The distinction between beliefs legitimizing aggression and deviant processing of social cues: Testing measurement validity and the hypothesis that biased processing mediates the effects

Journal of Personality and Social Psychology1999. Vol. 77, No. 1. 150-166

Copyright 1999 by Che American Psychological Association, Inc.0022-3514/99/S3.00

The Distinction Between Beliefs Legitimizing Aggression andDeviant Processing of Social Cues: Testing Measurement Validityand the Hypothesis That Biased Processing Mediates the Effects

of Beliefs on Aggression

Arnaldo Zelli and Kenneth A. DodgeVanderbilt University

John E. LochmanDuke University

Robert D. LairdVanderbilt University

Conduct Problems Prevention Research Group

In 2 studies the authors examined knowledge and social information-processing mechanisms as 2 distinctsources of influence on child aggression. Data were collected from 387 boys and girls of diverse ethnicityin 3 successive years. In Study I, confirmatory factor analyses demonstrated the discriminant validity ofthe knowledge construct of aggression beliefs and the processing constructs of hostile intent attributions,accessing of aggressive responses, and positive evaluation of aggressive outcomes. In Study 2, structuralequation modeling analyses were used to test the mediation hypothesis that aggression beliefs wouldinfluence child aggression through the effects of deviant processing. A stronger belief that aggressiveretaliation is acceptable predicted more deviant processing 1 year later and more aggression 2 years later.However, this latter effect was substantially accounted for by the intervening effects of deviantprocessing on aggression.

Recent social-cognitive formulations have contended that ha-bitual aggression is regulated by the interplay of two fundamentaltypes of mental processes (Crick & Dodge, 1994; Dodge, 1993;Huesmann, 1988, 1998). The first entails a sequence of biasedinferences and judgments made during the representation of prox-imal social stimuli, leading to aggression in interpersonal situa-

Amaldo Zelli, Kenneth A. Dodge, and Robert D. Laird, Department ofPsychology and Human Development, Vanderbilt University; John E. Loch-man, Department of Psychiatry, Duke University Medical Center; ConductProblems Prevention Research Group (CPPRG; members of the CPPRG are,in alphabetical order, Karen L. Bierman, Department of Psychology, Pennsyl-vania State University; John D. Coie, Department of Psychology, Social andHealth Services, Duke University; Kenneth A. Dodge, Center For Child andFamily Policy, Duke University; Mark T. Greenberg, Department of Psychol-ogy, Pennsylvania State University; John E. Lochman, Department of Psy-chology, University of Alabama; and Robert J. McMahon, Department ofPsychology, University of Washington).

Arnaldo Zelli and Kenneth A. Dodge are now at the Center for Child andFamily Policy, Duke University; John E. Lochman is now at the Depart-ment of Psychology, University of Alabama; Robert D. Laird is now at theDepartment of Human Development and Family Studies, University ofRhode Island.

This work was supported by National Institute of Mental HealthGrants R18MH48043, R18MH50951, R18MH50952, R18MH50593, andK05MH01027 and by U.S. Department of Education G • S184U30002.The Center for Substance Abuse Prevention has also si ted this workthrough a memorandum of agreement.

Correspondence concerning this article should be adc sed to ArnaldoZelli, who is now at the Center for Child and Fan Policy, DukeUniversity, Box 90264, Durham, North Carolina 2770* 164. Electronicmail may be sent to [email protected].

tions of conflict or provocation. The steps in this sequence havebeen ^articulated in models of social information processing(Dodge, 1993; Huesmann, 1988). The second comprises socialknowledge mechanisms that link in memory past negative socialexperiences to one's unique representations of current stimuli.These mechanisms of influence have been traditionally articulatedin models in which latent knowledge structures summarize one'spast experiences (Catrambone & Markus, 1987; Nelson, 1993),offer a model for future action (Abelson, 1981; Bandura, 1986;Fuhrman & Funder, 1995; Markus, 1983; Schneider, 1991), andcan therefore guide one's processing of social cues in any singleinstance or situation (see Dodge, 1993; Huesmann, 1988, 1998, forapplications to aggressive behavior).

This theorizing, and its critical distinction between social infor-mation processing and knowledge sources of influence on aggres-sive behavior, is relatively recent and is consistent with severaladvances in the study of social cognition (Bandura, 1986; Cervone& Williams, 1992; Mischel & Shoda, 1995). It also stands as alogical extension of early work that primarily sought to identify themental processes and judgments that are proximally responsiblefor the display of habitual deviant behavior within well-specifiedsituational domains (Dodge, 1986; Dodge & Feldman, 1990).Nonetheless, with the exception of a few recent studies that fo-cused on memory or priming effects on one's inference processes(Dodge & Tomlin, 1987; Graham & Hudley, 1994; Stromquist &Strauman, 1991; Zelli, Cervone, & Huesmann, 1996; Zelli, Hues-mann, & Cervone, 1995), there is a paucity of empirical testing ofthe hypothesized relations among social knowledge, habitual mal-adaptive mental processing of current social cues, and individualdifferences in aggressive behavior.

150

BELIEFS AND SOCIAL INFORMATION PROCESSING 151

In the present investigation, our goal was to examine the validityof a general model of aggressive behavior that draws a distinctionbetween social knowledge and deviant processing and that furtherposits that the influence of antisocial knowledge structures onaggressive behavior is at least partially contingent on its effects onhow one processes current social cues. We pursued such a goalwith respect to two classes of constructs that have been implicatedin the analysis of developmental processes leading to habitualaggression; namely, social knowledge variables such as childnormative beliefs about the social appropriateness of aggression(Huesmann & Guerra, 1997) and processing variables such as achild's tendency to attribute hostile motives in hypothetical inter-personal exchanges, to generate aggressive solutions to the sameevents, and to foresee meaningful benefits for the self if aggressivesolutions were to be enacted (Dodge, 1986; Dodge & Price, 1994).

In the first study, we examined the validity of a measurementmodel, hypothesized that a four-factor model (i.e., separate con-structs of normative beliefs, hostile attributions, aggressive re-sponse generation, and anticipation of positive consequences foraggressing) could represent the patterns of bivariate relationsamong the variables of interest, and estimated the goodness of fitof such a model for child data collected in three successive yearsof elementary school. We then compared each year's four-factormodel with a series of models positing fewer than four factors andtested the changes in model fitting associated with these alternativehypotheses. In the second study, we further tested construct valid-ity by specifically examining the hypothesis that deviant patternsof processing current social cues significantly mediate—and couldadequately describe—the effects of aggression beliefs on lateraggressive behavior. Both studies were formulated and conductedin the context of a large longitudinal investigation that providedreliable developmental data on the constructs of interest and tem-poral constraints on the direction of effects. In the remainingsections of this introduction, we provide further details on theissues that motivated the present investigation, and the generalresearch hypotheses and findings from which it originated.

THE STUDY OF SOCIAL KNOWLEDGE ANDINFORMATION PROCESSING

Social knowledge acquisition, inference processes, and contextor situational effects are fundamental research themes in bothsocial and personality psychology (Ross & Nisbett, 1991; Zelli &Dodge, 1999). Both fields have increasingly investigated the waysin which people's self- and social knowledge affect appraisalprocesses and human behavior within specific domains of experi-ences or tasks (Cantor & Kihlstrom, 1987; Cervone & Williams,1992; Fiske & Taylor, 1991; Higgins, 1990; Mischel & Shoda,1995). Research on impression formation, for instance, clearlysuggests that differences in social experiences give rise to differ-ences in implicit goal systems and expectations and that these, inturn, regulate differences in inference processes. Thus, the likeli-hood of making any inference depends in part on one's relativelychronic orientation toward—or goals in regard to—social interac-tions (Anderson & Deuser, 1993; Battistich & Aronoff, 1985;Kruglanski, 1990; Moskowitz, 1993). Likewise, the type and con-tent of inferences made in the presence of ambiguous informationseem to hinge on chronically accessible social constructs andexpectations (Bandura, 1986; Bargh, Lombardi, & Higgins, 1988;

Bargh & Pratto, 1986; Cantor & Kihlstrom, 1987; Higgins & King,1981; Lochman & Dodge, 1998; Markus, 1977).

The importance of linking goal or knowledge mechanisms toappraisal processes and behavior also extends to traditional do-mains of personality psychology (Cervone & Williams, 1992;Pervin, 1989). Persistent maladaptive behavior in academic con-texts arises in those who are chronically concerned with document-ing their level of competence and who, in turn, are particularlysensitive to self-appraisals of incompetence in the face of the manychallenges inherent to academic tasks (Dweck & Leggett, 1988).Similarly, coherence in self-efficacy judgments arises when con-sidering both one's personal strengths or weaknesses (i.e., self-knowledge) and situational beliefs about how relevant these per-sonal attributes are to performing challenging behaviors (Cervone,1997).

More broadly, there seems to be consensus that coherence inone's behavior and functioning arise from interactions amongpsychological mechanisms such as one's relatively accessible self-or social knowledge, one's encoding and interpretation of currentsituation(s), and the self-regulating standards one uses in evaluat-ing any behavior or solution to a present situation that may cometo mind (Cervone & Williams, 1992; Higgins, 1990; Shoda &Mischel, 1993; Zelli & Dodge, 1999). In the last decade, thisconceptual framework has influenced work in many areas ofapplied research (Andersen, Spielman, & Bargh, 1992; Penn,Corrigan, Bentall, Racenstein, & Newman, 1997; Skinner, 1996;Taylor & Brown, 1988; Tomaka & Blascovich, 1994) includingaggression research, to which we now turn.

SOCIAL COGNITION AND AGGRESSION

Early Social Information-Processing Models

The social-cognitive mechanisms and processes underlying in-dividual differences in aggressive experiences and behavior (es-pecially of aggression that occurs at young ages) are relatively wellunderstood (Dodge & Frame, 1982; Dodge, Bates, & Pettit, 1990;Dodge, Pettit, McClaskey, & Brown, 1986; Dodge, 1986; Dodge,Murphy, & Buchsbaum, 1984; Dodge, 1980; Slaby & Guerra,1988). A detailed review of this work is beyond the scope of thisarticle (for reviews, see Crick & Dodge, 1994; Dodge, 1993).Broadly speaking, Dodge and colleagues were interested in thespecific processing and judgment events leading to aggression insituations of interpersonal confrontation or conflict. They theo-rized that benign processing of social cues fosters competentbehavior, whereas hostile processing of the same stimuli fostersaggressive behavior. In this view, on-line processing involves asequence of distinct and relatively independent judgment tasks (orprocessing steps) ranging from interpreting one's motives to ac-cessing and deciding to enact a certain behavioral response. To theextent that each judgment task represents a new link to aggressivebehavior, aggression might be linked to deviant processing at anyor all of these steps, and multiple types of deviant processing mightcumulatively influence aggressive responding (Dodge, 1991,1993).

