Stability and change of moral disengagement and its impact on aggression and violence in late adolescence

Child Development, September/October 2008, Volume 79, Number 5, Pages 1288 – 1309

Stability and Change of Moral Disengagement and Its Impact on

Aggression and Violence in Late Adolescence

Marinella Paciello, Roberta Fida, Carlo Tramontano, Catia Lupinetti, andGian Vittorio CapraraSapienza University of Rome

Stability and change of moral disengagement were examined in a sample of 366 adolescents from ages 14 to 20years. Four developmental trajectories were identified: (a) nondisengaged group that started with initially lowlevels followed by an important decline, (b) normative group that started with initially moderate levels followedby a decline, (c) later desister group that startedwith initially high-medium levels followed by an increase from 14to 16 years and an even steeper decline from16 to 20 years, and (d) chronic group that startedwith andmaintainedmedium-high levels. The results attest that adolescents who maintained higher levels of moral disengagementwere more likely to show frequent aggressive and violent acts in late adolescence.

Among the mechanisms conducive to aggression,moral standards exert a notable influence in dictatingwhen it may be legitimate to resort to behavior thatmay hurt other people andwhen, on the contrary, onehas to refrain from detrimental conduct (Bandura,1986, 1991, 2001). People are active agentswhopursuetheir goals in accordancewith personal standards thatserve as guides and deterrents for action through self-reactive anticipatory evaluations. People monitortheir conduct and the conditions under which itoccurs, judging their actions in relation to their moralstandards and perceived circumstances, and regulatetheir actions by anticipating the consequences theywould apply to themselves. People do things that givethem satisfaction and refrain frombehaviors that bringself-censure. Anticipatory self-pride and self-blameare suggested to be human mind self-regulatorycapacities that keep behavior in line with personalstandards. Yet, this does not exclude the possibilitythat people, after having adopted personal standardsand despite being morally committed to ethicalprinciples, may enact behaviors that violate thosestandards, while continuing to profess those princi-ples and avoiding feelings of conflict, guilt, orremorse. The chronicles of recent atrocities attest tothe fact that even considerate people may become

engaged in inhumane behavior without experiencingany moral concern or discomfort (Bandura, 1999,2004; Zimbardo, 1995, 2004).

In this regard, Bandura’s (1986) social cognitivetheory provides a theory ofmoral agency that suppliesthe conceptual apparatus that is needed to clarify themechanisms by which people come to live in accor-dance with moral standards as well as the mecha-nisms that allow a kind of divorce between moralthought and moral actions (Bandura, 1991; Bandura,Barbaranelli, Caprara, & Pastorelli, 1996). In particu-lar, Bandura (1986, 1991) has introduced the constructof moral disengagement to explain the determinantsand mechanisms governing aggressive behaviors.Moral disengagement refers to individual’s tendencyto use mechanisms conducive to a selective disen-gagement of moral censure. This is achieved byreconstructing behavior, obscuring causal agency,misrepresenting injurious consequences, and blam-ing victims. These mechanisms allow individuals toengage in self-serving behavior that is in contrastwiththeir moral principles, while continuing to advocatethose principles and without incurring self-evaluativeemotional reactions such as guilt. It is likely that thesemechanisms become crystallized over timewhen deal-ing with transgressions in the pursuit of self-interest.Indeed, a previous study found that moral disengage-ment to bepositively correlatedwith self-enhancementand negatively correlated with self-transcendence(Caprara & Capanna, 2005).

The focus of the present study is to examine thestability and change of moral disengagement both at

This study was partially supported by grants from the SpencerFoundation and W. T. Grant Foundation to Albert Bandura, andfrom the Johann Jacobs Foundation and Ministero dell’Istruzionedell’Univesita e della Ricerca to G.V.C. (COFIN 1998, 2000, 2004)and to Eugenia Scabini (COFIN 2000 – 2002). We are grateful to theanonymous reviewers and Marie S. Tisak for their helpfulcomments and suggestions.

Correspondence concerning this article should be addressed toMarinella Paciello, Department of Psychology, SapienzaUniversityof Rome, Via deiMarsi 78, 00185 Rome, Italy. Electronicmailmay besent to [email protected].

# 2008, Copyright the Author(s)

Journal Compilation# 2008, Society for Research inChildDevelopment, Inc.

All rights reserved. 0009-3920/2008/7905-0006

a population level and at an interindividual level aswell as explore the extent to which different de-velopmental pathways are associated with differentaggressive and violent outcomes over the course ofadolescence. Over the past few decades, differentcognitive theories have examinedmoral developmentand the relation among moral thinking, social behav-iors, and social context (Bandura, 1986; Crick &Dodge, 1994; Dodge & Crick, 1990; Gibbs, Potter,Barriga, & Liau, 1996; Huesmann, 1988, 1998; Kohlberg,1969; Turiel, 1978, 1983). Furthermore, extensive knowl-edge has been accumulated on the development ofprocesses related to moral self-regulation, such as theconstruction of moral standards, the formation ofmoral judgments, and the exercise of self-influencethrough emotional self-reactions (Bandura, 1991;Eisenberg, 2000; Killen & Smetana, 2006; Tangney,Stuewig, & Mashek, 2007; Turiel, 2006). Accordingto social cognitive theory and adopting a develop-mental approach, we focus on self-exonerativemechanisms that exert a crucial role in attenuatingthe links between moral thought and moral actions.These mechanisms have been overlooked in theexamination of moral development. To our knowl-edge, this is the first study that longitudinally mapsindividual differences in moral disengagement overthe course of adolescence.

The present study focused on the adolescent tran-sition because various manifestations of aggressionmay dramatically change throughout childhood, ado-lescence, and early adulthood, and these changes arenot the same for all individuals (Brame, Nagin, &Tremblay, 2001; Loeber & Stouthamer-Loeber, 1998;Moffitt, 1993; Sampson&Laub, 2003; Tremblay, 2000).Most adolescents adopt antisocial behavior patternsbut a large part abandon them in adulthood and onlya small percentage become deeply and chronicallyengaged in risky aggressive and violent behaviors(Loeber, 1991; Loeber & Hay, 1997; Tremblay, 2000).Webelieve thatmoral disengagement can be crucial inexplainingmechanisms conducive to chronic engage-ment in aggressive and violent behavior (Bandura,1999; Bandura, Caprara, Barbaranelli, Pastorelli, &Regalia, 2001; Bandura et al., 1996; Zimbardo, 2004,2006).

Moral Disengagement

Over the course of development, the adoption ofmoral standards attests that the property of humanmind is guided from within in accordance withprinciples that derive from one’s own and others’experiences and that, in various ways, are at the coreof social life. Individuals’ increasing self-regulatory

capacities and the change from external to autonomyregulation make adolescence a period particularlysensitive for the study of moral functioning. Duringadolescence, the development of metacognitive abil-ities promotes the internalization of moral principles,and ultimately, the development of one’s own moralidentity and agency (Bergman, 2002; Blasi, 1984;Carlo, Eisenberg, & Knight, 1992; Hardy & Carlo,2005; Helwig & Turiel, 2003; Kohlberg & Candee,1984; Krettenauer, 2004; Swanson & Hill, 1993). Onceformed, moral standards dictate the goals to bepursued and the actions from which to refrain andexert their influence over behavior through antici-pated positive and negative self-evaluative reactions.Although self-sanctions carrying feelings of guilt areassociated with actions that violate individual moralstandards, positive reactions carrying pride, and self-worth are associated with actions that promote thosemoral standards (Bandura, 1991; Bandura & Walters,1959; Elkin &Westley, 1955; Emmons &Diener, 1986).Self-sanctions exert an important influence in refrain-ing individual’s detrimental conduct particularlywhen one’s interest is at stake, and detrimentalconductmay be instrumental in the pursuit of specificgoals. However, the adoption of moral abstract prin-ciples, although necessary, is not always sufficient torefrain from detrimental conduct. Through moraldisengagement, self-sanctions can be deactivatedand affective self-evaluative reactions are avoided,permitting different types of detrimental conductwhile saving the same moral standards.

In the social cognitive theory ofmoral agency, thereare four major points at which self-sanction can bedisengaged from detrimental conduct and eightmechanisms that have been corroborated experimen-tally, through which moral disengagement operates(Bandura, 1991). Different disengagement practicesmay focus on the behavior, on the sense of personalresponsibility, on the outcomes of behavior, and on therecipients of behavior. Thus, the first set of disengage-ment practices—focusing on behaviors—operates onthe cognitive construal of the behavior itself. Itincludes moral justification, euphemistic language, andadvantageous comparison. By moral justification, detri-mental conduct is made personally and sociallyacceptable (Kelman & Hamilton, 1989; Rapoport &Alexander, 1982; Reich, 1990; Sanford & Comstock,1971). A lot of aggressive behavior gets justified in thename of protecting honor and reputation (Cohen &Nisbett, 1994). Euphemistic language provides a con-venient way ofmasking immoral activities conferringa respectable status upon them (Bollinger, 1982;Diener, Dineen, Endresen, Beaman, & Fraser, 1975;Lutz, 1987). People behave much more aggressively

Moral Disengagement Development 1289

when assaulting a person is given a sanitized labelthan when it is called aggression. By advantageouscomparison, detrimental conduct can lose its repug-nancy by comparing it with more flagrant inhuman-ities (Bandura, 1991). The more flagrant the contrastedactivities, the more likely it is that one’s own injuriousconduct will appear trifling or even benevolent. Thesecond set of disengagement mechanisms—focusingon personal responsibility—operates by obscuring,minimizing, or disclaiming the agentic role in theharm that one causes. It includes displacement ofresponsibility and diffusion of responsibility by whichpeople view their actions as ordered by the socialpressures of others rather than as something forwhich they are personally responsible (Bandura,Underwood,&Fromson, 1975;Diener, 1977;Milgram,1974; Zimbardo, 1995). People arewilling to behave inways theynormally repudiate if a legitimate authorityaccepts responsibility for the effects of their actionsand if responsibility can be diffused (when everyoneis responsible, no one really feels responsible). Thethird set of disengagementmechanisms—focusing onoutcomes—avoids self-deterring reactions by disre-garding or distorting the consequences of action. Whenpeople are involved in activities harmful to others forpersonal gain or because of social inducements, theyavoid facing the harm they cause or they minimize it(Klass, 1978). They readily recall prior informationgiven to them about the potential benefits of thebehavior, but are less able to remember its harmfuleffects (Brock & Buss, 1962, 1964). In addition toselective inattention and cognitive distortion of ef-fects, themisrepresentationmay involve active effortsto discredit evidence of the harm they cause. The finalset of disengagement practices focusing on the recip-ients of detrimental acts includes dehumanization andattribution of blame. Dehumanization divests people ofhuman qualities or attributes bestial qualities to them.Once dehumanized, they are no longer viewed aspersons with feelings, hopes, and concerns but assubhuman objects (Haritos-Fatouros, 1988; Keen,1986; Kelman, 1973). It is difficult to mistreat human-ized persons without risking personal distress andself-censure. By attribution of blame, people viewthemselves as faultless victims driven to injuriousconduct by forcible provocation (Crick & Dodge,1994;Darley,Klosson,&Zanna, 1978; Ferguson&Rule,1983; Weiner, 1986). Punitive conduct thus becomesa justifiable defensive reaction to instigations.

A number of findings have shown that the aboveeight mechanisms can be traced back to a commonlatent variable that makes people more or lessinclined to use mechanisms of moral disengage-ment (Bandura et al., 1996, 2001; Caprara, Bandura,

Barbaranelli, & Vicino, 1996; Pelton, Gound, Fore-hand, & Brody, 2004). Thus, the proneness to moraldisengagement is assessed by the processes throughwhich it is presumed to operate. Appropriate assess-ment measures have already been developed in orderto investigate how the full set of moral disengage-ment mechanisms operate in concert (Bandura, 2004;Bandura et al., 1996; Caprara et al., 1996).

