Explaining community-level variance in group randomized trials

Between-community variance or community-by-time variance is one of the key factors driving the cost of conducting group randomized trials, which are often very expensive. We investigated empirically whether between-community variance could be reduced by controlling individual- and/or community-level covariates and identified these covariates from four large community-based group randomized trials or surveys: the Working Well Trial; Kaiser Adolescent Survey; Kaiser Adults Survey; and the Eating Patterns Study. We found that adjusting for covariates will often substantially reduce the between-community variance component. Therefore investigators could block the communities according to these covariates, or adjust for these covariates to improve the power of community trials. We found that the community-by-time variance components are always near zero in these data sets, especially for the surveys where a cohort was followed over time. The covariate adjustment had less impact on reducing the community-by-time variance for the cohort samples than for the cross-sectional samples. This suggests that blocking may not be necessary for the design of the group randomized trials where the change from baseline is of primary interest. The Working Well Trial data were used to illustrate this point.