kenny

et al., 1998; Preacher & Hayes, 2004, 2008). Particularly indevelopmental research, data show that the predictor may

not exert a direct relationship on the dependent variable, but

may operate only through a mediator that is more proximalto the outcome (Dearing & Hamilton, 2006).

Most importantly for this study, mediation of risk factors

through on-line social information processing is concep-tualized as working through multiple mediating processes

that operate together. Models incorporating relationships

among these processes operating together could be testedwith the described contemporary approaches to bootstrap-

ping in multiple mediation analysis. Importantly, using this

analytic framework, the effects of each variable on everyother variable (each path, or statistical relationship) in our

integrative path models were tested statistically controlling

for all other variables in that model. Consequently, itwas possible to examine and compare the unique contri-

bution of each variable to the outcome: That is, we tested

the unique contribution of each proposed mediator (eachon-line social information processing stage), and each

substantive risk factor (beliefs, callous-unemotional traits,

I/CP), to the outcome variable (aggression). Thus, eachtested model was based on theoretically driven hypotheses

about these relations. In addition, bootstrapping of standard

errors incorporated into the program using 5000 samples ofthe bootstrap estimates provided unbiased estimates of

standard errors for the indirect effects. Thus, an unbiased,

formal hypothesis test of mediation for each indirect path(controlling for all other variables) was incorporated into

the analysis. Statistical power was sufficient for theseanalyses. With a sample of 150, statistical power to detect

small, indirect or mediated effects (d = .15), while also

considering a moderate direct effect (d = .30), was ade-quate at .85. Power to detect a small, direct effect (d = .15)

was high at .99.

Following these contemporary approaches and bestpractices recommendations in statistical modeling, two

multiple mediator path analyses examined models testing

(a) the associations among aggression beliefs, multiplestages of social information processing biases, and

aggression, and (b) the hypothesis that callous-unemotional

traits show unique effects on aggressive behavior, aboveand beyond the direct and indirect effects of beliefs, on-line

social information processing, and impulsivity (I/CP).

Missing Data

There was a very small amount of missing data on the

study variables (\!% of observations); no case had morethan one missing value and only one variable had a missing

value on more than one case (it had 2 missing values).

Missing values were a consequence of entry error (researchassistant not clicking the response entry on computerized

forms before moving to the next item), and would have

resulted in the loss of five cases (and statistical power) foranalysis. Data were missing at random (MAR), and thus

appropriate for handling with multiple imputation (Allison,

2002; Little & Rubin, 1989; Schafer & Graham, 2002).Multiple imputation (MI) provides unbiased, generalizable

estimates of missing values and standard errors and is the

most highly praised statistical approach for handlingmissing data (McKnight, McKnight, Sidani, & Figueredo,

2007; Rubin, 1987; Schafer & Graham, 2002). MI for

missing values was implemented using Schafer’s NORMprocedure (Schafer, 1997). Analyses using listwise deletion

of missing values and those without missing values fol-

lowing MI were indistinguishable.

RESULTS

Descriptive statistics for the major study variables are dis-

played in Table 1. As expected, there were no significant

Law Hum Behav (2009) 33:515–529 521

123

differences on demographic or study variables between

youth in the two centers (other than gender). Consistent with

past research, boys had a significantly younger age of onsetof antisocial behavior than girls and significantly higher

levels of callous-unemotional traits. Girls were significantly

more likely to generate aggressive responses to provoca-tions. This difference was significant both with and without

including relationally aggressive responses as a component

of aggressive responses generated (Table 2).The first path model examined the direct and indirect

effects of aggression beliefs mediated through social infor-

mation processing biases of hostile attributions, aggressiveresponse intentions, prosocial response intentions, and

outcome expectancy (illustrated in Fig. 1). In addition,

gender, a dichotomous variable indicating whether youth

had a childhood onset (by age 10) or adolescent onset (age11 or older) of antisocial behavior, and ethnicity were

included as control variables. In particular, age of onset was

considered important because the effects of I/CP and cal-lous-unemotional traits are potentially confounded with

childhood onset of conduct disorder and because we wanted

to account for any variance attributable to age of onset. Cellsizes were too small to reliably test moderating effects of

ethnicity. It was beyond the scope of this study to adequately

examine gender differences in callous-unemotional traitsand other antisocial risk factors.

Table 1 Descriptive statistics for full sample and by gender

Variable Sample (n = 150) Girls (n = 60) Boys (n = 90)

M SD M SD M SD

Age of onset of antisocial behavior 10.72 3.38 12.03* 2.77 9.84 3.47

Hostile attributions 4.65 2.48 5.18 2.61 4.30 2.35

Positive outcome expectancy 5.35 2.91 5.42 2.97 5.30 2.89

Aggressive response intentions 28.01 22.82 34.05* 26.80 23.99 18.83

Prosocial response intentions 23.81 6.44 23.05 6.69 24.31 6.26

Callous unemotional traits 46.50 6.04 44.70* 6.90 47.70 5.09

Aggressive behavior 25.10 10.41 25.95 10.97 24.53 10.04

Beliefs about aggression 21.87 12.55 21.87 11.60 21.88 13.21

Impulsive conduct problems 7.93 1.46 7.85 1.42 7.98 1.49

Age 15.21 1.40 15.33 1.35 15.12 1.44

Note: Hostile attributions = Hostile Attributional Bias subscale of the Attribution and Response to Provocation Scale (ARAPS; Dodge, 1980);Positive outcome expectancy = Tangible and social rewards subscales of Outcome expectancy questionnaire (Perry et al., 1986); Aggressiveresponse intentions = total aggressive response subscale of ARAPS; Prosocial response intentions = problem solving subscale of ARAPS; CUtraits = Total score on Inventory of Callous Unemotional Traits (Essau et al., 2006); Aggressive behavior = composite score of aggressionsubscale of Youth Self Report (Achenbach, 1991a) and Teacher Report Form (Achenbach, 1991b); Beliefs about Aggression = Response toprovocation subscale of Normative Beliefs about Aggression scale (Guerra et al., 1995) and response to relational provocation subscale ofRelational Aggression Measure; Impulsive conduct problems = I/CP subscale of the Antisocial Process Screening Device (Frick & Hare, 2001).* Significant difference by gender, p \ .05

