About the speaker

Anne studied psychology at the University of Heidelberg and psychological research methods at the University of Glasgow, and worked in a developmental psychology lab at LMU Munich for two years. Her background is in infant research, but since she first learned about the “replication crisis” in psychology, she devoted more and more time to follow the discussions around ways to make research more transparent and reproducible (“open science”). Eventually this led her to switch tracks and turn to meta-science as her main research focus: In October 2017, she started her PhD in Daniël Lakens’ project “Increasing the reliability and efficiency of psychological science” at TU Eindhoven.

About the talk

Psychological theories typically predict the presence of an effect or relationship, which is commonly tested by setting up a null hypothesis of no effect or relationship and performing a significance test. One problem of this procedure is that researchers rarely specify the size of effects predicted by their theories. Hypotheses tested this way are difficult or impossible to falsify: A non-significant result could always be due to insufficient power and simply mean that an effect is smaller than expected (but not zero). This is why non-significant results merely represent the absence of evidence for an effect, yet they are often misinterpreted as evidence for the absence of an effect.

Equivalence testing can offer a solution to this problem: It allows researchers to test (and reject) the hypothesis that an effect is larger than a “smallest effect size of interest” (SESOI), and conclude that it is too small to care about — e.g., smaller than the effect predicted by a theory, smaller than an effect that would justify the costs for a new intervention, or simply smaller than effects that can meaningfully be studied with the resources a lab has available. The procedure has the added benefit of making hypotheses more falsifiable, because a more precise prediction than “not zero” has to be made.

Equivalence tests are based on a very simple technique (two one-sided tests) and have been used in medicine and biostatistics for several decades. Recently, new software solutions have been developed to make them more accessible to psychological researchers, providing a helpful addition to the frequentist tool kit. This talk will give an introduction to equivalence tests and how to use them in practice, and discuss approaches to defining the smallest effect size of interest for different research questions.