Difference in differences with unpoolable data

In this presentation, we describe a new Stata package called unpooled-DID. This procedure is useful when data from different jurisdictions cannot be combined for analysis because of legal restrictions or confidentiality laws. Through Monte Carlo simulation studies, this procedure has been shown to be equivalent to a variation of the conventional DID model when data are poolable. The canonical DID implicitly assumes that the data for the treated group and the control group can be combined. The combined dataset is used to generate a post and treat dummy variables, which are then interacted to estimate the ATT. We also require “poolable” data to verify parallel trends, a key assumption of DID. As a result, conducting DID analysis is nearly impossible using traditional methods when datasets are not combinable. The problem is pronounced for health economists, for whom legal restrictions in sharing administrative data can constrain DID analysis to learn of health systems. This package will make it easier for researchers who work with “unpoolable” data to conduct DID analysis. Furthermore, the package will also provide researchers with a plausibility check for pretreatment trends.