Table 2 Correlations among study variables

1 2 3 4 5 6 7 8 9 10

1. Age of onset –

2. HAB .18*

3. Outcome expectancy .24* .04

4. Aggressive responses -.01 .57* .17*

5. Relational aggression responses .07 .59* .07 .87*

6. Prosocial responses -.12 -.50* -.24* -.51* -.46*

7. CU traits -.30* -.01 .08 .24* .16* -.16*

8. Aggressive behavior -.23 .04 -.10 .20* .07 -.17* .36*

9. Aggression beliefs .04 .28* .42* .51* .36* -.42* .34* .03

10. I/CP -.05 .07 .02 .09 .04 -.12 .31* .44* .21*

11. Age .40* -.13 .17* -.19* -.19* -.09 -.30* -.20 .03 -.01

Note: * p \ .05

522 Law Hum Behav (2009) 33:515–529

123

Figure 2 displays the associations among aggression

beliefs, social information processing stages, and aggres-

sion. Standardized path coefficients are provided for allstatistically significant paths (p \ .05). The results of this

model for the associations among aggression beliefs,

social information processing, and aggressive behavior areconsistent with past studies. In particular, all stages of

social information processing were significantly and

strongly related to aggression beliefs. In addition, a socialinformation processing bias for generating aggressive

responses (intent to respond aggressively) to provocation

and prosocial responses were significant mediators ofbeliefs in predicting aggression. These paths indicate that

more frequent intentions to act aggressively when youthperceived provocation were associated with higher levels

of aggression. More frequent intentions to problem solve

or act prosocially were associated with lower levels ofaggression.

The second path model examined the direct and indirect

effects of aggression beliefs mediated through socialinformation processing biases of hostile attributions,

aggressive response intentions, prosocial response inten-

tions, and outcome expectancy, and the direct effects ofcallous-unemotional traits and I/CP on aggression. Figure 3

displays the associations among this set of variables. Cal-

lous-unemotional traits and I/CP showed strong andsignificant associations with aggressive behavior and with

aggression beliefs. Also consistent with past studies testing

multiple stage social information processing models (e.g.,Zelli et al., 1999), not all stages of social information pro-

cessing showed associations with aggressive behavior. As

in the previous model and in past studies using a multi-variate approach, aggressive response intentions showed a

strong positive association with aggressive behavior, but

none of the other stages did. The magnitude of those

relationships was, however, similar to those reported in

previous studies (b = .09-.13). The overall model accoun-ted for 34% of the variance in aggressive behavior.

Callous-unemotional traits and I/CP showed strong

positive associations with aggressive behavior and thesetwo variables accounted for the majority of observed var-

iance in aggression. As predicted, callous-unemotional

traits showed unique associations with aggressive behavior,above and beyond the effects of beliefs, social information

processing, and I/CP. Compared to the first multiplemediator model we tested (without callous-unemotional

traits and I/CP), controlling age of onset and ethnicity, the

model depicted in Fig. 3 including callous-unemotionaltraits and I/CP, accounted for approximately 3 times the

variance in aggressive behavior (R2 = .34 vs. R2 = .12)

compared to multiple social information processing biasesmediating beliefs alone.

There is evidence that the aggression beliefs variable

acts as the suppressor variable in the second path model(Cohen & Cohen, 1983; Maassen & Bakker, 2001). Its

zero-order correlation with aggression is small and posi-

tive, and its direct effect controlling for other variables inthe models is negative. Examining the relationships among

beliefs and the other predictors in the model through

individual regression and partial correlation analyses,it became clear that the beliefs variable has a small

suppression effect on callous-unemotional traits. The

Aggression Beliefs

HAB

Aggressive response

bias

Prosocial response

bias

Outcome Expectancy

Aggression

.26*

.52*

-.40*

.45*

.14 (.03)

.26*

-.19*

Fig. 2 Path model of aggression beliefs mediated by social infor-mation processing associated with aggressive behavior. Note: Age ofonset of antisocial behavior, gender, and ethnicity were controlled.Standardized path coefficients, * statistically significant path control-ling all other paths, p \ .05. R2 = .12, F = 3.52, p \ .001

Aggression Beliefs

HAB

Aggressive response access

Prosocial response access

Outcome Expectancy

CU Traits

I/CP

Aggression

.28*

.48*

-.38*

.46*

.-.26* (.03)

.35*

.33*

.26*

.22*

.17*

Fig. 3 Path model of CU traits, impulsive conduct problems, andmediated social cognitive process associated with aggressive behav-ior. Note: Model estimated includes the effect of CU and I/CP on allmediators—arrows & estimates omitted for readability. Age of onsetof antisocial behavior, gender, and ethnicity were controlled.Standardized path coefficients, * statistically significant path control-ling all other paths, p \ .05. R2 = .34, F = 12.15, p \ .0001

Law Hum Behav (2009) 33:515–529 523

123

suppressor relationship follows the form of classical sup-

pression (Cohen & Cohen, 1983; Maassen & Bakker, 2001)and removes noise (irrelevant variance) that reduces the

magnitude of the relationship between callous-unemotional

traits and aggression. The beta for callous-unemotionaltraits in explaining aggression increases by .04 in a model

including beliefs, compared to a model without beliefs. The

squared semi-partial correlation for callous-unemotionaltraits, however, is essentially equivalent in the two models.

