Intraclass correlation for measures from a worksite health promotion study: Estimates, correlates, and applications

Objectives. Investigators planning studies employing group-randomized designs need good estimates of the extra variation introduced as a result of correlated observations with-in units of assignment. We report intraclass correlation coefficients (ICCs) for a wide range of outcomes commonly employed in worksite studies and demonstrate analysis methods that can limit their deleterious impact. Methods. A sample of 11,711 employees of 24 firms recruited from the Minneapolis/St. Paul metropolitan area completed a mailed survey in the SUCCESS study, reporting on a broad array of outcomes. Applying mixed-model regression, we provide both crude and adjusted estimates of ICCs for 27 outcomes. Results. The crude ICCs were generally small, with a mean of .0163 and values ranging from 0 to .0650. Adjustment for demographics reduced the ICCs for 25 of the 27 outcomes, and adjustment for additional individual- level covariates further reduced the ICCs for 23 of the 27 outcomes. Conclusions. Our results suggest that worksite-level ICCs for a variety of outcomes are generally small and can generally be reduced by adjustment for individual-level characteristics. Incorporating this information in planning worksite studies can improve sample size calculations to avoid underpowered studies.