Title

Three essays in programme evaluation

Date of Award

5-2006

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Economics

Advisor(s)

Dan Black

Keywords

Program evaluation, Nonparametric regression, Regression discontinuity design, Unemployment insurance

Subject Categories

Econometrics | Labor Economics

Abstract

This thesis consists of three essays in programme evaluation. The first essay considers a semiparametric matching model with fixed effects that allows the identification and estimation of matching treatment impacts after differencing out the effect of problematic covariates that make the support and balancing test assumptions difficult to achieve. I use an experimental program, the Kentucky Working and Reemployment Services, to validate the effectiveness of the proposed estimator in eliminating selection bias. I find the proposed estimator yields small bias with respect to the benchmark experimental estimates and outperforms "naive" propensity score matching estimates.

In the second essay, which is a joint work with Dan Black and Jeffrey Smith, we use a "tie breaking" experiment, to investigate selection bias in the regression discontinuity approach. Two features characterize this program. First, the treatment is assigned as a discontinuous function of a profiling variable. Second, we deal with a discontinuity frontier rather than a discontinuity point, which allows the identification of treatment effects over a wide range of the support of the discontinuous variable. Using a variety of parametric and nonparametric estimators, we find that the RD estimates are sensitive to the sample used in the estimations, the outcome of interest, and the econometric models.

The third essay considers the problem of estimating optimal bandwidths in the context of average treatment impacts. Because precise estimation of counterfactuals in regions containing much of the probability mass of the treated units is more important than it is in regions where few treated units are located, I re-weight the data such that comparison individuals that get used intensively in the estimation of treatment impacts receive the largest weights in the estimation of the optimum bandwidths. Using standard cross-validation techniques, I applied the proposed method to data from the National Support Work Demonstration Program (NSW). Two main findings emerge. First, the magnitude of the optimal bandwidth varies depending on whether the support problem is considered or not. Second, local constant matching with Gaussian kernel functions yields the smoothest MSE-h relationship, whereas local linear matching with Epanechnikov kernel function yields local minima.

Access

Surface provides description only. Full text is available to ProQuest subscribers. Ask your Librarian for assistance.

http://libezproxy.syr.edu/login?url=http://proquest.umi.com/pqdweb?did=1232412471&sid=1&Fmt=2&clientId=3739&RQT=309&VName=PQD

Link to Full Text

Share

COinS