Partial correlation analyses showed a similar pattern. Thesuppressor effect is small, but it does indicate that the

direct effect of beliefs about aggression on ratings of

aggressive behavior should be considered non-significant(rather than negative). In the mediation model without

callous-unemotional traits noted above, both the total effect

and the direct effect of beliefs on aggression were small,and there was not a significant negative association with

aggressive behavior. The association between callous-

unemotional traits and aggressive behavior increases whencontrolling for beliefs, and beliefs has a non-significant

correlation with aggressive behavior with and with-

out controlling callous-unemotional traits. Although thispattern suggests that the correlation between callous-

unemotional traits and beliefs suppresses the association

between callous-unemotional traits and aggressive behav-ior, the effect on the callous-unemotional traits and

aggression path is small. The small effect does not change

the overall interpretation of the callous-unemotional traits–aggression relationship (Maassen & Bakker, 2001).

As with any theoretical model, alternative models may

account for the relations among variables. In order toexamine a likely alternative model suggested by Dodge and

Pettit (2003), we tested whether the relation between

callous-unemotional traits and aggressive behavior ismediated by social information processing. There were no

significant mediated relationships. It is also worth noting

that the covariances between beliefs and callous-unemo-tional traits, and beliefs and I/CP are mathematically

equivalent to mediational paths. There is, however, no

a priori theoretical basis for construing these two dimen-sions as mediators between beliefs and aggression.

DISCUSSION

Our primary aim in this study was to examine how multiplerisk factors, found separately to predict antisocial behavior,

are jointly and uniquely associated with antisocial out-

comes when considered together. To achieve this aim, weused path analyses examining the direct and mediating

effects of multiple risk factors on concurrent levels of

aggression. Specifically, we examined (a) the direct andindirect effects of social cognitive risk factors (online

social information processing biases, aggression beliefs) on

aggression, and (b) the direct and indirect effects of anemotional overcontrol risk factor (callous-unemotional

traits) and an emotional undercontrol risk factor (impul-

sivity), on aggression in a sample of adjudicatedadolescents. Although previous work has demonstrated that

all of these risk factors are concurrently and prospectively

associated with antisocial outcomes, it has been unclear theextent to which they are uniquely associated with these

outcomes.We drew on best practices and contemporary recom-

mendations for testing multiple mediation (Dearing &

Hamilton, 2006; Preacher & Hayes, 2004, 2008). The resultsproduced comparative models in which callous-unemo-

tional traits demonstrated unique, strong, and significant

associations with aggressive behavior. Furthermore, theseeffects for callous-unemotional traits were associated with

aggression above and beyond the effects of impulsivity,

aggression beliefs, multiple online stages of social infor-mation processing, and age of onset of antisocial behavior,

which are more established predictors of antisocial out-

comes. This overall set of findings supports our contentionthat callous-unemotional traits have unique and important

effects on antisocial outcomes such as aggression severity.

Strengthening the assertion that callous-unemotional traitsare uniquely associated with more severe aggression, online

social information processing biases did not mediate the

effects of callous-unemotional traits on aggression. Thus,those results did not support the hypothesis that online social

information processing biases are proximal mediators of allpredictors of antisocial outcomes (Dodge & Pettit, 2003). Inaddition, there were no significant indirect effects of cal-

lous-unemotional traits or I/CP on aggression operating

through online social information processing stages. Inmediation models testing Dodge and Pettit’s (2003)

hypothesis, callous-unemotional traits and I/CP had only

significant direct effects on aggressive behavior. Impulsivityas indexed by I/CP also demonstrated unique and strong

associations with social cognition and aggression. It is

especially striking that compared with a model examiningthe mediating effects of social information processing on

aggression beliefs, the path model also including callous-

unemotional traits and I/CP suggested that these two vari-ables were strongly associated with aggressive behavior.

These findings underscore the importance of the primary

conceptual basis for this article; integrative models com-bining multiple risk processes are necessary to understand

how risk factors work together to produce antisocial

outcomes.Moreover, these findings provide evidence consistent with

other studies indicating differing underlying dispositional

vulnerabilities for antisocial outcomes that may appear notonly as emotional undercontrol and dysregulation, but also as

524 Law Hum Behav (2009) 33:515–529

123

emotional overcontrol. These results are also consistent with a

growing literature that suggests not only equivalent vulnera-bility from emotional overcontrol, but a significantly

heightened vulnerability to severe and persistent antisocial

behavior, as compared to youth with emotional undercontrol(Flight & Forth, 2007; Frick & Morris, 2004; Frick et al., 2005;

Lynam, Caspi, Moffitt, Loeber, & Southamer-Loeber, 2007;

Salekin, 2008).There are several additional considerations suggested by

these findings, in addition to underscoring Dodge and Pet-tit’s (2003) emphasis on the importance of examining

integrative models of multiple risk factors associated with

antisocial outcomes. The pattern of results from this studyindicates that rather than each important risk factor operat-

ing through the same mechanism or pathway, there are

likely multiple mechanisms influencing antisocial acts. Thisidea is not new, but these findings add to other evidence

supporting the importance of considering equifinality, or

multiple pathways leading to the same outcome, in devel-opmental psychopathology research. Specifically, there is

considerable evidence that social information processing

biases associated with emotional undercontrol are one vul-nerability mechanism that is proximal to aggression in social

situations. The findings of this study, however, suggest that

emotional overcontrol associated with high levels of cal-lous-unemotional traits also exerts a proximal influence on

aggression. The results not only replicate findings showing

that high levels of callous-unemotional traits are associatedwith severe aggression, but extend that research by sug-

gesting that callous-unemotional traits confer unique risk for

this severity, above and beyond social information pro-cessing biases.

