Journal of Abnormal Child Psychology, Vol. 26, No. 6, 1998, pp. 431-440

Social-Cognitive and Behavioral Correlates ofAggression and Victimization in Boys9 Play Groups

David Schwartz,1,7 Kenneth A. Dodge,2 John D. Coie,3 Julie A. Hubbard,4 Antonius H.N. Cillessen,5 Elizabeth A. Lemerise,6 and Helen Bateman7

Received November 21, 1997; revision received April 21, 1998; accepted April 21, 1998

A contrived play group procedure was utilized to examine the behavioral and social-cognitivecorrelates of reactive aggression, proactive aggression, and victimization via peers. Elevenplay groups, each of which consisted of six familiar African-American 8-year-old boys, metfor 45-min sessions on five consecutive days. Social-cognitive interviews were conducted fol-lowing the second and fourth sessions. Play group interactions were videotaped and examinedby trained observers. High rates of proactive aggression were associated with positive out-come expectancies for aggression/assertion, frequent displays of assertive social behavior,and low rates of submissive behavior. Reactive aggression was associated with hostile at-tributional tendencies and frequent victimization by peers. Victimization was associated withsubmissive behavior, hostile attributional bias, reactive aggression, and negative outcome ex-pectations for aggression/assertion. These results demonstrate that there is a theoreticallycoherent and empirically distinct set of correlates associated with each of the examined ag-gression subtypes, and with victimization by peers.

KEY WORDS: Bullying; aggression; social cognition.

Our understanding of the social and psychologi-cal processes underlying aggression and victimizationin children's peer groups has been greatly enhancedby investigations conducted using contrived playgroup methodologies. In such procedures, smallgroups of children are brought together on a regularbasis over a short period of time and allowed to in-teract in a structured play setting (e.g., Coie & Ku-persmidt, 1983; Dodge, 1983). A high degree ofcontrol over the social context is maintained by theresearchers, and the children's interactions can be re-corded with hidden audiovisual equipment. Vide-

1University of Southern California, Los Angeles, California 90089.2Duke University, Durham, North Carolina 27708.3Duke University, Durham, North Carolina 27706.4University of Delaware, Newark, Delaware 19716.5University of Connecticut, Storrs, Connecticut 06268.6Western Kentucky University, Bowling Green, Kentucky 42101.7Address all correspondence regarding this article to D. Schwartz,Psychology Department, University of Southern California, SGM501, Los Angeles, California 90089.

otaped records can be examined at a later point, fa-cilitating detailed analysis of a relatively large num-ber of aggressive episodes, even though aggression isa low frequency behavior that occurs in limited socialcontexts (Coie & Dodge, in press; Parke & Slaby,1983).

In the current study, a contrived playgroup ap-proach was utilized to examine the social-cognitiveand behavioral correlates of aggression and victimi-zation in boys' peer groups. We were particularly fo-cused on the correlates of peer group victimizationbecause children who are maltreated by their peersmay be at risk for significant psychological and be-havioral difficulties (i.e., Boivin, Hymel, & Bukowski,1995; Schwartz, McFadyen-Ketchum, Dodge, Pettit,& Bates, 1998). In addition, a second area of focusin the current investigation was identification of thecognitive and behavioral correlates of specific sub-classes of aggressive behavior. There is evidence thatthere are distinct developmental outcomes associatedwith different subtypes of aggression (Dodge,Lochman, Harnish, Bates, & Pettit, 1997).

431

0091-0627/98/1200-0431$15.00/0 © 1998 Plenum Publishing Corporation

432 Schwartz et al.

Subtypes of Aggressive Behavior

In examining the social-cognitive and behavioralcorrelates of aggression, we made a distinction be-tween the two subtypes of aggressive behavior pos-ited by Dodge and Coie (1987): reactive aggressionand proactive aggression. According to these re-searchers, reactive aggression is a "hot-blooded" an-gry retaliatory response to a perceived provocationor frustration. Proactive aggression, on the otherhand, is a goal-directed behavior that is maintainedby positive environmental contingencies and is gen-erally not associated with underlying states of angeror frustration. Proactive aggression includes unpro-voked behaviors that are oriented toward specific so-cial goals as well as behaviors directed towardposition or object acquisition (Dodge, 1991).

Investigation of the social and psychologicalmechanisms underlying these subtypes of aggressionis an important task, given evidence that reactive andproactive aggression are associated with distinct de-velopmental outcomes. Dodge et al. (1997) foundthat reactively aggressive boys display higher rates ofother behavior problems than proactively aggressiveboys, and tend to be characterized by an earlier onsetof such difficulties. These researchers also reportedthat, among chronically assaultive youth, reactivelyaggressive boys experience psychiatric disturbancemore frequently than do proactively aggressive boys.Other investigators have found that reactive aggres-sion is more strongly associated with peer rejectionthan is proactive aggression (e.g., Price & Dodge,1989).

These findings not withstanding, a potentiallimitation of past investigations in this domain is thatteachers and parents have served as the most com-mon sources of information regarding the subtypesof aggression (e.g., Dodge & Coie, 1987). The result-ing assessments of reactive and proactive aggressionare often highly correlated. In the current investiga-tion, we relied on direct observations of reactive andproactive aggression, which yield more distinct esti-mates (Schwartz, Dodge, & Coie, 1993).

