Keywords: Cluster randomized trial, randomization inference, constrained randomization, community engagement, spillover
We illustrate the design, implementation, and analysis of a cluster randomized trial with three key features: 1) constrained randomization to balance groups on key baseline variables, 2) randomization inference to handle a relatively small number of clusters and potential spillover between clusters, and 3) adjustment for baseline prognostic variables that may improve precision. Despite the benefits of these three techniques, to the best of our knowledge they have not been utilized all at once for a cluster randomized trial. Methods are applied to a cluster randomized trial to assess the impact on health outcomes of an intervention designed to bridge existing medical and social services between community-based organizations (CBOs) in East Baltimore and The Johns Hopkins Health System.