The Average Predictive Validity of Intimate Partner Violence Risk Assessment Instruments

TLDR

Intimate partner violence risk assessment is a rapidly expanding field, largely studied in the past decade, yet it remains relatively new and often applied with proxy instruments, requiring consideration of setting, outcome, assessor skill, and information access. The study evaluates the average sample‑size‑weighted predictive validity of five standalone IPV risk assessment instruments using ROC AUC. The authors pooled data from multiple studies to compute a sample‑size‑weighted ROC AUC for each of five IPV risk assessment tools. ODARA achieved the highest average weighted AUC (.666), followed by SARA (.628), DA (.618), DVSI (.582), and K‑SID (.537), with overall effect sizes small except for ODARA’s medium effect, and only 45% of the 20 predictive validity measures were administered correctly.

Abstract

The field of intimate partner violence (IPV) risk assessment (predicting recidivism, lethality) is fast growing, and the majority of research examining the predictive validity of IPV risk assessment instruments has been conducted in the past decade. This study examines the average predictive validity weighted by sample size of five stand alone IPV risk assessment instruments that have been validated in multiple research studies using the Receiver Operating Characteristic Area Under the Curve (AUC). The Ontario Domestic Assault Risk Assessment (ODARA) has the highest average weighted AUC (=.666, k=5) followed, in order of most to least predictive, by the Spousal Assault Risk Assessment (SARA; AUC=.628, k=6), the Danger Assessment (DA; AUC=.618, k=4), the Domestic Violence Screening Inventory (DVSI; AUC=.582, k=3), and the Kingston Screening Instrument for Domestic Violence (K-SID; AUC=.537, k=2). The effect size for the average AUCs for IPV risk assessment instruments is small, with the exception of a medium effect size for the ODARA. Of the 20 measures of predictive validity included in this analysis, the risk assessment was administered correctly in nine (45%). IPV risk assessment is relatively new, and the use of proxy instruments and utilization of risk assessment instruments in settings for which they were not created is widespread. While waiting for a more rigorous body of research, factors in addition to predictive validity must be taken into consideration (e.g., setting, outcome, skills of the assessor, access to information) when choosing which risk assessment instrument is appropriate for use in a particular practice setting.