Relations Between Victimization and the Subtypes ofAggression

Because aggression and victimization are aspectsof a reciprocal dyadic behavioral exchange system(Dodge, Price, Coie, & Christopoulos, 1990), one ob-jective of this study was to examine the relation be-

tween aggression and victimization by peers. Rele-vant analyses conducted by previous investigatorshave produced an inconsistent pattern of findings. Intheir influential report on victimization in an elemen-tary school peer group, Perry, Kusel, and Perry(1988) concluded that aggression and victimizationare orthogonal aspects of a child's social experience.More recently, however, evidence has emerged sug-gesting positive correlations between victimizationand aggression, as well as other subtypes of external-izing behavior (e.g., Schwartz et al., 1998). A possibleexplanation for these apparent disparities might bethat there are qualitative differences in the nature ofthe relation between victimization and distinct sub-classes of aggressive behavior. Analyses in which ag-gression is conceptualized as a unidimensionalconstruct might produce ambiguous or inconsistentresults. In the present study, we examined relationsbetween victimization and distinct subtypes of ag-gression.

We hypothesized that reactive aggression wouldbe strongly associated with peer group victimization,but we did not expect an association between proac-tive aggression and victimization. Peers tend to evalu-ate reactive aggression more negatively thanproactive aggression (Price & Dodge, 1989), so thatreactively aggressive children may be at especiallyhigh risk for maltreatment by peers (Schwartz,Dodge, Pettit, & Bates, 1997). Likewise, maltreat-ment by peers may incite reactive aggression. Proac-tive aggression, in contrast, is a behavior that canoccur in the absence of a peer provocation.

Social Cognitive Correlates of Victimization andReactive and Proactive Aggression

A second objective of this study was to investi-gate the social cognitive correlates of peer group vic-timization. Associations between victimization bypeers and individual differences in social cognitivestyles have not been extensively examined in the ex-isting research. However, an understanding of the so-cial cognitive mechanisms underlying themaladaptive social behaviors displayed by victimizedchildren may be an important step toward the designof appropriate interventions (see Perry et al., 1988).

Our focus was on two specific dimensions of so-cial cognition, attributional tendencies (i.e., tenden-cies to interpret peer intent as either hostile orbenign) and outcome expectations (i.e., beliefs re-garding the probable outcomes of particular behav-

Social-Cognitive and Behavioral Correlates 433

ioral strategies), both of which have been linked tochildren's adjustment in the peer group by previousinvestigators (Crick & Dodge, 1994; Dodge &Schwartz, 1997). We hypothesized that there wouldbe a positive association between victimization andtendencies toward attribution of hostile peer intent,because children who are the persistent victims ofbullying could come to view their peers with somewariness. Moreover, hostile attribution al bias (Nasby,Hayden, & DePaulo, 1979) may lead to maladaptivedisplays of inappropriate angry behavior, which arepredictive of rebuff and maltreatment by peers (Coie,Dodge, Terry, & Wright 1991). In addition, we pre-dicted that victimization would be associated withnegative expectancies for the outcomes of assertiveand aggressive behavior. Children who hold negativeoutcome beliefs for aggressive and assertive strate-gies tend to display low rates of such behaviors(Perry, Perry, & Rasmussen, 1986) and, as a result,may be vulnerable to victimization by peers(Schwartz et al., 1993).

We also examined the social cognitive correlatesof proactive aggression and reactive aggression. Re-lated analyses have been conducted by previous in-vestigators. Crick and Dodge (1996), for example,examined the social cognitive attributes of extremesubgroups of children, rated by teachers as predomi-nantly reactively or proactively aggressive. Dodgeand Coie (1987) examined associations between hos-tile attributional biases (but not outcome expectan-cies) and aggression subtypes in contrived groups ofunfamiliar peers. However, the present investigationwill be the first to examine relations between multi-ple dimensions of children's social cognition and di-rectly observed displays of each of the aggressionsubtypes.

Because proactive aggression is conceptualizedas a behavioral process that is maintained by positiveenvironmental contingencies, we hypothesized thatthis subtype of aggression would be associated withpositive outcome expectations for aggressive behav-ior. We further expected that there would be no re-lation between interpretation of social cues and ratesof proactive aggression, because proactive aggressionis a nonangry behavior. In contrast, we hypothesizedthat reactive aggression is contingent on perceptionsof hostile peer intent. Thus, we predicted that hostileattributional tendencies would be associated withhigh rates of this subtype of aggression, but we didnot expect an association between reactive aggressionand outcome beliefs.

Behavioral Correlates of Reactive Aggression,Proactive Aggression, and Victimization

Further understanding of the processes underly-ing aggression and victimization also could comefrom examination of the behavioral contexts of thesebehaviors. Thus, a third area of inquiry was the be-havioral correlates of victimization, proactive aggres-sion, and reactive aggression. The behavioralcorrelates of victimization have been examined in anumber of previous investigations (e.g., Olweus,1978), although relatively little research has beenconducted in this domain using observational meth-ods (Schwartz et al., 1993). The behavioral patternsthat are associated with each of two aggression sub-types, in contrast, have not been extensively exam-ined in any existing study and analyses addressingthis issue were of particular interest.

