The worksite component of variance: design effects and the Healthy Worker Project.

Variance estimates in worksite health promotion studies depend partly on the intraclass correlation coefficient (ICC). ICC quantifies homogeneity of a variable within worksites. ICC would be zero for randomly formed worksites, but is generally positive because employees tend to share personal characteristics. The ratio comparing the variance estimated from worksite means with that estimated from individuals under simple random sampling is the design effect (DEFF). A DEFF of 1.0 indicates no excess variance due to worksite. The Healthy Worker Project (HWP) was a 32 worksite cross-sectional and longitudinal study of a weight and smoking intervention program. ICCs in cross-sectional surveys for health-related outcome variables ranged from 0.006 to 0.009, DEFFs from 2.0 to 2.6 ICCs/DEFF's in longitudinal analysis were smaller; ICCs ranged from -0.002 to 0.003, DEFFs from 0.7 to 1.5. Positive ICCs substantially increased variance estimates at a single measurement, yet variance of longitudinal analysis was less subject to worksite dependence. It is concluded the worksite component of variance is real and should not be ignored, although the worksite component of variance is small in these longitudinal analyses. This observation should be replicated before it is used in other worksite health promotion research.