A Bayesian approach to sample size estimation and the decision to continue program development in intervention research

Abstract

OBJECTIVE : In intervention research, the decision to continue developing a new program or treatment is dependent on both the change-inducing potential of a new strategy (i.e., its effect size) and the methods used to measure change, including the size of samples. This article describes a Bayesian approach to determining sample sizes in the sequential development of interventions. DESCRIPTION : Because sample sizes are related to the likelihood of detecting program effects, large samples are preferred. But in the design and development process that characterizes intervention research, smaller scale studies are usually required to justify more costly, larger scale studies. We present 4 scenarios designed to address common but complex questions regarding sample-size determination and the risk of observing misleading (e.g., false-positive) findings. From a Bayesian perspective, this article describes the use of decision rules composed of different target probabilities and prespecified effect sizes. Monte-Carlo simulations are used to demonstrate a Bayesian approach—which tends to require smaller samples than the classical frequentist approach—in the development of interventions from one study to the next.