Structural equation modeling (SEM) was used to perform the mediation analyses using AMOS 6.0 software (Arbuckle 2005). Analyses were conducted using the full information maximum likelihood estimation method, which estimates means and intercepts to handle missing data. A latent variable representing emotion dysregulation was created using the observed variables of poor emotional awareness, dysregulated expression of anger and of sadness, and rumination (McLaughlin et al. 2009). A multiply indicated latent variable was created for aggressive behavior and peer victimization using parcels of items from the RPEQ. Parcels were created using the domain representative approach, which accounts for the multidimensionality of aggressive behavior (i.e., overt, relational, and reputational aggression), such that each parcel included items from each of these subscales (Little, et al. 2002). The use of parcels to model constructs as latent factors, as opposed to an observed variable representing a total scale score, confers a number of psychometric advantages including greater reliability, reduction of error variance, and increased efficiency (Kishton and Wadaman 1994; Little et al. 2002). Consistent with previous research in this sample (McLaughlin and Hatzenbuehler 2009), stressful life events were modeled as an observed variable.