Because proactive aggression is hypothesized tobe associated with positive outcome expectancies forboth aggressive and assertive behavior (Crick &Dodge, 1996), we predicted that high rates of proac-tive aggression would be associated with assertive so-cial behavior. In contrast, reactive aggression wasexpected to be negatively associated with such be-havior. Price and Dodge (1989) reported that ratesof reactive aggression were negatively correlated withleadership ratings.

We examined relations between the aggressionsubtypes and submissive behavior in three specific so-cial contexts: rough-and-tumble play bouts, aggres-sive exchanges, and persuasion episodes. Previousresearch has demonstrated that behavior in thesecontexts can be an important predictor of social out-comes, particular in regard to aggression and victimi-zation (Schwartz et al., 1993). Rates of nonaggressivedominance attempts were also examined as indica-tors of assertive social behavior.

METHOD

Participants

Participants were identified as part of a largerstudy of aggression in dyadic peer relationships. Allboys in 11 third-grade classrooms (mean age of 8years old), in 11 separate Southeastern elementaryschools, were invited to participate. These schoolsserved predominately African-American populationsof lower-socioeconomic-class background. Approxi-

434 Schwartz et al.

mately 79% of the boys returned positive parentalpermission.

For purposes of analyses not directly relevant tothe current study, consenting boys were asked to ratethe mutual aggressiveness of dyads of boys in theirclassrooms (i.e., the frequency with which the twoboys display aggression toward each other) on a 5-point scale (never to always). The ratings were util-ized to select the most mutually aggressive dyad ineach classroom, and parental permission was re-quested for those two boys to participate in play-groups the following summer. In the event that eithermember of that dyad was unavailable, the next mostaggressive dyad was selected. Four other boys fromeach classroom were also randomly selected to par-ticipate. In total, 11 playgroups consisting of sixthird-grade African-American boys were formed (to-tal N of boys = 66).

Playgroup Administration

The playgroups met for five consecutive days inthe summer following the boys' third-grade year.Playgroup attendance was consistent, with only sixabsences occurring out of a possible 330 instances (6boys x 11 playgroups x 5 sessions). Boys were drivenfrom their home to a laboratory room on a universitycampus, where they participated in 45- min free-playsessions. The laboratory was equipped with age-ap-propriate toys and games. All interactions were re-corded with a video camera that was hidden behinda one-way mirror positioned at one end of the room.No adults were present in the playroom but the chil-dren were closely monitored via the camera.

Social-Cognitive Interviews

Following the second and fourth sessions, theboys participated in individual social- cognitive inter-views conducted by their drivers. Interview questionsassessed attribution styles, and outcome expectanciesfor aggressive or assertive behavior.

For assessment of attribution style, each boy wasread six vignettes (three following Session 2, threefollowing Session 4), which described ambiguousprovocation by a peer. Following each vignette (seeAppendix), the boys were asked to imagine that theprovoking peer was a specific member of their play-group. They were then asked to rate the hostility ofthe peer's intent in the situation, on a 1 to 3 scale(accident to being mean). Next, the boys were asked

to imagine that the vignette involved a differentmember of their playgroup and were administeredthe item again. This procedure was repeated until theboys had responded with an attribution for each oftheir playgroup peers, for each vignette. For lateranalysis, the mean of each subject's attributionsacross play group peers, and across vignettes, was cal-culated. The consistency of this score over the sixvignettes was alpha = .61.

Expectations for aggressive and assertive behav-ior were assessed in a similar manner. The boys wereread six vignettes (three following Session 2, threefollowing Session 4), each of which described a con-flict situation with a peer. They were then asked toimagine that the peer in each vignette was a specificmember of their playgroup, and to imagine makingan aggressive response and an assertive response tothat peer. Next, the boys completed a rating of theexpected instrumental effectiveness of the assertivestrategy and the aggressive strategy, using a 1 to 4scale (definitely not effective to definitely effective).Aggressive and assertive strategies were presented incounterbalanced order across vignettes. The boyswere then asked to imagine that the peer in thevignette was a different member of their playgroup,and then to complete the item once again. This pro-cedure was repeated until the subject had respondedwith an evaluation for each of his playgroup peersfor each of the vignettes. For later analysis, themeans for each subject's assertive and aggressive out-come expectancies across play group peers, andacross vignettes, were calculated. Across all sixvignettes, the consistency of assertive outcome expec-tations was alpha = .73, and the consistency of ag-gression outcome expectations was alpha = .72.

Observations

Videotaped records of playgroup interactionswere coded by trained observers according to twoseparate coding systems: an interval-based system de-signed to assess the frequency of aggression anddominance, and an event-based system that focusedon boys' responses to peer overtures. Five observers,who were trained over a period of 6 to 8 weeks,coded the playgroup interactions. Two of the ob-servers utilized the first coding scheme, and threeutilized the second scheme. Observers met regularlyduring the training period, and periodically duringthe coding, to review progress and discuss disagree-ments. Observers were randomly assigned playgroups

Social-Cognitive and Behavioral Correlates 435

to code. Approximately 15% of the sessions wererandomly selected for unannounced agreementchecks.

