Econometric Evaluation of Socio-Economic Programs

It is common practice for policymakers to perform ex post evaluation of the impact of economic and social programs via evidence-based statistical analysis. This effort is mainly devoted to measure the “causal effects” of an intervention on the part of an external authority (generally, a local or national government) on a set of subjects (people, companies, etc.) targeted by the program. Evidence-based evaluation is progressively becoming an integral part of many policies worldwide. The main motivation resides in the fact that, when a public authority chooses to support private entities by costly interventions, a responsibility towards taxpayers is assumed. This commitment, constitutionally recognized in several countries, draws upon the principle that, since many alternative uses of the same amount of money are generally possible, any misuse of it is seen as waste, especially under severe budget constraints.

This chapter deals with the estimation of average treatment effects (ATEs) under the assumption of “selection on observables”, and provides a systematic account of the meaning and scope of such an assumption in program evaluation analysis. It illustrates a number of econometric methods developed in the literature to provide correct inference for causal parameters under selection on observables. In particular, it illustrates and discusses the four most popular approaches used in applications: Regression-adjustment, Matching, Reweighting, and the Doubly-robust estimator. In the final part, the chapter also offers a number of applications of these econometric methods in a comparative perspective using built-in and user-written Stata commands on real datasets.

This chapter covers econometric methods for estimating average treatment effects (ATEs) of social and economic programs under the assumption of “selection on unobservables”. When nonobservable factors significantly drive the nonrandom assignment to treatment, recovering consistent estimations of average treatment effects relying only on observables is no longer possible. As a consequence, econometric methods only based on assuming “selection on observables” become inappropriate. This chapter illustrates methods suitable for dealing with unobservable selection, thus critically discussing various Instrumental-variables (IV) approaches, by introducing the Heckman Selection-model, and by illustrating the Difference-in differences (DID) estimator both in a repeated cross section and in a longitudinal data structure. The chapter concludes by focusing on a number of applications of previous methods using built-in and user-written Stata commands on real and simulative datasets.

This chapter addresses two different but related approaches, both widely used within the literature on the econometrics of program evaluation: the Local average treatment effect (LATE) and the Regression-discontinuity-design (RDD). Considered as nearly quasi-experimental methods, these approaches have recently been the subject of a vigorous interest as tools for detecting treatment effects within special statistical settings. The first part of the chapter covers the theory behind LATE, thus illustrating how such approach can be embedded within the setting of a randomized experiment with imperfect compliance. The discussion then goes on to present the Wald estimator of LATE, and to extend LATE to the case of multiple instruments and multiple treatments. The second part of the chapter illustrates the RDD econometric theoretical background; in particular, it discusses separately sharp RDD and fuzzy RDD, and suggests a protocol for the empirical implementation of such approaches. The chapter ends with some illustrative empirical implementation of both LATE and RDD performed using Stata on both real and simulative examples.