Evaluating Social Programs with Endogenous Program Placement and Selection of the Treated

This chapter considers methods for evaluating the impact of social programs in the presence of nonrandom program placement or program selection. It first presents the evaluation problem as a missing data problem and then considers various solutions proposed in the statistics and econometrics literature. For ex post evaluation, the following estimation methods are discussed: traditional regression methods, matching, control function methods, instrumental variable and local instrumental variable (LIV) methods, and regression-discontinuity. Alternative estimators are described along with their identifying assumptions, the behavioral implications of those assumptions, and the data requirements for implementation. The chapter also considers methods for ex ante evaluation, which can be used to assess the effects of programs prior to their implementation, for example, in trying to design a program that achieves some desired outcomes for a given cost. Throughout the chapter, numerous examples from the development literature illustrate applications of the different estimation methods and highlight factors affecting estimator performance.