Measuring the Sensitivity of Parameter Estimates to Sample Statistics

Empirical papers in economics often describe heuristically how their estimators map specific data features to parameters or other magnitudes of interest. We propose a formal, quantitative measure of this relationship that can be computed at negligible cost even for complex models. We show that the sensitivity of estimates to particular sample statistics can be informative about the importance of particular identifying assumptions, providing one rationale for the attention that sensitivity receives in applied research. We apply our measure to empirical papers in industrial organization, macroeconomics, public economics and finance. ∗Conversations with Kevin M. Murphy inspired and greatly improved this work. We are grateful also to Jos