A large body of research has demonstrated thedisinhibitory power of moral disengagement in fos-tering aggressive behavior (Andrus, 1969; Bandura,1990, 2006; Kelman & Hamilton, 1989; Rapoport &Alexander, 1982; Reich, 1990). In adolescence, differ-ent studies have shown strong associations betweenproneness tomoral disengagement andvariousmeas-ures of aggression and violence, aswell as other formsof antisocial conduct and bullying (Bandura, Caprara,& Zsolnai, 2000; Bandura et al., 1996, 2001; Capraraet al., 1996; Gini, 2006; Menesini et al., 2003). Longi-tudinal studies have found that high levels of moraldisengagement predict levels of violence, theft, andother forms of antisocial conduct (Bandura et al., 2001;Elliott & Rhinehart, 1995; Pepler, Jiang, Craig, &Connolly, 2008). On the one hand, high moral dis-engagers tend to be more irritable, more prone tovengeful rumination, more inclined toward physicaland verbal aggression, and more frequently involvedin violent episodes. On the other hand, high moraldisengagers are less troubled by anticipatory feelingsof guilt due to injurious conduct, that is, the higher themoral disengagement, the weaker the felt guilt andthe lower the need to undo any harm caused byaggressive conduct (Bandura et al., 1996). Highmoraldisengagers not only experience low guilt over inju-rious conduct but they are also less prosocial and lessable to resist peer pressure for transgressive activities(Bandura et al., 1996, 2001; Kwak & Bandura, 1997).

Moreover, boys are more likely than girls tobecome moral disengagers over the course of devel-opment, although gender differences do not existin the earlier years (Bandura, 2004; Bandura et al.,1996). Some findings from cross-cultural studiesindicate that some of the gender differences in aggres-sion may reside in the differential proclivity of disen-gaging moral self-sanctions from injurious conduct(Bandura, 1999, 2004).

Aim of Present Study: A Developmental Perspective

Our general aim was to examine the predictivepower of different moral disengagement develop-mental trends in adolescence that are conducive todifferent levels of later aggressive and violent behav-iors. To this aim, we used longitudinal structural

1290 Paciello, Fida, Tramontano, Lupinetti, and Caprara

equation modeling (i.e., growth latent curve andlatent class growth analysis), a procedure not typi-cally used in research onmoral development. In parti-cular, we first investigated the development of moraldisengagement at a population level using growthlatent curve analysis (Duncan & Duncan, 1994, 1995;Duncan, Duncan, & Hops, 1994; McArdle, 1988;McArdle & Anderson, 1989; McArdle & Hamagami,1991; Meredith & Tisak, 1990; Muthen & Khoo, 1998;Stoolmiller, 1994) among an Italian adolescent samplefrom 14 to 20 years old. We hypothesized a decline inmoral disengagement due to social, emotional, andcognitive skills that, over the course of development,enhance youths’ capacities to regulate their ownaffect, to commit themselves to moral principlesthat ban aggression and violence, to accord theirbehavior to social standards that decrease behaviorsthat offend others, and to interact effectively withothers (Eisenberg, Fabes, & Spinrad, 2006; Hart &Carlo, 2005; Tangney et al., 2007).

A second goal was to identify subgroups of ado-lescents following distinct developmental patterns ofmoral disengagement using latent class growth anal-ysis, a semiparametric group-based approach (Nagin,1999, 2005) that should be seen as a natural extensionof growthmodels (Hoeksma & Kelderman, 2006). Wehypothesized four trajectories, in accordance withwhat has been previously found in the aggressionand in the aggressogenic personality factors literature(Barker, Tremblay, Nagin, Vitaro, & Lacourse, 2006;Brame et al., 2001; Broidy et al., 2003; Caprara,Paciello, Gerbino, & Cugini, 2007; Cote, Tremblay,Nagin, Zoccolillo, & Vitaro, 2002; Nagin & Tremblay,1999; Tremblay & Nagin, 2005). We hypothesizeda ‘‘nondisengaged’’ group whose moral disengage-ment would start and remain low, a ‘‘chronic’’ groupwhose moral disengagement would start and remainhigh, and two trajectories that could either decrease(‘‘desister’’) or increase (‘‘escalator’’) over the courseof adolescence. To validate the trajectory model, thegroups were compared on aggressive and violentbehavior and a proxy of guilt across time (i.e., needof reparation; see Bandura et al., 1996; Caprara,Barbaranelli, Pastorelli, Cermak, & Rosza, 2001; Cap-rara, Manzi, & Perugini, 1992). We hypothesized thatthe chronic group would show higher levels ofaggressive and violent behavior and lower levels ofguilt and that the nondisengaged group would showlower levels of aggressive and violent behavior andhigher levels of guilt over time.

Our third goal was to explore how gender isassociated with moral disengagement development.We hypothesized that boys are more likely to exhibithigher levels ofmoral disengagement than girls in the

course of development (Bandura et al., 1996) and thatthe decline in moral disengagement would be stron-ger in girls than in boys due to the higher resistance ofgirls to resort to direct aggression (Archer, 2004). Asa consequence, we hypothesized that girls wouldhave a higher probability of beingmembers of a ‘‘non-disengaged’’ group, whereas boys would demon-strate a higher probability of being members ofa ‘‘chronic’’ or ‘‘escalator’’ group.

Our fourth goal was to explore howpeer evaluationof aggression in early adolescence is associated withmoral disengagement development. Because it ismoreprobable that adolescents who have already shownmore aggressive behaviors in early adolescence couldhave more opportunities to recourse to moral disen-gagement, we hypothesized that the more preadoles-cents are evaluated as aggressive by peers, the higherthe level of moral disengagement, the less the likeli-hood of declining during adolescence, and thus, thehigher theprobabilityof being amemberof a ‘‘chronic’’or ‘‘escalator’’ group and the lower the probability ofbeing a member of a ‘‘nondisengaged’’ group.

Our final goal was to explore the association ofmoral disengagement development with aggressiveand violent conduct in late adolescence, controllingfor earlier peer evaluations of aggression.We hypoth-esized that the earlier preadolescents are inclined touse moral disengagement, and the more persistentrecourse to it, the higher the engagement in aggres-sion and violence. As a consequence, adolescentswith a chronic high or increasing trajectory wouldshowhigher levels of aggression and violence relativeto nondisengaged adolescents.

Method

Sample and Data Collection

The participants were part of an ongoing longitu-dinal project that began in 1989.We implemented thisproject to investigate the main determinants andpathways of aggression and successful developmentand adjustment from late infancy to early adulthood.The longitudinal project followed a staggered, multi-ple cohort design, with two cohorts assessed at fivedifferent time points. The first cohort was age 12 in1994, the second cohort was age 12 in 1996 (Time 1).These participants were retested at age 14 (Time 2), atage 16 (Time 3), at age 18 (Time 4), and finally, at age 20(Time 5). Cohort effects were previously tested andwere found to be insignificant for sociodemographicand major study variables.

Three hundred and sixty-six adolescents (177 boysand 189 girls) were included in the study. Participants


attended sixth grade at Time 1 and eighth grade atTime 2. Most of them attended high school at Times 3and 4, whereas at Time 5, the majority were collegestudents (71%). At Time 1, the participating adoles-cents were drawn from two schools in a residentialcommunity located near Rome, Italy. The communitywas composed of families of skilledworkers, farmers,professionals, local merchants, and their service staff.Occupational socioeconomic distribution matchedthe national profile (Istituto Italiano di Statistica,2002). The composition of the families also matchednational data with regards to type of families andnumber of children. Most subjects were from intact(94.1%) and, on average, one-child families. Theparticipation rate was high during longitudinal datacollection, with 84.4% of the original sample (Time 1)returning at the final assessment time (Time 5). Inparticular, there were three patterns of missing data.Complete data were available for 65% of males and83% of females. Thirty percent of males and 16% offemales had missing data in one of the five waves.Finally, 5% of males and 1% of females had missingdata in two of the five waves.

Procedures

Young adolescents at Times 1 and 2 were admin-istered a set of scales tapping different types ofdimensions in their classrooms by two trained femaleexperimenters and were asked to complete the scalesindividually. Before starting, the experimenters ex-plained that their responses to the questionnaireswould be absolutely confidential. When necessary,the experimenters offered the children clarificationson the dimensions beingmeasured. At Times 3, 4, and5, participants were contacted by phone and invitedto participate in the study for which they receiveda small payment. A stringent consent procedure forthe researchwas followed including at various stages,parents’ consent and approval from school councils,while letting children decline their participation ifthey so chose.

Measures

Moral disengagement (Bandura et al., 1996; Capraraet al., 1996). From Time 2 to Time 5, moral disengage-ment was assessed by self-report. The scale assessesproneness to moral disengagement of different formsof detrimental conduct in diverse contexts and inter-personal relationships. The full set of 32 items, pre-sented in the Appendix, tap the eight differentmechanisms by which moral self-sanctions can bedisengaged from transgressive conduct as hypothe-

sized by Bandura (1990, 1999). For each of the 32items, adolescents rated on a 5-point Likert-type scale,their degree of acceptance of moral exonerations forsuch conduct on an agreement continuum (from 1 5

agree not at all to 5 5 completely agree).Physical and verbal aggression scale (Caprara &

Pastorelli, 1993). At Time 1, physical and verbalaggression was assessed by peer nominations. Chil-dren made their nominations from the roster ofclassmates in their classroom. Because this is a highlystable community, children were thoroughly ac-quainted with each other. Children were providedwith a booklet containing all classmates’ names. Therespondents selected the three classmates who mostfrequently exert each of three different forms ofphysical and verbal aggression (kick andhit or punch,insult other kids or call them names, and hurt otherkids). The numbers of nominations that each childreceivedwere summed separately and divided by thenumber of classmates (as a consequence this variableranged from 0 5 never nominated by the classmates to15 nominated by all classmates). Cronbach’s reliabilitycoefficient was .94.

From Time 2 to Time 5 physical and verbal aggres-sion was assessed by self-report. This scale measuredbehaviors aimed to physically and verbally hurtingothers. It includes three items for physical aggression(i.e., ‘‘I hurt others’’) and three items for verbalaggression (i.e., ‘‘I say bad things about others kids’’)using a 3-point scale (from 1 5 never to 3 5 often) atTime 2 and a 5-point scale (from 15 never to 55 often)from Time 3 to Time 5. Cronbach’s reliability coeffi-cients ranged from .74 to .82 for physical aggressionand from .68 to .72 for verbal aggression.

Violence (Caprara, Mazzotti, & Prezza, 1990). FromTime 3 to Time 5, violencewas assessed by self-report.This scale included 11 items aimed to assess the extentto which adolescents engage in violent conduct. Foreach item, adolescents rated how often they haveengaged in violent actions, such as fighting, vandal-ism, or weapon use on a 5-point scale (from 15 neverto 55 often). ‘‘Have youparticipated in violent actionsof ‘gangs’?’’ and ‘‘Have you ever had the occasion touse violence when there are arguments?’’ are twosample items. Cronbach’s reliability coefficientsranged from .87 to .90.

Need for reparation (Caprara et al., 1992). From Time2 to Time 5, need for reparation was assessed byself-report. This scale included five items measur-ing proneness to experience feelings of remorse,embarrassment, disturbance, tension, and desire forjustice that are linked to the need for reparation ofguilt-eliciting acts (i.e., ‘‘I feel that I have to make upfor the wrongs that I’ve done to others’’) using


a 6-point scale (from 1 5 completely true for me to 6 5

completely false for me). The Cronbach coefficients forthis scale ranged from .65 to .71.

Results

Preliminary Analysis: Longitudinal Factorial Analysis

As a preliminary analysis, we first examined thedimensionality of the moral disengagement scaleby using the exploratory factor analysis (EFA) ap-proach. Second, we examined the longitudinal facto-rial invariance of moral disengagement to test thedegree to which the construct is measured similarlyacross waves (Hoyle & Smith, 1994; Motl, Dishman,Birnbaum, & Lytle, 2005).

In accordance with previous studies on moraldisengagement (Bandura et al., 1996; Caprara et al.,1996), two alternative models for each wave werehypothesized: (a) an eight-factor solution, each factorrepresenting a distinct mechanism of moral disen-gagement, and (b) a one-factor solution, consistentwith previous studies.

To identify the best solution, we analyzed theeigenvalues (Cattell & Vogelmann, 1977) and consid-ered the standardized root mean square residual(SRMR; Joreskog & Sorbom, 1984) and the root meansquare error of approximation (RMSEA; Steiger &Lind, 1980) as indices of goodness of fit. Becauseseveral items presented a deviation from a normaldistribution, EFA was performed using maximumlikelihood parameter estimates with standard errorsand the chi-square test statistic, which are robust tononnormality (MLMV). Factors were rotated usingthe Promax procedure.