The results of this study have implications for social

information processing theory and for intervention withantisocial and aggressive youth. At a minimum, the results

suggest that social information processing models are

likely to benefit from incorporation of emotion processes.Whether emotion vulnerabilities related to undercontrol

and overcontrol operate primarily in parallel or interact

with cognitive vulnerabilities exhibited through socialinformation processing biases is somewhat unclear. Future

studies seeking clearer delineation of these associations

would benefit from incorporation of multiple emotionprocesses such as emotion regulation and callous-unemo-

tional traits. It would be particularly useful for such studies

to examine interactions among these processes to begindisentangling the extent to which emotion vulnerability and

cognitive vulnerability affect each other. In addition, how

these associations unfold and change across developmentaltrajectories of antisocial behavior can usefully inform

prevention and intervention strategies.

These results can also inform intervention that is spe-cifically focused on callous-unemotional traits and

psychopathy. It is worth noting that emerging evidence

supports the view that effective intervention for antisocialand aggressive youth with callous-unemotional traits is

possible. First, although youth with callous-unemotional

traits tend to have poorer outcomes than other conductproblem youth from many interventions, the emerging

picture is cause for guarded optimism. One program using

intensive residential treatment has demonstrated signifi-cantly lower aggressive and violent recidivism among

adolescent offenders with high levels of callous-unemo-tional traits (and other psychopathic features; Caldwell,

Skeem, Salekin, & Van Rybroek, 2006). Moreover, this

intensive residential treatment program provides a sub-stantial cost-benefit advantage over usual treatment within

the juvenile justice system (Caldwell, Vitacco, & Van

Rybroek, 2006). Adding to these findings, Hawes andDadds (2005) report that although youth with callous-

unemotional traits had poorer overall outcomes following

behavioral parent training, a subset of youth who exhibitedhigh levels of callous-unemotional traits at pre-treatment

showed both behavioral improvements and significantly

lower levels of callous-unemotional traits at post-treatmentand follow-up.

In the Hawes and Dadds (2005) study, reductions in

callous-unemotional traits and reductions in behavioralproblems (i.e., I/CP), were independent. That is, callous-

unemotional traits were not simply a marker of more severe

conduct problems, and improvements in one domain didnot depend on improvements in the other. Mechanisms

associated with these changes are unknown at this time.

Extending these findings, results from this study alsoindicate that associations among callous-unemotional traits

and aggression are independent of other risk processes

associated with aggression. That is, results from our studysuggest that aggression is independently associated with

social information processing biases, with I/CP, and with

callous-unemotional traits. Taken together, the evidenceemerging for antisocial youth with callous-unemotional

traits suggests that the best available approaches may

benefit from individualized treatment plans tailored tomodify specific risk processes.

Traditional evidence-based treatments apply a single

model for all persons with a particular disorder. However,these differing risk processes affecting aggressive behavior

indicate that applying single intervention packages based

on treatment outcome research is likely to continue to haveinconsistent results. Rather, treatments such as intensive

residential treatment, Multisystemic therapy (MST), and

Families and Schools Together (FAST Track) that aresuccessful in treating some aggressive youth, are likely to

benefit from identification of levels of callous-unemotional

traits (in addition to impulsivity and social informationprocessing biases), and from incorporating strategies aimed

Law Hum Behav (2009) 33:515–529 525

123

at the affective deficits associated with callous-unemo-

tional traits. In particular, a decreased emphasis onpunishment and negative consequences, shown to be inef-

fective for youth with high levels of callous-unemotional

traits, and an increased emphasis on empathy training andrewarding prosocial behavior may help to modify callous-

unemotional traits and to reduce antisocial acts (Stickle &

Frick, 2002; Wong & Hare, 2005).Although emerging evidence is promising and counters

views that the presence of callous-unemotional traits andother psychopathic features suggests treatment will neces-

sarily be unsuccessful, there are numerous challenges to

implementing individually tailored intervention approa-ches. Incorporating evidence-based assessment of risk

processes is a necessary first step. Even if efforts to dis-

seminate implementation of such assessments aresuccessful, cost-effective and efficient approaches to syn-

thesizing and incorporating information derived from such

assessments are not yet available. Moreover, the order inwhich treatments should target multiple risk processes that

may operate somewhat independently is unknown. Thus,

evidence-based, cost-effective, comprehensive, and indi-vidualized interventions need to be increasingly and

systematically studied (Henggeler, Schoenwald, Bordoin,

Rowland, & Cunningham, 1998). There are also numerouschallenges in transporting comprehensive treatment

approaches to community settings, where they appear most

effective. Despite these challenges, it is clear that under-standing and targeting multiple risk and vulnerability

factors that are associated with different developmental

trajectories to antisocial and aggressive behavior should bethe future focus of intervention for these problems.

Several limitations should be considered when inter-

preting the results of this study. Although we purposivelysampled to achieve tests at an extreme end of the range of

antisocial behavior, sampling in this way does present

some limits to the generality of the findings. It also isapparent that the girls in this sample are particularly severe

and observed gender differences would benefit from rep-

lication in a community sample. Some of the measurementstrategies might be strengthened in future studies. The age

of onset of antisocial behavior variable was based on youth

self-report. It is unclear how accurate their reporting is. It isstriking, though, that the reports of early onset problems

and delayed onset problems correspond closely with data

using prospective methods, lending some support to thevalidity of the self-report. Although we used multiple

informant measures, we used path analysis with index

scores of measured variables rather than SEM with latentvariables. The use of multiple measures and multiple

informants increases reliability and stability of estimates.

