Making Distinctions: A Multiple Models Approach to Assessing Risk of Repeat Domestic Violence

Lynette Martin, The Pennsylvania State University

The standard procedure in domestic violence recidivism research is to develop a single prediction model that can be used to assess risk of future violent behavior. However, most of these models are based on an additive main effects regression approach (e.g. logistic regression) and do not adequately reflect the contingent and interactive nature of the assessment processes. Further, the use of single models applies the belief that a single risk assessment model can capture the essential information needed to predict recidivism. I question this standard procedure by suggesting that the interactive complexities inherent in the prediction of violent behavior are such that no single risk assessment model is sufficient for the task. By combining statistically the predictions of several classification tree-based assessment models, I will demonstrate that this multiple models approach will increase the accuracy in assessing the risk of 2 types of violence: 1) situational couple violence; and 2) intimate terrorism.

(Return to Program Resources)

Updated 05/20/2006