For Time 2, the first 10 eigenvalues of the correlationmatrix were: 7.19, 2.21, 1.54, 1.36, 1.26, 1.12, 1.02, 0.98,and 0.94. The eight-factor solution did not converge.The analysis of eigenvalues suggested a one-factorsolution. Thismodel fit the data (v25 1192.19, df5 464,p , .001, RMSEA 5 .06, SRMR 5 .07), reproducingwith a good approximation the intercorrelation amongthe 32 items of the scale. Cronbach’s reliability coeffi-cient for this wave was .87.

For Time 3, the first 10 eigenvalues of the correlationmatrix were: 10.67, 2.17, 1.68, 1.40, 1.26, 1.07, 0.99, 0.91,0.88, and 0.82. The eight-factor solution did not con-verge. The analysis of eigenvalues suggested a one-factor solution. Thismodel fit the data (v25 1291.78, df5 464, p , .001, RMSEA 5 .07, SRMR 5 .07),reproducing with a good approximation the intercor-relation among the 32 items of the scale. Cronbach’sreliability coefficient for this wave was .93.

For Time 4, the first 10 eigenvalues of the correla-tion matrix were: 9.99, 2.20, 1.61, 1.42, 1.31, 1.24, 1.04,0.99, 0.96, and 0.90. Although the RMSEA and theSRMR of the eight-factor model resulted in a goodapproximation to the data (RMSEA 5 .03, SRMR 5

.03), the pattern of rotated factor loadings was notinterpretable or simple (Thurstone, 1947). The analy-sis of eigenvalues suggested a one-factor solution.This model fit the data (v2 5 1322.68, df 5 464, p ,

.001, RMSEA 5 .07, SRMR 5 .07), reproducing witha good approximation the intercorrelation among the32 items of the scale. Cronbach’s reliability coefficientfor this wave was .92.

For Time 5, the first 10 eigenvalues of the correla-tion matrix were: 9.15, 2.19, 1.95, 1.51, 1.39, 1.19, 1.08,1.01, 0.90, and 0.87. Although the RMSEA and theSRMR of the eight-factor model resulted in a goodapproximation to the data (RMSEA 5 .03, SRMR 5

.03), the pattern of rotated factor loadings was notinterpretable or simple (Thurstone, 1947). The analy-sis of eigenvalues suggested a one-factor solution.This model fit the data (v2 5 1293.32, df 5 464, p ,

.001, RMSEA 5 .08, SRMR 5 .09). Cronbach’s reli-ability coefficient for this wave was .91.

The above findings substantially corroborated pre-vious results (Bandura et al., 1996; Caprara et al., 1996)and attest to a main latent dimension, which includesall the items measuring the eight moral disengage-ment mechanisms. As a next step, we analyzed thelongitudinal factor structure of moral disengagementacross the four waves of data (14 – 20 years) testinga series of nested models, from less to more demand-ing ones. AsMeredith (1993) andWidaman and Reise(1997) have argued, there are different levels ofinvariance. Configural invariance (Thurstone, 1947)hypothesizes the equality of the overall structure(i.e., same factor and same patterns of fixed and freedparameters) across groups or time. Metric or patterninvariance (Thurstone, 1947) hypothesizes the equalityof the factor loadings across time or groups. Strongfactorial invariance (Meredith, 1993) hypothesizes theequality of the intercept of the measured variablesacross time or groups. Strict factorial invariance (Mer-edith, 1993) hypothesizes the equality of the unique-ness of the measured variables across time or groups.

Given the small size of the longitudinal sample andthe number of indicators of moral disengagement,the longitudinal invariance was analyzed via itemparceling (Anderson & Gerbing, 1988; Hoyle 1995;Martinez, Black, & Starr, 2002). Parceling typicallyexhibits a distribution that more closely approachesa normal distribution than using the original items asindividual indicators (West, Finch, & Curran, 1995).Thus,with theaimof reducing thevariance – covariance


matrix to be analyzed, we built four parcels for eachwave. We used item-scale correlation to select items.In particular, items were assigned to parcels so thatequivalent parcels in terms of homogeneity werecreated, and the monodimensionality of each parcelwas guaranteed as suggested by Bandalos andFinney (2001).

In all models, autocorrelations were estimatedamong all pairs of uniqueness because the same parcelswere used across the four time points (Motl et al., 2005;Pitts,West,&Tein, 1996). Tocompare the fit of thenestedmodels in the longitudinal invariance sequence, weused the chi-square difference test (Bollen, 1989; Fix,Hodges, & Lehmann, 1959) and the difference incomparative fit index (CFI; Bentler, 1989); as noted byCheungandRensvold (2002), adifference larger than .01inCFIwould indicate ameaningful change inmodel fit.

The fit indices for the models tested in the invari-ance analyses are presented in Table 1. As can benoted, configural and metric invariance were sup-ported by all fit indices. Although full strong invari-ance was not supported, support for partial stronginvariance came from differences in the CFI. Finally,support for a strict factorial invariance only camefrom differences in the CFI. Because the majority ofloadings were invariant, results confirmed that theconstructs were measured as comparable over time(Raykov, 2004; Raykov & Marcoulides, 2006).

Descriptive Statistics

Observed means of moral disengagement fromTime 2 to Time 5, peer evaluation of physical and

verbal aggression measured at Time 1, and the out-comes (violence and physical and verbal aggression)measured at Time 5, separately formales and females,are reported in Table 2.

Asshown, thereweremissingdata inall thevariables.As noted by Hanson, Tobler, and Graham (1990), inlongitudinal research, it is common to have subjectattrition, which produces missing values in the dataset. In the presence of missing data, estimation ofparameters must be adjusted accordingly. Among thedifferent methods used for taking into account missingdata, we used the maximum likelihood estimation ofparameters, a method widely accepted as appropriatefor handling missing data (Muthen & Shedden, 1999;Schafer &Graham, 2002) under the assumption that thedata aremissingat random(Arbuckle, 1996; Little, 1995).

Before proceeding with the analysis, the normalityof the variables was ascertained. Due to the non-normality of some measures (the skewness andkurtosis varied from 2.23 for physical aggression to10.79 for violence), we computed the inverse ofviolence and physical aggression and the root squareof peer evaluation of physical and verbal aggressionto normalize these variables as suggested by Tabach-nik and Fidell (1989). The skewness and kurtosis ofthe computed outcomes varied from �.572 for phys-ical aggression to �1.29 for violence.

Correlations betweenmoral disengagement acrossthe four time points, peer evaluation of physical andverbal aggression, and the three outcomes—physicaland verbal aggression and violence—for boys andgirls are provided in Table 3. As shown, they attestedto a high-medium stability of moral disengagement

Table 1

Fit Indices for the Model of the Invariance Routine

Model v2 df p CFI RMSEA SRMR

Baseline

14 years .806 2 .67 1.00 .00 (.00 – .08), p 5 .84 .005

16 years .877 2 .65 1.00 .00 (.00 – .08), p 5 .82 .003

18 years 3.696 2 .16 1.00 .05 (.00 – .12), p 5 .40 .007

20 years 1.591 2 .45 1.00 .00 (.00 – .10), p 5 .67 .006

Dv2 df p DCFI

Longitudinal invariance

Configural 80.993 74 .27 1.00 .02 (.00 – .03), p 5 1.00 .024

Metric 95.223 83 .17 1.00 .02 (.00 – .04), p 5 1.00 .040 Metric versus configural 14.23 9 .11 .00

Full strong 508.925 95 .00 0.91 .11 (.10 – .12), .00 .254 Full strong versus metric 412,80 12 .00 .09

Partial strong 151.322 89 .00 0.99 .04 (.03 – .05), p 5 .80 .058 Partial strong versus

full strong

56.10 6 .00 .01

Strict 174.952 101 .00 0.98 .04 (.03 – .06), p 5 .78 .062 Strict versus partial strong 23,63 12 02 .01

Note. CFI 5 comparative fit index; RMSEA5 root mean square error of approximation; SRMR5 standardized root mean square residual.


over time with lower correlations the longer thedistance of time. Finally, peer evaluation of physicaland verbal aggression and the outcomes were corre-lated with the moral disengagement variable, espe-cially in boys.

Growth Models

The analysis ofmoral disengagement developmentwas carried out within a latent variable framework.

We specified a multigroup growth curve model thatsimultaneously estimated the same pattern of rela-tionships among variables for males and females. Toexamine gender differences in the estimated param-eters,we constrained all parameters to be equal acrossgroups and used the chi-square difference test tocompare nested models. Modification indices wereused to assess the tenability of the equality constraintimposed across gender.

We defined two latent growth factors from multi-ple indicators, that is, the four repeated measures ofmoral disengagement. The first factor was labeledintercept and represents the initial starting point ofmoral disengagement. The second factor representsthe slope or the shape of the trajectory over time.The following equation shows the mathematicalrepresentation of the growth model:

yt 5 g0 þ g1xt þ et; where t 5 Times 2; 3; 4; 5;

where yt5 observed score at time t, g0 5 unobservedscore for the intercept factor, g1 5 unobserved scorefor the growth rate factor, and xt 5 factor loadingrelating yt to latent growth variables.

Because the factor loadings of the slope give theshape of the growth, a series of models were testedand compared with each other. In this way, we coulddetermine the parameterization that best fit the data.In all the models tested, we defined the intercept asmoral disengagement at age 14 by fixing the factorloading relating this variable to the slope at 0.

Following McArdle and Anderson (1989), the firstmodel testedwas a no-growthmodel (only intercept),v2(14, N 5 177, 188) 5 419.497, p , .001, CFI 5 .08,RMSEA 5 .398 (.366, .482), SRMR 5 .779. This modelassumed that the level of moral disengagement wasstable over time except for a random error componentat each time of evaluation. The second was a linearmodel representing a constant change over time, v2(8,N5 177, 188)5 70.668, p, .001, CFI5 .86, RMSEA5

.207 (.164, .253), SRMR5 .077. In this model, we fixedthe factor loadings on the slope at 0, 1, 2, and 3. Thethird model examined a nonlinear growth where theform of the change over time was not specifieda priori, v2(6, N 5 177, 188) 5 4.086, p 5 .67, CFI 51.00, RMSEA5 .000 (.000, .077), SRMR5 .040. Finally,the fourthmodelwas the nonlinearmodelwithout theintercept, v2(12, N 5 177, 188) 5 88.738, p , .001,CFI5 .83, RMSEA5 .187 (.152, .225), SRMR5 .230. Inparticular, we fixed the first factor loadings on theslope and estimated the others. In all the growthmodels, autocorrelations between the same measuresover time were estimated and for model identifica-tion, two uniquenesses and two covariances betweenuniquenesses were constrained to be equal among

Table 2

Observed Descriptive Statistics of Moral Disengagement, Predictor and

Outcomes Separately for Gender

Males Females

N M SD N M SD

MD14 174 2.68 0.45 187 2.50 0.46

MD16 168 2.34 0.63 178 2.02 0.52

MD18 172 2.23 0.55 188 1.90 0.50

MD20 137 2.11 0.50 172 1.81 0.41

AGG_PE 166 0.15 0.19 188 0.05 0.07

PHY 137 1.53 0.71 172 1.26 0.50

VERB 137 1.97 0.75 172 1.61 0.58

VIO 136 1.40 0.52 170 1.11 0.21

Note. MD14 5 moral disengagement at age 14; MD16 5 moraldisengagement at age 16; MD185moral disengagement at age 18;MD205 moral disengagement at age 20; AGG_PE5 physical andverbal aggression peer evaluation at age 12; PHY 5 physicalaggression at age 20; VERB 5 verbal aggression at age 20; VIO 5

violence at age 20.

