In 2010 I was asked to present my thoughts about propensity score matching (PSM) at the Agricultural and Applied Economics Association’s annual conference in July. I think the organizer anticipated that I would say good things about PSM because I am an “identification” guy. But being an identification guy is precisely why I didn’t say good things about PSM. Instead I discussed the strengths and weaknesses of OLS and PSM. I went on to show how to combine both approaches to acheive the best estimators. Here are my slides from that presentation:
Later that same year, Chris Blattman echoed my sentiments about PSM in the following blog post: The cardinal sin of matching. The bottom line is, “matching is not a solution to your endogeneity problem.” I continue to be shocked by researchers who, six years later, still haven’t figured that out.