The spatial modification effect of predictors on household level food insecurity in Ethiopia

Household food insecurity remains highly prevalent in developing countries (including in Ethiopia) and it has been recognized as a serious public health problem. Several factors such as demographic, economic, social, and clinical factors influence household food insecurity, and these vary geographically. In this work, we investigate the geographical modification of the effect of several factors on chronic food insecurity. The data is from the Ethiopia socioeconomic survey conducted by the Ethiopia Central Statistics Agency (ECSA) in collaboration with the World Bank. Ethiopia socioeconomic survey is a long-term project to collect nationally representative panel survey of over 6500 households. A geo-additive model which accounts the structured and unstructured special effect was adopted to estimate household food insecurity risk factors. The study also revealed significant spatial variations on household food insecurity among administrative zones. Mainly, household living in the Sidama, Gamo Gofa, Shinille, Basketo, Wolyita, Wag Hemira, Liben, Awi, Eastern Tigray and West Harerghe zones, having higher food insecurity than the other zones in Ethiopia. Moreover, the analysis also showed that availability of credit services, proximity to service centers, average years of schooling of members of the household, and household assets are negatively associated with household food insecurity, whereas shocks, age, and dependency ratio increase the odds of a household to be food insecured. The generalized geo-additive mixed-effects model enables simultaneous modeling of spatial correlation, heterogeneity and possible nonlinear effects of covariates. Our study investigated the spatial heterogeneity of household level food insecurity, and its association with shocks, age, dependency ratio, availability of credit services, average years of schooling, and household assets. Our findings have also an important implication for planning as well as in the search for the variables that might account for the residual spatial patterns.