A Multiple Imputation Approach to Improving Health Data Accuracy in Pooled Cross-Sectional Analysis

Romuald Daniel Boy-ngbogbele (Department of Mathematics, Pan African University, Nairobi- 6200000200 (Kenya))
Oscar Ngesa (Department of Mathematics, Taita Taveta University, Nairobi-63580300 (Kenya))
Thomas Mageto (Department of Statistics and Actuarial Sciences, Jomo Kenyatta University of Agriculture and Technology, Nairobi-6200000200 (Kenya))
Celestin C Kokonendji (Department of Mathematics, Marie and Louis Pasteur University, Cedex-25030 (France))

Journal of Multidisciplinary Applied Natural Science
Vol. 6 No. 3 (2026): Journal of Multidisciplinary Applied Natural Science

Pooled cross-sectional health survey data are commonly utilized to examine trends in child malnutrition; however, their precision is usually undermined by measurement errors and absent data. This study assesses the efficacy of a Multiple Imputation (MI) methodology in rectifying these biases and enhancing the reliability of malnutrition prevalence estimates for children under five in Cameroon. Utilizing simulated data that mirrors the structure of four rounds of Demographic and Health Surveys (DHS) from Cameroon (2004, 2011, 2018, 2022), we provide a logistic regression model including measurement error correction by multiple imputation. The efficacy of the MI-corrected model is evaluated against an uncorrected model using several measures, such as prevalence estimates, classification accuracy, precision, and parameter stability across different sample sizes. The MI-corrected model consistently yielded lower and more precise estimates of malnutrition prevalence than the uncorrected model, with reductions of up to 11.34 percentage points in 2004. Classification accuracy increased by 3–4 percentage points across survey waves, with a corresponding improvement in precision. Increasing the sample size reduced the variability of the parameters, as shown by lower standard deviations and coefficients of variation. This made the regression estimates more stable. The model exhibited enhanced resilience in situations including absent data and inaccurately assessed variables. Multiple Imputation successfully rectifies measurement errors and addresses missing data in pooled cross-sectional surveys, resulting in more credible estimates of child malnutrition prevalence. These improved estimates offer a more accurate assessment of public health needs, better targeting of dietary interventions, and more dependable monitoring of trends over time. The results support the implementation of MI-based corrections in national health survey analyses to enhance evidence-based policy and resource distribution in resource-constrained environments.