When nothing happens: Bayesian approaches to testing for "no effect"

When nothing happens: Bayesian approaches to testing for "no effect"

We are delighted to have Mircea Zloteanu for the first Bayesian Mixer of 2026, please register here: https://www.tickettailor.com/events/bayesianmixer/2066946

Abstract: Non-significant p-values do not constitute evidence for the absence of an effect, yet researchers often wish to make claims about “no difference.” Often this will lead to publications that make incorrect statements (i.e., claim no difference when the data is just uncertain) or even deter researchers from pursuing specific hypotheses or publish their work.

This talk presents Bayesian alternatives for evaluating null results. First, I demonstrate Bayes Equivalence Tests for interval (composite) hypotheses. Second, I illustrate the default Bayesian t-test method relying on model comparison (using the JZS Bayes factor), showing how routine analyses can quantify evidence for or against a point null. Third, I introduce posterior-based equivalence testing using the Region of Practical Equivalence (ROPE) via bayestestR. Together, these approaches provide principled tools for publishing credible null findings.

Bio: Mircea Zloteanu is a Lecturer in Psychology and Criminology at King’s College London, where his focus is on Research Methods and Statistics. He received his PhD in Experimental Psychology from University College London and has previously held posts at UCL, Teesside University, and Kingston University. His recent work focuses on meta-research, statistical methodology, and improving research practice, with particular emphasis on Bayesian methods, equivalence testing, and transparent reporting. Mircea is actively involved in the open science movement as UK Reproducibility Network Local Co-Lead and a member of ReproducibiliTea, promoting reproducible, rigorous, and cumulative psychological science.