Abstract
Sample sizes in cluster surveys must be greater than those in surveys using simple random sampling in order to obtain similarly precise prevalence estimates, because results from subjects examined in the same cluster cannot be assumed to be independent. Therefore, a crucial aspect of cluster sampling is estimation of the intracluster correlation coefficient (ρ): the degree of relatedness of outcomes in a given cluster, defined as the proportion of total variance accounted for by between-cluster variation. In infectious disease epidemiology, this coefficient is related to transmission patterns and the natural history of infection; its value also depends on particulars of survey design. Estimation of ρ is often difficult due to the lack of comparable survey data with which to calculate summary estimates. Here we use a parametric bootstrap model to estimate ρ for the ocular clinical sign “trachomatous inflammation—follicular” (TF) among children aged 1–9 years within population-based trachoma prevalence surveys. We present results from a meta-regression analysis of data from 261 such surveys completed using standardized methods in Ethiopia, Mozambique, and Nigeria in 2012–2015. Consistent with the underlying theory, we found that ρ increased with increasing overall TF prevalence and smaller numbers of children examined per cluster. Estimates of ρ for TF were independently higher in Ethiopia than in the other countries.
Intracluster correlation coefficient in multicenter childhood trauma studies