Nevertheless, use of SEM with latent variables in a largersample would provide more stable estimates of the

relations observed here by reducing effects of measurement

error.Although we followed the recommendations of pub-

lished manuals (Frick & Hare, 2001) and best practices in

methodology (Piacentini et al., 1992) for scoring ourmultiple informant measures, some of the associations

among callous-unemotional traits, I/CP, and aggressive

behavior may be attributable to method variance. It isworth noting that these three scales measure different

aspects of antisocial development and that the item contentdoes not overlap. Moreover, callous-unemotional traits and

I/CP showed similar magnitudes of association with both

aggression beliefs and social information processing stagesas they did with aggression. The social cognition variables

are distinct constructs and were not scored in the same

fashion as callous-unemotional traits and I/CP. Thus,although there may be some method variance, any method

effect appears to be small.

The study was cross-sectional and the associationsamong variables observed here should be considered con-

current, or perhaps in some cases, postdictive. These

associations may also differ across time. The possibilitythat callous-unemotional traits and I/CP act as mediators of

other risk factors for aggression such as beliefs, in partic-

ular, needs to be examined with longitudinal data to morerigorously test this possibility. Some of the social infor-

mation processing scales had Cronbach’s Alpha estimates

just below .70, exhibiting somewhat low internal consis-tency in this sample. It is worth noting, however, that

prosocial responses, which showed among the lowest

alphas, did demonstrate one of the strongest associationswith aggression. The extent to which this low reliability

may have attenuated relations with aggression is unknown,

but it is plausible that it may have affected results. Nev-ertheless, the overall pattern and magnitude of associations

among social cognition and aggression, and among callous-

unemotional traits, I/CP, and aggression in this study areconsistent with those reported in prospective, longitudinal

studies (e.g., Frick et al., 2005; Zelli et al., 1999).

It should also be noted that reported analyses used theI/CP scale of the APSD alone, rather than as part of the

APSD total score. There is limited research on using this

subscale in isolation rather than as an aspect of a totalpsychopathy score. The I/CP scale was used because there

is increasing evidence that callous-unemotional traits and

various other dimensions of antisocial behavior (criminal-ity, impulsivity) have additive effects on antisocial

outcomes. That is, the combination of high levels of both I/

CP and callous-unemotional traits appears most related tosevere and persistent antisocial behavior (e.g., Frick et al.,

2005; Salekin, 2008). Moreover, to better inform both

theory of antisocial development and intervention devel-opment, clearer understanding of the extent to which

526 Law Hum Behav (2009) 33:515–529

123

outcomes are independently related to these risk dimen-

sions is important. To the extent that outcomes aredifferentially related to these dimensions, interventions can

be individually tailored to target the underlying vulnera-

bilities of emotional and cognitive deficits.Finally, the path model including all the risk processes

accounted for 34% of the variance in aggression. This is a

large effect size (d = 1.4; Cohen, 1988). Despite thiseffect, it should be noted that two-thirds of the variance in

aggression was not associated with callous-unemotionaltraits, I/CP, and social cognition. The practical and clinical

implications of lacking an explanation for the other 2/3 of

the variance are important and underscore the fact thatantisocial outcomes such as aggression are the result of a

complex set of interacting risk factors. Thus, despite

improvement over models examining single risk domains,the current results also indicate that a more complete

understanding of these problems will require incorporation

of other important risk factors such as genetic vulnerability,and gene 9 environment interactions, among others.

Despite limitations, the current results add to a growing

number of findings emphasizing the importance of callous-unemotional traits in understanding aggression and anti-

social behavior in youth. Moreover, future research on the

development and expression of antisocial behavior shouldemphasize multiple mediating and moderating processes in

producing antisocial outcomes. Past work provides exten-

sive evidence for many risk factors in antisocialdevelopment. How these risk factors and processes affect

each other and youth with heterogeneous risk factors

should have high priority in future research.In conclusion, individual differences in callous-unemo-

tional traits and I/CP accounted for the majority of the

observed variance in aggressive behavior after accountingfor the more established risk factors of aggression beliefs,

online social information processing biases, and age of

onset of antisocial behavior. Callous-unemotional traitshave been demonstrated to be consistently associated with

increased stability and severity of antisocial behavior and

aggression in both children and adolescents (Dadds, Fraser,Frost, & Hawes, 2005; Frick et al., 2005; Lynam et al.,

2007; Salekin, 2008). These results support previous find-

ings indicating that the combination of high levels ofcallous-unemotional traits and I/CP is associated with

especially severe antisocial behavior (Frick et al., 2005;

Lynam et al., 2007; Salekin, 2008), that the presence ofcallous-unemotional traits confers unique risk for severe

aggression in youth, and provides incremental concurrent

and postdictive value above and beyond other risk factorsin understanding antisocial outcomes.

Acknowledgments This research was funded in part by a grantfrom Child and Adolescent Psychology Training and Research

Foundation, by financial support from University of Vermont Dean’sFaculty Fund, and Linda Brittain, awarded to the first author. Theauthors thank Judith Christensen and detention center staff andteachers for their generous support and their time, and Paul Frick, RexForehand, Annie Murray-Close, and Heather Bouchey for helpfulcomments on an earlier draft of this article.

REFERENCES

Achenbach, T. M. (1991a). Manual for the Youth Self-Report and1991 profile. Burlington, VT: University of Vermont Departmentof Psychiatry.

Achenbach, T. M. (1991b). Manual for the Teacher Report Form and1991 profile. Burlington, VT: University of Vermont Departmentof Psychiatry.

Achenbach, T. M. (1991c). Manual for the Child Behavior Checklist/4–18 and 1991 profile. Burlington, VT: University of VermontDepartment of Psychiatry.

