New publication | Nonstandard Errors
Dreber and Johannesson with co-authors designed a multi-analyst study where over 160 independent research teams in finance tested the same six hypotheses on the same data. The dataset was kindly provided by Deutsche Börse and spanned 17 years of trading in Europe's most actively traded instrument, the EuroStoxx 50 index futures. The research teams tested hypotheses about trends in the data such as if there was a change in market effficiency or realized bid-ask spread in the studied period. For all the tested hypotheses, the results across research teams varied from statistically significant in one direction to statistically significant in the other direction. The results suggests that the analytical variation in how to test an hypothesis contributes substantial uncertainty in results for this type of research, not taken into account in the reporting of results of indvidual studies or in standard hypothesis testing.
View the publication
Abstract
In statistics, samples are drawn from a population in a data-generating process (DGP). Standard errors measure the uncertainty in estimates of population parameters. In science, evidence is generated to test hypotheses in an evidence-generating process (EGP). We claim that EGP variation across researchers adds uncertainty—nonstandard errors (NSEs). We study NSEs by letting 164 teams test the same hypotheses on the same data. NSEs turn out to be sizable, but smaller for more reproducible or higher rated research. Adding peer-review stages reduces NSEs. We further find that this type of uncertainty is underestimated by participants.
