Objective: Alcohol intervention studies that allocate intact social groups to study conditions require adjustment to the usual analytic methods to account for the positive intraclass correlation that exists in such groups. This article presents intraclass correlations for measures related to alcohol use among young adults and discusses the use of those estimates to plan new studies.
Method: Young adults aged 18-20 were selected at random from driver's license lists in each of the 15 communities participating in the Communities Mobilizing for Change on Alcohol project. Respondents were surveyed by telephone to assess their drinking habits and other factors related to alcohol use. Community-level intraclass correlations were computed for those measures, both prior to and after adjustment for person- and community-level covariates.
Results: The community-level intraclass correlations tend to be small, with larger values for belief and attitude items than for self-reported behaviors. Even so, correlations of this magnitude can have important deleterious effects on the power to detect important treatment effects in an otherwise well-designed and well-executed study. Adjustment for person-level covariates often reduced those correlations, and adjustment for community-level covariates often reduced them substantially.
Conclusions: There is measurable variation in measures related to alcohol use among young adults that is attributable to their community of residence. With adjustment for selected person- and community-level covariates, the magnitude of these correlations can be sharply reduced allowing the investigator to plan a more efficient community trial.