Why optional stopping can be a problem for Bayesians

Rianne de Heide and Peter Grünwald

What is this paper about?

This paper asks when optional stopping is harmless for Bayesian hypothesis testing and when it is not. It shows that robustness can fail for the default or pragmatic priors commonly used in applications and separates several qualitatively different types of prior behaviour.

Summary

Recently, optional stopping has been debated in the Bayesian psychology community. This article asks whether optional stopping is problematic for Bayesian methods, and specifies under which circumstances and in which sense it is and is not. By varying and extending earlier experiments, we show that as soon as parameters of interest are equipped with default or pragmatic priors—as in many practical applications of Bayes-factor hypothesis testing—resilience to optional stopping can break down. We distinguish three types of default priors with different optional-stopping behaviour, ranging from no problem to severe problems.

Topics

Bayesian learning and generalized Bayes · E-values and anytime-valid inference · Foundations of statistics, probability and learning

← Research