Making null effects informative using Bayes factors and equivalence tests

Publication bias, in which non-significant results are less likely to be published, is a considerable issue for many areas of psychology. At least within a hypothetico-deductive framework, the capacity to falsify a hypothesis is critical for testing predictions. Thus, one contributing factor to publication bias is that traditional frequentist p-values cannot be used to provide evidence for the absence of an effect, as a non-significant p-value (regardless of its size) could either be consistent with the absence of an effect or that the test lacked statistical power.

In this talk, two alternative approaches for making null effects informative will be presented: Equivalence tests and Bayesian hypothesis tests. These methods will be demonstrated in a non-technical manner using JASP and JAMOVI, which are open source point-and-click software packages. In sum, this talk will be useful for anyone that wants to expand their statistical toolbox with methods that can accurately evaluate the absence of effects.

This talk was delivered and recorded at the Department of Psychology, University of Oslo, as part of its open science drop-in series, on 11 June, 2024.

Slidedeck: https://osf.io/7bdhk
Data: https://osf.io/fa7w6