SEMINAR: Modelling Body Mass Index Distribution using Maximum Entropy Density

The objective of this paper is to model the conditional distribution of Body Mass Index (BMI) by examining the relations between a set of covariates and the moments of the BMI distribution. While BMI is often seen as a leading indicators of health, most studies on the distribution of BMI did not model beyond the second order moments. This makes it difficult to examine the determinants of obesity as the mean and variance do not contain sufficient information about the tail of the distribution. This paper applies the Maximum Entropy Density framework to examine the relations between a set of covariates and the higher order moments of the BMI distribution. The aim is to provide a more accurate description on the relations between a set of determinant and the shape of the BMI distribution. Theoretically, the paper derives the asymptotic properties of the maximum likelihood estimator of the proposed density, including consistency and asymptotic normality. Empirically, this paper applies the proposed framework to an Australian dataset. The results demonstrate how different covariates affect different moments of the BMI distribution.