A0478
Title: Using extreme bounds analysis to assess reproducibility
Authors: Markus Fritsch - University of Passau (Germany)
Andrew Adrian Pua - Xiamen University (China) [presenting]
Joachim Schnurbus - University of Passau (Germany)
Abstract: Suppose we are willing to accept a more tolerable standard for reproducibility. Suppose we find that we cannot exactly reproduce numerical findings, but are willing to give the maximum benefit of the doubt to a published result. We show how to extend and apply extreme bounds analysis to provide a quick check of whether a published result involving instrumental variable and generalized method of moments estimates may tentatively be accepted. The minimal ingredients we need for the check are (1) output from an instrumental variables regression of the outcome variable on the regressors and the excluded instruments and (2) the cleaned data along with the published coefficient of interest and its standard error. Extending and applying extreme bounds analysis in this manner is compatible with the spirit of existing tools such as the GRIM test or statcheck.