Logit Models for Household Food Insecurity Classification

Abstract

Micro-level measurement of food insecurity is a necessary approach towards a more feasible solution to the global problem for proper classification of households by food insecurity status. Measurement of food insecurity is a challenge because it is a multi-faceted latent and continuous phenomenon explained by a wide range of both quantitative and qualitative variables. In this paper, we examined the quantitative variables and applied exploratory factor analysis to identify which of them significantly influence household food insecurity. Logit models were then developed using the variables identified. Further, empirical data obtained from Tororo and Busia rural households in Uganda were used to fit the models. Four logit models based on four scenarios were developed and compared. The key findings pointed to the fact that if households were to be correctly analyzed and classified into the right food security category, a hybrid dependent variable that represents as many aspects of food insecurity as possible should be used. The model correctly classified 90 % of the combined households for two districts. However, when fitted for separate districts, it was established that 99% of households in Busia and 96% in Tororo district respectively, were found to be food insecure