Table 3

Correlations BetweenMoral Disengagement FromAges 14 to 20, Physical

and Verbal Aggression at Age 12, and Aggressive and Violent Conduct at

Age 20, Separately for Gender

1 2 3 4 5 6 7 8

1. AGG_PE .09 .11 .27 .21 .12 .16 .17

2. MD14 .21 .51 .40 .43 .11 .01 .05

3. MD16 .29 .49 .66 .52 .17 .15 .21

4. MD18 .21 .44 .58 .64 .22 .21 .22

5. MD20 .18 .39 .37 .65 .41 .32 .42

6. VERB .15 .39 .25 .37 .50 .33 .41

7. PHY .16 .31 .32 .42 .53 .67 .57

8. VIO .25 .21 .32 .37 .40 .45 .53

Note. Values below the diagonal are for males. Values above thediagonal are for females. All correlation coefficients were signifi-cant at least at p, .05, except for those that are in italics, which arenot significant. Agg_PE 5 physical and verbal aggression peerevaluation at age 12;MD145moral disengagement at age 14;MD165moral disengagement at age 16;MD185moral disengagement atage 18; MD20 5 moral disengagement at age 20; PHY 5 physicalaggression at age 20; VERB 5 verbal aggression at age 20; VIO 5

violence at age 20.


males and females. Because these models are nested,we performed a chi-square differences test to com-pare themodels (Bollen, 1989; Fix et al., 1959). This testrevealed that the nonlinear model provided the bestfit to the data compared to the no-growth model,Dv2(8) 5 415.41, p , .001; the linear growth model,Dv2(2) 5 66.58, p , .001; and the nonlinear withoutintercept model, Dv2(6) 5 84.65, p , .001. Then, wetested if there were any differences among males andfemales on the free factor loadings, constraining theseparameters to be equal among the two groups andused the chi-square difference test to compare theconstrained model with the nonconstrained one.Results showed that the free factor loadings on theslope (x4 5 1.282, x5 5 1.513) were the same for malesand females, Dv2(2) 5 1.57, p 5 .46, and weresignificantly smaller than the linear loadings (x4 , 2with p, .001, x5, 3with p, .001), indicating that thedecrease in moral disengagement from ages 14 to 16was larger than both the decrease from ages 16 to 18and the decrease from ages 18 to 20.

The results of the finalmodel, v2(9,N= 177)5 5.096,188,p5 .83, CFI5 1.00, RMSEA5 .000, CI5 .000, .050,SRMR 5 .065, summarized in Table 4 and Figure 1,suggest that there were gender differences in theinitial level of moral disengagement; that is, malesexhibited higher levels of moral disengagement thanfemales at age 14—we compared the model with themean of the intercept constrained to be equal amongmales and females with the nonconstrained model,Dv2(1)5 13.03, p, .001. Furthermore, the mean of theslope indicated a significant decrease from ages 14 to16. In particular, females exhibited a greater declinethan males—we compared the model with the meanof the slope constrained to be equal amongmales andfemales with the nonconstrained model, Dv2(1) 5

7.64, p , .001. The variances of the growth factors

were also estimated and they suggested that therewas a significant variation in individual differences inthe initial status and in the growth rate and that thesewere the same for both males and females—wecompared themodelwith the variance of the interceptand the variance of slope constrained to be equalamong males and females with the nonconstrainedmodel, Dv2(2) 5 .08, p 5 .96. Finally, results demon-strate that the covariance between the initial level andthe slope was significantly negative and equal formales and females, that is, the higher the level ofmoral disengagement at age 14, the less the change ofmoral disengagement during adolescence—we com-pared themodel with the covariance among interceptand slope constrained to be equal among males andfemales with the nonconstrained model, Dv2(1)5 .59,p 5 .44.

1,5

2

2,5

3

MD14 MD16 MD18 MD20

Males Females

Figure 1. Moral disengagement development.Note. MD145moral disengagement at age 14; MD165moral disengagement at age 16; MD185moral disengagement at age 18; MD205moral disengagement at age 20.

Table 4

Growth Curve Parameters

Growth parameters

Males Females

Effect t value Effect t value

Mean

Intercepta 2.68 78.25 2.51 75.51

Slopea �0.36 �11.67 �0.46 �13.61

Variances

Intercept 0.19 3.66 0.19 3.66

Slope 0.08 2.61 0.08 2.61

Correlation

Intercept4slope �.061 �2.06 �0.61 �2.06

Note. The tvalues greater than 1.96 (1.65 for variances) inmagnitudeindicate a parameter estimate that is significantly different fromzero (for p , .05). Parameters estimated for correlation (4) arepresented in standardized form. All other parameter estimates arepresented in unstandardized form.aParameter is different for males and females.


After ascertaining the best fitting growth curve, weadded the covariate peer evaluation of physical andverbal aggression at age 12 and the three outcomes toexamine the influence exerted by the covariate onmoral disengagement at age 14, on its developmentduring adolescence, and on physical and verbalaggression and violence at age 20 separately formalesand females. Furthermore,we examined the influenceexerted by moral disengagement at age 14 and itsdevelopment during adolescence on physical andverbal aggression and violence at age 20. The fitindices of themodel suggested a good approximationof the model to the data. In particular, v2(34,N5 177,188)5 28.077, p5 .75, CFI5 1, RMSEA5 .000 (.000�.039), SRMR 5 .082.

The results of the regression part of this model,summarized in Table 5, indicate that peer evalua-tion of physical and verbal aggression at age 12significantly predicted a higher level of moraldisengagement at age 14 for both males and fe-males. Peer evaluation of physical and verbalaggression at age 12 affected physical and verbalaggression at Time 5 only indirectly through moral

disengagement at age 14 (intercept; for physicalaggression: male, b5 .17, z5 3.415; female, b5 .11,z 5 2.647; for verbal aggression: male, b 5 .18, z 5

3.472; female, b 5 .12, z 5 2.783). Moreover, peerevaluation of physical and verbal aggression at age12 directly affected violence at Time 5 for malesonly and indirectly through moral disengagementat age 14 (intercept) both for males and females(male: b5 .11, z 5 3.421; female: b5 .14, z 5 2.954).Furthermore, moral disengagement measured atage 14 significantly predicted physical and verbalaggression and violent behavior 6 years later forboth males and females. In particular, the higherthe level of moral disengagement at age 14, thehigher the levels of physical and verbal aggressionand violence problems at age 20. In addition,results suggested that the more moral disengage-ment decreased from age 14 to 16, the lower theexpected levels of physical and verbal aggressionand violence at age 20 for both males and females.The model accounted for 33% and 24% of thevariance of verbal aggression, 34% and 25% of thevariance of physical aggression, and 28% and 34%of the variance of violence for males and females,respectively.

Trajectory Model

A semiparametric group-based approach (Nagin,1999, 2005) was then used to identify developmentaltrajectories. The Bayesian information criterion (BIC;Schwarz, 1978) was used to determine the optimalnumber of groups. This method allows: (a) the iden-tification of subgroups of individuals followingdistinct developmental trajectories for a given dimen-sion, (b) the estimation of the proportion of individ-uals following each trajectory, and (c) the estimationof the stability of each subgroup over time (i.e., theshape of each trajectory over time). Censored normalmodels were estimated. A detailed description of thestatistical rationale underlying the trajectory estima-tion procedure is given elsewhere (e.g., Jones &Nagin, 2007; Jones, Nagin, & Roeder, 2001; Nagin,1999, 2005).

The analysis proceeded in four stages. First, theoptimal number of groups was identified andmodelswith various specifications for stable, linear, qua-dratic, or cubic shapes of the moral disengagementtrajectory groupswere estimated. In the second stage,after ascertaining the best trajectory model, we testedwhether and how much gender and peer evaluationof physical and verbal aggression at age 12 affectedthe probability of group membership. In the thirdstage, we validated the trajectory model examining

Table 5

The Effect of the Growth Curve of Moral Disengagement on Aggressive

Behavior

Males Females

Effect t value Effect t value

Correlation

VIO4PHY .18 4.17 .28 4.17

VIO4VERB .12 3.20 .21 3.20

VERB4PHYa .28 3.97 .18 2.52

Regression

Agg_Pe/Intercepta .26 3.98 .25 3.33

Agg_Pe/Slope .00 1.69 .14 1.68

Agg_Pe/PHY .00 0.35 .02 0.35

Agg_Pe/VERB .00 0.01 .00 0.01

Agg_Pe/VIOa .21 2.92 .00 0.03

Intercept/PHYa .66 6.58 .42 4.44

Slope/PHY .54 5.42 .60 5.42

Intercept/VERBa .67 6.90 .47 5.15

Slope/VERB .49 5.60 .59 5.60

Intercept/VIO .39 6.54 .54 6.54

Slope/VIO .49 5.48 .69 5.48

Agg_Pe/VIO .16 2.20 .00 0.07

Note. The values greater than 1.96 (1.65 for variances) in magnitudeindicate a parameter estimate that is significantly different fromzero (for p , .05). Parameters estimated for correlation (4) andregression (predictor / outcome) are presented in standardizedform. PHY 5 physical aggression at age 20; VERB 5 verbalaggression at age 20; VIO5 violence at age 20; Agg_Pe5 physicaland verbal aggression peer evaluation at age 12.aParameter is different for males and females.


the profiles of each group on four correlates acrosstime. In particular, we considered physical aggres-sion, verbal aggression, violence, and need for repa-ration. Finally, the linkage between trajectories ofmoral disengagement and aggressive and violentoutcomes at age 20 was tested (Nagin, 1999).

First stage: Identification of the developmental model.A four-group unconditional model showed the bestfit to the data (BIC5�867.58). The four-groupmodelwas respecified considering gender and peer evalua-tion of physical and verbal aggression as predictors ofthe trajectory group membership probability. Thefour-group conditional model was confirmed as thebestmodel (BIC5�832.89). Figure 2 presents the sizeand the shapes of the trajectory groups.

The developmental trajectory model suggested thefollowing groups:

d Nondisengaged group: composed of 37.9% ofthe adolescents characterized by a quadratictrend with initially low levels of moral disen-gagement followed by a significant decline;

d Normative group: composed of 44.5% of theadolescents characterized by a cubic trend withinitially moderate levels of moral disengage-ment followed by a significant decline;

d Later desister group: composed of 6.9% of theadolescents characterized by a cubic trend withinitially high-medium levels ofmoral disengage-ment followed by a significant increase fromages 14 to 16 and an even more significantdecline from ages 16 to 20;

d Chronic group: composed of 10.7% of the ado-lescents with constant medium-high levels ofmoral disengagement.

Table 6 presents the correlations between theactual grouping variables (dichotomous) and theposterior group membership probabilities.

With regard to the gender distribution of the grouptrajectories, there were a higher proportion of boys inthe chronic group and a higher proportion of girls inthe nondisengaged group (see Figure 2).

Second stage: Linking group membership to gender andaggressive predictor. Table 7 presents the results of themultinomial logit regression used to examinewhether and how much gender and peer evaluationof physical and verbal aggression affected the prob-ability of group; membership. The first panel ofTable 4 shows coefficient estimates and t statistics.The normative group served as the contrast group.Relative to the normative group, gender significantlyincreased the probability of membership in the non-disengaged group; that is, females had a higherprobability of being in this group than did males.Furthermore, higher levels of peer evaluation onphysical and verbal aggression significantly in-creased the probability of membership in the chronicand later desister groups (a 5 .05, one-tailed test).

The second panel of Table 7 shows the predictedprobabilities of group membership based on thecoefficient estimates. Both males and females whowere never nominated by their classmates had ahigh probability of belonging to the normative or

Figure 2. Moral disengagement trajectories.Note. The continuous lines represent trajectories of actual moral disengagement calculated as mean scores for adolescents in groupsidentified by Proc traj procedure. The broken lines represent predicted moral disengagement calculated with model’s coefficient estimates.Genderdistribution in each trajectory group: nondisengaged group: 23.1%male and 76.9% female (of the total sample, 18.7%male and 54.8%female); normative group: 54.7% male and 45.3% female (of the total sample, 52.4 % male and 38.3% female); chronic disengaged group:89.7%male and10.3% female (of the total sample, 21.1%male and2.1% female); laterdesistergroup: 59.1%male and40.9% female (of the totalsample, 7.8% male and 4.8% female).


nondisengaged groups (.86 and .96, respectively), butwhile females had a higher probability of beinga member of the nondisengaged group (.61), maleshad a higher probability of being a member of thenormative group (.55).

Considering a high-risk scenario in which adoles-cents were nominated by all classmates, the predictedprobability formales of being amember of the chronicgroup was .38, whereas it was .10 in the previousscenario. In a similar way, the predicted probabilityfor females of being a member of the chronic groupwas .14, whereas it was .02 in the previous scenario.Furthermore, both males and females had a high-predicted probability to belong to the later desister ornormative group (.58 and .70, respectively). Finally,the predicted probability of males being a member ofthe nondisengaged groupwas .04 instead of .31 in theprevious scenario. In a similar way, the predictedprobability for females to be a member of the non-disengaged group was .15, whereas it was .61 in theprevious scenario.

Third stage: Trajectories profiles on aggression, violenceand need for reparation. Figure 3 shows the profiles inphysical aggression,verbal aggression,violent conduct,and need for reparation for each group across time.

