Components of Variance and Intraclass Correlations for the Design of Community-based Surveys and Intervention Studies: Data from the Health Survey for England 1994

• Published in: American journal of epidemiology Vol. 149; no. 9; pp. 876 - 883
• Main Authors: Gulliford, Martin C., Ukoumunne, Obioha C., Chinn, Susan
• Format: Journal Article
• Language: English
• Published: Cary, NC Oxford University Press 01.05.1999
• Subjects: Biological and medical sciences; Cluster Analysis; Cross-Sectional Studies; data collection; Data Interpretation, Statistical; England - epidemiology; epidemiologic methods; Epidemiologic Studies; Epidemiology; General aspects; Health Status; Health Surveys; Humans; intraclass correlation; Life Style; Medical sciences; Methodology; Public health. Hygiene; Public health. Hygiene-occupational medicine; random allocation; Research Design; Risk Factors; sampling studies; statistics; variance; England; United Kingdom; Great Britain; Europe; Data logging; Human; Correlation; Statistical analysis; Methodology; Epidemiology; Public health; Variance; Geographical variation
• ISSN: 0002-9262; 1476-6256
• DOI: 10.1093/oxfordjournals.aje.a009904

The authors estimated components of variance and intraclass correlation coefficients (ICCs) to aid in the design of complex surveys and community intervention studies by analyzing data from the Health Survey for England 1994. This cross-sectional survey of English adults included data on a range of lifestyle risk factors and health outcomes. For the survey, households were sampled in 720 postal code sectors nested within 177 district health authorities and 14 regional health authorities. Study subjects were adults aged 16 years or more. ICCs and components of variance were estimated from a nested random-effects analysis of variance. Results are presented at the district health authority, postal code sector, and household levels. Between-cluster variation was evident at each level of clustering. In these data, ICCs were inversely related to cluster size, but design effects could be substantial when the cluster size was large. Most ICCs were below 0.01 at the district health authority level, and they were mostly below 0.05 at the postal code sector level. At the household level, many ICCs were in the range of 0.0–0.3. These data may provide useful information for the design of epidemiologic studies in which the units sampled or allocated range in size from households to large administrative areas. Am J Epidemiol 1999;149:876–83.