Allison, P. D. (2002). Missing data. Thousand Oaks, CA: Sage.Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator

variable distinction in social psychological research. Journal ofPersonality and Social Psychology, 51, 1173–1182. doi:10.1037/0022-3514.51.6.1173.

Barry, T. D., Barry, C. T., Deming, A. M., & Lochman, J. E. (2008).Stability of psychopathic characteristics in childhood: Theinfluence of social relationships. Criminal Justice and Behavior,35(2), 244–262.

Blair, R. J. R. (1999). Responsiveness to distress cues in the childwith psychopathic tendencies. Personality and Individual Dif-ferences, 27, 135–145. doi:10.1016/S0191-8869(98)00231-1.

Burke, J. D., Loeber, R., & Lahey, B. B. (2007). Adolescent conductdisorder and interpersonal callousness as predictors of psychop-athy in young adults. Journal of Clinical Child and AdolescentPsychology, 36, 334–346.

Caldwell, M., Skeem, J., Salekin, R., & Van Rybroek, G. (2006).Treatment response of violent offenders with psychopathyfeatures: A 2-year follow-up. Criminal Justice and Behavior,33(5), 571–596.

Caldwell, M. F., Vitacco, M., & Van Rybroek, G. J. (2006). Areviolent delinquents worth Treating? A cost-benefit analysis.Journal of Research in Crime and Delinquency, 43, 148–168.doi:10.1177/0022427805280053.

Caputo, A. A., Frick, P. J., & Brodsky, S. L. (1999). Family violenceand juvenile sex offending: Potential mediating role of psycho-pathic traits and negative attitudes towards women. CriminalJustice and Behavior, 26, 338–356. doi:10.1177/0093854899026003004.

Christian, R. E., Frick, P. J., Hill, N. L., Tyler, L., & Frazer, D. R.(1997). Psychopathy and conduct problems in children: II.Implications for subtyping children with conduct problems.Journal of the American Academy of Child and AdolescentPsychiatry, 36, 233–243. doi:10.1097/00004583-199702000-00014.

Cleckley, H. (1976). The mask of sanity. St. Louis: Mosby.Cohen, J. (1988). Statistical power analysis for the behavioral

sciences (2nd ed.). Mahwah, NJ: Erlbaum.Cohen, J., & Cohen, P. (1983). Applied multiple regression/correla-

tion analysis for the behavioral sciences (2nd ed.). Mahwah, NJ:Erlbaum.

Crick, N. R. (1995). Relational aggression: The role of intentattributions, feelings of distress, and provocation type. Develop-ment and Psychopathology, 7, 313–322.

Crick, N. R., & Dodge, K. A. (1994). A review of reformulation of socialinformation-processing mechanisms in children’s social adjustment.

Law Hum Behav (2009) 33:515–529 527

123

Psychological Bulletin, 115, 74–101. doi:10.1037/0033-2909.115.1.74.

Crick, N. R., Grotpeter, J. K., & Bigbee, M. A. (2002). Relationallyand physically aggressive children’s intent attributions andfeelings of distress for relational and instrumental peer provo-cations. Child Development, 73, 1134–1142. doi:10.1111/1467-8624.00462.

Dadds, M. R., Fraser, J., Frost, A., & Hawes, D. (2005). Disentanglingthe underlying dimensions of psychopathy and conduct problemsin childhood: A community study. Journal of Consulting andClinical Psychology, 73, 400–410. doi:10.1037/0022-006X.73.3.400.

Dadds, M. R., Whiting, C., & Hawes, D. J. (2006). Associationsamong cruelty to animals, family conflict, and psychopathictraits in childhood. Journal of Interpersonal Violence, 21(3),411–429.

Dearing, E., & Hamilton, L. C. (2006). Contemporary advances andclassic advice for analyzing mediating and moderating variables.Monographs of the Society for Research in Child Development,71, 88–104.

Dodge, K. A. (1980). Social cognition and children’s aggressivebehavior. Child Development, 51, 162–170. doi:10.2307/1129603.

Dodge, K. A., & Pettit, G. S. (2003). A biopsychosocial model of thedevelopment of chronic conduct problems in adolescence. Devel-opmental Psychology, 39, 349–371. doi:10.1037/0012-1649.39.2.349.

Dubow, E. F., & Reid, G. J. (1994). Risk and resource variables inchildren’s aggressive behavior: A two-year longitudinal study. InL. R. Huesmann (Ed.), Aggressive behavior: Current perspec-tives (pp. 187–214). New York: Plenum.

Erdley, C., & Asher, S. R. (1998). Children’s beliefs about thelegitimacy of aggression and their behavior. Social Development,7, 321–339. doi:10.1111/1467-9507.00070.

Essau, C. A., Sasagawa, S., & Frick, P. J. (2006). Callous-unemotional traits in a community sample of adolescents.Assessment, 13, 464–469. doi:10.1177/1073191106287354.

Eisenberg, N., Fabes, R. A., Guthrie, I. K., & Reiser, M. (2000).Dispositional emotionality and regulation: Their role in predict-ing quality of social functioning. Journal of Personality andSocial Psychology, 78, 136–157. doi:10.1037/0022-3514.78.1.136.

Flight, J. I., & Forth, A. E. (2007). Instrumentally violent youths: Theroles of psychopathic traits, empathy, and attachment. CriminalJustice and Behavior, 34, 739–751. doi:10.1177/0093854807299462.

Frick, P. J., Bodin, S. D., & Barry, C. T. (2000). Psychopathic traits andconduct problems in community and clinic-referred samples ofchildren: Further development of the Psychopathy ScreeningDevice. Psychological Assessment, 12, 382–393. doi:10.1037/1040-3590.12.4.382.