As shown, the four trajectory groups significantlydiffered on physical aggression (Duncan post hoctest). At ages 14, F(3, 229) = 40.02, p , .000, g2 5 .21,and 16, F(3, 317) = 42.47, p , .001, g2 5 .29, all fourgroups were different from each other. In particular,the later desister group showed the highest level ofphysical aggression, whereas the nondisengagedgroup showed the lowest level. At age 18, F(3, 330)= 44.94, p , .001, g2 5 .29, the nondisengaged groupshowed the lowest level of physical aggression,whereas the later desister and chronic groups didnot differ from each other and showed the highestlevel. Finally, at age 20, F(3, 299) = 22.02, p, .001,g25

.18, the chronic group showed the highest level ofphysical aggression.

With regard to verbal aggression, the four trajec-tory groups were significantly different (Duncan posthoc test). In particular, at age 14, F(3, 330) = 20.01, p,.000, g2 5 .15, the later desister and chronic groupsdid not differ from each other and showed the highestlevel of verbal aggression. At age 16, F(3, 317) = 37.38,p , .001, g2 5 .26, the later desister and chronicgroups did not differ from each other and showed thehighest level of verbal aggression and the nondisen-gaged showed the lowest. At age 18, F(3, 330) = 42.53,

Table 6

Correlation Between the Actual Grouping Variables and the Posterior Group Membership Probabilities

Normative PRB Nondisengaged PRB Later desister PRB Chronic PRB

Normative .94 �.62 �.24 �.28

Nondisengaged �.65 .95 �.26 �.32

Later desister �.25 �.23 .95 �.01

Chronic disengaged �.30 �.32 .05 .95

Note. All correlation coefficients were significant at least at p , .05, except for those in italics, which are not significant. PRB 5 posteriorprobabilities.

Table 7

The Impact of Gender and Physical and Verbal Aggression Peer Evaluated on Group Membership Probabilities

Groups

Later desister disengaged Chronic disengaged Normative Nondisengaged

Multinomial logit coefficients (with 6 statistics given in parenthesis)

Constant �2.67 (�3.87) �2.84 (�3.45) — 0.56 (1.86)

Gender �0.05 (0.09) �1.15 (�1.37) — �1.12 (�3.38)

Agg_Pe 2.78 (2.11) 2.01 (1.89) — �1.38 (�1.46)

Predicted membership probabilities based on multinomial model coefficient estimates

Males and Agg_Pe 5 0 .04 .10 .55 .31

Females and Agg_Pe5 0 .02 .02 .35 .61

Males and Agg_Pe 5 1 .30 .38 .28 .04

Females and Agg_Pe5 1 .37 .14 .33 .15

Note. Agg_Pe5physical andverbal aggression peer evaluation at age 12.Agg_Pe5 0: never nominated by classmates; Agg_Pe5 1: nominated byall classmates.


p, .001, g2 5 .28, all four groups were different fromeach other. In particular, the chronic group showedthe highest level of verbal aggression and the non-disengaged the lowest. Finally, at age 20, F(3, 299) =21.53, p, .001,g25 .18, the chronic group showed thehighest level of verbal aggression, whereas the nor-mative and later desister groups did not differ fromeach other.

With regards to violence, the four trajectorygroups were significantly different (Duncan posthoc test). In particular, at age 16, F(3, 317) = 48.02,p , .000, g2 5 .31, all four groups were differentfrom each other. In particular, the later desistergroup showed the highest level of violence. At age18, F(3, 330) = 24.42, p , .001, g2 5 .18, the chronicand later desister groups did not differ from eachother and showed the highest level of violence.Finally, at age 20, F(3, 286) = 27.33, p , .001, g2 5

.22, the chronic group showed the highest level ofviolence, whereas the later desister group showedhigher levels of violence than the normative andnondisengaged groups.

Finally, the four trajectory groups significantlydiffered on need for reparation (Duncan post hoctest). In particular, at ages 14, F(3, 246) = 5.44, p, .000,

g25 .05, and 16, F(3, 327) = 8.92, p, .001,g25 .08, thenondisengaged group showed the highest level ofneed for reparation. At age 18, F(3, 342) = 15.02, p ,

.001, g2 5.12, the nondisengaged group showed thehighest level and the chronic showed the lowest.Finally, at age 20, F(3, 299) = 12.60, p , .001, g2 5

.11, the chronic group showed the lowest level of needfor reparation, whereas the other three groups did notdiffer.

In each figure, the correlations between moraldisengagement and each variable were reported. Asshown, moral disengagement was positive andhighly correlatedwith physical and verbal aggressionand violence and negatively correlated with need ofreparation in each time point.

Fourth stage: The impact of moral disengagementtrajectories on aggressive and violent conduct. Table 8presents the correlations between the posterior prob-abilities of groupmembership for each individual andaggressive and violent conduct at age 20.

As shown, aggressive and violent outcomes werepositively and significantly correlated with the pos-terior probability of being in the chronic or laterdesister groups. On the contrary, the outcomes werenegatively and significantly correlated with the

-1

0

1

2

-1

0

1

2

-1

0

1

2

-1

0

1

a

b

c

d

a

b

c

d

a

a

b

b

a

bbc

c

aa

b

b

a

a

b

c

a

b

c

d

a

b

bcc

a

aa

b

a

b

b

c

a

bb

b

a

bbb

a

b

c

c

a

a

b

b

a

b

c

d

r= .53 r =.58 r=.4 =.34 r=.5 =.52

r=-.15 r=-.19 r=-.32 r=-.32r=.49 r=.48 r=.46

-1

0

1

2

-1

0

1

2

-1

0

1

2

-1

0

1

-1

0

1

2

PHY 14 PHY 16 PHY 18 PHY 20-1

0

1

2

VERB 14 VERB 16 VERB 18 VERB 20

-1

0

1

2

VIO 16 VIO 18 VIO 20-1

0

1

GUILT 14 GUILT 16 GUILT 18 GUILT 20

a

b

c

d

a

b

c

d

a

a

b

b

a

bbc

c

aa

b

b

a

a

b

c

a

b

c

d

a

b

bcc

a

aa

b

a

b

b

c

a

bb

b

a

bbb

a

b

c

c

a

a

b

b

a

b

c

d

r=.41 r=.53 r=.58 r=.47 r=.34 r=.51 r=.52 r=.50

r=-.15 r=-.19 r=-.32 r=-.32r=.49 r=.48 r=.46

Normative Non-Disengaged Chronic DisengagedLater Desister

Figure 3. Trajectories profiles on aggressive behavior, violence, and need for reparation.Note. PHY14 5 physical aggression at age 14; PHY16 5 physical aggression at age 16; PHY18 5 physical aggression at age 18; PHY20 5

physical aggression at age 20; VERB145 verbal aggression at age 14; VERB165 verbal aggression at age 16; VERB185 verbal aggression atage 18; VERB205 verbal aggression at age 20; VIO165 violence at age 16; VIO185 violence at age 18; VIO205 violence at age 20; GUILT145need for reparation at age 14; GUILT16 5 need for reparation at age 16; GUILT18 5 need for reparation at age 18; GUILT20 5 need forreparation at age 20. In each figure correlations between moral disengagement and each variable are reported for each time point.


posterior probability of being in the nondisengagedgroup. Moreover, correlations were not significantbetween outcomes and the posterior probability ofbeing in the normative group.

To test the linkage between trajectories of moraldisengagement and the three outcomes (i.e., physicalaggression, verbal aggression, and violent behaviorsat age 20), three hierarchical regression analyses wereconducted. In these models, gender, peer evaluationon physical and verbal aggression at age 12, and theposterior probabilities of group membership wereconsidered independent variables. In entering theposterior probabilities, we excluded the probabilityof group membership for one of the groups becausethe posterior probabilities of group membership ineach group added to 1 for a given individual (e.g., theindependent variableswould be perfectly correlated).We decided to exclude the probability of being in thenormative group.

As a preliminary step, we evaluated whether theseanalyses needed to be conducted separately formalesand females. In this regard, the interaction of Gender

� Peer Evaluation on Physical and Verbal Aggressionand the interaction of Gender � The Different GroupMembership Probabilities were included as indepen-dent variables in the hierarchical regressionmodel. Inparticular, gender was entered at Step 1, peer evalu-ation of physical and verbal aggressionwas entered atStep 2, the interaction of Gender� Peer Evaluation ofPhysical and Verbal Aggressionwas entered at Step 3,the different group membership probabilities wereentered at Step 4, and the interaction of Gender� TheDifferent Group Membership Probabilities wereentered at Step 5. The analysis revealed no statisticallysignificant interactions; therefore, our subsequenthierarchical regression analyseswere performedwithgender as a factor rather than separately for the twogender groups.

The final hierarchical regression models are pre-sented in Table 9. The results of the three hierarchicalmodels showed that gender significantly predictedthe three outcomes, especially violence (17% of ex-plained variance). Moreover, peer nomination ofphysical and verbal aggression at age 12 significantlypredicted higher levels of physical and verbal aggres-sion and violent conduct at age 20 (the explainedvariance ranged from 2% to 5%).

In addition to the contribution of gender andearlier peer evaluations of aggressive behavior, theposterior probability of group membership signifi-cantly predicted the three outcomes (the explainedvariance ranged from 10% to 13%). In particular, theprobability of being in the chronic group was posi-tively and significantly linkedwith verbal aggression,physical aggression, and violence, whereas the prob-ability of being amemberof the nondisengaged groupwas negatively and significantly linked with aggres-sion and violence.

Table 8

Correlations Between the Posterior Probabilities of Group Membership

and Physical and Verbal Aggression and Violence at Age 20

Posterior probabilities

Verbal

aggression

Physical

aggression Violence

Normative .07 .04 .10

Nondisengaged �.35 �.33 �.42

Later desister .10 .17 .20

Chronic .39 .36 .39

Note. All correlation coefficients were significant at least at p, .05,except the one that is underlined, which is significant at p, .10, andthose that are italics, which are not significant.

Table 9

Results of Hierarchical Regression of Trajectories on Physical and Verbal Aggression and Violence at Age 20

Step

Verbal aggression Physical aggression Violence

b1 b2 b3 R2 change b1 b2 b3 R2 change b1 b2 b3 R2 change

1 Gender .16*** .19** ns .07*** .23*** .15** ns .05*** .41*** .30*** .18*** .17***

2 Aggr_Pe .16** ns .02*** .20*** .11y .04*** .25*** .16*** .05***

3 Nondisengaged �.20*** .13*** �.18** .11*** �.22*** .10***

Chronic disengaged .28*** .25*** .20**

Later desister ns ns ns

R2 .21 .19 .32

Note. b1 5 beta coefficients before entering trajectories probabilities and levels of outcomes at Time 1; b2 5 coefficients before enteringtrajectories probabilities; b35 final beta coefficients;R2 change5 change inR2—for the first step it refers toR2; Agg_Pe5 physical and verbalaggression peer evaluation at age 12.ySignificant at p �. 07. **Significant at p �. 01. ***Significant at p �. 001.


Discussion

As previous studies have highlighted (Bandura, 1999;Bandura et al., 1996, 2001; Caprara et al., 1996), thelongitudinal findings of the present study attested tothe important predictive role that moral disengage-ment plays in the study of aggressive and violentbehaviors. At the population level, for both boys andgirls, the developmental model attested to the generaltendency of moral disengagement to decline overtime. In particular, moral disengagement decreasedstrongly between ages 14 and 16 and less evidentlyuntil age 20. The general decrease in moral disen-gagement could be specific to the developmentalperiod from early adolescence (junior high school)to adolescence (high school), which is characterizedby new challenges related to educational and socialrole transitions. It could reflect a change in cognitiveand social structures and processes through develop-ment of the capacity to assign meaning, to anticipateoutcomes, to plan actions, and to learn from socialexperiences the value of assigning different behav-iors. Generally, over the course of maturation, indi-viduals’ social adjustment and self-regulatoryabilities improve and their ability to infer the perspec-tive of others increases (Bandura, 1991, 2004), promo-ting moral reasoning and moral agency (Eisenberg,2000; Eisenberg et al., 2006). Indeed, adolescents areexposed to novel ‘‘role-taking opportunities’’ andmoral dilemmas with personally relevant consequen-ces to the self and to others (Hart &Atkins, 2002; Hart,Atkins, Markey, & Youniss, 2004). It may be that theserefinements in skills are indispensable in preventingdisengagement in ‘‘moral life’’ during adolescence,a timewhen individuals start to becomemore respon-sible and agentic in their social life than was typicallypossible in childhood (Hart & Carlo, 2005).

