ABSTRACT
As was shown in the previous chapter a cluster randomized trial is less efficient in terms of estimating the treatment effect than a simple randomized trial with the same total sample size. The degree of efficiency loss increases when the cluster size and the intraclass correlation coefficient increase. One strategy to increase design efficiency is using larger sample sizes at the subject and/or cluster level. The drawback of this strategy is that it will most likely result in increasing study costs, from adding more clusters and subjects within clusters. The aim of this chapter is to discuss alternative strategies to improve statistical power in cluster randomized trials. Taking account of relevant covariates in the multilevel model is a relatively easy method. Relevant covariates are correlated with the outcome variable and they explain part of the unexplained variance. This will in general result in a lower standard error of the treatment effect estimator, provided the covariate is uncorrelated with treatment condition. For a large number of clusters, random assignment of clusters to treatment condition will most likely ensure treatment condition and the covariate will be uncorrelated. For a small number of clusters, one must rely on other randomization procedures to achieve covariate balancing such as minimization, matching and pre-stratification.
