Infant mortality is a serious problem in India. In order to better understand the problem previous research has looked at the topic using sophisticated multivariate models assuming that infant mortality is either Gaussian or LogGaussian. In this paper we argue that infant mortality is Log-Gaussian distributed with non-constant variance and that making such an assumption leads to more efficient estimates and a better fit to the data. Using infant mortality data from the National Health Survey in Bihar, India we compare two distributions--Log Gaussian and Gamma—and find that the Log-Gaussian non-constant variance model does indeed lead to more efficient estimates and a better fit to the data.
Download Full PDF Version (Non-Commercial Use)