Inference for Welfare Quality Control Programs

Federal and state agencies estimate rates of payment error made by state caseworkers in determining benefits under the Aid to Families with Dependent Children, Food Stamp, and Medicaid programs. Estimated payment error rates are used in turn to estimate program penalties that are a function of the difference between payment error rates and target rates. This article first outlines the statistical methodology currently used in welfare quality control systems, namely, the double sampling scheme for data collection, the regression estimator of the payment error rate, and the penalty system. Stratification of the data is suggested as a method for addressing obvious modeling problems with the regression estimator, and the notion of penalty bias and its properties is discussed. Alternative estimators of payment error rates, such as hierarchical Bayes and empirical Bayes estimators, are next proposed, which may reduce bias and error through combining information. The article is illustrated by real-data examples. Directions for future research are also pointed out.