Multilevel analysis of factors associated with child mortality in Uganda

Abstract

The purpose of this study was to examine the effect of factors associated with child mortality in Uganda. Demographic and Health Survey data for 2006 were used to investigate these factors. This data set had a hierarchical structure. To account for this nested data, a hierarchical / random regression model was fitted to find the significant factors affecting child mortality.

Sex of a child, duration of breastfeeding, birth weight, Education level, age of mother, household wealth were found to be important predictors of child mortality in the two models. However, controlling for mother level factors in model one, the within childhood characteristics were seen to be highly correlated. In a concept from an explicit multilevel analytic framework, the study demonstrated that individual (child) and mother level characteristics are independent predictors of child mortality, and that there is significant variation in odds of reporting child mortality, even after controlling for effects of both individual- and mother-level characteristics.

Results as by the Standard Logistic regression model (model II) were almost the same as the results by the random effects model (model I). However, the p - values in the random effects model were small compared to the p – values of a standard logistic model. Hence the random effects model are more statistically significant than those in a standard logistic regression model due to its lack of independence within variables. In this setting, random effects regression model is recommended as an appropriate alternative to standard logistic regression to account for variations due to a hierarchical structure in the data used in this study.