This research has identified several processing variables orjudgment dimensions along which individuals' performances varyreliably with individual differences in aggressive behavior. Rela-tive to their nonaggressive peers, highly aggressive children and

152 ZELLI ET AL.

youths presume hostile motives in their peers' actions more oftenand more consistently, both when these actions are ambiguous(i.e., situations that do not explicitly cue any particular motive;Dodge & Tomlin, 1987; Feldman & Dodge, 1987: Milich &Dodge, 1984; Waas, 1988) and when they are (normatively speak-ing) nonthreatening or benign to the self (i.e., hostile attributionalbias; Nasby, Hayden, & DePaulo, 1979; Dodge et al., 1984;Dodge, Bates, et al., 1990; Dodge, Price, Bachorowski, & New-man, 1990). Highly aggressive persons also access more incom-petent, action-oriented solutions to a variety of interpersonal situ-ations (Dodge, 1993; Rubin, Bream, & Rose-Krasnor, 1991). Insituations of direct provocation, they are more prone to generatesolutions of immediate retaliation (Slaby & Guerra, 1988; Waas,1988). Likewise, in situations in which social status is threatenedor in situations of open conflict with peers, highly aggressivepersons primarily consider manipulation and coercive responses oradult punishment (Dodge et al., 1986). Finally, habitually aggres-sive persons also depart from normative processing at the time ofevaluating the utility or effectiveness of a variety of aggressivebehavioral solutions along several dimensions of social outcomes(Crick & Dodge, 1996; Crick & Ladd, 1990; Perry, Perry, &Rasmussen, 1986). They tend to judge aggression quite positively,and this evaluation stems from favorably construing the moralimplications of acting aggressively (i.e., aggression is not bad),from praising the highly instrumental value of behaving aggres-sively (i.e., aggression leads to more material gain), from notforeseeing negative sanctions as likely to follow (i.e., aggressioncould not lead to punishment), or from feeling confident aboutacting aggressively (i.e., it is relatively easy to be aggressive).

As evidenced by several longitudinal studies, these patterns ofdeviant processing lead to, rather than merely correlate with,aggressive responding across development (see Crick & Dodge,1994, for an extensive review). Furthermore, the acquisition ofdeviant processing patterns seems to be the critical link in under-standing (and mediating) the long-lasting effects that particularsocializing experiences, such as early child abuse or parentalrejection and punishment, reliably exert on later aggressive re-sponding (Dodge, Bates, & Petit, 1990; Dodge, Pettit, Bates, &Valente, 1995).

The Role of Organized Knowledge

Despite these findings, researchers have begun to acknowledgethat the understanding of habitual aggression can also benefit froma careful analysis of one's organized (self- and social) knowledge(Crick & Dodge, 1994; Dodge, 1993; Huesmann, 1998). Severalissues bear upon and benefit from this conceptual analysis. A firstgeneral issue is in regard to the links between specific deviantprocessing patterns and the well-established stability or cross-situational consistency of aggressive behavior (Caspi, Elder, &Bern, 1987; Eron & Huesmann, 1990; Huesmann, Eron, Lefko-witz, & Walder, 1984; Olweus, 1979). It seems clear that deviantprocessing is reliably consistent across a set of stimuli only orespecially when the stimuli belong to the same situational domain,and that processing patterns may vary substantially when consid-ering different situational domains (Dodge & Feldman, 1990). Tothe extent that social information processing is a proximal ante-cedent to enacting a certain behavioral response in single in-stances, one must seek additional mechanisms to model the social-

cognitive processes contributing to the situational consistency ofsocial behavior (Dodge, 1993). Early information-processing re-search was not explicitly conceived for this level of analysis. Italso did not embody constructs of latent cognitions linking mech-anisms of mental categorization, mental representation, or memorystorage to one's predisposition to process and respond in determi-nate ways to a variety of social situations (Dodge, 1993).

Social knowledge mechanisms also may need to be positedwhen considering the direction of causal relations between one'ssocial experiences and social information processing (see Crick &Dodge, 1994; Huesmann, 1998). One's behavior not only influ-ences, but also is influenced by, the most immediate social envi-ronment (Bandura, 1986). Thus, aggression is likely to elicit avariety of responses that can range from social reward and recog-nition to, alternatively, social rejection, neglect, or isolation (Crick& Dodge, 1994; Lochman, 1987). As Crick and Dodge (1994) alsosuggested, these environmental responses become critical to one'sfuture behavior to the extent that they further reinforce or weakengeneral prior convictions, beliefs, and expectations about the socialworld (Coie, Dodge, & Kupersmidt, 1990; Lochman & Dodge,1994; Lochman & Lampron, 1986). Thus, for instance, if a child'saggression leads to social rejection by his or her peers, this childmight increasingly endorse beliefs or expectations that de-emphasize the importance of social relations and peers, and thesecognitions, in turn, may further promote processing actions lead-ing, or conducive, to aggression (Crick & Dodge, 1994; Hues-mann, 1998).

Laboratory Studies of Memory and Priming Effects

These issues and hypotheses have prompted work that hasfocused on the relatively accessible knowledge constructs thatseem critical to the interpretive processes of chronically aggressivepersons. Thus, for instance, when compared with average children,aggressive children who deliberate upon others' behaviors andmotives reach their judgments by using personal past experiences(i.e.,self-schemas) rather than current situational cues (Dodge &Tomlin, 1987). Aggressive people tend to make more extremejudgments of presumed hostility than do nonaggressive personseven when both groups are primed for unintentionality; this sug-gests that aggressive persons carry with them causal beliefs thatanticipate others' malevolent motives (Graham & Hudley, 1994).Children who are rated as maladjusted and antisocial hold highlyaccessible social constructs (i.e., representations of relevant refer-ence groups) that are permeated with more antisocial andunlikable-trait attributes (Stromquist & Strauman, 1991). Finally,participants' recall performance for trait-relevant behaviors showsthat highly aggressive persons make hostile inferences even whenthey seemingly have no inferential goal, thus suggesting thathighly accessible hostile configurations or schemas affect infer-ences at encoding (Zelli et al., 1995, 1996).

This and other recent work (Anderson, 1997; Bushman, 1998) isconsistent with the hypothesis that chronically or highly accessibleaggressive constructs exert an influence on how aggressive peopleprocess and interpret current social information. However, thiswork relied on hypotheses and methodologies that primarily fo-cused on assessing highly specific judgments (e.g., "spontaneous"hostility inferences; Zelli et al., 1996) or knowledge constructs

BELIEFS AND SOCIAL INFORMATION PROCESSING 153

(e.g., a child's memory representation of peers; Stromquist &Strauman, 1991).

Individual Differences in Normative BeliefsAbout Aggression

Huesmann and colleagues (Huesmann, 1982, 1988; Huesmann& Eron, 1984) have presented a more integrated view of how latentknowledge structures influence one's social information process-ing and aggressive behavior. In their model, "scripts," or programsof behavior, determine what behaviors one ultimately enacts, andself-regulating beliefs directly influence one's active response-decision process at the time of dealing with a certain situation(Huesmann & Guerra, 1997). Thus, habitual aggression arisesfrom learning, using, and increasingly reinforcing aggressivescripts: "aggressive scripts are . . . stored in a person's memoryand are used as guides for behavior and social problem solv-ing . . . [they] suggest what events are to happen in the environ-ment, how the person should behave in response to those events,and what the likely outcome of those behaviors would be" (Hues-mann, 1988, p. 15). Within this framework, habitually aggressiveindividuals have encoded in memory more extensive networks ofbehavioral scripts suggesting aggressive problem solving than dononaggressive individuals. They also have acquired individualbeliefs that legitimize aggression (i.e., the belief that aggression isokay) and that filter the specific (aggressive) behaviors that maycome to mind at the time of dealing with a given situation (Ban-dura, 1986; Bandura, Barbanelli, Caprara, & Pastorelli, 1996;Huesmann, 1998; Huesmann & Guerra, 1997).

In a recent longitudinal study with a large metropolitan childsample, Huesmann and Guerra (1997) found clear evidence insupport of their hypotheses that there exist individual differencesin child individual beliefs legitimizing aggression and that thesebeliefs exert an influence on aggressive behavior. The 1-yearstability of beliefs increased with age, and it reached moderate butmeaningful levels by third or fourth grade. Despite their transientstatus in earlier years, a stronger approval of aggression wasassociated with more actual aggressive behavior throughout theelementary years. The meaning of this relation seems to changewith development. Thus, behaving more aggressively at veryyoung ages (e.g., around age 6) leads to the acquisition of strongerbeliefs condoning aggression, whereas a stronger approval ofaggression leads to more aggressive behavior at later ages (e.g.,around ages 8 or 9). In both cases, the early variable predicted"growth," or changes, in the subsequent variable (Huesmann &Guerra, 1997).

These findings are consistent with the general hypothesis thatone may learn aggressive scripts through a sequence of reciprocalprocesses in which aggressive behavior initially promotes beliefsthat are consistent with such behavior and in which, eventually,beliefs begin fostering further aggression by virtue of encouragingor not condemning it (Huesmann, 1998).

THE PRESENT INVESTIGATION

As Huesmann and Guerra (1997) acknowledged, their modelrests on the assumption that normative beliefs about aggressionaffect behavior by also influencing the ways one processes andtherefore responds to current social situations. For instance, it is

easy to conceive that one who believes aggression is appropriatemay be more likely to judge an aggressive solution that comes tomind as an appealing route for dealing with a situation at hand (i.e.,an influence on the processing step of response evaluation). In-deed, the function of filtering behavioral solutions that come tomind is paramount to the self-regulating value that normativebeliefs about aggression seemingly have in one's functioning(Huesmann & Guerra, 1997). Their effects, however, may wellextend to other steps of mental processing. One who believesaggression is appropriate may perceive aggression as relativelynormative and this, in turn, may enhance the chances of perceivingothers' behaviors as being motivated by hostile or provocativeintent (i.e., an influence on the processing step of interpretation).Finally, a stronger belief that aggression is a legitimate behaviormay represent a stronger internal cue for the retrieval of aggressivesolutions from long-term memory and, thus, enhance how readilyaggression comes to mind (i.e., an influence on the processing stepof response access).

Lacking empirical evidence, these hypotheses of mediated in-fluences represent theoretical propositions. Furthermore, the heu-ristic value of distinguishing between knowledge constructs suchas one's normative beliefs and processing constructs such as one'sinterpretation of others' intent, response access, and responseevaluation remains uncertain. The present investigation was de-signed to address both issues.

We assessed normative beliefs approving of aggression in asample of children three times, starting at the end of Grade 3school year and then 12 months and 24 months later. These agesand time lags were chosen to be consistent with Huesmann andGuerra.'s (1997) finding that normative beliefs begin to exertinfluence on future behavior at about these ages. As part of thesame interview sessions, we assessed children's processing ofage-appropriate hypothetical situations of conflict or confronta-tion. More specifically, we measured children's inferences aboutthe motives underlying behaviors depicted in stories, their ten-dency to anticipate aggressive solutions for the self when askedwhat they would do in those situations, and their tendency to havea relatively positive evaluation of aggressive solutions to thesituations at hand. Finally, we considered the children's level ofaggressive behavior as indexed by behavioral ratings provided byschool teachers, children, and their parents.

In the first study, we tested the validity of a measurement modelpositing clear distinctions among the four constructs of aggressionbeliefs, hostile biases, response accessing, and response evalua-tion. In the second study, we further examined construct validityby testing three specific hypotheses; namely, that beliefs approvingof aggression influence one's processing of social cues at a latertime, that deviant patterns of processing information influenceaggressive behavior at a later time, and that these two effectspartially account for the longitudinal relations between aggressionbeliefs and aggressive behavior.

