“…Most of the trials used worksite, or whatever unit of cluster randomization, as the unit of analysis (8,11,13,22,26,27,39). Those using workers as the unit of analysis took into account the effect of the cluster randomization procedure by incorporating a nested design structure (29,36,37,51) or by using other procedures (18,30,31,35,47). Seven trials did not take cluster randomization into account when analyzing their data (12,16,24,28,38,47,49).…”