Robust Spatial Regression Modeling Based on Combined Estimating Equations: Analysis of Malaria Incidences in Liberia and Madagascar

Malaria is a major obstacle to socio-economic development in Sub Sahara Africa, with about 90% of all recorded cases worldwide. The disease accounted for nearly one million deaths in 2008, mostly among children living in Africa. Furthermore malaria is a leading cause of under-five deaths in SSA where a child dies every 45 seconds of Malaria. It is of high importance to properly identify risk factors that are associated with the incidence of Malaria. Analyzing spatial data must be done with caution as observations may now be correlated, hence ordinary statistical methods assuming independence of observations are no longer valid. Ignoring the structure of the data may result in asymptotically biased parameter estimates. A crucial step in modeling spatial data is the specification of the spatial dependency, by choosing the correlation function. However, often the choice for a particular application is unclear and diagnostic tests will have to be carried out following fitting of a model. To resolve this problem, we adopt a more robust method for modeling spatial correlation by simultaneously solving the combined estimating equations using different working correlation structures. We illustrate our method by modeling the spatial correlation of malaria incidence in Liberia and Madagascar.