PAULETTL COOPER.AND PERRY riable of nd uadratic term.as described next).we exa ined its pattern If athway was ted if high scores on nder ure o way interaction and the tw ctions of sex with to omote hacas nent of gender- ning p variable in turn (in order of Table 4 9 variahle this was done by runnine s ent models tha nt of gender. ning peers,but there was the tw main effects e evidence r the overconfide ntial mod If this der identity variable with thre 10n c self-app elf ef was he i inte ith self-e uded at Level one predicting the Bias Self-Esteen included such an ea cept m Leve ollo towith fal (19 d at 1 and SD of t n.Ou influ nces of cognition on trait with s ng ingroup fay tism (p 001 simulta d thei rd h s and for to be moder ately effects.Th attr rate sim wi h other-s 003 ut not signif ing child) arget Because the b towa nent of s ing peers That childr high in bot ion toward othe sex p ports the overconfide t pathway:that childr ted few r set of analyses We first summ ergrolf-di s but low win self-efficacy also do so supports the ggre sion to ard gende onconforming ame. ex peers pressure for er ntiation Felt pres ssure fo edict in ever,the inte on of sex and self-esteem wa cks on gender-nor onforming peers.These interac e-estem fore 081D0291 in a level 2 equation predictine slopes ure.but it is more sistent with th To evalu cure ant pathwa than ray interactions list that high s res on nigh felt was more conducive than lov felt e gender identity variable combined with high scores on anothe ictimization of gender-nonconforming peers (b 0.75.p- effects of the two cognitive variables and their product. If the interaction was significant (and not qualified by child sex or by a quadratic term, as described next), we examined its pattern. If, however, the three-way interaction of child sex and the cognitive variables was significant (in a subsequent model that included the three-way interaction and the two-way interactions of sex with each cognitive variable), we evaluated the interaction of the cognitive variables separately for each sex. Also for each two-way interaction, we examined whether the first-named member of each variable pair listed in Table 4 moderated a quadratic effect of its partner variable; this was done by running subsequent models that included not only the two main effects and their interaction but also the quadratic effect of the partner variable and the interaction of this quadratic variable with its potential moderator. If this interaction was significant, we examined its pattern. (We also ran models to see if any such interactions were qualified by child sex; none was.) When a Level 2 equation is used to predict the slopes from a Level 1 equation, the model may be misspecified unless a second equation is included at Level 2— one predicting the intercepts from Level 1 (from the same terms as those predicting the slopes). We therefore always included such an equation at Level 2. An intercept from Level 1 estimates a participant’s general (trait) aggression in the spring. Because this extra equation predicts the intercepts from cognition with fall general aggression controlled, the coefficients for the cognitive terms indicate whether cognition predicts change over the year in trait aggression. Our hypotheses do not address influences of cognition on trait aggression, but we later comment on these effects. Although we have described the analysis as proceeding in two steps (from Level 1 to Level 2), in reality the HLM program estimates all equations simultaneously. Each HLM analysis included the data for boys’ aggression toward boys and for girls’ aggression toward girls. Participant sex is therefore confounded (deliberately) with target sex, but as noted, sex was examined both as a main effect and as a moderator of cognitive effects. This analytic strategy is appropriate when testing hypotheses expected to operate similarly in both boys’ and girls’ interactions with same-sex peers but not necessarily in their interactions with other-sex peers (e.g., Rodkin & Berger, 2008; Sainio, Veenstra, Huitsing, & Salmivalli, 2012). We also ran a parallel series of analyses examining children’s aggression toward othersex targets. Because the base rate of children’s aggression toward other-sex peers is very low and because there is less reason to expect gender identity to affect aggression toward other-sex peers, we expected few results for this set of analyses. We first summarize results for same-sex targets. Aggression toward gender-nonconforming same-sex peers. No cognitive variable predicted change in children’s maltreatment of gender-nonconforming peers over the school year as a main effect. However, the interaction of sex and self-esteem was significant (1.40, p .009, t 2.64). For boys only, low self-esteem forecast increased targeting of gender-nonconforming peers (0.81, p .029, t 2.22). (The gamma symbol denotes an unstandardized beta coefficient from a cognitive term in a Level 2 equation predicting slopes.) To evaluate the overconfident and insecure pathways to aggression toward gender-nonconforming peers, we tested the 20 twoway interactions listed in Table 4. Evidence that high scores on a gender identity variable combined with high scores on another variable to encourage maltreatment of gender-nonconforming peers was taken as support for the overconfident pathway. The insecure pathway was seen as supported if high scores on a measure of between-gender identity or low scores on a measure of within-gender identity combined with low scores on another variable to promote harassment of gender-nonconforming peers. Of the 20 interactions tested, 10 were significant, only one of which was moderated by child sex. We describe the significant interactions for each gender identity variable in turn (in order of Table 4). As we shall see, most results supported the insecure pathway to maltreatment of gender-nonconforming peers, but there was also some evidence for the overconfident route. Intergroup bias. Intergroup bias did not interact with any other gender identity variable but did interact with three gendernonspecific self-appraisals—self-esteem, popularity self-efficacy, and attractiveness self-efficacy—to predict increased victimization of gender-nonconforming peers. Because the interaction of intergroup bias with self-esteem was qualified by child sex (three-way interaction 2.63, p .018, t 2.40), the two-way interaction of Intergroup Bias Self-Esteem was examined for each sex. This interaction was significant for boys (p .001) but not girls (details of significant interactions are in Table 4). Significant interactions were followed up using procedures recommended by Aiken and West (1991), that is, the effect of one variable was estimated at 1 and 1 SD of the other (conceptualized as the moderator). As seen in Figure 1A, low self-esteem predicted victimization of gender-nonconforming peers only for boys with strong ingroup favoritism (p .001). This supports the insecure, self-discrepant pathway to harassment of gendernonconforming peers: Boys who saw their own gender as superior but nonetheless questioned their own worth grew increasingly mean to gender-nonconforming peers with time. (This was the only interaction to be moderated by child sex; all remaining interactions in this section capture patterns for the total sample.) Intergroup bias also interacted with the quadratic form of both popularity self-efficacy and attractiveness self-efficacy (see Figures 1B and 1C). In each case, the quadratic effect of the selfefficacy variable was highly significant for children with strong intergroup bias (both ps .003) but not significant for children with little intergroup bias: For children with high intergroup bias, either very low or very high self-efficacy led to selective harassment of gender-nonconforming peers. That children high in both intergroup bias and self-efficacy for a demonstration goal harass these peers supports the overconfident pathway; that children high in intergroup bias but low in self-efficacy also do so supports the insecure, self-discrepant pathway. Felt pressure for gender differentiation. Felt pressure for gender differentiation interacted with three variables—felt gender typicality, gender contentedness, and self-esteem—to predict increased attacks on gender-nonconforming peers. These interactions are depicted in Figures 2A, 2B, and 2C. Felt pressure interacted with the quadratic form of felt typicality (p .012; see Figure 2A). The pattern does not provide strong support for either pathway because the quadratic effect of felt typicality is not significant at high felt pressure, but it is more consistent with the insecure, self-discrepant pathway than with the overconfident pathway. This is because for children who felt gender atypical, high felt pressure was more conducive than low felt pressure to victimization of gender-nonconforming peers (b 0.75, p This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 852 PAULETTI, COOPER, AND PERRY