If you test predictions, but your test can’t be wrong, what’s the value of demonstrating you are right? Error statistical approaches to philosophy of science put more trust in claims that have passed a severe test – the claim was supported, despite a high probability of being falsified. So how do we make sure predictions are falsifiable? One approach in frequentist statistics is to specify a smallest effect size of interest, and to consider a prediction falsified if you can reject the presence of any effect that is large enough to be interesting. I will discuss why falsifiability is desirable, explain the general approach to falsify predictions, and provide some specific examples researchers can apply to their own research questions.
Daniel Lakens is an experimental psychologist working at the Human-Technology interaction group at Eindhoven University of Technology. In addition to his empirical work in cognitive and social psychology, he works actively on improving research methods and statistical inferences. His lab is funded until 2022 by a VIDI grant from the Dutch science funder NWO on a project that aims to improve the reliability and efficiency of psychological science. He teaches a popular online course ‘Improving Your Statistical Inferences’ for which he received the Leamer-Rosenthal Prize for Open Social Science in 2017.