Aggression/Dominance Codes, Aggression anddominance were assessed with a focal-dyad codingsystem. Observers coded the interactions of one dyadof boys at a time. Each 45-min playgroup session wassegmented into 270 ten-second intervals. At the endof each interval, observers recorded the occur-rence/nonoccurrence of an initiation from one of theboys in the dyad toward the other. The followingclasses of initiation were coded:

1. Proactive aggression included nonangry goal-oriented aggressive behaviors. This categorywas coded when a boy teased, made fun of,physically abused his dyadic partner, or usedaversive means to reach an external materialgoal (e.g., acquisition of an object).

2. Reactive aggression included angry aggressivebehaviors. This category was coded when aboy responded to a peer overture with hos-tility and retaliatory counterattacking behav-iors. Signs of overt hostility or anger wereoften readily observable.

3. Dominance included forceful dominance-ori-ented behaviors that were not aggressive innature. This category was coded whenever aboy attempted to control or alter the ongo-ing behavior of a peer with assertive (butnonaggressive) social behavior.

For later analysis, the number of intervals inwhich each boy initiated dominance or one of theaggression subtypes toward his play group peers wascalculated. Similarly, the number of intervals inwhich each boy was targeted for either subtype ofaggression was calculated. These scores were cor-rected for instances in which boys were off camera.

Assessment of interobserver agreement focusedon occurrence/nonoccurrence of a behavioral initia-tion, initiation type, and initiator identity. Kappa (K)statistics were utilized as the index of agreement (Co-hen, 1960). Agreement was K = .76 for proactive ag-gression, K = .71 for reactive aggression, and K =.67 for dominance.

Event/Response Codes

The second coding system was derived from theevent-sampling system described by Schwartz et al.(1993). This focal-child system examined boys' be-

havior during specific classes of social exchange.Each boy was observed for the middle 20-min periodin each of the five sessions. Observers recorded anevent whenever the focal boy initiated an overturetoward a playgroup peer. Three mutually exclusiveclasses of overture were coded:

1. Proactive aggression included person-ori-ented and object-oriented aggressive over-tures that were goal driven and nonangry.Behaviors coded included unprovoked physi-cal bullying, verbal teasing, and position orobject struggles.

2. Rough-and-tumble play included nonaggres-sive physical or verbal attempts to involvepeers in quasi-agonistic play activities (e.g.,play boxing, "kung fu" fighting, play wres-tling). These overtures were often accompa-nied by signs of positive emotional states(e.g., laughter, smiles) or signs of prosocialintent (e.g., positive vocal tone).

3. Persuasion attempt included forceful, butnonaggressive, attempts to alter or controlthe behavior of a peer. Behaviors coded in-cluded verbal demands, assertive requests,low-intensity threats, or other dominance-oriented persuasion attempts.

For each of these three types of events, ob-servers recorded the names of the initiator and tar-get, overture type, and time of start and termination.An interaction was considered terminated wheneither the participants did not interact for a 10-s in-terval, or the participants stopped interacting witheach other and began interacting with other boys.

Coders also recorded the response(s) of the tar-gets to each overture, using an exhaustive codingscheme that assessed a number of different dimen-sions of response behavior. Submissive responses,which were of primary interest in the current study,were coded whenever a target boy allowed himselfto be dominated or rewarded a peer's coercive over-ture. Behaviors in this category included showingpain, requesting cessation, and relinquishing objectsor position. For later analysis, a submission score wasgenerated for each boy by dividing the frequency ofsubmission to an overture from a peer by the totalfrequency of received overtures.

Assessment of interobserver agreement focusedon time of overture initiation (within a 5-s period) andidentities of initiating and target boys. Agreement oninitiations was calculated by dividing the total numberof agreements on event occurrence by the average

436 Schwartz et al.

number of events recorded across observers (Coie etal., 1991), yielding a statistic of .70. Agreement on theoccurrence/nonoccurrence of submissive responses toovertures, indexed with a K statistic calculated contin-gent on observer concordance on the occurrence ofan overture, was moderate with K = .57.

RESULTS

Overview

In order to control the influence of intragroupdependencies on analysis results,8 mean-centeredscores were created for each of the variables of in-terest (i.e., the difference between each boy's scoreon a variable and the mean for that variable withinthe boy's play group). Means and standard deviationsfor the variables are presented in Table I.

Logarithmic and square-root transformationswere utilized to reduce skewness in the variable dis-tributions (Neter, Wasserman, & Kutner, 1989), butanalyses conducted with and without transformationyielded identical patterns of results. For ease of pres-entation, the analyses conducted without transforma-tion will be described. Type I error rates for allanalyses were controlled using Holm's (1979) correc-tion procedure. Separate corrections were appliedwithin each series of hypothesized relations.

