nicholasmwanza/ITN-Malaria-Kenya

Causal effect of ITN use on malaria in Kenyan children under five # Estimating the Causal Effect of ITN Use on Malaria Infection Among Children Under Five in Kenya **Author:** Nicholas Mwanza **Supervisor:** Dr. Haisa Osmanli **Institution:** African Institute for Mathematical Sciences (AIMS) Cameroon **Date:** April-May 2026 --- ## Abstract Malaria remains a leading cause of child mortality in sub-Saharan Africa. While insecticide-treated nets (ITNs) are the primary prevention tool, estimating their real-world effectiveness using observational data is challenging due to confounding from environmental, socioeconomic, and demographic factors. This study estimates the causal effect of ITN use on malaria infection among children under five in Kenya using KMIS 2015 and 2020 data. The analytical sample comprises 6,544 children across 528 clusters. Six causal inference methods are implemented and compared: logistic regression, generalized linear mixed models (GLMM), propensity score matching (PSM), inverse probability of treatment weighting (IPTW), doubly robust (DR) estimation, and double machine learning (DML). Key findings show substantial confounding. Unadjusted models show no association, but fully adjusted models reveal protective effects with odds ratio 0.777 and 95 percent confidence interval 0.605 to 0.999. GLMM-based causal methods produce stronger cluster-specific effects with odds ratios ranging from 0.392 to 0.562, representing 44 to 61 percent reduction. Double machine learning estimates population-average effects with odds ratio 0.972, representing a 2.8 percent reduction. The discrepancy is consistent with non-collapsibility given high intracluster correlation of 50 to 58 percent. E-value analysis gives 3.05, suggesting moderate robustness to unmeasured confounding. Heterogeneity analysis reveals stronger effects among poorer households, urban areas, older children aged 36 to 59 months, and high-rainfall regions. --- ## Research Question **Primary question:** What is the causal effect of ITN use on malaria infe …