| This thesis deals with the use of econometric methods to address problems of program evaluation. Program evaluation refers broadly to the assessment of the impact of a program or policy variation, referred to as "the treatment", on some outcomes of interest; examples of treatments include labor training programs, the adoption of new high-yield seeds, and a change in minimum wage regulations.;The first chapter considers methods that can be used to obtain unbiased estimates of treatment impacts when the available data are from non-experimental studies. Data from randomized experiments are considered ideal, because random assignment into treatment and control implies that simple mean comparisons of outcomes of interest across treatment and control yield unbiased estimates of the treatment impact. Instead when data is gathered non-experimentally, many factors can confound estimates of the treatment impact, a leading instance being sample selection bias. The first chapter demonstrates the use of propensity score methods to adjust in a flexible way for those sources of bias that are attributable to observable differences between the treatment and control groups. Using data from the National Supported Work Program, I demonstrate that propensity score methods succeed in yielding accurate estimates of the treatment effect.;In Chapter 2, using data from the Greater Avenues for Independence (GAIN) experiment, I argue for the use of Bayesian decision theory to set up and solve the decision problems implicitly motivating the program evaluation. There are two advantages to this approach. First, when using standard methods, one concludes that the impact of GAIN is small and statistically insignificant. Instead, in terms of the decision problem, the impact is economically significant, i.e., any risk-averse or -neutral agent prefers the distribution of outcomes under GAIN. Second, I show that a decision-theoretic approach allows us to evaluate hypothetical policies, such as allowing career counselors to assign individuals into the programs, alongside the standard policies of assigning everyone into treatment or control. By assigning only a subset of individuals into treatment, such policies turn out both to be less costly and to yield higher earnings for the participants. |
