Spatial Bayesian Modeling of Child Malnutrition with Measurement Error in Demographic Health Survey Data

Background: Childhood malnutrition remains a major public health challenge in low- and middle-income countries, particularly in Sub-Saharan Africa. Conventional statistical models often ignore spatial dependence and measurement error, potentially leading to biased estimates and reduced predictive performance. Methods: This study proposes a spatial Bayesian logistic regression model that incorporates both measurement error and spatial dependence for analyzing pooled cross-sectional data. A simulation study, designed to mimic Demographic and Health Survey (DHS) data, was conducted using multiple clusters and survey periods. Covariates were generated with realistic measurement error, and spatial correlation was introduced through a distance-based covariance structure. The performance of the spatial model was compared with a non-spatial Bayesian model using accuracy, area under the curve (AUC), precision, recall, F1-score, and estimated prevalence. Results: The spatial Bayesian model consistently outperformed the non-spatial model across all evaluation metrics. Improvements were observed in accuracy (approximately 3--5\%), AUC (0.05--0.07 increase), precision, recall, and F1-score. Additionally, the spatial model produced lower and more stable prevalence estimates, indicating improved calibration and reduced bias. These results demonstrate the advantages of incorporating spatial structure in modeling childhood malnutrition. Conclusion: Incorporating spatial dependence and measurement error within a Bayesian framework significantly enhances model performance and provides more reliable estimates of childhood malnutrition. The proposed approach offers a valuable tool for improving health data analysis and supports more effective targeting of public health interventions.