STUDY 1

Method

Participants

The participants of this study were 387 children who were participantsin a longitudinal multisite investigation of the development and prevention

154 ZELLI ET AL.

of conduct problems in children. The details of this investigation have beendescribed elsewhere (Conduct Problems Prevention Research Group, 1992;Lochman & Conduct Problems Prevention Research Group, 1995). Thechild sample was representative of the population in the high-risk schoolstargeted in the larger investigation. Sites included Durham, North Carolina;Nashville, Tennessee; rural central Pennsylvania; and Seattle, Washington.Within each site, schools were randomly assigned to receive intervention orserve as a no-treatment control. During the spring of kindergarten, all thechildren enrolled in the control schools were rated by their teachers for thepresence of behavioral problems. A normative sample of 100 children ateach of the four sites was then obtained by randomly selecting 10 childrenfrom each decile of the teacher rating score distributions (Teacher ScreenProblem Behavior; Lochman & Conduct Problems Prevention ResearchGroup, 1995). This selection respected the race and sex group compositionobtained within each of the Teacher Screen deciles. One site providedonly 87 children because one of its schools was dropped from the studyduring the first year. Across all sites, the sample mean age in Year 1 (i.e.,when the children entered first grade) was 6.36 years (SD = 0.44), 50% ofthe children were boys, and 49% of the sample had a minority ethnicbackground (43% African American and 6% other).

Children were interviewed at home by trained interviewers for annual2-hr individualized, summer assessment sessions. The present study reliedon normative belief and social information-processing data collected ineach of three successive years from 331 third graders (86% of originalsample), 340 fourth graders (88%), and 328 fifth graders (85%). A sub-stantial component of the original sample (300 children, 78%) providedcomplete 3-year data on the cognitive variables of interest.

Cognitive Assessment

Normative Beliefs

Children's normative beliefs about aggression were assessed on a 20-item scale originally developed and refined over several studies by Hues-mann and colleagues (Huesmann & Guerra, 1997; Huesmann, Guerra,Miller, & Zelli, 1992; Huesmann, Guerra, Zelli, & Miller, 1992). Twelveitems measured approval for retaliating to either a weak (e.g., "suppose aboy says something bad to another boy") or a strong (e.g., "suppose a boyhits another boy") prior provocation. In particular, for each of four weaksituations, children indicated whether it was okay to scream back (total of 4items) or hit back (total of 4 items), whereas for each of four strongsituations, children indicated whether it was okay to hit back (total of 4items). The remaining 8 items were included to measure the children'smore general normative beliefs about aggression (e.g., "it is usually okayto hit others"). Children responded to each belief item by indicatingwhether the behavior was perfectly okay (0), sort of okay (1), sort of wrong(2), or really wrong (3). Item responses were reversed so that higher scoreswould indicate greater endorsement of aggression.

Each belief dimension was measured reliably at each of the 3 years ofassessment, and scale Cronbach's alpha reliability did not appear to varywith child gender or ethnicity. At each year, the 12-item coefficient forretaliation beliefs was slightly higher (i.e., .85, .86, and .87 for Grade 3,Grade 4, and Grade 5, respectively) than the 8-item coefficient calculatedfor general approval beliefs (i.e., .74, .77, and .82 for Grade 3, Grade 4, andGrade 5, respectively). These reliability estimates were consistent withthose reported by Huesmann and Guerra (1997).

In the present study, we used scale mean scores for general aggressionand for situational aggression associated with both a weak and a strongprovocation (all three scores could range from 0 to 3).

Social Information-Processing Patterns

We assessed (a) children's interpretation of peers' motives in hypothet-ical situations, (b) their mental access of behavioral responses to each ofthe same situations, and (c) evaluation of aggressive solutions to hypothet-

ical problems that the interviewer provided and that children evaluatedwith respect to their possible instrumental or social repercussions. Twoseparate instruments were used in the first interview year (i.e., whenchildren ended Grade 3), whereas a single instrument was used in each ofthe two successive years (i.e., at the end of Grades 4 and 5). The instru-ments varied across years to depict developmental^ appropriate socialsituations.

Children were asked to imagine being personally involved in hypothet-ical scenarios. Stories depicted either ambiguous situations of harm orprovocation (e.g., hit in the back by a ball), problematic peer-group entry(e.g., not being allowed to join a group playing basketball), obstruction ofthe child's desired goal (e.g., while in line for recess, a peer demands thechild's spot), or situations in which the child interacts with adults or peersfor being wrongly accused of something (e.g., throwing rocks in the streetor stealing something from a store). Studies investigating the links betweendeviant processing of social cues and conduct problems have shown thatthese types of situations are particularly relevant and problematic forelementary school-age children (see Dodge, 1993 for review; also Dodge,McClaskey, & Feldman, 1985; Dodge et al., 1995).

Home Interview With the Child. This instrument provided measures ofchildren's intent attributions and response access. After each of eightstories, third graders verbalized the reasons for why the peer in the storyacted the way he or she did. For each story, the interviewer recordedimmediately the child's response by indicating 1 (a benign intent expla-nation), 2 (a non-applicable/don't know explanation), or 3 (a hostile intentexplanation). The 8-item reliability coefficient calculated for dichotomizeditems (i.e., whether the child made a hostile attribution) was high (a = .80).In the current study, we used four separate hostile intent scores, each ofwhich was the sum of two (a potential provocation and a group-entryconflict) nonoverlapping stories (i.e., each score could range from 2 to 6).

For each story, the child then verbalized possible ways to respond to thehypothetical situation. Again, the child's verbalizations were coded imme-diately *a's to whether the child would choose to do nothing, ask why/askagain, command the peer, threaten/seek adult punishment, or directlyretaliate. The last two types of responses were conceived to indicateaggressive solutions, and 8-item reliability for dichotomized responses(i.e., whether a response generated by the child was coded as aggressive)was high (a = .74). As in the intent attribution scoring, we created fourtwo-story scores, each of which indicated the number of times a coding ofdirect retaliation or adult punishment was assigned across the two stories(i.e., each score could range from 0 to 2). In this scoring, the pairing of thestories matched with the pairing used in the four intent attribution scores.

For both processing variables, a second trained interviewer coded theresponses of a randomly selected subsample of 100 children (i.e., 25 foreach of the four geographic sites). We estimated interrater agreement byusing a kappa coefficient, thus correcting for skewness in the distributionof the coding classification. As in past studies (Pettit; Dodge, & Brown,1988), we found quite substantial interrater agreement for both intentattribution (K = .94) and response access (K = .92).

Things That Happen to Me. This instrument provided a measure ofthird graders' response-evaluation patterns. After each of eight stories,third graders were presented with an aggressive solution and asked toindicate by yes or no whether this solution would be instrumental inobtaining a desired outcome, whether it would affect the child's friendshipwith the peer in the story, and whether it would be accepted by otherchildren in general. For each story, we summed the positive (yes) re-sponses. We then created four response evaluation scores, each of whichwas calculated by summing across two (one provocation and one group-entry failure) nonoverlapping stories (i.e., each score could range from 0 to6). Reliability across the eight stories was high (8-item a = .89).

What Do You Think. This instrument provided data on all three pro-cessing variables (i.e., intent interpretation, response access and responseevaluation) for fourth and fifth graders, and children were presented withthe same six stimulus stories at each year.

BELIEFS AND SOCIAL INFORMATION PROCESSING 155

After each story children were first provided with a benign and a hostileintent account of the event, and asked to indicate the extent to which eachaccount motivated or elicited the event. Responses were reported on a5-point likelihood rating scale ranging from 1 (yes, definitely) to 5 (no,definitely not). We first reversed the hostile intent rating score for eachstory and then created three hostile intent scores, each of which was theaverage rating across two nonoverlapping stories (i.e., each score couldrange from 1 to 5). Reliability was acceptable across the six stories, 6-itemas = .52 and .53 for Grades 4 and 5, respectively.

For each story, children then verbalized a possible way to respond to thehypothetical problem. Children's responses were coded into one of eightpossible response categories (i.e., direct aggressive, assertive/competent,authority-punish, authority-intervene, passive-avoidant, self-control, other,and irrelevant/unable to respond), and aggressive response access scores werethen calculated exclusively on the basis of either direct aggressive orauthority-punish coding classification. Alpha reliability for six dichotomizeditems (i.e., whether an aggressive category was used) was moderately high ateach year (as = .66 and .69 for Grades 4 and 5, respectively). For the presentstudy, we created 3 two-story scores, each of which indicated the number oftimes either an aggressive or an authority-punish coding was assigned acrosstwo nonoverlapping stories. Scores were created across the same three pairs ofstories represented in the hostile intent scores (i.e., each score could rangefrom 0 to 2). The interrater agreement for this coding classification wasextremely high (K = .93 for both school years).

Finally, for each story, the interviewer provided an aggressive solutionto the situation at hand, and the child evaluated it with regard to whetherhe or she could enact it, whether it would lead to a desired outcome, andwhether enacting it would lead to acceptance by peers. Children's re-sponses were recorded on a five-point Likert scale ranging from 1 (yes,definitely) to 5 (no, definitely not), and then reversed scored so that higherscores would indicate a stronger judgment that aggression would lead topositive outcomes (6-item as = .68 and .65 for Grades 4 and 5, respec-tively). Again, we created three response evaluation mean scores, each ofwhich was the average rating across two randomly selected stories (i.e.,each score could range from 1 to 5).

Results and Discussion

We used structural equation modeling with latent variables andperformed a series of confirmatory factor analyses at each year ofmeasurement. We first estimated the measurement parameters andgoodness of fit of a four-factor model that distinguished among thefactors of Aggression Beliefs, Hostile Intent Attributions, Accessingof Aggressive Responses, and Evaluation of Aggressive Solutions.We then adopted a nested-model approach and further evaluated themeasurement validity of the four-factor model by examining theextent to which there would be a decline in model fitting as a result ofhypothesizing (a) no distinction among the three social information-processing factors (i.e., a theory-driven two-factor model), (b) a singlegeneralized cognitive factor (i.e., a theory-driven one-factor model),and (c) no distinction between aggression beliefs and the single socialinformation-processing factor that appeared to be most highly corre-lated with Aggression Beliefs (i.e., a data-driven three-factor model).

Analysis of Measurement Validity

For each of the 3 years, we used confirmatory factor analysis totest a measurement model that (a) allowed each item to loadexclusively on its hypothesized factor, (b) included a normativebelief latent factor and three social information-processing latentfactors (i.e., Intent Attributions, Response Access, and ResponseEvaluation), and (c) posited interfactor correlations among latent

constructs, as well as correlated error terms across same-storyindicators (i.e., to control for possible systematic variance associ-ated with stimulus stories). Table 1 shows the standardized factorloadings, the measurement error terms, and goodness-of-fit indicesfor the four-factor solutions obtained in each year.

All four latent factors appear to have been well measured at eachyear. All the standardized loadings for each latent factor at eachyear are statistically nonzero. Across the three models, item load-ings ranged from .40 to .90, and 77% of them were .60 or greater.The fitted estimates for each model could substantially reproducethe patterns of correlations observed in its respective year, thussubstantiating measurement validity. In particular, despite statisti-cally significant chi-squares (sensitive to relatively large samplesizes), the ratio between each chi-square statistic and its degrees offreedom in each case was close to or smaller than 2, suggesting anadequate fit (Bollen, 1989; Marsh, Balla, & McDonald, 1988).Likewise, each model's goodness-of-fit index was .96 or higher,and each model's unstandardized root-mean-square residual (i.e.,an averaged index of discrepancy between the observed and esti-mated pattern of covariances) was .05 or smaller, again suggestingvery adequate model fitting. Thus, the four-factor model appears torepresent the data quite well in each year of measurement.

We then compared each year's four-factor solution to each ofthree alternative models nested within the hypothesized model(i.e., nested models are models that are hierarchically related toone another in the sense that particular parameters that are freelyestimated in one model are then fixed to some predetermined valuein a second model). These alternative solutions posited fewer thanfour factors, and their plausibility was evaluated in terms of theresulting.- changes in model chi-square (i.e., changes in modelfitting). We hypothesized that these alternative solutions wouldprovide significantly worse model fit and therefore expected sta-tistically significant increases in model chi-square or a decline ingoodness of fit. The bottom of Table 1 lists the changes in modelchi-square that we obtained for each year when we contrasted thehypothesized four-factor model with its nested variations.

