Joint copula‐based beta regression with covariate‐dependent dispersion for malaria indicators

Abstract

Malaria surveillance relies on multiple indicators that reflect different aspects of transmission and disease burden, yet these outcomes are often analyzed separately under constant‐variance assumptions. Such approaches may obscure residual dependence and heterogeneity in variability. Using Demographic and Health Survey data from Burkina Faso (2005–2020), linked to area‐level estimates of Plasmodium falciparum parasite prevalence in children aged 2–10 years and annual clinical malaria risk, we study copula‐based joint regression models with covariate‐dependent dispersion. We compare independent and joint models under constant and varying dispersion structures. Joint modelling improves model fit and estimation of joint exceedance probabilities, and allowing dispersion to vary enhances uncertainty quantification and identification of areas with elevated combined risk. Simulation results show that ignoring heteroskedasticity can lead to biased inference.