TWEEDIE REGRESSION ANALYSIS OF DETERMINANTS OF BIRTH WEIGHT IN NAVRONGO

Abstract

This study models infant birth weight and investigate maternal predictors using lifetime data by taking into account the non-zero probability of zero occurrences of the response variable. The birth weights were found to be positively skewed and leptokurtic in nature. The average birth weight was found to be 2.9898kg. In modelling the determinants of infant birth weight, the Tweedie regression model was compared with Gaussian, Gamma and inverse Gaussian regression models. The performance of these models was assessed using the Akaike Information Criterion. Based on the Akaike Information Criterion, the Tweedie regression model showed superiority in modelling the determinants of infant birth weight as compared to the Gaussian, Gamma and inverse Gaussian regression models respectively. This showcases the added advantages of the Tweedie regression models having the capacity of handling non-negative highly right-skewed and continuous data with probability mass at zero. The parameter estimates of the Tweedie regression model revealed that the gender of the child, religion, marital status and educational level of the mother significantly contributes to the infant birth weight. Based on the findings of the study, it was recommended that except for the status of antenatal care, emphasis must be placed on establishing the reasons for the decrease in infant birth weight as a result of the maternal factors; parity, delivery type and the age of the mother.