The Use of Risk Assessment to Predict Recurrent Maltreatment: A Classification and Regression Tree Analysis (CART)

Research has suggested that recurrent maltreatment may be best predicted by a combination of factors that vary across families. The present study set out to determine whether a pattern-centered analytic approach would better predict families at high risk for recurrence when compared to logistic regression methods. Archival data from substantiated investigations during 2003 were collected from a Connecticut Department of Children and Families county branch. Families (n = 244) with a substantiated index case were followed for 18 months to identify the presence of additional substantiated cases within the CPS system. Classification and Regression Tree (CART) analyses revealed that prior CPS involvement was the best predictor of recurrent maltreatment. Further, risk items that were associated with recurrence were different for families with and without previous CPS investigations. Families with only prior unsubstantiated CPS investigations and poor child visibility within the community were at high risk for recurrence. Families without prior CPS involvement that were not actively involved in case planning and had a history of domestic violence were at high risk for recurrence. These findings suggest that pattern-centered analyses may be a useful approach to informing site-specific predictors of maltreatment recurrence by creating clear decision points that delineate high risk subgroups.