I believe our work has been highly cited because Sadek Wahba and I were able to make a timely contribution to the growing literature on propensity score methods. We tried to be as transparent about the method, its assumptions, and its potential as possible, and used a classic data set from economics to illustrate these points.

  Does it describe a new discovery, methodology, or synthesis of knowledge?

“We tried to be as transparent about the method, its assumptions, and its potential as possible, and used a classic data set from economics to illustrate these points.”

I view our work as methodological. Since propensity score methods were originated by Paul R. Rosenbaum and Donald B. Rubin in "The Central Role of the Propensity Score in Observational Studies for Causal Effects," (Biometrika, [Vol. 70], No. 1, pp. 41-55, April, 1983), they had not been widely applied in economics. By working through the method on a well known data set, we tried to extend the understanding of these methods.

  Would you summarize the significance of your paper in layman’s terms?

Using propensity score methods, the paper revisits a classic data set and conundrum in program evaluation. Robert J. Lalonde’s 1983 paper used the combination of an experimental treatment group (from a labor training program, the National Supported Work Demonstration) and non-experimental comparison groups to evaluate how well econometric models work in estimating the treatment effect of a program. His conclusion was that, in general, econometric models are not very effective.

Wahba and I revisit this result, and use the propensity score, first, to show that the non-experimental comparison groups used by Lalonde, a professor in the Harris School of Public Policy at the University of Chicago, are very different from the experimental treatment group and, second, to match the treatment group to the subset of the comparison groups that are most similar based on observable characteristics. When we do this, we show that we are able robustly to estimate the treatment effect.

The paper thus underlines the importance in non-experimental settings of finding comparison groups that are similar to the treatment group in terms of observable characteristics and offers one possible method for doing this.

  How did you become involved in this research and were any particular problems encountered along the way?

Wahba and I studied with Donald B. Rubin, the John L. Loeb Professor of Statistics at Harvard University, who was one of the originators of propensity score methods, and also with Guido W. Imbens, Professor of Economics at the Harvard University, who was Donald Rubin’s first collaborator in economics. Robert Lalonde kindly provided data from his classic, original study.

  Where do you see your research leading in the future?

I continue to believe that propensity score methods are a useful tool in evaluating treatment effects in many settings, and am interested in understanding the application of these methods in a wide range of settings.

  Are there any social or political implications for your research?

Since the 1980s, empirical researchers in the social sciences, statistics, and epidemiology have come to view randomized trials as the gold standard in program evaluation. But, in many settings, it is difficult, as well as both socially and ethically impossible or even undesirable, to engage in randomized trials, in which some individuals are denied potentially valuable treatments or others exposed to placebos that are known to have no effect. This work is a small contribution to a larger intellectual effort that is trying to carefully chart out methods for program evaluation when randomized trials are not an option.