The case of a single generalized cognitive factor was consideredfirst. We fixed to one each of the six covariance paths linking the fourlatent factors, thus testing the hypothesis of a single source of sys-tematic variance in children's cognitive scores (i.e., or, equivalently,the null hypothesis of no differences between within- and across-construct correlations). As Table 1 shows, this model yielded adramatic increase in chi-square and represented a statistically signif-icant decline in model fitting relative to the hypothesized four-factormodel. Next, we considered the measurement validity of a two-factormodel comprising exclusively a latent factor of aggression beliefs anda latent factor of deviant processing that did not distinguish amonghostile attributions, response access, and response evaluation (i.e., seethe two-factor model statistics in Table 1). Accordingly, we fixed toone the three covariance paths linking the three processing factors.Again, the model fit dramatically worsened relative to the hypothe-sized four-factor solution.

Finally, we examined changes in model fitting as a result ofmodifying the model in line with empirical rather than theoret-ical criteria. In particular, we fixed to one the covariance pathlinking the latent factor of Aggression Beliefs to the processinglatent factor for which we obtained the highest estimate ofinterfactor correlation. As such, this nested model represented aquite conservative test of the measurement validity of the

156 ZELLI ET AL.

Table 1Standardized Factor Loadings and Indicator Error Terms, as Well as Correlations of LatentFactors, Goodness-of-Fit Indices (GFIs), and Changes in Chi-Square for the Four-FactorModels Tested in Grades 3, 4, and 5

Factor

Aggression BeliefsGeneral approvalSituational approval

Weak provocationStrong provocation

Intent Attributionsabcd

Response Accessabcd

Response Evaluationabcd

Between Agression Beliefs and:Intent AttributionsResponse AccessResponse Evaluation

Between Intent Attributions and:Response AccessResponse Evaluation

Between Response Access and:Response Evaluation

)C(df)GFIRMSR

^-change test (df)One-factor model (6)Two-factor model (3)Three-factor model (1)

Grade 3{N = 331)

Standardized factor loadinj

.57 (.81)

.77 (.64)

.70 (.71)

.59 (.81)

.57 (.82)

.66 (.75)

.65 (.76)

.66 (.75)

.71 (.71)

.66 (.75)

.73 (.68)

.77 (.64)

.71 (.71)

.85 (.53)

.85 (.53)

Grade 4(N = 340)

.63 (.77)

.83 (.56)

.68 (.73)

.45 (.89)

.62 (.78)

.43 (.90)—

.71 (.71)

.76 (.65)

.43 (.90)—

.79 (.62)

.90 (.45)

.72 (.69)—

Within-year interfactor correlations

.15

.16*

.52**

.47**

.09

.15

Goodness of fit

106.1 (80).96.04

641.4524.3116.4

.54**

.44**

.52**

.66**

.67**

.50**

86.8 (39).96.02

270.1120.619.6

Grade 5(N = 328)

.70 (.71)

.77 (.64)

.70 (.71)

.51 (.86)

.73 (.68)

.40 (.92)—

.75 (.66)

.73 (.68)

.52 (.85)—

.82 (.57)

.85 (.53)

.69 (.72)—

.17*

.58**

.50**

.37**37**

.57**

62.8 (39).97.01

317.7167.789.4

Note. Error terms for the measurement loadings and x degrees of freedom are listed in parentheses; a, b, c,and d = two-story item scores; RMSR = root mean square residual. Dashes indicate the lack of a fourthtwo-story item score for Grade 4 and Grade 5 (i.e., in these 2 years, social information-processing patterns wereassessed across six, rather than eight, stories).* p < .05. **p < .01.

hypothesized four-factor structure, in that it minimized thechances of detecting a statistically significant decline in chi-square statistics. As one can see from Table 1, the highestinterfactor correlation estimates were with Response Evaluationat Grade 3 (.52), with Hostile Intent Attributions at Grade 4(.54), and with Response Access at Grade 5 (.58). Despite thesemoderately high correlations, fixing the respective covariancepaths to one significantly worsened the model fitting at eachyear (i.e., three-factor model statistics in Table 1).

Testing Moderating Gender and Ethnicity Effects

We next examined whether the same four-factor model similarlyrepresented the relations observed in each gender and ethnicitygroup (thus testing for moderating effects). Child ethnicity wasdichotomously classified, with "other" ethnicity groups excluded.We again employed structural equation modeling confirmatoryfactor analysis. For each child characteristic, we estimated thefour-factor model parameters across the two groups simulta-

BELIEFS AND SOCIAL INFORMATION PROCESSING 157

neously. This simultaneous parameter estimation was performedtwice. In the first analysis, measurement loadings and interfactorcorrelations were set as equality constraints. This amounted totesting whether the model's estimates could reliably reproduce thecorrelation matrix of each group. We then relaxed these constraintsand tested the four-factor model again. As such, these analysestested two nested models, and could therefore be compared interms of changes in model's chi-square.

For Grade 3 data, we obtained nonsignificant changes in modelchi-square for both gender, equality constraint, ^(ISl, N = 331)= 198.6, no equality constraint, / (160 , N = 331) = 185.07,A^(21, TV = 331) = 13.57, and ethnicity, ^ ( 181, TV = 320) =218.0, A^(160, N = 320) = 194.2, and A^(21, N = 320) = 23.8.We also obtained nonsignificant changes in model chi-squarewhen we evaluated differences between boys and girls in Grades 4and 5, model A * ^ , N = 340) = 28.1 and A/(18 , N = 328)= 24.7, respectively. In these two latter years, the analysis of childethnicity differences yielded significant results, Ax^lS, TV = 325)= 57.4, p < . 0 0 1 for Grade 4, and Ax2(18,A^= 317) = 32.9, p<.01 for Grade 5, respectively. In each case, however, only 1 ofthe 18 estimates that were fixed (and then relaxed) seemed to beresponsible for such a significant change in model chi-squares. Inparticular, the interfactor correlation between Aggression Beliefsand Response Access was .55 for African American fourth grad-ers, whereas it was .01 for Caucasian fourth graders; likewise, theinterfactor correlation between Aggression Beliefs and HostileAttributions was .26 for African American fifth graders, whereas itwas —.06 for Caucasian fifth graders. We reasoned that this resultmay have reflected measurement fluctuations and concluded that,in general, the same four-factor model's estimates could reliablyreproduce the observed correlations in the two ethnic groups (i.e.,we found no evidence of moderating effects).

In sum, we found clear support for a model positing distinctionsamong the cognitive constructs of normative beliefs legitimizingaggression, hostile attributional biases, mental accessing of aggres-sive responses, and evaluation of aggressive solutions to hypothet-ical social problems. This support was exemplified not only in thismodel's adequate measurement parameters and goodness of fit ateach of three separate years, but also in its greater effectiveness inrepresenting the factor structure underlying the data relative toalternative nested solutions. Finally, we found little evidence sug-gesting that the bivariate relations between aggression beliefs anddeviant processing factors varied significantly with child charac-teristics such as gender and ethnicity.

STUDY 2

Despite corroborating the measurement validity of distinguish-ing between beliefs legitimizing aggression and deviant processingpatterns, Study 1 provided no information about how aggressionbeliefs and deviant processing operate together in influencingaggressive behavior and whether deviant processing mediates theeffects that beliefs exert on aggression, as hypothesized.

In the present study, we addressed these issues and tested thespecific hypotheses that (a) early beliefs approving of aggressionwould predict later aggression (thus seeking a replication of part ofHuesmann and Guerra's, 1997, findings), (b) early beliefs approv-ing of aggression would predict later on-line information-processing associated with interpreting, responding, and evaluat-

ing responses to a current social event, (c) deviant processing earlyon would predict aggressive responding at a later time (thusseeking a replication of findings reviewed by Crick & Dodge,1994), and (d) the longitudinal relation in (a) is partially mediatedby the relations in (b) and (c) when deviant processing is assessedat an interpolated time. As a corollary goal, we also exploredwhether these hypothesized relations would appear for both situ-ational (i.e., retaliation) and general approval beliefs. We thereforeperformed analyses separately for the two types of beliefs.

Finally, we evaluated the merits of a reversed sequence ofeffects in which aggression beliefs mediate the effects of deviantprocessing on later aggressive behavior. We expected this test toyield null findings.

Method

Participants

The participants in this study belonged to the same pool of 387 norma-tive sample children described in Study 1. In the present study, weanalyzed data for two samples; namely, the sample of 300 children (80%of the total) who completed cognitive interviews at Grades 3, 4, and 5, anda mediation sample of 279 children (72% of the total) for whom we alsoobtained Grade 5 self-, parent, and teacher ratings of aggressive behavior.Preliminary analyses yielded no significant relation between attrition andvariables of interest.'

Cognitive Assessment

As in the first study, we relied on children's measures of normativebeliefs legitimizing aggression and measures of hostile intent attribution,response access of aggressive solutions, and response evaluation. For childaggression beliefs, we again used the three scores described in Study 1 plusa legitimacy of retaliation score that represented average ratings acrossthe 12 items of weak and strong provocation. For each of the threeprocessing factors, we again used the two-story item scores described inStudy 1. In addition, however, we also calculated scale scores that eitheraveraged or summed across these two-story items. In particular, mean scalescores were calculated for hostile attributions at each year, and for responseevaluation for two of the years (i.e., Grades 4 and 5). Item scores wereinstead summed for response access at each grade so that the score wouldindicate the number of times either an aggressive or an adult-punish codingwas assigned across stories (i.e., the Grade 3 score could range from 0 to 8,whereas scores for the next 2 years could range from 0 to 6). Finally, itemscores were also summed for Grade 3 response evaluation, and this score

1 The 279 fifth graders who composed the mediation sample appeared torepresent the full sample with respect to child aggressive behavior. Inparticular, the score distributions and levels of child aggression (as mea-sured by parent, teacher, and self-report) for this sample were compared tothe respective indices calculated for those fifth graders who were excludedfor lack of data on the cognitive variables of interest. Attrition did notappear to restrict the range of aggression scores; that is, a test for equalityof variances yielded nonsignificant results for self-ratings, F(l,327) = 1.10; parent ratings, F(l, 335) = 0.002; and teacher ratings, F(l,326) = 0.39, all ps > .30. Nor did attrition appear to affect the mean levelsof aggression scores of the two groups. In fact, aggression appeared to beslightly higher in the mediation sample than in the excluded group for selfratings (M = .18 vs. M = .13) and teacher ratings (M = 9.9 vs. M = 8.6).However, there were no statistically significant differences on aggressionratings between groups, t(\) = —1.57 for self-ratings, t(l) = 0.27 forparent ratings, and r(l) = —0.74 for teacher ratings, all ps > .10.

158 ZELLI ET AL.

indicated the extent to which a child evaluated aggression as having animpact on the stimulus stories (score could range from 0 to 24).

Assessment of Child Behavior Problems

Teacher Ratings

When the children completed fifth grade, teachers completed theTeacher Rating Form (TRF) of the Child Behavior Checklist (Achenbach& Edelbrock, 1986), one of the most commonly used checklists of childbehavior problems. For each of 112 behavioral items, teachers responded 0if the problem statement was not true for the child, 1 if it was somewhattrue, and 2 if it was very or often true. The TRF rating instrument is scoredon several dimensions of psychological or conduct problems, includingaggression. Analysis of nationally normed 3"-scores (M = 50, SD =10) forthese scales has yielded high 15-day test-retest reliability, 2-month stabil-ity, cross-teacher agreement, and validity (Achenbach, 1991; Achenbach &Edelbrock, 1986). The present report relied exclusively on the TRF Ag-gression Scale score.