Frick, P. J., Cornell, A. H., Barry, C. T., Bodin, S. D., & Dane, H. A.(2003). Callous-unemotional traits and conduct problems in theprediction of conduct problem severity, aggression, and self-report of delinquency. Journal of Abnormal Child Psychology,31, 457–470. doi:10.1023/A:1023899703866.

Frick, P. J., & Dickens, C. (2006). Current perspectives on conductdisorder. Current Psychiatry Reports, 8, 59–72.

Frick, P. J., & Hare, R. D. (2001). The antisocial process screeningdevice. Toronto: Multi-Health Systems.

Frick, P. J., & Morris, A. S. (2004). Temperament and developmentalpathways to conduct problems. Journal of Clinical Child andAdolescent Psychology, 33, 54–68. doi:10.1207/S15374424JCCP3301_6.

Frick, P. J., O’Brien, B. S., Wootton, J. M., & McBurnett, K. (1994).Psychopathy and conduct problems in children. Journal of

Abnormal Psychology, 103, 700–707. doi:10.1037/0021-843X.103.4.700.

Frick, P. J., Stickle, T. R., Dandreaux, D. M., Farrell, J. M., & Kimonis,E. R. (2005). Callous-unemotional traits in predicting the severityand stability of conduct problems and delinquency. Journal ofAbnormal Child Psychology, 33, 471–487. doi:10.1007/s10648-005-5728-9.

Frick, P. J., & White, S. F. (2008). Research review: The importanceof callous-unemotional traits for developmental models ofaggressive and antisocial behavior. The Journal of ChildPsychology and Psychiatry, 49(4), 359–375.

Giancola, P. R., Mezzich, A. C., & Tarter, R. E. (1998). Executivecognitive functioning, temperament, and antisocial behavior inconduct disordered adolescent females. Journal of AbnormalPsychology, 107, 629–641. doi:10.1037/0021-843X.107.4.629.

Glenn, A. L., Raine, A., Venables, P. H., & Mednick, S. A. (2007).Early temperamental and psychophysiological precursors ofadult psychopathic personality. Journal of Abnormal Psychol-ogy, 116(3), 508–518.

Guerra, N. G., Huesmann, L. R., Tolan, P. H., Van Acker, R., & Eron,L. D. (1995). Stressful events and individual beliefs as correlatesof economic disadvantage and aggression among urban children.Journal of Consulting and Clinical Psychology, 63, 518–528.doi:10.1037/0022-006X.63.4.518.

Hall, J. A., & Skrowronski, K. J. (1998). Outcome expectancies andoutcome values as predictors of children’s aggression. AggressiveBehavior, 24, 439–454. doi :10.1002/(SICI)1098-2337(1998)24:6\439::AID-AB4[3.0.CO;2-Q.

Hare, R. D. (1998). Psychopaths and their nature: Implications for themental health and criminal justice systems. In T. Millon, E.Simonson, M. Burket-Smith, & R. Davis (Eds.), Psychopathy:Antisocial, criminal, and violent behavior (pp. 188–212). NewYork, NY: Guilford Press.

Hawes, D. J., & Dadds, M. R. (2005). The treatment of conductproblems in children with callous-unemotional traits. Journal ofConsulting and Clinical Psychology, 73, 737–741. doi:10.1037/0022-006X.73.4.737.

Hemphill, J. F. (2007). The Hare Psychopathy Checklist and recidi-vism: Methodological issues and critically evaluating empiricalevidence. In H. Herve & J. C. Yule (Eds.), The psychopath: Theoryresearch, and practice (pp. 141–170). Mahwah, NJ: Earlbaum.

Henggeler, S. W., Schoenwald, S. K., Bordoin, C. M., Rowland, M. D.,& Cunningham, P. B. (1998). Multisystemic treatment of antiso-cial behavior in children and adolescents. New York: Guilford.

Huesmann, L. R. (1988). An information processing model for thedevelopment of aggression. Aggressive Behavior, 14, 13–24.doi :10.1002/1098-2337(1988)14:1\13::AID-AB2480140104[3.0.CO;2-J.

Huesmann, L. R. (1998). The role of social information processingand cognitive schema in the acquisition and maintenance ofhabitual aggressive behavior. In E. Donnerstein (Ed.), Humanaggression: Theories, research, and implications for socialpolicy (pp. 73–109). San Diego, CA: Academic Press, Inc.

Huesmann, L. R., & Guerra, N. G. (1997). Children’s normativebeliefs about aggression and aggressive behavior. Journal ofPersonality and Social Psychology, 72, 408–419. doi:10.1037/0022-3514.72.2.408.

Kagan, J., & Snidman, N. (1991). Temperamental factors in humandevelopment. The American Psychologist, 46, 856–862. doi:10.1037/0003-066X.46.8.856.

Kazdin, A. E., & Weisz, J. R. (1998). Identifying and developingempirically supported child and adolescent treatments. Journalof Consulting and Clinical Psychology, 66, 19–36. doi:10.1037/0022-006X.66.1.19.

Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis insocial psychology. In D. Gilbert & S. F. G. Lindzey (Eds.), The

528 Law Hum Behav (2009) 33:515–529

123

handbook of social psychology (pp. 233–265). Boston, MA:McGraw-Hill.

Kochanska, G. (1993). Toward a synthesis of parental socializationand child temperament in early development of conscience.Child Development, 66, 325–347. doi:10.2307/1131254.

Little, R., & Rubin, D. B. (1989). The analysis of social science datawith missing values. Sociological Methods & Research, 18, 292–326. doi:10.1177/0049124189018002004.

