848 PAULETTL COOPER.AND PERRY SPSS Program which At each time,each child's raw ent of target-specific aggression.We needed a mea was a for the child).we assessed global aggression ony (i.ebeing mean ation by female peers was similarly computed. Results Descriptive Statistics ificant ways.The following instructions ere read to the h Table I gives the mean and standard deviation of each measure h kid subsequent analyses,me aggression and victim c m an to other kids.For ple,a kid could hit or pu nch so e measure re ted in Table2 ble3 fo nc nted a hoklet Ea had the erm gender of children'v bcha conformiry to refer to oral gender noncon lace for stion Four pag lly,each model involved two steps.First 14-15 bed bo which the child'sg on loward pee need over the ad pr al 10n target gender r 4 sion to ard his or her classmate gender nonconformity in t child's ntage nomina his or her aggression to vard gend step of th were ion tos ard bovs early all children (17 f195. n 0r91.8%) toward girls.)Tw rally victimized by peers (to hetic target influ mcan numbe sin the fall (spring)respectively. mate ion).A Level I equation was used to compute ildren' red victims,each child's cHLM model,the within-child slope d as the depend in analyses i al linear modeling (HLM) analyses eg Nezlek 2011)Thu thes were between-child 02).The analyses cont hild 63(3fo encra (36)for boys aggr on tov ard girls,53 (.22)for girls a lso ex ined).A vel this pur ose.A sion toward boys,and 46(.15)for girls'aggression towar d girls effect of the cognitive term in indicate ysis on the spring data yielded a nearly identical structure. Com￾ponent scores were computed for both the fall and the spring by the SPSS program, which yields uncorrelated factors. Children’s scores for gender nonconformity were of particular interest in this study. Assessment of target-specific aggression. We needed a mea￾sure of each child’s aggression toward each classmate. Because obtaining such data is time consuming and burdensome (especially for the child), we assessed global aggression only (i.e., being mean to someone); we did not repeat the assessment process for different subtypes of aggression (e.g., physical, relational). This procedure allowed children to consider a wide range of acts as mean, pre￾sumably increasing the likelihood that children would nominate all peers whom they perceived to be mean to someone in one or more significant ways. The following instructions were read to the child: We want to find out which kids sometimes do mean things to other kids, and who they are doing it to. There are many different ways to be mean to other kids. For example, a kid could hit or punch someone, tease someone or say something mean about them, or do something mean over the cell phone or the Internet. For each kid, we want you to tell us whether he or she is sometimes mean to each kid on the list. Children were then presented a booklet. Each page had the name of a different peer at the top, followed by a list of all of that peer’s classmates. Next to each of the classmates’ names there was a place for the child to check either “Yes” or “No” in response to the question, “Is (name of the peer at the top of the page) mean to (name of the classmate)?” Thus, the peer whose name appeared at the top of the page was a potential aggressor, and the children whose names followed were potential victims. Fourth and fifth graders’ booklets had a page for each classmate, and sixth and seventh graders had a page for each of the 14 –15 peers for whom they had provided the nominations described above. However, all children’s lists of potential victims included all of the potential aggressor’s classmates (both male and female). The mean number of peers listed as potential victims was 28.5 (minimum 14, maximum 43, minimum number of peers of a given sex 7). Participants’ own names were never in the booklet, so they did not nominate themselves as aggressors or as victims. A child’s raw aggression toward a peer was the percentage of nominators who named the child as mean to that peer. These scores were used in the first step of the multilevel analyses described later. A child’s raw victimization by a peer was the percentage of nominators who named the peer as mean to the child. Nearly all children (179 of 195, or 91.8%) were nominated by at least one classmate (in either fall or spring) as mean to a peer. Of the aggressor–victim pairs so identified, the minimum, maximum, and mean numbers of nominators in the fall (spring), respectively, were one (one), 13 (16), and 2.01 (2.19). To estimate stability over the year in children’s profiles of preferred victims, each child’s profile of raw aggression toward classmates in the spring was predicted from the child’s profile of raw aggression toward the same classmates in the fall, using the hierarchical linear modeling (HLM) program (Raudenbush & Bryk, 2002). The mean (and standard deviation) of these within￾child betas were .63 (.31) for boys’ aggression toward boys, .54 (.36) for boys’ aggression toward girls, .53 (.22) for girls’ aggres￾sion toward boys, and .46 (.15) for girls’ aggression toward girls. These moderate coefficients indicate there was some stability but also room for change in children’s targets over the year. At each time, each child’s raw aggression toward male peers was averaged to estimate the child’s general aggression toward male peers; a measure of general aggression toward female peers was similarly computed. Also at each time, each child’s raw victimization by male peers was averaged to estimate the child’s general victimization by male peers; a measure of general victim￾ization by female peers was similarly computed. Results Descriptive Statistics Table 1 gives the mean and standard deviation of each measure in both the fall and the spring of the school year, as well as the stability of each measure from fall to spring. (To be consistent with subsequent analyses, measures of general aggression and victim￾ization are given separately in relation to same-sex and other-sex peers.) Correlations among the measures are presented in Table 2 for the fall and in Table 3 for the spring. Henceforth, for clarity, we use the term felt gender typicality to refer to children’s self-rated gender typicality and the term gender nonconformity to refer to peers’ perceptions of children’s overt, behavioral gender noncon￾formity. Does Gender Identity Predict Children’s Aggression Toward Gender-Nonconforming Peers? Analysis strategy. To answer this question, we ran a series of HLM models. Conceptually, each model involved two steps. First, a measure was computed for each child capturing the degree to which the child’s aggression toward peer targets changed over the school year as a function of target gender nonconformity. That is, a within-child beta coefficient (slope) was calculated for each child predicting the child’s raw aggression toward his or her classmates in the spring from the classmates’ gender nonconformity in the fall. The beta controlled for the child’s raw aggression toward each classmate in the fall, so in effect it assessed change over the school year in the child’s propensity to attack peers based on their gender nonconformity. The larger the slope, the more the child increased his or her aggression toward gender-nonconforming peers (relative to other targets) over the year. (These within-child associations were computed separately for children’s aggression toward boys and children’s aggression toward girls.) Two other variables were controlled in these betas—the degree to which a classmate was generally victimized by peers (to remove nomothetic target influ￾ences) and classmates’ aggression toward the participant (to re￾duce the possibility that participants’ aggression was reactive to classmates’ aggression). A Level 1 equation was used to compute these within-child betas. In the second step of each HLM model, the within-child slopes just described served as the dependent variable in analyses in which the cognitive variables were predictors (slopes-as-outcomes analyses; e.g., Nezlek, 2011). Thus, these were between-child analyses. These analyses controlled for child general aggression (trait aggression), age, and sex (though interactions with sex were also examined). A Level 2 equation served this purpose. A signif￾icant effect of the focal cognitive term in this equation indicated 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. 848 PAULETTI, COOPER, AND PERRY