Bayesian Sample Size Determination for Multilevel Models – the BayesSSD Package - Ulrich Lösener (Utrecht University)

11.05.2026 12:15 – 13:00

Sample size determination (SSD) is key in designing efficient and sufficiently powered studies. This step can be particularly complex in studies with hierarchical data, such as cluster-randomised trials (CRTs) and longitudinal intervention studies. Most existing methods for SSD are designed for null hypothesis significance testing, an approach with numerous practical and theoretical limitations that has been subjected of considerable criticism over the past few decades. An alternative approach is the Bayesian hypothesis evaluation using the Bayes Factor, which quantifies the relative evidence for competing hypotheses, providing more direct and informative answers to the question of which hypothesis the data supports. The current methods for SSD with the Bayes factor, however, are limited to non-hierarchical models. To address this gap, we developed the R package ‘BayesSSD’, which performs simulation-based Bayesian SSD for hierarchical data. The use of the package is demonstrated a range of study designs, from a simple two-arm CRT with binary and continuous outcomes to more complex scenarios involving multiple outcomes, longitudinal studies with participant dropout, and studies with more than two treatment conditions. We illustrate the impact of key design elements, such as the intraclass correlation coefficient, effect sizes, and variance components, on the required sample size. We conclude by providing practical recommendations for researchers designing studies with hierarchical data.

Lieu

Bâtiment: Pinchat

Salle 010

entrée libre