# Sexual Aggression and Victimization Among Adolescents in School: Using a MixIRT Analysis to Examine Measurement Equivalence

## TL;DR

This study examines whether measurements of sexual aggression and victimization among Israeli adolescents are consistent across different groups using a MixIRT analysis.

## Contribution

The study introduces a MixIRT approach to assess measurement equivalence in sexual aggression and victimization among adolescents.

## Key findings

- Four latent classes of adolescents were identified based on their responses to sexual aggression and victimization scales.
- Most items showed configural, metric, and scalar equivalence across latent classes.
- The study found that the measures did not fully capture the range of item difficulty for all latent classes.

## Abstract

Studies examining the frequency of sexual aggression and victimization in schools have compared different groups of respondents based on age, gender, or involvement in other types of school aggression. Between‐group comparisons assume measurement equality. We examine this assumption of measurement equality using a MixIRT analysis, which combines a latent profile analysis with a Rating Scale Model Item Response Theory analysis to determine whether sexual aggressors and victims can be divided into latent classes and whether the latent traits of sexual aggression or victimization have configural, metric and scalar equivalence and through an examination of differential item functioning (DIF). This is a secondary analysis of 3746 Israeli adolescents responding to a self‐report questionnaire regarding sexual aggression and victimization. Data analyses proceeded in five steps, and the unit of analysis was each respondent's responses to the aggressor and victim scales. We conducted a series of exploratory and confirmatory analyses of the aggression/victimization scale to examine configural equivalence, followed by a series of Latent Profile Analyses to determine metric and scalar equivalence. Finally, we examined DIF and Wright Maps using a Rating Scale IRT model. Four latent classes were identified. All items showed configural equivalence and most exhibited metric and scalar equivalence. An examination of DIF and Wright Maps showed that the structures of the latent traits for each latent class were fairly similar. However, for all latent classes, measures of sexual aggression and victimization failed to sample the full range of item difficulty (or endorseability).