Parent Ratings

During the summer that followed Grade 5 school year, parents com-pleted the Child Behavior Checklist (Achenbach & Edelbrock, 1986). Thenumber of items and item rating scales of this instrument are similar tothose described for the TRF instrument. Reliability and validity of thisparent instrument have aiso been amply demonstrated over the years(Achenbach, 1991). The present report relied exclusively on the analysis ofAggression Scale score.

Self-Report Ratings of Aggression

As part of the child summer interview, fifth graders answered 27questions concerning delinquent activities. Most of the delinquency items

were derived from the National Youth Survey, and their psychometric andtheoretical utility has been documented over the years (Elliott. Ageton, &Huizinga, 1985; Elliott, Huizinga, & Menard, 1989). The present reportrelied exclusively on the analysis of five items concerning acts of aggres-sion perpetrated against other persons. In particular, children were asked toindicate how many times during the last year they have "hit, slapped, orshoved other kids or gotten into a fight," "threatened to hit someone inorder to get something," "attacked someone," "thrown objects such asrocks or bottles at people," and "been involved in a gang fight." Becauseof extremely high positive skewness in the frequency distributions of theitems, a scale score was calculated by first dichotomizing each item as towhether the child committed the depicted act (i.e., scores greater than zerowere coded as 1), and then averaging across the five dichotomized scores.Each scale score thus represented the proportion of acts a child reportedhaving committed during the past year (M = .17, SD = .21, 5-item a =.64, for 329 fifth graders who provided self-report ratings of delinquency).

Results and Discussion

Table 2 lists the interscale correlations between the two types ofaggression beliefs and the three processing variables of intentattribution, response access, and response evaluation within andacross the three measurement points. The table also shows thedescriptive statistics for each variable and the year-to-year stabil-ities for aggression beliefs.

Group Differences in Aggression Beliefs

Overall, children did not believe aggression is a highly accept-able course of action (i.e., for each year, the two belief mean scoreswere less than 1). However, children endorsed aggression rela-tively more when it is a response to a prior provocation. We

Table 2

Scale Scores' Descriptive Statistics, Correlations of General and Retaliation Approval Beliefs

With the Processing Variables of Hostile Intent, Aggressive Response Access, Aggressive

Response Evaluation Within and Across Time, and Stability of Normative Beliefs

Variable and year

Intent attributionsGrade 3Grade 4Grade 5

Response accessGrade 3Grade 4Grade 5

Response evaluationGrade 3Grade 4Grade 5

M (SD)Aggression beliefs' stability coefficients

Grade 3Grade 4Grade 5

General

Grade 3

.01

.17**

.06

.07

.09

.23**

.38**

.21**27**.28 (.39)

—.18**.31**

approval of a

Grade 4

.0429**.12*

.13*22**.18**

.08

.27**24**.24 (.33)

—.34**

ggression

Grade 5

.01

.19**

.09

.01

.15**

.36**

.16**

.14**

.36**

.23 (.33)

—

Beliefs

Grade

.11*

.26**

.18**

.11*

.21**

.25**

.39**

.27**

.20**

Approval of retaliation

3 Grade 4

.04

.36**

.05

.09

.32**

.13*

.21**

.43**

.17**.64 (.60) .65 (.53)

.44**3 7**

—.40**

Grade 5

.09

.17**

.08

.10*

.18**

.37**

.15**

.17*

.32**

.68 (.54)

—

M

4.502.74 •2.76

1.74.65.57

5.902.312.42

SD

1.010.650.60

1.711.081.07

5.15.70.62

Note. N = 300 for all statistics.*p < .05, one-tailed. ** p < .01, one-tailed.

BELIEFS AND SOCIAL INFORMATION PROCESSING 159

performed a repeated measure analysis of variance to examinewhether there were developmental, gender, or ethnicity groupdifferences in aggression approval.2 As one can see in Table 2,aggression approval did not change much as children grew older,F(2, 576) = 1.90 for general approval, and F(2, 576) = 0.86 forretaliation approval. This (lack of a) trend did not significantlyvary with child gender or ethnicity. When averaged over measure-ment points, however, aggression approval varied with child gen-der, F(l, 288) = 4.35, p < .03 for general approval, and F(\,288) = 5.77, p < .02 for retaliation approval, and with childethnicity, F(l, 288) = 9.08, p < .001 (for general approval); F(l,288) = 19.38, p < .001 (for retaliation approval). Boys condonedgeneral aggression (M = .28) and retaliation (M = .73) signifi-cantly more than did girls (M = 0.22 and M— 0.61, respectively).Likewise, African American children legitimized both aggression(M = .30) and retaliation (M = 0.77) significantly more than didCaucasian children (M = 0.21 andM = 0.56, respectively). Theseresults are consistent with gender and ethnicity differences foundin past aggression research (Guerra, Huesmann, Tolan, Van Acker,& Eron, 1995; Huesmann & Guerra, 1997).

Stability of Individual Differences in Aggression Beliefs

Individual differences in normative beliefs appeared to be rela-tively stable across the late years of elementary school. Childapproval beliefs concerning retaliation seemed to be relativelymore stable than child general approval beliefs. This seemedparticularly so for the stability estimated between the first twomeasurement points (i.e., r = .44 and r = .18, respectively), andthis result was virtually identical to the stability results Huesmannand Guerra reported for their Grade 4 sample (Huesmann &Guerra, 1997, p. 413). Given that retaliation beliefs and generalapproval beliefs were quite comparable in scale reliability (seeStudy l's Method section), the higher stabilities for retaliationbeliefs could reflect greater stability in true scores, and indicatethat children may first acquire relatively firm beliefs about theappropriateness of situational aggression and then consolidate theirbeliefs about the appropriateness of aggression in general onlysome time later.

Bivariate Relations Between Aggression Beliefs andDeviant Processing

As we hypothesized (and anticipated after the analyses reportedin Study 1), greater approval of aggression was significantly as-sociated with higher (more deviant) processing. As one can see inTable 2, 14 of the 18 within-year correlations were statisticallysignificant (i.e., one-tailed test). Descriptively, this pattern of bi-variate correlations was especially evident for the relations be-tween both types of aggression beliefs and response evaluationscores (i.e., all six of the correlations were significant in this case,versus five of six and three of six of the coefficients concerningresponse access and hostile intent attributions, respectively). Thepattern of correlations was also particularly evident for retaliationbeliefs (i.e., eight of nine correlations were significant in this case)relative to general approval beliefs (i.e., six of nine correlationswere statistically significant). Thus, the more children believedthat aggression is okay, the more they engaged in deviant process-ing, and this relation was especially evident for individual differ-

Tabie 3Descriptive Statistics of Grade-5 Aggression and Its Concurrentand Longitudinal Bivariate Correlations With AggressionApproval Beliefs and Deviant Processing

Variable and year

Aggression beliefsGeneral approval

Grade 3Grade 4Grade 5

Retaliation approvalGrade 3Grade 4Grade 5

Deviant processing patternsIntent attributions

Grade 3Grade 4Grade 5

Response accessGrade 3Grade 4Grade 5

Response evaluationGrade 3Grade 4Grade 5

MSD

Teacherratings

.04

.06

.21**

.18**

.16**

.16**

.11**

.19**

.21**

.07

.34**

.33**

.08

.12*

.18**9.92

11.89

Grade-5 aggression

Parentratings

.03

.04

.16**

.02

.03

.07

.08

.06

.13**

.08

.03

.18**

.00

.02

.11*7.606.90

Self-reportratings

.12*

.05

.24**

.17**

.09

.28**

.14**

.07

.24**

.18** -

.18**

.33**

.17**

.12*

.24**

.18

.14

Note. N = 279 for all statistics.* p < .05, one-tailed. ** p < .01, one-tailed.

ences in retaliation beliefs and differences in how children evalu-ated the outcomes of behaving aggressively.

These patterns were further qualified by positive across-year(i.e., longitudinal) correlations. Twenty-six of 36 coefficients werestatistically significant, and most of these coefficients (i.e., 15 of26) represented the longitudinal effect that early beliefs possiblyexerted on later deviant processing. The magnitude of these lon-gitudinal correlations between early aggression beliefs and laterdeviant processing was also somewhat stronger than those corre-lations linking early deviant processing and later aggression beliefs(i.e., correlations ranged from .12 to .27, median r = .21 in the firstcase, and from .10 to .21, median r = .16 in the latter case).Children who endorsed aggression early on were more likely toengage in deviant processing at a later time. The opposite case didnot seem to hold as strongly.

Bivariate Relations Between Cognitive Patterns andAggressive Behavior

We next examined the relations between cognitive patterns andthe three measures of child aggressive behavior. Table 3 lists thedescriptive statistics for Grade 5 aggression ratings and interscale

2 Ethnicity group differences were tested only for African American andCaucasian children. As a result, 8 of the 300 cases who had completed3-year data were removed from the group difference analysis.

160 ZELLI ET AL.

correlations. Overall, we substantiated the findings reported byHuesmann and colleagues (Huesmann & Guerra, 1997; Huesmannet al., 1992) with respect to the positive correlations betweenaggression beliefs and aggressive behavior. Nine of the 18 coef-ficients were statistically significant and in the expected positivedirection. These coefficients were modest or low in size, theconcurrent (i.e., Grade 5) correlations were among the highest, andparental ratings of child aggressive behavior were the least likelyto be correlated with child aggression beliefs.

Social information-processing patterns were also significantlycorrelated with aggressive behavior. Eighteen of the 27 correla-tions were significant and in the expected direction, includingcorrelations for all three aspects of processing. Overall, thesefindings are consistent with what is commonly reported in thedevelopmental literature on child aggression (Crick & Dodge,1994; Guerra et al., 1995).

A Mediational Model Linking Normative Beliefs,Deviant Processing, and Aggressive Behavior

We then turned toward our core mediation hypothesis thataggression beliefs would exert an influence on aggressive behaviorthrough the intervening effect of deviant processing. We used astructural equation modeling analysis, and performed both amediation-effect analysis and a direct-effect analysis according toa three-wave panel design. Illustratively, the mediation effectanalysis tested two longitudinal regression models simultaneously;one model in which Grade 3 Aggression Beliefs, Grade 4 HostileAttributions, Grade 4 Access of Aggressive Responses, andGrade 4 Evaluation of Aggressive Responses predicted Grade 5aggressive behavior, and a second model in which Grade 3 Ag-gression Beliefs predicted each of the three Grade 4 deviantprocessing factors. The direct-effect analysis was identical exceptthat the first regression model did not include the regression pathsfor the three deviant processing variables. As such, we tested twonested models, and we were interested in comparing the overall fitof the two models, in examining whether the direct effect modelwould produce a substantial increase in the coefficient estimatedfor the path linking beliefs to aggression, and in indexing thechanges in model fit across the two structural equation models.