Furthermore, results attested to the significantvariation in individual differences in the initial statusand in the general developmental pathway of moraldisengagement, supporting the importance of furtherexploring interindividual differences during devel-opment. At this level, the developmental trajectoriesmodel confirmed what was found at the populationlevel and provides a more comprehensive picture ofthe development of moral disengagement. In fact,although distinct groups of adolescents followeddifferent developmental trends, most adolescents(89%) decreased their level of moral disengagementduring this time. As expected, the best developmentaltrajectory model included four different trajectorygroups in accordance with the existing aggressionliterature (Barker et al, 2006; Brame et al., 2001; Broidyet al., 2003; Cote et al., 2002; Nagin & Tremblay, 1999;

Tremblay & Nagin, 2005) and other developmentalstudies on aggressogenic personality factors such asirritability and hostile rumination, which have beenshown to be related to moral disengagement andaggression (Caprara et al., 2007). The largest groupmay be considered as representing the ‘‘normative’’group and started with medium levels of moraldisengagement, which subsequently decreased dur-ing development. A similar trendwas followed by thegroup called ‘‘nondisengaged,’’ which started withlower levels of moral disengagement than the norma-tive group. The last two smaller groups started withinitial high-medium levels of moral disengagementbut followed different developmental trends. The‘‘later desister group’’ showed a discontinuous pat-tern of moral disengagement decrease. In particular,their level of moral disengagement increased fromages 14 to 16 and then strongly decreased fromages 16to 20. In contrast, the ‘‘chronic group’’ showed stablelevels of moral disengagement during adolescence.Although these two groups demonstrated similarphysical and verbal aggression and violence profilesfrom ages 14 to 16; from ages 18 to 20, they showeddifferent developmental trends: The chronic groupmaintained their high levels of aggression, whereasthe later desister group demonstrated decreasinglevels of aggression. These two groups were similarto findings of previous longitudinal studies that haveidentified a group of youth who show chronicallyhigh levels of antisocial and aggressive behaviorduring adolescence and adulthood (Brame et al.,2001; Moffitt, Caspi, Harrington, & Milne, 2002) anda group that later decrease their level of problematicbehaviors during adolescence (Moffitt, 1993; Nagin,Farrington, & Moffit, 1995).

Both the later desister and chronic groups call one’sattention to two possible developmental trends foradolescents that showhigh levels ofmoral disengage-ment at age 14. Later desister adolescents decreasetheir level of moral disengagement probably byimproving, over the course of maturation, their cog-nitive and emotional skills, even if they continued toshowhigher levels of violence than their normative ornondisengaged peers. For chronic disengaged ado-lescents, moral disengagement could instead reflecta strategy of adaptation that is embedded into a sys-tem of beliefs about the self and others and leads toperceive aggression and violence as appropriatemeans to pursue one’s own goals. Adolescents whoconsider aggression amorally acceptable behavior aremore likely to act persistently aggressive (Crane-Ross,Tisak, & Tisak, 1998; Huesmann & Guerra, 1997;Keltikangas-Jarvinen, 2001) and to avoid affectiveself-evaluative reactions like guilt (Bandura, 1991;


Bandura et al., 1996). Profile analysis demonstratedthat at ages 14 and 16, the chronic group showed thesame level of need for reparation as did the normativeand the later desister groups.With increasing age, thechronic group decreased their level of need forreparation, whereas the later desister group contin-ued to show the same level as the normative group.These results highlight how affective self-evaluativereactions were associated with different moral disen-gagement trends. The need for reparation associatedwith the negative outcomes of guilt-eliciting actionsindicated that self-sanctions even existed at age 14,especially for the low disengaged group. During thecourse of adolescence, the chronic recourse to moraldisengagement mechanisms consented people toattenuate self-sanctions and guilt feelings.

In accordance with previous findings (Bandura,1999, 2004; Bandura et al., 1996), boys showed higherlevels of moral disengagement than did girls anda lower decline from ages 14 to 16. Furthermore,adolescents in the chronic group were mostly boys,whereas most of nondisengaged group were girls.Several studies have attested that girls have highermoral emotional competence than boys (Eisenberg,2000; Ferguson&Crowley, 1997; Lennon&Eisenberg,1987; Mills, Pedersen, & Grusec, 1989) and lowerlevels of moral disengagement (Bandura, 2004; Ban-dura et al., 1996; Pelton et al., 2004) and aggressivebehaviors (Bettencourt & Miller, 1996; Knight, Gu-thrie, Page, & Fabes, 2002). Furthermore, resultsshowed that boys evaluated as aggressive by theirpeers already in early adolescenceweremore at risk torecourse to moral disengagement than girls. More-over, both boys and girls that were never nominatedby their classmates had a higher probability of beinga member of both the normative and the nondisen-gaged groups. Thus, it is more probable that adoles-cents, especially boys, who have already shownaggressive behaviors in early adolescence, have moreopportunities to recourse to moral disengagement.

Regarding the outcomes, the stability and changeofmoral disengagement significantly affected aggres-sive and violent behaviors. In fact, on the one hand,decreasing levels of moral disengagement at thepopulation level were significantly linked with lowerlevels of physical and verbal aggression and violencein late adolescence. On the other hand, the analysis ofthe predictive power of moral disengagement trajec-tories showed that different trends were differentlyassociated with aggressive and violent outcomes. Forthe nondisengaged youth (above all girls), the declineover time was negatively associated with aggressiveand violent behavior. Instead, for the later desisteryouth,whichweremore likely to be at risk, the decline

over time was not significantly associated with ag-gressive and violent outcomes. In this last case (7% oftotal sample), we cannot exclude the beneficial effectdue to a later development of social emotional skills.For the chronic group, who showed high stable levelsof moral disengagement over the course of time andthe lowest levels of self-sanction for their offensivebehaviors in late adolescence, the relation betweenmoral disengagement development and aggressiveand violent outcomes was particularly problematic.In fact, the results indicated that adolescents witha high probability of being a member of this groupwere more likely to show frequent aggressive andviolent behavior in late adolescence, controlling forgender and peer nomination of physical and verbalaggression at age 12. For these youth, the stabletendency toward moral disengagement may reflecta crystallization ofmoral disengagementmechanismsover time that subsequently legitimatizes the recourseto aggressive and violent behaviors. The influencethat moral disengagement exerted on engagement inaggression and violent episodes for these adolescence(above boys) calls one’s attention to the psychologicalprocesses and social models that provide the cogni-tive framework within which retaliation, revenge,and violence appear appropriate and acquire legiti-macy (Arsenio & Lemerise, 2004; Caprara, 1996).

In sum, the longitudinal findings of the presentstudy indicated that: (a) most adolescents exhibiteddeclining levels of moral disengagement over time;(b) adolescents who exhibited lower levels of moraldisengagement than the normative group were typi-cally girls, less likely to show aggression and violentacts in late adolescence and more likely to feel guilt;(c) adolescents who showed high levels of aggressionin early adolescence were more likely to recourse tomoral disengagement; (d) adolescents who main-tained higher levels of moral disengagement weretypically boys, were more likely to show frequentphysical and verbal aggression and violent acts, andwere less likely to feel guilt in late adolescence; and (e)adolescents who were more likely to exhibit aggres-sion and violent acts in late adolescence were typi-cally boys evaluated as aggressive by peers in earlyadolescence.

Limitations and Future Directions

The identification of developmental pathways ofmoral disengagement represents a first step in under-standing the processes that drive some people toresort to aggression and violence more frequentlythan others in late adolescence. However, clarificationof some issues remains necessary. First, our findings


call for some caution in light of the possible limita-tions of the study due to the specific cultural contextwhere the research was conducted and the use of self-report data. Second, because there are different ap-proaches in studying moral development, futureresearch should combine the study of moral disen-gagement (see Bandura, 1991) with the examinationof moral reasoning and moral judgments (Kochan-ska & Aksan, 2006; Yau & Smetana, 2003) over thecourse of development. Third, future studies shouldexamine the role of self-evaluative reactions inpreventing the chronic recourse to moral disengage-ment mechanisms that may get crystallized overtime. In this respect, different studies have high-lighted the importance of studying affective reac-tions such as guilt, remorse, and empathy inassociation with moral development (Eisenberg,2000; Hoffman, 1979; Kochanska & Aksan, 2006;Tangney et al., 2007). Finally, further studies shouldclarify how differences in moral disengagement areassociated with emotional, cognitive, and social pro-cesses implied in moral self-regulation (Hart &Carlo, 2005; Turiel, 2006) and how individual skillsand environmental factors could affect both individ-ual differences and aggressive and violent outcomes.In this regard, we believe that it is necessary tocombine the systematic study of individual differ-ences with the study of external situational variableswithin a longitudinal perspective. The study of bothinternal mechanisms and relevant environmentalvariables that operate at stimulus, social, contextual,structural, and cultural levels to influence individ-ual, and group behaviors are necessary in designingappropriate interventions aimed at preventingaggressive and violent outcomes.

References

Anderson, J. C., & Gerbing, D. W. (1988). Structuralequation modeling in practice: A review and recom-mended two-step approach. Psychological Bulletin, 103,411 – 423.

Andrus, B. C. (1969). The infamous of Nuremberg. London:Fravin.

Arbuckle, J. L. (1996). Full information estimation in thepresence of incomplete data. In G. A. Marcoulides & R. E.Schumacker (Eds.), Advanced structural equation modeling:Issues and techniques (pp. 243 – 277).Mahwah, NJ: Erlbaum.

Archer, J. (2004). Sex differences in aggression in real-world settings: A meta-analytic review. Review of GeneralPsychology, 8, 291 – 322.

Arsenio, W., & Lemerise, E. (2004). Aggression and moraldevelopment: Integrating the social information process-ing and moral domain models. Child Development, 75,987 – 1002.

Bandalos, D. L., & Finney, S. J. (2001). Item parceling issuesin structural equation modeling. In G. A. Marcoulides &R. E. Schumacker (Eds.), Advanced structural equationmodeling: New developments and techniques (pp. 269 –296). Mahwah, NJ: Erlbaum .

Bandura, A. (1986). Social foundation of thought and action: Asocial cognitive theory. Englewood Cliffs, NJ: Prentice Hall.

Bandura, A. (1990). Mechanism of moral disengagement.In W. Reich (Ed.), Origin of terrorism: Psychologies,ideologies, theologies, states of mind (pp. 161 – 191). Cam-bridge, UK: Cambridge University Press.

Bandura, A. (1991). Social cognitive theory of moralthought and action. In W. M. Kurtines & J. L. Gewirtz(Eds.), Handbook of moral behaviour and development (Vol. I,pp. 45 – 103). Hillsdale, NJ: Erlbaum.

Bandura, A. (1999). Moral disengagement in the perpetra-tion of inhumanities. Personality and Social PsychologyReview, 3, 193 – 209.

Bandura, A. (2001). Social cognitive theory: An agentic pers-pective. Annual review of psychology (Vol. 52, pp. 1 – 26).Palo Alto, CA: Annual Reviews.

Bandura, A. (2004). Selective exercise of moral agency. InT. A. Thorkildsen & H. J. Walberg (Eds.), Nurturingmorality (pp. 37 – 57). Boston: Kluwer Academic.

Bandura, A. (2006). Toward a psychology of humanagency. Perspectives on Psychological Science, 1,164 – 180.

Bandura, A., Barbaranelli, C., Caprara, G. V., & Pastorelli,C. (1996). Mechanism of moral disengagement in theexercise of moral agency. Journal of Personality and SocialPsychology, 71, 364 – 374.

Bandura, A., Caprara, G. V., Barbaranelli, C., Pastorelli, C.,& Regalia, C. (2001). Sociocognitive self-regulatorymechanisms governing transgressive behavior. Journalof Personality and Social Psychology, 80, 125 – 135.

Bandura, A., Caprara, G. V., & Zsolnai, L. (2000). Corpo-rate transgressions through moral disengagement. Jour-nal of Human Values, 6, 57 – 64.

Bandura, A., Underwood, B., & Fromson, M. E. (1975).Disinhibition of aggression through diffusion of respon-sibility and dehumanizaion of victims. Journal of Researchin Personality, 9, 253 – 269.

