Some Finite Sample Properties and Assumptions of Methods for Determining Treatment Effects: Ordinary Least Squares Regression, Propensity Score Matching, and Inverse Probability Weighing Compared

Erik Petrovski

Research output: Working paperResearch


There is a growing interest in determining the exact effects of policies, programs, and other social interventions within the social sciences. In order to do so, researchers have a variety of econometric techniques at their disposal. However, the choice between them may be obscure. In this paper, I will compare assumptions and properties of select methods for determining treatment effects with Monte Carlo simulation. The comparison will highlight the pros and cons of using one method over another and the assumptions that researchers need to make for the method they choose. To limit the scope of this paper, three popular methods for determining treatment effects were chosen: ordinary least squares regression, propensity score matching, and inverse probability weighting. The assumptions and properties tested across these methods are: unconfoundedness, differences in average treatment effects and treatment effects on the treated, overlap, and robustness.
Original languageEnglish
Number of pages15
Publication statusPublished - 2016

Cite this