Against normality testing

TL;DR

The paper argues against the necessity of normality testing in statistical analysis, claiming it is theoretically unjustified and often a waste of resources, without invalidating results.

Contribution

It challenges the validity of normality tests and advocates for abandoning them based on theoretical reasoning, simplifying statistical workflows.

Findings

Normality tests are theoretically unjustified.

Rejecting normality does not invalidate results.

Avoiding normality testing saves time and effort.

Abstract

I reject the following null hypothesis: {H0: your data are normal}. Such drastic decision is motivated by theoretical reasons, and applies to your current data, the past ones, and the future ones. While this situation may appear embarrassing, it does not invalidate any of your results. Moreover, it allows to save time and energy that are currently spent in vain by performing the following unnecessary tasks: (i) carrying out normality tests; (ii) pretending to do something if normality is rejected; and (iii) arguing about normality with Referee #2.