Bandura, A., & Walters, R. H. (1959). Adolescent aggres-sion: A study of the influence of child-training practicesand family interrelationships. Oxford, UK: Ronald Press.

Barker, E. D., Tremblay, R. E., Nagin, D. S., Vitaro, F., &Lacourse, E. (2006). Development of male proactive andreactive physical aggression during adolescence. Journalof Child Psychology and Psychiatry, 47, 783 – 790.

Bentler, P. M. (1989). EQS Structural Equations ProgramManual. Manual for Version 3.0. Los Angeles: BMDPStatistical Software.

Bergman, R. (2002). Why be moral? A conceptual modelfrom developmental psychology. Human Development,45, 104 – 124.

Bettencourt, A. B., & Miller, N. (1996). Gender differencesin aggression as a function of provocation: A meta-analysis. Psychological Bulletin, 119, 422 – 447.


Blasi, A. (1984). Moral identity: Its role in moral function-ing. In W. W. Kurtines & J. L. Gewirtz (Eds.), Morality,moral behavior, and moral development (pp. 128 – 139). NewYork: Wiley.

Bollen, K. A. (1989). Structural equations with latent variables.New York: Wiley.

Bollinger, D. (1982). Language: The loaded weapon. London:Longman.

Brame, B., Nagin, D. S., & Tremblay, R. E. (2001). Devel-opmental trajectories of physical aggression from schoolentry to late adolescence. Journal of Child Psychology andPsychiatry, 42, 503 – 512.

Brock, T. C., & Buss, A. H. (1962). Dissonance, aggression,and evaluation of pain. Journal of Abnormal and SocialPsychology, 65, 197 – 202.

Brock, T. C., & Buss, A. H. (1964). Effects of justification foraggression and communication with the victim onpostaggression dissonance. Journal of Abnormal and SocialPsychology, 68, 403 – 412.

Broidy, L. M., Nagin, D. S., Tremblay, R. E., Bates, J. E.,Brame, B., Dodge, K. A., et al. (2003). Developmentaltrajectories of childhood disruptive behaviors and ado-lescent delinquency: A six-site, cross-national study.Developmental Psychology, 39, 222 – 245.

Caprara, G. V. (1996). Structures and processes in person-ality psychology. European Psychologist, 1, 14 – 26.

Caprara, G. V., Bandura, A., Barbaranelli, C., & Vicino, S.(1996). LaMisura del Disimpegnomorale [The assessmentofmoraldisengagement].Rassegna diPsicologia,13, 93 – 105.

Caprara, G. V., Barbaranelli, C., Pastorelli, C., Cermak, I., &Rosza, S. (2001). Facing guilt: Role of negative affectivity,need for reparation, and fear of punishment in leadingto prosocial behaviour and aggression. European Journalof Personality, 15, 219 – 237.

Caprara, G. V., & Capanna, C. (2005). Moral-civic disen-gagement and values. Ricerche di Psicologia, 28(1), 67 –83.

Caprara, G. V., Manzi, J., & Perugini, M. (1992). Investi-gating guilt in relation to emotionality and aggression.Personality and Individual Differences, 13, 519 – 532.

Caprara, G. V., Mazzotti, E., & Prezza, M. (1990). Una scalaper la misura dell’atteggiamento verso la violenza [Ascale for the measurement of attitude toward violence].Giornale Italiano di Psicologia, 17, 107 – 120.

Caprara, G. V., Paciello, M., Gerbino, M., & Cugini, C.(2007). Individual differences conducive to aggressionand violence: Trajectories and correlates of irritabilityand hostile rumination through adolescence. AggressiveBehavior, 33, 359 – 374.

Caprara, G. V., & Pastorelli, C. (1993). Early emotionalinstability. Prosocial behavior and aggression: Somemeth-odological aspects.European Journal of Personality, 7, 19 – 36.

Carlo, G., Eisenberg, N., & Knight, G. P. (1992). Anobjective measure of adolescents’ prosocial moral rea-soning. Journal of Research on Adolescence, 2, 331 – 349.

Cattell, R. B., & Vogelmann, S. (1977). A comprehensivetrial of the scree and KG criteria for determining the

number of factors. Multivariate Behavioral Research, 12,289 – 325.

Cheung, G. W., & Rensvold, R. B. (2002). Evaluatinggoodness-of-fit indexes for testing measurement invari-ance. Structural Equation Modeling, 9, 233 – 255.

Cohen, D., & Nisbett, R. E. (1994). Self-protection and theculture of honor: Explaining southern violence. Person-ality and Social Psychology Bulletin, 20, 551 – 567.

Cote, S., Tremblay, R. E., Nagin, D. S., Zoccolillo, M., &Vitaro, F. (2002). Childhood behavioral profiles leadingto adolescent conduct disorder: Risk trajectories for boysand girls. Journal of the American Academy of Child andAdolescent Psychiatry, 41, 1086 – 1094.

Crane-Ross, D., Tisak, M. S., & Tisak, J. (1998). Aggressionand conventional rule violation among adolescents:Social-reasoning predictors of social behavior. AggressiveBehavior, 24, 347 – 365.

Crick, N. R., & Dodge, K. A. (1994). A review andreformulation of social information-processing mecha-nism in children’s social adjustment. Psychological Bulle-tin, 115, 74 – 101.

Darley, J. M., Klosson, E. C., & Zanna, M. P. (1978).Intentions and their contexts in the moral judgments ofchildren and adults. Child Development, 49, 66 – 74.

Diener, E. (1977). Deindividuation: Causes and consequen-ces. Social Behavior and Personality, 5, 145 – 156.

Diener, E., Dineen, J., Endresen, K., Beaman, A. L., &Fraser, S. C. (1975). Effects of altered responsibility,cognitive set, and modeling on physical aggressionand deindividuation. Journal of Personality and SocialPsychology, 31, 328 – 337.

Dodge, K. A., & Crick, N. R. (1990). Social information-processing bases of aggressive behavior in children.Personality and Social Psychology Bulletin, 16, 8 – 22.

Duncan, S. C., & Duncan, T. E. (1994). Modeling incom-plete longitudinal substance use data using latent vari-able growth curve methodology. Multivariate BehavioralResearch, 29, 313 – 338.

Duncan, T. E., & Duncan, S. C. (1995). Modeling theprocesses of development via latent growth curve meth-odology. Structural Equation Modeling, 2, 187 – 213.

Duncan, T. E., Duncan, S. C., & Hops, H. (1994). The effectsof family cohesiveness and peer encouragement on thedevelopment of adolescence alcohol use: A cohort-sequential approach to the analysis of longitudinal data.Journal of Studies on Alcohol, 55, 588 – 599.

Eisenberg, N. (2000). Emotion, regulation, and moraldevelopment. Annual Review of Psychology, 51, 665 –697.

Eisenberg, N., Fabes, R. A., & Spinrad, T. L. (2006).Prosocial development. Hoboken, NJ: John Wiley & Sons.

Elkin, F., & Westley, W. A. (1955). The myth of adolescentculture. American Sociological Review, 20, 680 – 684.

Elliott, D. S., & Rhinehart, M. (1995). Moral disengagement,delinquent peers and delinquent behaviour. Unpublishedmanuscript, University of Colorado, Institute of Behav-ioral Science.


Emmons, R. A., & Diener, E. (1986). Situation selection asa moderator of response consistency and stability. Jour-nal of Personality and Social Psychology, 51, 1013 – 1019.

Ferguson, T. J., & Crowley, S. L. (1997). Gender differences inthe organization of guilt and shame. Sex Roles, 37, 19 – 43.

Ferguson, T. J., & Rule, B. G. (1983). An attributionalperspective on anger and aggression. In R. G. Geen &E. I. Donnerstein (Eds.), Aggression: Theoretical and empir-ical reviews (Vol. 1, pp. 41 – 74). New York: AcademicPress.

Fix, E., Hodges, J. L., & Lehmann, E. L. (1959). Therestricted chi-square test. In U. Grenanader (Ed.), Prob-ability and statistics: The herald Cramer volume (pp. 92 – 97).New York: Wiley.

Gibbs, J. C., Potter, G. B., Barriga, A. Q., & Liau, A. K.(1996). Developing the helping skills and prosocialmotivation of aggressive adolescents in peer groupprograms. Aggression and Violent Behavior, 1, 283 – 305.

Gini, G. (2006). Social cognition and moral cognitionin bullying: What’s wrong? Aggressive Behavior, 32,528 – 539.

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

Hardy, S. A., & Carlo, G. (2005). Identity as a source ofmoral motivation. Human Development, 48, 232 – 256.

Haritos-Fatouros, M. (1988). The official torturer: A learn-ing model for obedience to the authority of violence.Journal of Applied Social Psychology, 18, 1107 – 1120.

Hart, D., & Atkins, R. (2002). Civic competence in urbanyouth. Applied Developmental Science, 6, 227 – 236.

Hart, D., Atkins, R., Markey, P., & Youniss, J. (2004). Youthbulges in communities: The effects of age structure onadolescent civic knowledge and civic participation.Psychological Science, 15, 591 – 597.

Hart, D., & Carlo, G. (2005). Moral developmentin adolescence. Journal of Research on Adolescence, 15,223 – 233.

Helwig, C. C., & Turiel, E. (2003). Children’s social andmoral reasoning. In P. K. Smith & C. H. Hart (Eds.),Blackwell handbook of childhood social development (pp.476 – 490). Malden, MA: Blackwell.

Hoeksma, J. B., & Kelderman, H. (2006). On growth curvesand mixture models. Infant and Child Development, 15,627 – 634.

Hoffman, M. L. (1979). Development of moral thought,feeling, and behavior. American Psychologist, 34, 958 – 966.

Hoyle, R. H. (1995). Structural equation modeling: Concepts,issues, and applications. Thousand Oaks, CA: Sage.

Hoyle, R. H., & Smith, G. T. (1994). Formulating clinicalresearch hypotheses as structural equation models: Aconceptual overview. Journal of Consulting and ClinicalPsychology, 62, 429 – 440.

Huesmann, L. R. (1988). An information processing modelfor the development of aggression. Aggressive Behavior,14, 13 – 24.

Huesmann, L. R. (1998). The role of social informationprocessing and cognitive schema in the acquisition andmaintenance of habitual aggressive behavior. San Diego,CA: Academic Press.

Huesmann, L. R., & Guerra, N. G. (1997). Children’snormative beliefs about aggression and aggressivebehavior. Journal of Personality and Social Psychology, 72,408 – 419.

Istituto Italiano di Statistica. (2002). Annuario statisticoItaliano 2000. [Italian yearbook of statistics 2002]. Rome,Italy: Author.

Jones, B. L., & Nagin, D. S. (2007). Advances in group-based trajectory modeling and an SAS procedure forestimating them. Sociological Methods and Research, 35,542 – 571.

Jones, B. L., Nagin, D., & Roeder, K. (2001). A SASprocedure based on mixture models for estimatingdevelopmental trajectories. Sociological Methods andResearch, 29, 374 – 393.

Joreskog, K. G., & Sorbom, D. (1984). LISRELVI: Analysis oflinear structural relationships by method of maximum likeli-hood. Chicago: National Educational Resources.

Keen, S. (1986). Faces of the enemy. San Francisco: Harper &Row.

Kelman, H. C. (1973). Violence without moral restraint:Reflections on the dehumanization of victims and vic-timizers. Journal of Social Issues, 29, 25 – 61.

Kelman, H. C., & Hamilton, V. L. (1989). Crimes of obedience:Toward a social psychology of authority and responsibility.New Haven, CT: Yale University Press.

Keltikangas-Jarvinen, L. (2001). Aggressive behaviourand social problem-solving strategies: A review of thefindings of a seven-year follow-up from childhood tolate adolescence. Criminal Behaviour and Mental Health,11(4), 236 – 250.

Killen, M., & Smetana, J. G. (2006). Handbook of moraldevelopment. Mahwah, NJ: Erlbaum.

Klass, E. T. (1978). Psychological effects of immoral actions:The experimental evidence. Psychological Bulletin, 85,756 – 771.

Knight, G. P., Guthrie, I. K., Page, M. C., & Fabes, R. A.(2002). Emotional arousal and gender differences inaggression: A meta-analysis. Aggressive Behavior, 28,366 – 393.

Kochanska, G., & Aksan, N. (2006). Children’s conscienceand self-regulation. Journal of Personality, 74, 1587 – 1617.

