This paper examines the effectiveness of propensity score matching estimators in evaluating the impact of social programs using data from the National Supported Work (NSW) Demonstration. The authors apply both cross-sectional and longitudinal propensity score matching estimators to the NSW data, which has been previously analyzed by LaLonde (1986) and Dehejia and Wahba (1999, 2002). They find that the estimates of the NSW impact based on propensity score matching are highly sensitive to the variables included in the scores and the specific analysis sample used. Among the estimators studied, the difference-in-differences (DID) matching estimator performs best, as it eliminates potential sources of temporally-invariant bias, such as geographic mismatch and differences in how dependent variables are measured for participants and non-participants. The paper concludes that while propensity score matching is a useful econometric tool, it does not provide a general solution to the evaluation problem and that the optimal non-experimental evaluation strategy depends on the available data and the institutions governing selection into the program.This paper examines the effectiveness of propensity score matching estimators in evaluating the impact of social programs using data from the National Supported Work (NSW) Demonstration. The authors apply both cross-sectional and longitudinal propensity score matching estimators to the NSW data, which has been previously analyzed by LaLonde (1986) and Dehejia and Wahba (1999, 2002). They find that the estimates of the NSW impact based on propensity score matching are highly sensitive to the variables included in the scores and the specific analysis sample used. Among the estimators studied, the difference-in-differences (DID) matching estimator performs best, as it eliminates potential sources of temporally-invariant bias, such as geographic mismatch and differences in how dependent variables are measured for participants and non-participants. The paper concludes that while propensity score matching is a useful econometric tool, it does not provide a general solution to the evaluation problem and that the optimal non-experimental evaluation strategy depends on the available data and the institutions governing selection into the program.