Verifying the internal validity of a flagship RCT : A review of Crépon , Devoto , Duflo and Parienté ” : A rejoinder

In a recent paper, Bédecarrats, Guerin, Morvan-Roux and Roubaud (2019) reanalyze the data from a randomized controlled trial of the impact of the program of Al Amana, a microcredit organization in Morocco, which we published in 2015 (Crépon, Devoto, Duflo, and Parienté, 2015). They make a number of strong claims about the validity of our approach, and the lack of robustness of our results to alternative assumptions. In this paper, we argue that their own quantitative results in fact demonstrate the robustness of our initial analysis. We question several of their suggested modifications to our analysis. Finally, in the spirit of serious scholarship, we provide our own candid re-analysis of our data, with more robust methods. Overall, our careful analysis of their paper, their code, and our own data suggest that the results we presented in our 2015 paper are in fact quite robust, contrary to what is claimed in their paper. In particular, we find strong support for large impacts on top quantiles of business profits, assets, and sales, no effect at the lower quantiles, a result that is robust to the method used for inference. Due to fat upper tail of the distribution of profit, the average treatment effect on profit is noisier than standard inference methods imply. Impacts on average treatment effects for the other business variables remain significant no matter the method used for inference. We continue to find off-settings effect on labor supply and no impact on average per capita consumption.