Kohlberg, L. (1969). Stage and sequence: The cognitive-developmental approach to socialization. In D. A. Goslin(Ed.), Handbook of socialization theory and research.Chicago: McNally.

Kohlberg, L., & Candee, D. (1984). The relationship ofmoral judgement to moral action. In W. M. Kurtines &J. Gewirtz (Eds.), Morality, moral behavior, and moraldevelopment (pp. 52 – 73). New York: Wiley.

Krettenauer, T. (2004). Metaethical cognition and epistemicreasoning development in adolescence. InternationalJournal of Behavioral Development, 28, 461 – 470.


Kwak, K., & Bandura, A. (1997). Role of perceived self-efficacyand moral disengagement in antisocial conduct. Unpub-lished manuscript, Osan College, Seoul, Korea.

Lennon, R., & Eisenberg, N. (1987). Gender and agedifferences in empathy and sympathy. In N. Eisenberg& J. Strayer (Eds.), Empathy and its development (pp. 195 –217). New York: Cambridge University Press.

Little, R. J. (1995). Modeling the dropout mechanism inrepeated-measures studies. Journal of the American Sta-tistical Association, 90, 1112 – 1121.

Loeber, R. (1991). Antisocial behavior: More enduring thanchangeable? Journal of the American Academy of Child andAdolescent Psychiatry, 30, 393 – 397.

Loeber, R., & Hay, D. (1997). Key issues in the develop-ment of aggression and violence from childhood to earlyadulthood. Annual Review of Psychology, 48, 371 – 410.

Loeber, R., & Stouthamer-Loeber, M. (1998). Developmentof juvenile aggression and violence: Some commonmisconceptions and controversies. American Psychologist,53, 242 – 259.

Lutz, W. D. (1987). Language, appearance, and reality:Doublespeak in 1984. ETC, 44, 382 – 391.

Martinez, M. L., Black, M., & Starr, R. H. (2002). Factorialstructure of the perceived neighbourhood scale (PNS): Atest of longitudinal invariance. Journal of CommunityPsychology, 30, 23 – 43.

McArdle, J. J. (1988). Dynamic but structural equation mod-elling of repeated measures data. In R. B. Cattel & J.Nesselroade (Eds.), Handbook of multivariate experimentalpsychology (2nded.,pp.561 – 614).NewYork:PlenumPress.

McArdle, J. J., & Anderson, E. R. (1989). Latent growthmodel for research on aging. In L. E. Biren & K. W.Schaie (Eds.), The handbook of the psychology of aging (3rded., pp. 21 – 44). San Diego: Academic Press.

McArdle, J. J., & Hamagami, F. (1991). Modeling incompletelongitudinal and cross-sectional data using latent growthstructural models. In L. M. Collins & J. C. Horn (Eds.),Best methods for the analysis of change (pp. 276 – 304).Washington, DC: American Psychological Association.

Menesini, E., Sanchez, V., Fonzi, A., Ortega, R., Costabile, A.,& Lo Feudo, G. (2003). Moral emotions and bullying: Across-national comparison of differences between bullies,victims and outsiders. Aggressive Behavior, 29, 515 – 530.

Meredith, W. (1993). Measurement invariance, factor anal-ysis and factorial invariance. Psychometrika, 58, 525 – 543.

Meredith, W., & Tisak, J. (1990). Latent curve analysis.Psychometrika, 55, 107 – 122.

Milgram, S. (1974). Obedience to authority: An experimentalview. New York: Harper & Row.

Mills, R. S., Pedersen, J., & Grusec, J. E. (1989). Sexdifferences in reasoning and emotion about altruism.Sex Roles, 20, 603 – 621.

Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A developmental taxon-omy. Psychological Review, 100, 674 – 701.

Moffitt, T. E., Caspi, A., Harrington, H. L., & Milne, B. J.(2002).Males on the life-coursepersistent andadolescence-

limited antisocial pathways: Follow-up at age 26. Develop-ment and Psychopathology, 14, 179 – 207.

Motl, R. W., Dishman, R. K., Birnbaum, A. S., & Lytle, L. A.(2005). Longitudinal invariance of the center for epide-miological studies-depression scale among girls andboys in middle school. Educational and PsychologicalMeasurement, 65, 90 – 108.

Muthen, B. O., & Khoo, S. (1998). Longitudinal studies ofachievement growth using latent variable modeling.Learning and Individual Differences, 10, 73 – 101.

Muthen, B., & Shedden, K. (1999). Finite mixture model-ling with mixture outcomes using the EM algorithm.Biometrics, 6, 463 – 469.

Nagin, D. S. (1999). Analyzing developmental trajectories:A semiparametric, group-based approach. PsychologicalMethods, 4, 139 – 157.

Nagin, D. S. (2005). Group-based modeling of development.Cambridge, MA: Harvard University Press.

Nagin, D. S., Farrington, D. P., & Moffit, T. E. (1995). Life-course trajectories of different types of offenders. Crim-inology, 33, 111 – 139.

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

Pelton, J., Gound, M., Forehand, R., & Brody, G. (2004). Themoral disengagement scale: Extension with an Americanminority sample. Journal of Psychopathology and Behav-ioral Assessment, 26, 31 – 39.

Pepler, D., Jiang, D., Craig, W., & Connolly, J. (2008).Developmental trajectories of bullying and associatedfactors. Child Development, 79, 325 – 338.

Pitts, S. C., West, S. G., & Tein, J. (1996). Longitudinalmeasurement models in evaluation research: Examiningstability and change. Evaluation and Program Planning,19, 333 – 350.

Rapoport, D. C., & Alexander, Y. (1982). The morality ofterrorism: Religious and secular justification. Elmsford, NY:Pergamon Press.

Raykov, T. (2004). Behavioral scale reliability and measure-ment invariance evaluation using latent variable mod-eling. Behavior Therapy, 35, 299 – 331.

Raykov, T., & Marcoulides, G. A. (2006). A first course instructural equation modeling (2nd ed.). Mahwah, NJ:Erlbaum.

Reich, W. (1990). Origins of terrorism: Psychologies, ideologies,theologies, states of mind. Cambridge, UK: CambridgeUniversity Press.

Sampson, R. J., & Laub, J. H. (2003). Life-course desisters?Trajectories of crime among delinquent boys followed toage 70. Criminology, 41, 555 – 592.

Sanford, N., & Comstock, C. (1971). Sanction for evil. SanFrancisco: Jossey-Bass.

Schafer, J. L., & Graham, J. W. (2002). Missing data: Our viewof the state of the art. Psychological methods, 7, 147 – 177.

Schwarz, G. (1978). Estimating the dimension of a model.Annals of Statistics, 6, 461 – 464.


Steiger, J. H., & Lind, J. C. (1980, May). Statistically basedtests for the number of common factors. Paper presented atthe annual meeting of the Psychometric Society, IowaCity, IA.

Stoolmiller, M. (1994). Antisocial behavior, delinquent peerassociation, and unsupervised wandering for boys:Growth and change from childhood to early adoles-cence. Multivariate Behavioral Research, 29, 263 – 288.

Swanson, L. H., & Hill, G. (1993). Metacognitive aspectsof moral reasoning and behavior. Adolescence, 28, 711 –736.

Tabachnick, B. G., & Fidell, L. S. (1989). Using multivariatestatistics. New York: Harper & Row.

Tangney, J. P., Stuewig, J., & Mashek, D. J. (2007). Moralemotions and moral behavior. Annual Review of Psychol-ogy, 58, 345 – 372.

Thurstone, L. L. (1947). Multiple-factor analysis. Chicago:University of Chicago Press.

Tremblay, R. E. (2000). The development of aggressivebehavior during childhood: What have we learned inthe past century? International Journal of Behavioral Devel-opment, 24, 129 – 141.

Tremblay, R. E., & Nagin, D. S. (2005). The developmentalorigins of physical aggression in humans. In R. E.Tremblay, W. W. Hartup, & J. Archer (Eds.), Developmen-tal origins of aggression (pp. 83 – 106). New York: GuilfordPress.

Turiel, E. (1978). The development of concepts of socialstructure: Social convention. In J. Glick & K. A. Clarke-Stewart (Eds.), The development of social understanding.New York: Gardner Press.

Turiel, E. (1983). The development of social knowledge: Moral-ity and convention. Cambridge, UK: Cambridge Univer-sity Press.

Turiel, E. (2006). Thought, emotions, and social interactional pro-cesses in moral development. Mahwah, NJ: Erlbaum.

Weiner, B. (1986). An attributional theory of motivation andemotion. New York: Springer-Verlag.

West, S. G., Finch, J. F., & Curran, P. J. (1995). Structuralequation models with nonnormal variables: Problems andremedies. Thousand Oaks, CA: Sage Publications.

Widaman, K. F., & Reise, S. P. (1997). Exploring themeasurement invariance of psychological instruments: Appli-cations in the substance use domain. Washington, DC:American Psychological Association.

Yau, J., & Smetana, J. G. (2003). Conceptions of moral,social-conventional, and personal events among Chinesepreschoolers in Hong Kong. Child Development, 74, 647 –658.

Zimbardo, P. G. (1995). The psychology of evil: A situa-tionist perspective on recruiting good people to engagein anti-social acts. Research in Social Psychology, 11, 125 –133.

Zimbardo, P. G. (2004). A situationist perspective on thepsychology of evil: Understanding how good people aretransformed into perpetrators. In A. G. Miller (Ed.), Thesocial psychology of good and evil (pp. 21 – 50). New York:Guilford Press.

Zimbardo, P. G. (2006). The psychology of power: To theperson? To the situation? To the system?. In D. L. Rhode(Ed.), Moral leadership: The theory and practice of power,judgment and policy. Warren Bennis Signatures Series (pp.129 – 157). San Francisco: Jossey-Bass.

Appendix

Moral Disengagement Scale

1. It is alright to fly off the handle to protect yourfriends.

2. Slapping and shoving someone is just a way ofjoking.

3. Damaging some property is no big deal when youconsider that others are beating people up.

4. A member of a group should not be blamed fortrouble the group causes.

5. If youth are living under bad conditions in theirneighbourhood they cannot be blamed for behav-ing aggressively.

6. It is not serious to tell small lies because they don’thurt anybody.

7. Some people deserve to be treated like animals.8. If people fight andmisbehave in school or at work

it is their teacher’s /superior’s fault.9. It is alright to beat someonewho badmouths your

family.10. To hit obnoxious friends is just giving them "a

lesson."11. Stealing some money is not too serious compared

to those who steal a lot of money.12. A person who only suggests breaking rules

should not be blamed if others go ahead and do it.13. If youth are not disciplined at home they should

not be blamed for misbehaving.14. People do notmind being teased because it shows

interest in them.15. It is okay to treat somebody badly who behaved

like a ‘‘worm.’’16. If people are careless about where they leave their

things it is their own fault if they get stolen.17. It is alright to fight when your group’s honour is

threatened.18. Taking someone’s motorcycle or car without their

permission is just ‘‘borrowing it.’’19. It is not serious to insult a friend because beating

him/her up is worse.20. If a group decides together to do something

harmful it is unfair to blame a single member ofthe group for it.

21. Youths cannot be blamed for using bad wordswhen all their friends do it.


22. Teasing someone does not really hurt him/her.23. Someonewho is detestable does not deserve to be

treated like a human being.24. People who get mistreated usually do things that

deserve it.25. It is alright to lie tokeepyour friends out of trouble.26. It is not a bad thing to ‘‘get drunk’’ once in awhile.27. Compared to the illegal things people do, taking

some things from a store without paying for themis not very serious.

28. It is unfair to blame a single person who had onlya small part in the harm caused by a group.

29. Youth cannot be blamed for misbehaving if theirfriends pressured them to do it.

30. Insults among peers do not hurt anyone.31. Some people have to be treated roughly because

they lack feelings that can be hurt.32. Youths are not at fault for misbehaving if their

parents are too restrictive (severe, and they don’tallow them any freedom).

The following items correspond to the variousmechanisms of moral disengagement. Moral justifica-tion: 1, 9, 17, 25. Euphemistic language: 2, 10, 18, 26.Advantageous comparison: 3, 11, 19, 27. Displacement ofresponsibility: 5, 13, 21, 29.Diffusion of responsibility: 4, 12,20, 28.Distorting consequences: 6, 14, 22, 30.Attribution ofblame: 8, 16, 24, 32. Dehumanization: 7, 15, 23, 31.


Stability and change of moral disengagement and its impact on aggression and violence in late adolescence

Documents