Estimation techniques in generalized linear mixed models with application to disease impact modelling

Abstract

This study applies the theory of Generalised linear Mixed Models (GLMMs) to survey data on disease impact in Ghana. It determines the variables that are responsible for making dependent members of households feel the impact of illness and/or death for three identified types of households. It assesses four models in terms of these variables generated using the Maximum Mean Pseudo-Likelihood (MMPL) and the Residual Mean Pseudo-Likelihood (RMPL) techniques in SAS. For all four models considered, the MMPL produces more suitable models than the RMPL. The impact of illness and/or death on HIV/AIDS, Other Illness/Deaths or No Illness/Death households is felt in the areas of reallocation of dependents’ time, dependents having to work harder to substitute for lost income, dependents leaving work to care for the sick, and household reducing expenditure as a result of illness and/or death. It is found that the degree of impact depends on marital status, sex or tribe of household headship, remoteness of occurrence of mortality/morbidity, total asset value, and level of annual adult’s health expenditure. It is also found that in almost all cases examined, the HIV/AIDS household suffers a significant impact compared to No/Other disease household. The findings indicate that there should be a continued effort at reducing not only the incidence but also the impact of HIV/AIDS on households.