The measurement model was identical in the two analyses. Inparticular, hostile attribution, response access, and response eval-uation were measured by the 3 two-story indicators that we used inthe confirmatory factor analysis presented in Study 1. Retaliationbeliefs were measured by the two approval scores concerningone's retaliation to a prior weak and strong provocation, respec-tively. In separate analyses, general approval beliefs were mea-sured by two scores representing the averaged ratings across twosets of 4 randomly selected items, respectively (r = .58, p < .001between the two scores). Finally, child aggression was measuredby Grade 5 teacher, parental, and self-report ratings (r = .38 andr = .34 for the bivariate relations of teacher ratings with parentratings and self-ratings, respectively, and r = .24 for the bivariaterelation between parent and self ratings, allps < .001). By virtueof measuring aggressive behavior from multiple sources, any lon-gitudinal effect on aggressive behavior would exclusively tap intocommon variance among the three measurements, thus controllingconceivably for method or context confounds (e.g., school).3

The two structural equation modeling analyses were first per-formed for retaliation beliefs. The factor loadings and error termsfor the measurement model are listed in the Appendix. The medi-ation model's latent path estimates, interfactor correlations amongthe three latent processing constructs, and goodness-of-fit indicesare presented in Figure 1. As one can see, stronger beliefs aboutthe legitimacy of retaliating in Grade 3 significantly predictedhigher (more deviant) Hostile Attributions, Accessing of Aggres-sive Responses, and positive Evaluation of Aggressive Solutions ayear later (/3s = .41, .27, and .33, respectively, all ts > 3.2 andps < .001). Likewise, greater access of aggressive responses fromlong-term memory uniquely predicted higher ratings of aggressivebehavior almost a year later (J3 = .44). Finally, a stronger belief inthe legitimacy of retaliating did not significantly predict higherratings of aggressive behavior in Grade 5 when its effects weretested controlling for mediation (/3 = .15). Overall, the mediationmodel appeared to fit the data quite well, as a relatively small )f:dfratio (i.e., 91.5:58 = 1.58, Marsh et al., 1988) and a close-to-onegoodness-of-fit index would suggest.

As expected, relaxing (i.e., not estimating) the paths linkingdeviant processing to aggressive behavior led to significantchanges in model estimation. The model fit worsened (e.g.,goodness-of-fit index = .94 vs. .96, and unstandardized root-mean-square residual = .32 vs. .12), and the regression coefficientfor the (total) effect of retaliation beliefs'on aggression nearlydoubled and became highly significant (j3 = .29, p < .01, vs. jS =.15), as did the model's chi-square, that is, X*(6\, N = 279) =115.6, vs. ^ (58 , N = 279) = 91.5. Finally, the mediation modelfit the data significantly better than did the direct effect model,A^(3, JV = 279) = 24.1, p < .001.

Thus, the results clearly indicated that the effect of retaliationbeliefs on aggressive behavior could be explained in the context ofa mediation-effect model significantly better than in the context ofa direct-effect model. The processing step of accessing aggressiveresponses from long-term memory contributed uniquely to thismediation, and nearly 50% of the effect of retaliation beliefs onaggressive behavior was mediated through deviant processingwhen hostile attributions, response access, and response evaluationwere considered as a set (i.e., tested simultaneously).

As one could have anticipated after examining the bivariaterelations between beliefs and aggressive behavior, this pattern ofpositive results did not hold for general approval of aggression. Astronger belief that aggression in general is acceptable did notsignificantly predict more child aggressive behavior nearly 2 yearslater (j3 = .10), t = 1.2. Without such a finding, a test of howdeviant processing variables would mediate—and reliably re-duce—such a direct effect was deemed unnecessary. Strongerbeliefs at the end of Grade 3, however, still significantly anduniquely predicted higher (more deviant) Hostile Attributions(j8 = .24), t = 2.33, and positive Evaluation of Aggressive Re-sponses O = .27), t = 3.47, a year later.

Test of Moderation by Gender and Ethnicity

As we did in the case of testing measurement validity (see Study1), we examined the extent to which our mediation model for

3 We thank an anonymous reviewer for highlighting the need to correctfor possible school effects.

BELIEFS AND SOCIAL INFORMATION PROCESSING 161

/ Grade 3 \I Retaliation !\ Beliefs /

.15 (.29**)

.41

/ Grade 4 \i Hostile Intent |\ Attributions

.27 :

.06

Grade 5 \—>; Child )

\ Aggression /

.33**

.44'

/ Grade 4 \Mental Access 1

\

/ Grade 4 \Positive \Evaluation of j

Figure 1. Mediation model: Longitudinal effect of retaliation approval beliefs on child aggression is mediatedthrough patterns of deviant processing. Latent path coefficients are presented in standardized units. Thecoefficient in parentheses is the latent "total effect" coefficient. Model ;r(58, N = 279) = 91.5, p £.001. Goodness-of-fit index = .96. *p< .05. ** p < .01.

retaliation beliefs could account for the relations specific to eachgender and ethnicity group (i.e., the possibility of moderatingeffects). We again adopted a nested model approach, and indexedchanges in model chi-square between a model that assumed iden-tical estimates across groups (i.e., a model with equality con-straints) and a model that relaxed such an assumption (i.e., a modelwithout equality constraints). The results again clearly indicatedthat our mediation model could represent the relations in eachgroup equally well, and that therefore neither gender nor ethnicitysignificantly moderated such relations: ethnicity, A^2(17, N =272) = 13.28, and gender A/(17, N = 279) = 18.73 (p > .20for both statistics).

Test of Alternate Models

Finally, our findings in support of a mediation model for retal-iation beliefs do not appear to merely reflect statistical artifact ormulticollinearity among measurements. Such a conclusion waswarranted in light of a series of analyses in which, again, weindexed the changes in model fitting as a result of omittingmediating paths, but in which we hypothesized the alternative casethat beliefs mediate, rather than are mediated by, the influence ofdeviant processing on aggressive behavior. In the first analysis, wetested a time-impossible model in which Grade 3 retaliation beliefsmediated the effects that Grade 4 deviant processing exerted onGrade 5 aggressive behavior. In the second analysis, we tested anidentical model where variables were, however, timely ordered(i.e., we used Grade 3 deviant processing variables and Grade 4retaliation belief scores). If our previous findings merely reflectedmeasurement artifact or multicollinearity, we reasoned that theinclusion of a mediation path from retaliation beliefs to aggressivebehavior should significantly improve model fitting as indexed bychanges in model chi-square. This however was not the case in

either analysis, A^2(l, N = 279) = 2.5, p > .10 for the time-illogical test, and A ^ ( l , N = 279) = 3.6, p > .05, for the timelyordered test. In the end, these results strongly suggest that ourmodel linking retaliation beliefs to later aggression through inter-vening deviant processing patterns was very plausible, and that itseffectiveness in fitting our longitudinal data was a reliable effectthat could not easily be dismissed as a case of measurementidiosyncracies.

CONCLUSIONS

In much of the current research on how individual differences inaggressive behavior develop and crystalize, researchers have rec-ognized the importance of distinguishing between two sources ofinfluence on behavior; namely, one's organized knowledge and thesocial information-processing operations that one may adopt indealing with a current situation. There also is consensus withrespect to a general hypothesized model in which knowledgeinfluences current behavior by regulating the ways one encodes,interprets, and evaluates responses to any current situation (Crick& Dodge, 1994; Dodge, 1993; Huesmann, 1998). In the currentinvestigation, we tested the validity of both propositions withrespect to two classes of constructs for which prior research hasclearly established their separate influences on child aggressivebehavior. One refers to individual normative beliefs; that is, one'sself-regulating standards about the appropriateness of social be-haviors (Huesmann & Guerra, 1997). The other refers to theinformation-processing judgments one may make at the time ofinterpreting others' intentions, of thinking of possible ways torespond to a current situation, and of evaluating the effectivenessor value of any response option that may come to mind (see Crick& Dodge, 1994). We measured individual beliefs by asking chil-dren to express their opinions about the appropriateness of aggres-

162 ZELLI ET AL.

sive behavior, both in the context of a prior provocation and asidefrom any apparent triggering social event (Huesmann & Guerra,1997). We measured on-line processing by assessing whetherchildren would attribute hostile intent to others' behaviors inhypothetical situations, would primarily think of aggressive re-sponses to such situations, and would positively judge the out-comes of acting aggressively in those situations.

Taken together, the results of these two studies provide clearsupport for the validity of distinguishing between latent knowledgeand processing operations as two mechanisms regulating aggres-sion, and for the validity of a mediation model in which knowledgeeffects operate through proximal mechanisms of processing cur-rent social cues.

In the first study, we used structural equation modeling withlatent factors and performed a series of confirmatory factor anal-yses to examine the measurement validity of a four-factor modelcomprising a construct of Aggression Beliefs, and the three pro-cessing constructs of Hostile Intent Attributions, Accessing ofAggressive Responses, and Evaluation of Aggressive Solutions.The findings of this study suggested that a four-factor modelrepresented a reliable and valid structure to describe the pattern ofstatistical relations among variables we observed in three succes-sive years of measurement. In addition to conforming to conven-tional criteria of adequate model fitting, a model comprising afactor of Aggression Beliefs, and the three processing factors ofHostile Attributions, Response Access, and Response Evaluationalso prevailed over alternative solutions that posited fewer factors,thus demonstrating discriminant validity in the hypothesized con-structs. In particular, the steps of either omitting clear distinctionsamong the three processing variables or considering a singlegeneralized cognitive construct produced a decline in model fittingrelative to the adequacy of our hypothesized four-factor model.

In the second study, we extended our analysis of aggressionbeliefs and deviant processing by also examining how these vari-ables are related to individual differences in aggressive behaviorand whether aggression beliefs influence behavior indirectlythrough the mediating effects of intervening deviant processingpatterns. Several findings emerged from such an analysis. Overall',children did not believe that aggression is an appropriate course ofaction, whether it was evaluated as a response to certain situationalinstigators or aside from prior provocations, and this remainedvirtually unchanged as children grew older. Individual differencesin aggression beliefs appeared relatively stable across the years oflate elementary school, especially with respect to beliefs about theappropriateness of retaliation, thus suggesting that a child maybegin forming individual convictions about social behavior withinthe context of discrete and relatively salient situations.

Novel to the present investigation, the results of the secondstudy demonstrated that beliefs about aggression are linked to andexert an influence on children's patterns of processing currentsocial cues. We hypothesized that a child who believes it isappropriate to act aggressively is also more likely to interpretcurrent encounters as hostile, to readily access aggressive re-sponses or scripts (Huesmann, 1988, 1998) as conceivable solu-tions to the problem at hand, and to misconstrue any or all of thebenefits (instrumental, intrapersonal, and social) that would accruefrom acting aggressively (e.g., such as getting one's way, gainingfriends' respect, or being well-liked). Our findings clearly substan-tiated these hypotheses. Longitudinally, stronger aggression be-

liefs correlated with more deviant processing at the mental steps ofinterpreting, accessing responses to, and evaluating hypotheticalsolutions to a variety of social problems. This pattern of relationsappeared to be stronger than the opposite pattern of a longitudinalinfluence of deviant processing on aggression beliefs, thus docu-menting the plausibility of the hypothesized model of effects. Theresults of a series of structural equation modeling analyses withlatent factors further substantiated our model. Individual differ-ences in both retaliation beliefs and general approval beliefsamong third graders uniquely predicted how children interpretedothers' intents as hostile, how readily they thought of aggressivesolutions (i.e., only for retaliation beliefs), and how positively theyevaluated the outcomes of hypothetical aggressive solutions a yearlater.

Perhaps more important, the SEM findings from Study 2 alsosupported quite clearly a view that deviant processing of currentsocial cues may act as a proximal mechanism through which one'sorganized knowledge affects subsequent behavior (Crick &Dodge, 1994; Huesmann, 1998). Individual differences in retalia-tion approval among third graders predicted individual differencesin fifth graders' aggressive behavior, and nearly 50% of this effectcould be attributed to the intervening effects that one's hostileintent attributions, mental accessing of aggressive responses, andaggressive response evaluation exerted on aggression. This reduc-tion was statistically significant, and a model positing these me-diating effects represented a statistically significant improvementover a model that omitted them. Thus, our findings are clear indocumenting the validity of a theoretical model in which socialknowledge, once established, shapes the manner in which a childprocesses information in current or future interactions. Informationprocessing of discrete social cues, in turn, is proximally responsi-ble for which behavior is ultimately enacted.

