The article discusses methods for designing and analyzing three-arm cluster randomized trials with small numbers of clusters. The researchers evaluated different randomization methods and data modeling techniques. They found that small-sample corrections are needed for certain values of the intraclass correlation coefficient. Restricted randomization methods can improve power in single post-intervention trials. Matched-group randomization enhances power but is not as effective as covariate-constrained randomization. Adjusting for fewer covariates than used in randomization can lead to errors. Overall, the methods perform similarly to two-arm cluster trials.