b Heritability estimates and 95% confidence intervals (in parentheses) based on variance components from restricted maximum likelihood estimators.
We adjusted for clustering using a method described by Gilles Lamothe based on a variance estimator to weight the Mantel Haenszel odds ratio for cluster-correlated data.
We adjusted the CI for clustering (CIca), using a method described by Gilles Lamothe (2011) based on a variance estimator to weight the Mantel Haenszel OR for cluster-correlated data (Bieler & Williams, 1995; Williams, 2000).
We adjusted the CI for clustering within sites (CIca), using a method described by Gilles Lamothe [ 13], based on a variance estimator to weight the Mantel Haenszel Odds Ratio for cluster-correlated data [ 14, 15].
We adjusted for clustering using a method produced by Gilles Lamothe based on a variance estimator to weight the Mantel Haenszel odds ratio for cluster-correlated data, described elsewhere [ 35].
21 We adjusted these bivariate estimates of association by country and for clustering (at school level) using a method described by Lamothe 22 23 based on a variance estimator to weight the Mantel–Haenszel OR for cluster-correlated data.
We examined associations between HIV status and potential risk factors in bivariate then multivariate analysis using the Mantel–Haenszel procedure [ 42] with an adjustment for clustering described by Gilles Lamothe based on a variance estimator to weight the Mantel–Haenszel odds ratio for cluster-correlated data [ 43, 44].
Further analysis examined factors related to outcomes of interest, in bivariate analysis and then in multivariate analysis using the Mantel Haenszel procedure [ 7] with an adjustment for clustering described by Gilles Lamothe, based on a variance estimator to weight the Mantel Haenszel odds ratio for cluster-correlated data [ 8, 9].
Figure 1 illustrates the relative bias, Figure 2 the bias of the variance estimators and Figure 3 the true coverage of the 95% confidence interval based on the robust variance estimator.
P-values and CI are based on a standard variance estimator.
