Evaluating program evaluations: new evidence on commonly used nonexperimental methods.
利用1980年代多个州的社会实验数据，比较两种非实验评估方法与实验结果的差异，为无法进行实验时选择评估策略提供依据。
In this paper we follow previous research by using experimental data to assess the two nonexperimental evaluation approaches. Our data are from a series of social experiments conducted in several states during the 1980s to evaluate programs aimed at helping welfare recipients find jobs. We simulate the two nonexperimental approaches by creating comparison groups from the true control groups and by comparing the resultant nonexperimental estimates of program effects with experimentally derived estimates of program effects. Although experimental data are required to provide an assessment of the nonexperimental approaches they are not required to create the comparison groups in practice. Thus our results have direct implications for choosing a nonexperimental evaluation strategy when an experiment is not feasible. The remainder of this paper is organized as follows. Section I describes the social experiments and the various comparison groups created for the analysis. Section II discusses the methods used to generate and assess the nonexperimental estimates. Section III presents the empirical results and Section IV offers some conclusions. (excerpt)