Assessing Measurement Invariance of Adolescent Violent Behavior Across Genders: Are We Comparing Apples and Oranges?
Background/Purpose: Measurement invariance (i.e., that a given measure functions equivalently across groups) is a necessary prerequisite to cross-group comparisons. Despite this, violent behavior measures are often used to compare rates of violence between males and females without establishing measurement invariance. This practice can lead to biased cross-group comparisons. Further, according to social role theory, attitudes and behaviors are influenced in part by social norms, suggesting that gender norms may influence individual’s responses to survey questions. The current study examines the measurement invariance of a modified version of the Violent Behavior Checklist across genders.
Methods: Using multiple group confirmatory factor analysis, measurement invariance of a 6-item version of the Violent Behavior Checklist was assessed in a sample of racially/ethnically diverse middle and high school students (N=4,128) in two rural counties. Specifically, configural invariance, which indicates that the same factor model exists across groups; metric invariance, which indicates that all factor loadings are equivalent across groups; and equivalence of individual factor loadings were tested. Based on social role theory, it was hypothesized that results of the current study would reveal partial measurement invariance across genders on the Violent Behavior Checklist-Modified. Specifically, it was hypothesized that items assessing more severe forms of violence (i.e., “beaten somebody up,” “used any weapon in a fight,” “gotten involved in a fight with one group of kids fighting another group of kids”) would be non-invariant across genders.
Results: Results indicated that the Violent Behavior Checklist-Modified has partial measurement invariance across genders. Specifically, four out of six items were non-invariant across genders. In line with our hypothesis, items assessing more severe forms of violence (i.e., “beaten somebody up,” “used any weapon in a fight,” “gotten involved in a fight with one group of kids fighting another group of kids”) were non-invariant across groups, such that stronger factor loadings were observed for males compared to females. The item “pushed or shoved someone” was also non-invariant across genders such that stronger factor loadings were observed for females compared to males.
Conclusion/Implications: Overall, findings suggest that the latent factor of violence may be qualitatively different between males and females based on gender norms. According to social role theory, gender norms influence behavior and beliefs (Eagly 1987). Gender norms may create unequal thresholds dictating what kind of behavior is acceptable for males versus females. For instance, males engaging in relatively less serious behaviors such as pushing/shoving may be considered normative “roughhousing” (the “boys will be boys” adage). At the same time, these same behaviors among females may be considered egregious. The current study has salient implications for adolescent violence research and underlines the necessity of establishing measurement invariance prior to making cross-gender comparisons to avoid biased conclusions.