8Relations among the variables were also examined with a seriesof multilevel models that included both random and fixed effects(Bock, 1989; Goldstein, 1987). For each of these models, predic-tor/outcome relations (i.e., relations between the aggression sub-types, victimization, and behavioral or social-cognitive correlates)were specified as fixed effects and group was specified as a ran-dom effect. Because such multilevel models incorporate simulta-neous estimation of both fixed and random effects (see Bryk &Raudenbush, 1992), this approach allowed us to examine directlythe influence of intergroup differences on the correlates of theaggression subtypes.

This series of analyses failed to yield any marginal or sig-nificant random effects. Thus, the slopes of the individual pre-dictor/outcome relations do not appear to have variedsystematically as a function of group (see Burnstein, 1980). More-over, the pattern of fixed effects yielded by these analyses wasconsistent with the pattern of results yielded by correlationalanalyses of the mean-centered scores.

We complemented these multilevel models with an addi-tional series of analyses in which separate correlation coefficientswere generated within each group. Mean correlation coefficientsacross groups were then tested to determine whether they weresignificantly different from zero (df = 10). This highly conserva-tive analytic strategy also produced a pattern of findings that wasidentical to pattern yielded by the mean-centered correlationalanalyses

Correlations Among the Predictor Variables

A series of correlations was computed to exam-ine relations among the predictor variables. As is de-picted in Table II, associations among the predictorvariables were generally moderate in magnitude. Be-cause there was a strong positive correlation betweenaggressive and assertive outcome expectancies, asummary outcome expectancy variable was createdfrom the mean of the two scores.

Correlations Among Victimization, ReactiveAggression, and Proactive Aggression

Correlations among victimization and the sub-types of aggression are summarized in Table III. Ashypothesized, there was a significant positive corre-lation between victimization and reactive aggression,whereas the correlation between victimization andproactive aggression was not significant. Moreover, at-test for dependent rs (Steiger, 1980) indicated thatthe correlation between victimization and reactiveaggression was significantly different from the corre-lation between victimization and proactive aggres-sion, ?(63) = -6.11, p < .0001. As in previous

Table I. Descriptive Statistics for Social-Cognition, Aggression,Victimization, and Social Behaviora

Variable Mean

Social cognition

Outcome Expectancy Aggression (1- to 4-point rating)

Outcome Expectancy Assertion (1- to 4-point rating)

Hostile Attributional Tendency (1- to 3-point rating)

2.50

2.83

1.69

Aggression and victimization

Reactive Aggression (rate)Proactive Aggression(rate)Victimization (rate)

Social behavior

Dominance (rate)Submission (%)

.03

.09

.12

.02

.25

Standarddeviation

.59

.48

.32

.03

.06

.09

.03

.14aOutcome expectancies are 1 to 4-point ratings with 4 being themost positive evaluation. Attribution is a 1 to 3-point rating with3 being the most hostile attribution. Variables noted with "rate"are computed as the proportion of observed 10-s intervals duringwhich the behavior occurred. Submission is a ratio calculated asthe frequency of submissive responses to peer overtures dividedby the frequency of peer overtures.

Social-Cognitive and Behavioral Correlates 437

tile attributional tendencies were positively corre-lated with victimization, and marginally positivelycorrelated with reactive aggression, but not signifi-cantly correlated with proactive aggression. Out-come expectations for aggressive/assertive behaviorwere negatively correlated with victimization, andpositively correlated with proactive aggression, butnot significantly correlated with reactive aggression.For both classes of social cognition, the correlationcoefficient for reactive aggression was significantlydifferent from the correlation coefficient for proac-tive aggression.

Behavioral Correlates of Victimization, and Reactiveand Proactive Aggression

A similar analytic strategy was adopted to ex-amine the behavioral correlates of victimization andthe two subtypes of aggression. As is depicted in Ta-ble IV, dominance was negatively correlated with vic-timization and positively correlated with proactiveaggression, but was not significantly correlated withreactive aggression. Submission was positively corre-lated with victimization, and negatively correlatedwith both proactive and reactive aggression (the cor-relation for reactive aggression failed Type I errorcorrection). For both classes of social behavior, thecorrelation coefficient for reactive aggression was sig-nificantly different than the correlation coefficientfor proactive aggression.

investigations (e.g., Price & Dodge, 1989), there wasa positive correlation between reactive aggressionand proactive aggression, although the coefficientfound in our analyses appeared to be considerablylower than has been reported in studies conductedusing teacher or peer reports (for which rs are oftengreater than .70; see Dodge et al., 1997).

Social Cognitive Correlates of Victimization andReactive and Proactive Aggression

The social cognitive correlates of victimizationand the two aggression subtypes were examinedwith a series of Pearson correlation coefficients. Inaddition, t-tests for dependent rs (Steiger, 1980)were conducted, in order to determine whether thecorrelation coefficients found for reactive andproactive aggression were significantly different. Asis depicted in Table IV, the overall pattern of cor-relations was consistent with our hypotheses. Hos-

Table in. Correlations Among Subtypes of Aggression andVictimization (N = 66)

Variable

VictimizationReactive Aggression

Victimization/aggression variable

Victimization

—

—

ReactiveAggression

.53a

-

ProactiveAggression

-.07.48a

aIndicates a significant effect at the .0005 level.

