Estimating Empirical Cutoffs for Identifying Victims of Bullying in Middle School

Peer victimization is a noted antecedent of poor wellbeing in children. Recent meta-analyses have supported linkages between peer victimization and internalizing problems.  Professionals have expressed the importance of a multi-tier approach to prevention and intervention. Within this approach all students and even personnel receive intervention, while students who are most vulnerable receive additional layers of services. This type of intervention system is dependent upon the appropriate classification of children in need of services. However, one of the struggles in peer victimization research is differentiating normative levels of peer conflict from persistent and debilitating levels. Recent advances in scale development offer solutions for differentiating these children.

     The purpose of this proposal is to integrate three methods for establishing empirical cutoffs: 1) basing cutoffs on raw percentiles, 2) basing cutoffs on percentiles using a latent factor score, and 3) receiver-operating curve characteristics. Data for these analyses were collected as part of an ongoing evaluation of an intervention for victims of bullying. At pretest middle school students (N= 551, 59% female) completed a previously validated, 17 item measure of peer victimization, as well as other measures of psychosocial health.  Confirmatory factor analysis in Mplus v6.1 was used to replicate the original structure of the inventory. Model fit indices suggest measurement validity (CFI = 0.98, RMSEA = .05). Also, victimization was a significant predictor of the internalizing T score from the BASC-II SRP (β = 0.61, p < .001). Cutoffs were examined and validated by dichotomizing the internalizing T score at the 70^th percentile. Cutoffs were estimated from the raw victimization scores that corresponded to a T score of 50, 55, 60 and 70. These same T scores were used for the victimization factor score. Both methods suggested a raw score of 28, the cutoff that best maximized sensitivity and specificity. The ROC analysis indicated that the victimization measure operates as a good screener for internalizing complaints (AUC = 0.92, SE = .02, CI_95% = 0.88-0.97). ROC curve analyses suggested a cutoff score of 29.  While the cutoff provided by ROC maximizes the effectives of the victimization scale as a screener, the multiple cutoffs provided by the other methods could be invaluable when determining when preventative services are warranted versus intervention. The presentation will focus on these issues, as well as broader practices in measuring bullying-victimization behaviors.