Keywords: Cluster Randomization Trials, Sample Size Calculation, Budget Constraint
In cluster randomization trials(CRTs), groups are randomized to treatments rather than individuals to minimize experimental contamination, since individuals assigned to different treatments are less likely to interact when they are nested in different clusters. CRTs are also adopted to avoid ethical issues or to control costs. But with a common interest in individual-level inference, CRTs could be less efficient than randomizing individuals directly since individuals in the same cluster might be correlated. With such unique design, sample size calculation for CRTs calls for special attention. Many papers propose formulas relying on the asymptotic approximation of different test statistics under the randomization and sampling distributions. Commonly in CRTs, only a small number of units are actually randomized, and thus asymptotic approach may not be appropriate, especially for discrete outcomes. We propose to calculate the required sample size for CRTs by approximating the randomization and sampling distributions using simulation, with options to incorporate budget constraints. This approach is general, non-parametric, and does not limit to any specific type of outcome or estimand.