Ex Ante Policy Evaluation, Structural Estimation, and Model Selection
区分了事后与事前政策评估，指出事前评估需依赖参数和行为假设，并讨论了结构估计与模型选择在评估新政策（如TANF、Medicare药品福利、Progresa）中的应用。
A number of major social policy interventions have been introduced recently in the United States. The new Temporary Assistance for Needy Families (TANF) program, introduced in 1996, was advertised changing the welfare system as we know it. The new Medicare prescription drug benefit, introduced in 2006, was the largest expansion of Medicare in its history. Developing countries are also fertile ground for innovative new policies. Mexico introduced a program in 1997 (Progresa) that provided subsidies to poor rural households contingent upon the school attendance of their children. The distinction between ex post and ex ante policy evaluation is important. Ex post policy evaluation occurs upon or after the policy has been implemented. It is ubiquitous in the social sciences. Such studies make use of existing policy variation. Examples include the study of minimum wage effects on labor market outcomes, the study of the impact of welfare benefits on labor market and demographic outcomes, and the study of how divorce laws affect marital stability. The development of methodological approaches to ex post program evaluation using nonexperimental methods is an active area of research (Petra Todd 2006). There is little methodological or applied research explicitly concerned with ex ante policy evaluation using nonexperimental methods, which is perhaps surprising given its potential value. Interventions that require ex ante evaluation are those that are outside the historical experience. These include a large change in the parameters of existing programs such doubling the (real) minimum wage, adding new features to an existing program such the Medicare drug benefit program, or introducing a completely new program such Progresa. The nonexperimental approach to ex ante policy evaluation must be an extrapolation from existing policy or policy-relevant variation.' Because of that, and unlike ex post evaluation, ex ante valuation must rely on parametric and/or behavioral assumptions (theory).