These conclusions are warranted, and our findings do not appearto merely reflect measurement collinearity. In fact, when we testedthe statistical plausibility of an alternative model in which beliefsabout retaliation were the mediating mechanism accounting for dieeffects of deviant processing on aggressive behavior, the mediationmodel did not provide a statistically significant improvement inmodel fitting relative to a direct effect model. We also note that ourfindings were quite robust across both studies in that they did notseem to be sensitive to or vary with child characteristics such aschild gender or ethnicity. In Study 1, a model in which AggressionBeliefs, Hostile Intent Attributions, Accessing of Aggressive Re-sponses, and Evaluation of Aggressive Solutions were clearlydistinct constructs could simultaneously and adequately accountfor the bivariate relations observed either in boys and girls or inAfrican American and Caucasian children. Likewise, in Study 2,we found no evidence of gender or ethnicity moderating effects inthat our mediation model for retaliation beliefs could simulta-neously and adequately account for the longitudinal relations ob-served in each group among beliefs, deviant processing, and ag-gressive behavior.

In sum, both studies supported the construct validity of distin-guishing between aggression beliefs and deviant processing oper-ations such as interpreting social situations as hostile encounters,thinking readily of aggressive solutions to a current problem, andevaluating the social benefits of behaving aggressively. The find-ings of Study 1 documented the measurement and discriminantvalidity of the hypothesized distinction. The findings of Study 2

BELIEFS AND SOCIAL INFORMATION PROCESSING 163

demonstrated that the effects of child beliefs on child aggressivebehavior can be understood more adequately within a context ofintervening or mediating processing patterns than within a contextof mere direct effects or alternative causal pathways. They thusdocumented that the four constructs represent relatively distinctmechanisms associated with individual differences in aggressivebehavior.

The findings of the present investigation must be evaluated withrespect to several caveats, however. First and foremost, our inves-tigation relied on correlational methods, and any empirical supportto the causal model linking aggression beliefs to aggressive be-havior through the intervening effects of deviant processing musttherefore be interpreted with some caution. Furthermore, it wouldbe premature to believe that deviant processing is the only mech-anism through which individualistic beliefs about aggression exertan influence on child aggressive behavior, and additional mecha-nisms (e.g., affective states) need to be posited and tested in futureinquiries.

It is equally important to recognize that our investigation did notdirectly address the issue of how aggression beliefs and deviantprocessing contribute to the development of habitual aggression.That is, although our findings can conceivably be integrated in amodel that posits acquired memories as enduring knowledge struc-tures guiding information processing patterns in discrete situations(Crick & Dodge, 1994; Huesmann, 1998), the findings say nothingabout the extent to which aggression beliefs influence changes orgrowth in aggressive behavior. Thus, for instance, it remains to beseen in future inquires whether the mediation model supported inthe present investigation can pass a test in which the hypothesizedeffects on late aggressive behavior are estimated after statisticallycontrolling for differences in early aggression. As a corollary tothis issue, our findings concerning possible differences betweensituational and generalized beliefs about aggression seem sugges-tive of developmental differences in the acquisition of normativebeliefs about the appropriateness of social conduct. Early on, wefound that generalized beliefs about the acceptability of aggressionwere measured less reliably and showed less temporal stabilitythan situational (i.e., retaliation approval) beliefs. By the timechildren ended primary school, however, generalized beliefsreached comparable levels of internal consistency and statisticalstability with the situational retaliation beliefs. Furthermore, al-though individual differences in generalized beliefs did not predictdifferences in aggressive behavior over time, differences in retal-iation beliefs clearly did. On the basis of these longitudinal effects,we were able to test the plausibility of our mediation model onlyfor retaliation beliefs. Thus, we note that generalized beliefs mayconceivably lag behind the development of situational beliefs, andtherefore contribute to the development or maintenance of aggres-sive behavior in different ways or at different time points. Thispossibility highlights another important consideration. We haveshown that the relation between retaliation beliefs and aggressivebehavior is mediated by intervening deviant processing patterns.This is not to say that self-regulating normative beliefs developbefore the acquisition of social information-processing styles.Rather, these cognitive mechanisms and processes exert a recip-rocal influence over time, and the possibility that processing pat-terns might contribute to well-formulated beliefs at younger agesawaits future inquiry.

Finally, it is also important to recognize that the findings of thepresent investigation do not automatically extend to other impor-tant knowledge and processing constructs that contribute to indi-vidual differences in aggressive behavior. Thus, for instance,whether deviant processing adequately accounts for the relationsbetween a child's antisocial knowledge representations of his orher peers (Stromquist & Strauman, 1991) and aggressive behaviorremains a possible goal for future research. Likewise, future effortsmay unveil the extent to which a model of mediating effects alsoapplies to one's goal or outcome orientations, an integral part ofthe processing mechanisms proximally responsible for one's ac-tions (see Crick & Dodge, 1994, for a review).

In the end, the present investigation supports a general model ofsocial behavior which emphasizes how enduring knowledge struc-tures guide and shape one's processing in discrete situations (Hig-gins, 1996; Mischel & Shoda, 1995). With regard to social-cognitive models of aggressive behavior, our findings areconsistent with a model in which child aggression is regulated byindividual beliefs about the appropriateness of social conduct, aswell as by proximal information-processing operations that culmi-nate in choosing to act aggressively.

References

Abelson, R. P. (1981). The psychological status of the script concept.American Psychologist, 36, 715-729.

Achenbach, T. M. (1991). Integrative guide for the 1991 CBCL/4-18, YSR,and TRF Profiles. Burlington: University of Vermont.

Achenbach, T. M, & Edelbrock, C. (1986). Manual for the Teacher'sReport Form and the Teacher Version of the Child Behavior Profile.Burlington: University of Vermont.

Andersen, S. M., Spielman, L. A., & Bargh, J. A. (1992). Future-eventschemas and certainty about the future: Automaticity in depressives'future-event predictions. Journal of Personality and Social Psychol-ogy, 63, 711-723.

Anderson, C. A. (1997). Effects of violent movies and trait hostility onhostile feelings and aggressive thoughts. Aggressive Behavior, 23, 161-178.

Anderson, C. A., & Deuser, W. E. (1993). The primacy of control in causalthinking and attributional style: An attributional functionalism perspec-tive. In G. Weary, F. Gleicher, & K. L. Marsh (Eds.), Control motivationand social cognition (pp. 94-121). New York: Springer-Verlag.

Bandura, A. (1986). Social foundations of thought and action: A socialcognitive theory. Englewood Cliffs, NJ: Prentice Hall.

Bandura, A., Barbaranelli, C, Caprara, G. V., & Pastorelli, C. (1996).Mechanisms of moral disengagement in the exercise of moral agency.Journal of Personality and Social Psychology, 71, 364-374.

Bargh, J. A., Lombardi, W. J., & Higgins, E. T. (1988). Automaticity ofchronically accessible constructs in Person X Situation effects on personperception: It's just a matter of time. Journal of Personality and SocialPsychology, 55, 599-605.

Bargh, J. A., & Pratto, F. (1986). Individual construct accessibility andperceptual selection. Journal of Experimental Social Psychology, 22,292-311.

Battistich, V. A., & Aronoff, J. (1985). Perceiver, target, and situationalinfluences on social cognition: An interactional analysis. Journal ofPersonality and Social Psychology, 49, 788-798.

Bollen, K. A. (1989). Structural equations with latent variables. NewYork: Wiley-Interscience.

Bushman, B. J. (1998). Priming effects of media violence on the accessi-bility of aggressive constructs in memory. Personality and Social Psy-chology Bulletin, 24, 537-545.

164 ZELLI ET AL.

Cantor, N., & Kihlstrom, J. F. (1987). Personality and social intelligence.Englewood Cliffs, NJ: Prentice Hall.

Caspi, A.. Elder, G., & Bern, D. (1987). Moving against the world: Lifecourse patterns of explosive children. Developmental Psychology, 23,308-314.

Catrambone, R., & Markus, H. (1987). The role of self-schemas in goingbeyond the information given. Social Cognition, 5, 349-368.

Cervone, D. (1997). Social-cognitive mechanisms and personality coher-ence: Self-knowledge, situational beliefs, and cross-situational coher-ence in perceived self-efficacy. Psychological Science, 8, 43-50.

Cervone, D., & Williams, S. L. (1992). Social cognitive theory andpersonality. In G. V. Caprara & G. L. Van Heck (Eds.), Modernpersonality psychology: Critical reviews and new directions (pp. 200—252). New York: Harvester Wheatsheaf.

Coie, J. D., Dodge, K. A., & Kupersmidt, J. (1990). Group behavior andsocial status. In S. R. Asher & J. D. Coie (Eds.), Peer rejection inchildhood: Origins, consequences, and intervention (pp. 17—59). NewYork: Cambridge University Press.

Conduct Problems Prevention Research Group. (1992). A developmentaland clinical model for the prevention of conduct disorders: The FASTTrack Program. Development and Psychopathology, 4, 509-527.

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

Crick, N. R., & Dodge, K. A. (1996). Social information-processingmechanisms in reactive and proactive aggression. Child Develop-ment, 67, 993-1002.

Crick, N. R., & Ladd, G. W. (1990). Children's perceptions of the out-comes of aggressive strategies: Do the ends justify being mean? Devel-opmental Psychology, 26, 612-620.

Dodge, K. A. (1980). Social cognition and children's aggressive behavior.Child Development, 51, 162-170.

Dodge, K. A. (1986). A social information processing model of socialcompetence in children. In M. Perlmutter (Ed.), Minnesota Symposiumon Child Psychology (Vol. 18, pp. 77-125). Hillsdale, NJ: Erlbaum.

Dodge, K. A. (1991). The structure and function of reactive and proactiveaggression. In D. J. Pepler & K. H. Rubin (Eds.), The development andtreatment of childhood aggression (pp. 201-218). Hillsdale, NJ: Erl-baum.

Dodge, K. A. (1993). Social-cognitive mechanisms in the development ofconduct disorder and depression. Annual Review of Psychology, 44,559-584.

Dodge, K. A., Bates, J. E., & Pettit, G. S. (1990, December 21). Mecha-nisms in the cycle of violence. Science, 250, 1678-1683.

Dodge, K. A., & Feldman, E. (1990). Issues in social cognition andsociometric status. In S. R. Asher & J. D. Coie (Eds.), Peer rejection inchildhood: Origins, consequences, and intervention (pp. 119-155). NewYork: Cambridge University Press.

Dodge, K. A., & Frame, C. L. (1982). Social cognitive biases and deficitsin aggressive boys. Child Development, 53, 630-635.

Dodge, K. A., McClaskey, C. L., & Feldman, E. (1985). A situationalapproach to the assessment of social competence in children. Journal ofConsulting and Clinical Psychology, 53, 344-353.

Dodge, K. A., Murphy, R. R., & Buchsbaum, K. (1984). The assessment ofintention-cue detection skills in children: Implications for developmentalpsychopathology. Child Development, 55, 163-173.

Dodge, K. A., Pettit, G. S., Bates, J. E., & Valente, E. (1995). Socialinformation-processing patterns partially mediate the effect of earlyphysical abuse on later conduct problems. Journal of Abnormal Psy-chology, 104, 632-643.

Dodge, K. A., Pettit, G. S., McClaskey, C. L., & Brown, M. (1986). Socialcompetence in children. Monographs of the Society for Research inChild Development (51, Serial No. 213).

