ABSTRACT
This article introduces classification trees as aviable alternative for classification modeling with social service data. As an empirical application of tree-based methods, the performance of a classification tree is com pared to logistic regression in the identification of recidivistic welfare cases. Classification trees offer statistical advantages such as an ability to in corporate all types of predietor variables and to account for complex nonadditive behavior. This allows easy interpretability and the identification of specific profiles of likely recidivist cases. In general, classification trees such as the one in this study can be developed as empirical guidelines to aiddecision making for a wide range of social work issues.