Fig. 1. Simplified comparison of a ‘traditional approach’ and a ‘Bayesian approach’ that combines a mini-experiment design with Bayesian updating.

Whereas in the traditional approach sample sizes are estimated a priori using frequentist power calculations, sample sizes can be adjusted over the course of a running experiment by conducting interim analyses in the Bayesian approach. Please note that in the latter approach, the interim analyses should ideally be conducted by a person different to the experimenter to avoid introducing any experimenter bias due to the knowledge gain over the course of the running experiment.