Abstract
Abstract
Objectives: To estimate how much unexplained variation in childhood stunting in Ghana occurs at the household and survey-community levels, and to assess whether that pattern remains after accounting for water, sanitation, maternal education, and core sociodemographic characteristics.
Methods: We analysed 4,928 children aged 0 -59 months from the 2022 Ghana Demographic and Health Survey, representing 3,545 households and 616 survey communities. Survey-weighted prevalence was estimated using the DHS design. Bayesian hierarchical logistic models included household and community random intercepts. We assessed the final model using eight chains and a series of checks covering alternative priors, survey weights, missing data, spline-based age modelling, detailed WASH coding, posterior predictive checks, PSIS-LOO, and independent GEE analyses.
Results: Survey-weighted stunting prevalence was 17.4% (95% CI 15.8 -18.9). In the final model, the household SD was 1.23 and the community SD 0.39; the corresponding household VPC and community ICC were 0.31 and 0.03. Boys had higher odds of stunting than girls (OR 1.50, 95% posterior interval 1.24 -1.82), as did children aged 24 -35 versus 0 -5 months (OR 2.91, 2.06 - 4.21). Unimproved drinking-water source had OR 1.50 (1.15 -1.99), whereas higher maternal education had OR 0.37 (0.20 - 0.65). Sensitivity analyses preserved the main household-versus-community pattern.
Conclusions: Most of the residual clustering in these data occurred at the household rather than the survey-community level. Because the study is observational, the results do not establish causal effects; instead, they point to household conditions that standard DHS covariates may not fully capture.