Table III. Correlations Among the Predictor Variables (N = 66)

Variable

Outcome Expectancy for AggressionOutcome Expectancy for AssertionHostile Attribution Bias

Dominance

Social cognitive variableOutcome

ExpectancyAggression

OutcomeExpectancyAssertion

Social cognition--

-

.83a

--

Social behavior

- -

HostileAttributional

Bias

-.41d

-37c

-

-

Social behavioral variable

Dominance Submission

.20 -.38d

.10 -.28b,c

-.30 .16

-.38d

aAll correlations were conducted using scores centered on the play group mean. See Table I for a description ofvariable calculation.

bIndicates an effect that failed the Type I error correction procedure.cIndicates a significant effect at the .05 level.dIndicates a significant effect at the .005 level.eIndicates a significant effect at the .0005 level.

438 Schwartz et al.

Table lV.a

Predictor variable

Outcome ExpectancyHostile Attribution

DominanceSubmission

Subtype ofaggression

Reactive Proactive

Social cognition

-.16 .26d

.18b -.12

Social behavior

.09 .54f-.26c,d -.71f

t

3.60d

-2.47d

4.34e

-4.94e

Victimization

-.42.26d

~.32e

.36e

aAll correlations were conducted using scores centered on the playgroup mean. The /-statistic is generated from a test of differencesin dependent rs (df = 63), as specified by Steiger (1980).Significance levels for the correlations are calculated as one-tailedeffects, based on directional hypotheses presented in the text.

bIndicates a marginally significant effect at the .075 level.cIndicates an effect that failed the Type I error correctionprocedure.dIndicates a significant effect at the .05 level.

eIndicates a significant effect at the .005 level.fIndicates a significant effect at the .0005 level.

DISCUSSION

The results of this investigation extend currentunderstanding of the correlates of victimization andaggression. In this study, objective observers madereliable distinctions between reactive aggression,proactive aggression, and victimization. Despite themoderate correlation between the two aggressionsubtypes, each was associated differentially with chil-dren's behavioral and social-cognitive attributes. Thepattern of correlates that emerged for each of theaggression subtypes, and for victimization, providesimportant clues regarding the underlying psychologi-cal and social processes.

Relations Between Victimization and the Subtypes ofAggression

Our analyses indicated that there are theoreti-cally consistent differences in the nature of the rela-tion between victimization and each of the twosubtypes of aggression. As hypothesized, high ratesof reactively aggressive behavior were associated withfrequent victimization by peers. In contrast, proactiveaggression was not significant correlated with victimi-zation. Aggressive behavior that is goal-oriented andnonangry appears to have different implications fora child's social adjustment with peers than angry re-taliatory behavior. Proactive aggression may havenegative implications for long-term adjustment

(Dodge et al., 1997), but displays of this subtype ofaggression do not appear to be associated with vic-timization by peers.

What underlies the relation between reactive ag-gression and victimization? It may be that reactivelyaggressive behavior leads to negative peer group at-titudes (Coie et al., 1991) which are manifested inmaltreatment by peers. Alternatively, children whoare the frequent targets of aggressive overtures (i.e.,victims) may simply have more opportunities to en-gage in retaliatory behavior than their peers (Patter-son, Littman, & Bricker, 1967).

The Social Cognitive Correlates of Victimization andSubtypes of Aggression

Victimization was also positively correlated withhostile attributional tendencies. Boys who were thefrequent targets of peer group aggression tended toview the behavior of their peers as provocative. Thus,hostile attributional bias appears to be associatedwith a victimized/submissive behavioral profile aswell as angry reactive aggression. Persistent maltreat-ment by peers might lead a child to view his or herpeers as hostilely motivated. That child might cometo be characterized by hyper-vigilance and hostile at-tributions, and frequent displays of angry retaliatorybehavior. Consistent with this suggestion, Dodge etal. (1997) reported that early rebuff by peers predictslater reactively aggressive behavior. In contrast, achild who has experienced repeated success withdominance-oriented behavior would be more likelyto develop positive expectations for the outcomes ofaggressive behavior, and might therefore be more in-clined toward bullying and instrumentally aggressivebehavior (Patterson et al., 1967).

Underlying the submissive social behavior associ-ated with frequent receipt of peer aggression(Schwartz et al., 1993) may be a social-cognitive pat-tern of negative expectations for the outcomes of ag-gressive and assertive behavior. Negative outcomeexpectancies for aggression and assertion were stronglyassociated with victimization by peers. Boys who nega-tively evaluate aggressive and assertive behavioralstrategies might tend to engage in high rates of sub-missive behavior. In turn, these boys might frequentlyemerge as persistent targets of peer victimization.

Analyses of the social-cognitive correlates ofdisplayed reactive and proactive aggression revealedimportant differences in the social-information-proc-essing (SIP) mechanisms underlying each subtype of

Social-Cognitive and Behavioral Correlates 439

aggression. In these playgroups, rates of displayed re-active aggression were marginally associated with hos-tile attributions but were not related to outcomeexpectations. Conversely, rates of proactive aggressionwere not significantly correlated with attributional ten-dencies but were significantly linked to outcome ex-pectations. This pattern of findings is consistent withpast research (Crick & Dodge, 1996) and is supportiveof our hypothesis that reactively aggressive behavioremerges from the representation and interpretationstages of SIP, whereas proactive aggression is moreclosely associated with processing at the responseevaluation stage.

