Sequential Analyses using Effect Size Confidence Intervals: A Simulation-Based Approach

Sequential Bayesian analyses have been suggested as a means for efficient hypothesis-testing within psychology, often substantially reducing the sample sizes required. In the context of clinical psychology, trial designs using sequential Bayesian analyses have been proposed as a way to accelerate psychological treatment development. However, a focus on Bayesian analyses as a means to conduct sequential testing may hinder uptake of such methods due to a lack of familiarity amongst many researchers. Demonstrating how equivalent sequential analyses are possible using non-Bayesian (e.g., frequentist) analyses could increase the accessibility of these methods and thus the extent to which psychological treatment development research can benefit from more time and resource-efficient trial designs. This paper demonstrates a simple method for sequential testing based on confidence intervals around effect size estimates, and illustrates comparative efficiency to an approach based on Bayes factors.