Lynam, D. R., Caspi, A., Moffitt, T. E., Loeber, R., & Southamer-Loeber, M. (2007). Longitudinal evidence that psychopathyscores in early adolescence predict adult psychopathy. Journal ofAbnormal Psychology, 106, 155–165. doi:10.1037/0021-843X.116.1.155.

Lynam, D. R., Loeber, R., & Stouthamer-Loeber, M. (2008). Thestability of psychopathy from adolescence into adulthood: Thesearch for moderators. Criminal Justice and Behavior, 35, 228–243. doi:10.1177/0093854807310153.

Maassen, G. H., & Bakker, A. B. (2001). Suppressor variables in pathmodels. Sociological Methods & Research, 30, 241–270. doi:10.1177/0049124101030002004.

MacKinnon, D. P. (2000). Contrasts in multiple mediator models. In J.Rose, L. Chassin, C. C. Presson, & S. J. Sherman (Eds.),Multivariate applications in substance use research: New methodsfor new questions (pp. 141–160). Mahwah, NJ: Erlbaum.

McKnight, P. E., McKnight, K. M., Sidani, S., & Figueredo, A. J.(2007). Missing data: A gentle introduction. New York: Guilford.

Moffitt, T. E. (1993). Adolescent-limited and life-course-persistentantisocial behavior: A developmental taxonomy. PsychologicalReview, 100, 674–701. doi:10.1037/0033-295X.100.4.674.

Mullins-Nelson, J. L., Salekin, R. T., & Leistico, A. R. (2006).Psychopathy, empathy, and perspective-taking ability in acommunity sample: Implications for the successful psychopathyconcept. International Journal of Forensic Mental Health, 5,133–149.

Pardini, D. A., Lochman, J. E., & Frick, P. J. (2003). Callous/unemotional traits and social-cognitive processes in adjudicatedyouths. Journal of the American Academy of Child andAdolescent Psychiatry, 42, 364–371. doi:10.1097/00004583-200303000-00018.

Perry, D. G., Perry, L. C., & Rasmussen, P. (1986). Cognitive sociallearning mediators of aggression. Child Development, 57, 700–711. doi:10.2307/1130347.

Piacentini, J. C., Cohen, P., & Cohen, J. (1992). Combiningdiscrepant diagnostic information from multiple sources: Arecomplex algorithms better than simple ones? Journal of Abnor-mal Child Psychology, 20, 51–63. doi:10.1007/BF00927116.

Porter, S., & Woodworth, M. (2006). Psychopathy and aggression. InC. J. Patrick (Ed.), Handbook of psychopathy (pp. 481–496).New York: Guilford Press.

Preacher, K. J., & Hayes, A. F. (2004). SPSS and SAS procedures forestimating indirect effects in simple mediation models. BehaviorResearch Methods, Instruments, & Computers, 36, 717–731.

Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resamplingstrategies for assessing and comparing indirect effects in simpleand multiple mediator models. Behavior Research Methods, 40,879–891. doi:10.3758/BRM.40.3.879.

Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys.New York: Wiley.

Salekin, R. T. (2006). Psychopathy in children and adolescents: Keyissues in conceptualization and assessment. In C. J. Patrick (Ed.),Handbook of psychopathy (pp. 389–414). New York: Guilford.

Salekin, R. T. (2008). Psychopathy and recidivism from mid-adolescence to young adulthood: Cumulating legal problemsand limiting life opportunities. Journal of Abnormal Psychology,117, 386–395. doi:10.1037/0021-843X.117.2.386.

Salekin, R. T., Neumann, C. S., Leistico, A. R., DiCicco, T. M., &Duros, R. L. (2004). Psychopathy and comorbidity in a youngoffender sample: Taking a closer look at psychopathy’s potentialimportance over disruptive behavior disorders. Journal ofAbnormal Psychology, 113, 416–427. doi:10.1037/0021-843X.113.3.416.

Salekin, R. T., Rogers, R., & Sewell, K. W. (1996). A review andmeta-analysis of the Psychopathy Checklist-Revised: Predictivevalidity of dangerousness. Clinical Psychology: Science andPractice, 3, 203–213.

Salekin, R. T., Rosenbaum, J., & Lee, Z. (2008). Child and adolescentpsychopathy: Stability and change. Psychiatry, Psychology andLaw, 15, 224–236. doi:10.1080/13218710802014519.

Schafer, J. L. (1997). Analysis of incomplete multivariate data.London: Chapman & Hall.

Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of thestate of the art. Psychological Methods, 7, 147–177. doi:10.1037/1082-989X.7.2.147.

Skeem, J. L., & Cauffman, E. (2003). Views of the downwardextension: Comparing the youth version of the psychopathychecklist with the youth psychopathic traits inventory. Behav-ioral Sciences and the Law, 21, 737–770.

Stickle, T. R., & Frick, P. J. (2002). Developmental pathways tosevere antisocial behavior: Interventions for youth with callous-unemotional traits. Expert Review of Neurotherapeutics, 2, 511–522. doi:10.1586/14737175.2.4.511.

Wilkinson, G. S. (1993). Wide range achievement test—Revision 3.Wilmington, DE: Jastak Association.

Wong, S., & Hare, R. D. (2005). Guidelines for psychopathytreatment program. Toronto, ON: Multi-Health Systems.

Zelli, A., Dodge, K. A., Lochman, J. E., Laird, R. D., & ConductProblems Prevention Research Group. (1999). The distinctionbetween beliefs legitimizing aggression and deviant processingof social cues: Testing measurement validity and the hypothesisthat biased processing mediates the effects of beliefs onaggression. Journal of Personality and Social Psychology, 77,150–166. doi:10.1037/0022-3514.77.1.150.

Law Hum Behav (2009) 33:515–529 529

123