Mostly Harmless Econometrics? Statistical Modeling Paradigms in the ‘Top Five’ From 2000 to 2018

Econometric standard practice in model building and selection operates under the null hypothesis statistical testing (NHST) paradigm. We show why the information-theoretic (IT) paradigm is more appropriate for econometrics, and follow up with a survey of all papers published in the period 2000 – 2018 in the Top Five economics journals. Our systematic review reveals that the NHST paradigm prevails far and wide, with the share of IT-based methodology never larger than 5% in any year. We discuss our findings in light of current methodological discussions in economics and science in general, and develop an agenda to move forward