Evaluating profiling as a means of allocating government services

This paper considers the use of statistical profiling to allocate persons to alternative options within government programs, or to participation or non-participation in programs. Profiling has been used in the United States to allocate unemployment insurance (UI) claimants to reemployment services based on the predicted duration of their UI claim. We place profiling in the context of the choice among alternative assignment mechanisms. Different mechanisms have different costs and benefits - any one mechanism, whether profiling or something else, may not be optimal for every program. Within profiling systems, we highlight the need for clarity regarding the objective of the assignment mechanism, e.g. equity or efficiency, and we discuss situations in which equity and efficiency goals may conflict. In relation to UI profiling in the United States, we provide empirical evidence from the state of Kentucky on two important questions. First, we demonstrate that it is possible to effectively predict the duration of UI spells, but that effectively doing so requires using more covariates than many US states presently do. This finding is important because effective prediction of the profiling variable is a necessary but not sufficient condition for the success of a profiling system. Second, we show that the impact of reemployment services does not appear to vary with expected duration of the UI spell, indicating that UI profiling in Kentucky does not advance the goal of efficiency, though it may advance equity goals.