Predicting Child Welfare Future Placements for Foster Youth

PROBLEM: Limited understanding of factors that lead to placement disruption and entry into higher levels of care has been a longstanding problem in child welfare research and practice. While prior research has successfully identified some variables that are associated with placement instability, these findings are limited by methodological shortcomings and limited evidence of predictive utility. METHOD: This study attempts to use child, caseworker, and caregiver factors to predict placement type and change in level of care over an 18 month period using random forest modeling. Data from the NSCAW I LTFC sample were used to train and evaluate predictive models. RESULTS: Models predicting placement type performed fairly, while models attempting to predict changes in level of care were unsuccessful. CONCLUSIONS: Future research should continue to consider nonlinear methods for evaluating child welfare outcomes. Consideration of a broader range of variables, localized data, and alternative measurement approaches are suggested.