The Behavioral Correlates of Victimization, ReactiveAggression, and Proactive Aggression

Our findings also extend the work of previousinvestigators by demonstrating that there is a distinctpattern of behavioral correlates associated with eachsubtype of aggression. Proactive aggression was nega-tively correlated with submissiveness and positivelycorrelated with assertiveness. Although proactivelyaggressive behavior is often linked to negative socialoutcomes in the peer group (Coie, Dodge, & Ku-persmidt, 1990), it appears that this subtype of ag-gression is also associated with positive socialattributes in some peer groups (e.g., leadership).Price and Dodge (1989) reached similar conclusionsbased on peer ratings of social behavior and obser-vations of proactive aggression.

In contrast, our analyses did not yield a strongpattern of behavioral correlates for reactive aggres-sion. This outcome may reflect limitations in theclasses of behavior assessed. We suspect that reactiveaggression is more closely associated with other im-pulsive or emotionally dysregulated behaviors thanthe assertive and submissive dimensions of behaviorassessed in the current investigation (Hubbard &Coie, 1992).

Future Directions and Caveats

Although the findings of this study extend cur-rent understanding of the processes that may under-lie distinct subtypes of aggression, additionalinvestigation will be needed to address a number ofunanswered questions. One important task for futureresearchers will be further identification of the socialbehaviors than are associated with reactive aggres-sion. New observational coding schemes will need to

be developed that focus on provocative, irritating be-haviors as well as relevant dimensions of emotionallydysregulated behaviors (Hubbard & Coie, 1992).Coding systems and observational methods that allowfor sequential analysis of behavior would be particu-larly informative.

Future efforts should also focus on more natu-ralistic settings. Previous researchers have demon-strated the ecological validity of contrived play groupapproaches (e.g., Dodge, 1983). However, our under-standing of the underlying processes and the out-comes associated with each subtype of aggressioncould be extended by research that focuses on natu-rally occurring group processes and relevant contex-tual influences.

A significant limitation of the current researchis that our focus was on boys only. Thus, it is notclear whether the observed patterns of relations be-tween the subtypes of aggression and social cogni-tion/behavior will generalize to mixed-gender peergroups, or girls' peer groups. Further research, in-cluding both genders will be needed. This work maybe complicated by the fact that girls often utilizeforms of aggression that can be difficult to observe(Crick & Grotpeter, 1995).

Finally, we hope that the emerging body of find-ings on the correlates and predictors of distinct sub-types of aggression will serve as a useful foundationfor future intervention efforts. For example, the cor-relational research presented here, as well as Crickand Dodge's (1996) focus on extreme subgroups ofreactively and proactively aggressive children, couldpotentially assist in the design of interventions thattarget the specific social cognitive mechanisms un-derlying each subtype of aggression. This could proveto be an important task, given the evidence that re-active and proactive aggression are associated withdissimilar psychosocial outcomes (Dodge et al.,1997).

APPENDIX

Example of attribution vignette:

Pretend that you were lining up to take a shot inthe basketball game and another boy in the groupcame up behind you and bumped into you just asyou were about to shoot. You missed the shot andfell down because of him. Sometimes when this hap-pens the person is trying to be mean and sometimesit is an accident. Suppose the other kid wasDo you think that was being mean, or do youthink that it was an accident, or is it hard to tell?

440 Schwartz et al.

Example of outcome expectancy vignette:

Pretend one day that one of the adults came in witha plate of candy bars. There was just enough foreach kid to have one candy bar and another kid inthe group took two candy bars leaving none for you.Now, pretend that kid was . Suppose you askednicely for the candy (assertive strategy)? Wouldkeep the candy bar or give it back to you? Now,supposed you shoved aggressive strategy).Would keep the candy bar or give it back to you?

ACKNOWLEDGMENTS

This investigation was supported by NIMH grant38765 to J. Coie and K. Dodge. We are grateful toE. Valente, S. Schilling, S. McFadyen-Ketchum, andM. Appelbaum for consultation regarding statisticalanalysis. The contributions of A. Johns, K. Skow, andV Tidwell are also acknowledged.

REFERENCES

Bock, R. D. (1989). Multilevel analysis of educational data. San Di-ego, CA: Academic Press.

Boivin, M., Hymel, S., & Bukowski, W. M. (1995). The roles ofsocial withdrawal, peer rejection, and victimization by peersin predicting loneliness and depressed mood in children. De-velopment and Psychopathology, 7, 765-786.

Burnstein, L. (1980). The analysis of multilevel data in educationalresearch and evaluation. Review of the Research in Education,8, 158-233.

Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear mod-els: Applications and data analysis methods. London, England:Sage.

Cohen, J. A. (1960). A coefficient of agreement of nominal scales.Educational and Psychological Measurement, 20, 37-46.

Coie, J. D., & Dodge, K. A. (in press). Aggression and antisocialbehavior. In W. Damon (Series Ed.) & N. Eisenberg (Vol.Ed.). Handbook of child psychology, Vol. 3; Social, emotional,and personality development (pp. 547-641). New York: Wiley.

