A simple way to assess inference methods

We propose a simple way to assess the quality of asymptotic approximations required for inference methods. Our assessment can detect problems when the asymptotic theory that justifies the inference method is invalid and/or provides a poor approximation given the design of the empirical application. It can be easily applied to a wide range of applications. If widely used by applied researchers, this assessment has the potential of substantially reducing the number of papers that are published based on misleading inference. We analyze in detail the cases of differences in differences with few treated cluster, stratified experiments, shift-share designs, and matching estimators.