Dodge, K. A., & Price, J. M. (1994). On the relation between social

information processing and socially competent behavior in early school-aged children. Child Development, 65, 1385-1398.

Dodge, K. A., Price, J. M., Bachorowski, J., & Newman, J. P. (1990).Hostile attributional biases in severely aggressive adolescents. Journalof Abnormal Psychology, 99, 385-392.

Dodge, K. A., & Tomlin, A. M. (1987). Utilization of self-schemas as amechanism of interpretational bias in aggressive children. Social Cog-nition, 5, 280-300.

Dweck, C. S., & Leggett, E. L. (1988). A social-cognitive approach tomotivation and personality. Psychological Review, 95, 256-273.

Elliott, D. S., Ageton, S. S., & Huizinga, D. (1985). Explaining delin-quency and drug use. Beverly Hills, CA: Siegel.

Elliott, D. S., Huizinga, D., & Menard, S. (1989). Multiple problem youth:Delinquency, substance use, and mental health. New York: Springer-Verlag.

Eron, L. D., & Huesmann, L. R. (1990). The stability of aggressivebehavior—even unto the third generation. In M. Lewis & S. M. Miller(Eds.), Handbook of developmental psychopathology (pp. 147-156).New York: Plenum.

Feldman, E., & Dodge, K. A. (1987). Social information processing andsociometric status: Sex, age, and situational effects. Journal of AbnormalChild Psychology, 15, 211-227.

Fiske, S. T., & Taylor, S. E. (1991). Social cognition (2nd ed.). New York:McGraw-Hill.

Fuhrman, R. W., & Funder, D. C. (1995). Convergence between self andpeer in the response-time processing of trait-relevant information. Jour-nal of Personality and Social Psychology, 69, 961-974.

Graham, S., & Hudley, C. (1994). Attributions of aggressive and nonag-gressive African-American male early adolescents: A study of constructaccessibility. Developmental Psychology, 30, 365-373.

Guerra, N. G., Huesmann, L. R., Tolan, P. H., Van Acker, R., & Eron,L. D. (1995). Stressful events and individual beliefs as correlates ofeconomic disadvantage and aggression among urban children. Journal ofConsulting and Clinical Psychology, 63, 518-528.

Higgins, E. T. (1990). Personality, social psychology, and person-situationrelations: Standards and knowledge activation as a common language. InL. A. Pervin (Ed.), Handbook of personality: Theory and research (pp.301-338). New York: Guilford Press.

Higgins, E. T. (1996). Knowledge activation: Accessibility, applicability,and salience. In E. T. Higgins & A. W. Kruglanski (Eds.), Socialpsychology: Handbook of basic principles (pp. 133-168). New York:Guilford Press.

Higgins, E. T., & King, G. A. (1981). Accessibility of social constructs:Information-processing consequences of individual and contextual vari-ability. In N. Cantor & J. F. Kihlstrom (Eds.), Personality, cognition,and social interaction (pp. 69-122). Hillsdale, NJ: Erlbaum.

Huesmann, L. R. (1982). Television violence and aggressive behavior. InD. Pearl, L. Bouthilet, & J. Lazard (Eds.), Television and behavior: Tenyears of scientific programs and implications for the 80s (pp. 126-137).Washington, DC: U.S. Government Printing Office.

Huesmann, L. R. (1988). An information-processing model for the devel-opment of aggression. Aggressive Behavior, 14, 13—24.

Huesmann, L. R. (1998). The role of social information processing andcognitive schema in the acquisition and maintenance of habitual aggres-sive behavior. In R. G. Geen & E. Donnerstein (Eds.), Human aggres-sion: Theories, research, and implications for policy (pp. 73-109). NewYork: Academic Press.

Huesmann, L. R., & Eron, L. D. (1984). Cognitive processes and thepersistence of aggressive behavior. Aggressive Behavior, 10, 243—251.

Huesmann, L. R., Eron, L. D., Lefkowitz, M. M., & Walder, L. O. (1984).The stability of aggression over time and generations. DevelopmentalPsychology, 20, 1120-1134.

Huesmann, L. R., & Guerra, N. G. (1997). Children's normative beliefs

BELIEFS AND SOCIAL INFORMATION PROCESSING 165

about aggression and aggressive behavior. Journal of Personality andSocial Psychology, 72, 408-419.

Huesmann, L. R., Guerra. N. G., Miller. L., & Zelli, A. (1992). The role ofsocial norms in the development of aggression. In H. Zumkley & A.Fraczek (Eds.), Socialization and aggression (pp. 139-151). New York:Springer-Verlag.

Huesmann, L. R., Guerra, N. G., Zelli, A.. & Miller, L. (1992). Differingnormative beliefs about aggression for boys and girls. In K. Bjorkqvist& P. Niemela (Eds.), Of mice and women: Aspects of female aggression(pp. 77-87). San Diego, CA: Academic Press.

Kxuglanski, A. W. (1990). Motivations for judging and knowing: Implica-tions for causal attribution. In E. T. Higgins, & R. M. Sorrentino (Eds.),Handbook of motivation and cognition: Foundations of social behavior(Vol. 2, pp. 333-368). New York: Guilford Press.

Lochman, J. E. (1987). Self- and peer perceptions and attributional biasesof aggressive and nonaggressive boys in dyadic interactions. Journal ofConsulting and Clinical Psychology, 55, 404-410.

Lochman, J. E., & Conduct Problems Prevention Research Group. (1995).Screening of child behavior problems for prevention programs at schoolentry. Journal of Consulting and Clinical Psychology, 63, 549 -559.

Lochman, J. E., & Dodge, K. A. (1994). Social-cognitive processes ofseverely violent, moderately aggressive, and nonaggressive boys. Jour-nal of Consulting and Clinical Psychology, 62, 366-374.

Lochman, J. E., & Dodge, K. A. (1998). Distorted perceptions in dyadicinteractions of aggressive and nonaggressive boys: Effects of priorexpectations, context, and boys' age. Development and Psychopathol-ogy, 10, 495-512.

Lochman, J. E., & Lampron, L. B. (1986). Situational social problemsolving skills and self-esteem of aggressive and non-aggressive boys.Journal of Abnormal Child Psychology, 14, 605-617.

Markus, H. (1977). Self-schemata and processing information about theself. Journal of Personality and Social Psychology, 35, 63-78.

Markus, H. (1983). Self-knowledge: An expanded view. Journal of Per-sonality, 51, 543-565.

Marsh, H. W., Balla, J. R., & McDonald, R, P. (1988). Goodness-of-fitindexes in confirmatory factor analysis: The effect of sample size.Psychological Bulletin, 103, 391-410.

Milich, R., & Dodge, K. A. (1984). Social information processing patternsin child psychiatric populations. Journal of Abnormal Child Psychol-ogy, 12, 471-490.

Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory ofpersonality: Reconceptualizing situations, dispositions, dynamics, andinvariance in personality structure. Psychological Review, 102, 246-286.

Moskowitz, G. B. (1993). Individual differences in social categorization:The influence of personal need for structure on spontaneous trait infer-ences. Journal of Personality and Social Psychology, 65, 132-142.

Nasby, W., Hayden, B., & DePaulo, B. M. (1979). Attributional biasamong aggressive boys to interpret unambiguous social stimuli as dis-plays of hostility. Journal of Abnormal Psychology, 89, 459-468.

Nelson, K. (1993). Developing self-knowledge from autobiographicalmemory. In T. K. Srull & R. S. Wyer (Eds.), Advances in socialcognition (Vol. 5, pp. 111-122). Hillsdale, NJ: Erlbaum.

Olweus, D. (1979). Patterns of aggressive reaction in males: A review.Psychological Bulletin, 86, 825-875.

Penn, D. L., Corrigan, P. W., Bentall, R. P., Racenstein, J. M., & Newman,L. (1997). Social cognition in schizophrenia. Psychological Bulletin,121, 114-132.

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

Pervin, L. A. (Ed.). (1989). Goal concepts in personality and socialpsychology- Hillsdale, NJ: Erlbaum.

Pettit, G. S., Dodge, K. A., & Brown, M. M. (1988). Early familyexperience, social problem solving patterns, and children's social com-petence. Child Development, 59, 107-120.

Ross, L., & Nisbett, R. E. (1991). The person and the situation: Perspec-tives of social psychology. New York: McGraw-Hill.

Rubin, K. H., Bream, L. A., & Rose-Krasnor, L. (1991). Social problemsolving and aggression in childhood. In D. J. Pepler & K. H. Rubin(Eds.), The development and treatment of childhood aggression (pp.219-246). Hillsdale, NJ: Erlbaum.

Schneider, D. J. (1991). Social cognition. Annual. Review of Psychol-ogy, 42, 527-561.

Shoda, Y., & Mischel, W. (1993). Cognitive social approach to disposi-tional inferences: What if the perceiver is a cognitive social scientist?Personality and Social Psychology Bulletin, 19, 574-585.

Skinner, E. A. (1996). A guide to constructs of control. Journal of Per-sonality and Social Psychology, 71, 549-570.

Slaby, R. G., & Guerra, N. G. (1988). Cognitive mediators of aggression inadolescent offenders: I. Assessment. Developmental Psychology, 24,580-588.

Stromquist, V. J., & Strauman, T. J. (1991). Children's social constructs:Nature, assessment, and association with adaptive versus maladaptivebehavior. Social Cognition, 9, 330-358.

Taylor, S.--E., & Brown, J. D. (1988). Illusion and well-being: A social-psychological perspective on mental health. Psychological Bulletin, 103,193-210.

Tomaka, J., & Blascovich, J. (1994). Effects of justice beliefs on cognitiveappraisal of and subjective, physiological, and behavioral responses topotential stress. Journal of Personality and Social Psychology, 67,732-740.

Waas, G. A. (1988). Social attributional biases of peer-rejected and ag-gressive children. Child Development, 59, 969-992.

Zelli, A., Cervone, D., & Huesmann, L. R. (1996). Behavioral experienceand social inference: Individual differences in aggressive experience andspontaneous versus deliberate trait inference. Social Cognition, 14, 165—190.

Zelli, A., & Dodge, K. A. (1999). Personality development from the bottomup. In D. Cervone, and Y. Shoda (Eds.), The coherence of personality:Social-cognitive bases of personality consistency, variability, and orga-nization (pp. 94-126). New York: Guilford Press.

Zelli, A., Huesmann, L. R., & Cervone, D. (1995). Social inference andindividual differences in aggression: Evidence for spontaneous judg-ments of hostility. Aggressive Behavior, 21, 405-417.

(Appendix follows)

166 ZELLI ET AL.

Appendix A

Measurement Model's Standardized Factor Loadings and Error Terms ObtainedWhen Mediation Was Tested for Retaliation Beliefs

1. Retaliation BeliefsWeak ProvStrong Prov

2. Hostile Intent AttributionsItem aItem bItem c

3. Accessing of Aggressive ResponsesItem aItem bItem c

.67 + .74

.85 + .53

.42 + .91

.55 + .83

.40 + .90

.67 + .74

.40 + .92

.73 + .68

4. Positive Evaluation of Aggressive SolutionsItem aItem bItem c

5. Child aggressionParent ratingsTeacher ratingsSelf-report ratings

.87 + .49

.71 + .70

.78 + .63

.43 + .90

.88 + .47

.41 + .92

Note. Weak Prov = approval of retaliation to weak provocation; Strong Prov = approval of retaliation to strong provocation; Items a, b, and c = two-storyscores.

Received August 25, 1997Revision received June 15, 1998

Accepted December 3, 1998