Abstract:
|
Design of cluster randomized trials requires determination of sample sizes at multiple levels. For the common two-level design, in which clusters of individuals are randomized to different intervention groups, statistical power depends on both the number of clusters per group and the cluster sizes. It is well known that, due to the effect of intraclass correlation, the statistical power can be improved more efficiently by selecting a larger number of clusters than larger cluster sizes. However, in practice, the number of candidate clusters is usually limited. We derive the formula for calculation of the minimum required number of clusters to achieve adequate power, allowing the cluster sizes to vary, for both two- and three-level designs. We then investigate the effect of the variation in cluster sizes on the minimum required number of clusters and discuss the implications of the minimum required number of clusters when designing a cluster randomized trial.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.