A TECHNIQUE FOR MEASURING EPIDEMIOLOGICALLY USEFUL FEATURES OF BIRTH-WEIGHT DISTRIBUTIONS

Birthweight distributions have been conceptualized as a predominant Ga ussian distribution contaminated in the tails by an unspecified 'resid ual' distribution. Acknowledging this idea, we propose a technique for measuring certain features of birthweight distributions useful to epi demiologists: the mean and variance of the predominant distribution; t he proportions of births in the low- and high-birthweight residual dis tributions, and the boundaries of support for these residual distribut ions. Our technique, based on an underlying multinomial sampling distr ibution, involves estimating parameters in a mixture model for the mul tinomial bin probabilities after having chosen the support of the resi dual distribution with a model selection criterion. A modest simulatio n study and experience with a few actual datasets indicate that use of a Bayesian information criterion (BIG) as model selection criterion i s superior to use of Akaike's information criterion (AIC) in this appl ication.