Coie, J. D., Dodge, K. A., & Kupersmidt, J. B. (1990). Peer groupbehavior and social status. In S. R. Asher & J. D. Coie (Eds.),Peer rejection in childhood. New York: Cambridge UniversityPress.

Coie, J. D., Dodge, K. A., Terry, R., & Wright, V. (1991). Therole of aggression in peer relations: An analysis of aggressionepisodes in boys' play groups. Child Development, 62, 812-826.

Coie, J. D., & Kupersmidt, J. B. (1983). A behavioral analysis ofemerging social status in boys' groups. Child Development, 54,1400-1416.

Crick, N. R., & Dodge, K. A. (1994). A review and reformulationof social information-processing mechanisms in children's so-cial adjustment. Psychological Bulletin, 115, 74-101.

Crick, N. R., & Dodge, K. A. (1996). Social information-process-ing mechanisms in reactive and proactive aggression. ChildDevelopment, 67, 993-1002.

Crick, N. R., & Grotpeter, J. K. (1995). Relational aggression, gen-der, and social- psychological adjustment. Child Development,66, 710-722.

Dodge, K. A. (1983). Behavioral antecedents of peer social status.Child Development, 54, 1386-1399.

Dodge, K. A. (1991). The structure and function of reactive andproactive aggression. In D. Pepler & K. Rubin (Eds.), Thedevelopment and treatment of childhood aggression (pp. 201-218). Lawrence Erlbaum: Hilsdale, NJ: Erlbaum.

Dodge, K. A., & Coie, J. D. (1987). Social information-processingfactors in reactive and proactive aggression in children's peergroups. Journal of Personality and Social Psychology, 53, 389-409.

Dodge, K. A., Lochman, J. E., Harnish, J. D., Bates, J. E., & Pettit,G. S. (1997). Reactive and proactive aggression in school chil-dren and psychiatrically impaired chronically assaultive youth.Journal of Abnormal Psychology, 106, 37-51

Dodge, K. A., Price, J. M., Coie, J. D., & Christopoulos, C. (1990).On the development of aggressive dyadic relationships inboys' peer groups. Human Development, 33, 260-270. .

Dodge, K. A., & Schwartz, D. (1997). Social information process-ing mechanisms in aggressive behavior. In J. Breiling & J.Maser (Eds.), Handbook of antisocial behavior (pp. 171-180).New York, Wiley.

Goldstein, H. (1987). Multilevel models in educational and socialresearch. Oxford: London, England: Oxford University Press.

Holm, S. (1979). A simple sequentially rejective multiple test pro-cedure. Scandinavian Journal of Statistics, 6, 65-70.

Hubbard, J. A. & Coie, J. D. (1994). Emotional correlates of socialcompetence in children's peer relations. Merrill-Palmer Quar-terly, 40, 1-20.

Nasby, W., Hayden, B., & DePaulo, B. M. (1979). Attributionalbiases among aggressive boys to interpret unambiguous socialstimuli as displays of hostility. Journal of Abnormal Psychology,89, 459-468.

Neter, J., Wasserman, W., & Kutner, M. H. (1989). Applied linearregression models. Homewood, IL: Irwin.

Olweus, D. (1978). Aggression in the schools: Bullies and their whip-ping boys. Washington, DC: Hemisphere.

Parke, R. D., & Slaby, R. G. (1983). The development of aggres-sion. In P. H. Mussen (Series Ed.) & E. M. Hetherington(Vol. Ed.), Handbook of child psychology: Vol. 4: Socialization,personality, and social development (pp. 547-641). New York:Wiley.

Patterson, G. R., Littman, R. A., & Bricker. W. (1967). Assertivebehavior in children: A step toward a theory of aggression.Monographs of the Society for Research in Children Develop-ment, 35 (5, Serial No. 113).

Perry, D. G., Kusel, S. J., & Perry, L. C. (1988). Victims of peeraggression. Developmental Psychology, 24, 807-814.

Perry, D. G., Perry, L. C, & Rasmussen, P. (1986). Cognitive so-cial learning mediators of aggression. Child Development, 57,700-711.

Price, J. M., & Dodge, K. A. (1989). Reactive and proactive ag-gression in childhood: Relations to peer status and social con-text dimensions. Journal of Abnormal Child Psychology, 17,455-471.

Schwartz, D., Dodge, K. A., & Coie, J. D. (1993). The emergenceof chronic peer victimization in boys' play groups. Child De-velopment, 64, 1755-1772.

Schwartz, D., Dodge, K. A., Pettit, G. P., & Bates, J. E. (1997).The early socialization of aggressive victims of bullying. ChildDevelopment, 68, 665-675.

Schwartz, D., McFadyen-Ketchum, S. A., Dodge, K. A., Pettit, G.P., & Bates, J. E. (1998 ). Peer victimization as a predictorof behavior problems at home and in school. Development andPsychopathology, 10, 87-100.

Steiger, J. H. (1980). Tests for comparing elements of a correlationmatrix. Psychological Bulletin, 87, 245-251.