A Robust Measurement Model for Use of the Conflict in Adolescent Dating Relationships Inventory (CADRI)

Teen dating violence (TDV) is a significant public health problem and is associated with many negative risk outcomes.  Although the Conflict in Adolescent Dating Relationships Inventory (CADRI) is used widely to assess TDV, limited work has examined the factor structure across time or across subgroups of gender, ethnicity, or grade. A better understanding of this measure can inform results regarding TDV and its association with other adolescent risk behaviors.

This study utilizes data from 5,797 6^th, 7^th, and 8^th grade youth (41.9% female, 27.7% Hispanic) attending 58 middle schools participating in the Dating Matters^®: Strategies to Promote Healthy Teen Relationships evaluation. The Dating Matters^® evaluation involves a cluster randomized controlled trial being conducted in four sites comparing the Dating Matters^® comprehensive approach to a “standard of care” approach (Safe Dates). The present study focuses on student survey data from the first two school years of the evaluation–baseline and follow-up (2012-2013) and fall and spring follow-up (2013-2014).

Analyses utilized a structural equation modeling (SEM) approach. Missing data (range 0-26.9%) were addressed using multiple imputation procedures in the newly developed Quark package of the statistical package “R.” Model fit was deemed appropriate when Confirmatory Factor Analysis methods specified five latent constructs according to subscales specified in the original measure (original CADRI subscales are Threatening Behaviors, Verbal/Emotional Aggression, Physical Aggression, Relational Aggression, and Sexual Aggression), with separate models for victimization and perpetration. Model fit for both models was appropriate (CFI > .90, RMSEA < .05). Next, we specified bifactor models to examine specific versus common variance for an overarching latent construct for TDV and manifest indicators of each subscale. An additional latent construct (severe violence items from Safe Dates Scales) tested construct validity. Multiple group measurement invariance models were tested across time and subgroups of gender, grade and ethnicity to examine construct validity in the measurement model.

Results demonstrated that the subscales contributed significant variance to the model, but common variance for the overarching TDV construct was more salient for both victimization and perpetration. Severe violence was correlated with overall TDV (p<.001) but not with any subscales (all p > .05). Models demonstrated good fit across four waves of data. Measurement invariance was detected across time, gender and grade with partial invariance for ethnicity. Results indicate that future research using the CADRI could consider this measurement model with one overarching construct for TDV when examining TDV among middle school youth.