This is a hybrid event.

Bayesian alternatives to null hypothesis significance testing

Null hypothesis significance testing (NHST) and its p-value are ubiquitous in scientific practice. However, misuse and misinterpretation of these crucial tools is well documented. Various suggestions to fix, or replace, NHST have been offered. In this talk we will discuss some options that fall under the Bayesian inferential framework. We will introduce null hypothesis Bayesian testing (NHBT) and its Bayes factor as the direct Bayesian analogues to NHST and the p-value, respectively. Important differences between the two approaches will be highlighted. Furthermore, we will emphasize that, just as NHST needs to be accompanied by effect size estimates, so does NHBT. We will recall a simple relation between Bayesian estimation of (posterior distributions of) effect sizes and NHBT, and its implications. This will lead to the realization that estimation can be seen as a workhorse for various alternative types of hypothesis testing. Indeed, by combining ideas by Kruschke (2018) and Wellek (2010) with Smiley, Glazier and Shoda’s (2023) framework for statistical inference, it will be shown how all methods in that framework can be dealt with in a Bayesian way.

This lecture is made possible in part by a generous endowment from Professor Allen L. Edwards.

Q&A and light refreshments to follow.