A Methodology for Studying Child Mortality Differentials in Populations with Limited Death Registration

Claire M. Noel-Miller, University of Pennsylvania
Douglas Ewbank, University of Pennsylvania

The paper presents an alternative to the Preston-Trussell method for analyzing child mortality differentials when the only available information is the number of children ever born and deceased and various characteristics of their environment. We apply the method to the 1993 Gambia census. Our model assumes a binomial dichotomous dependent variable recording failures (survivorships) and successes (deaths) following a number of trials (children ever born). It is based on a skewed logistic regression (scobit), an elaboration upon the standard logit model which relaxes the assumption that individuals with probability of experiencing a success equal to 0.5 are most sensitive to independent variable changes. We offer a second improvement by relying on Brass multipliers more carefully tailored to local mortality and fertility conditions. Using Monte Carlo simulations we compare the performance of the scobit model to more simple models in small samples (N=1,000 and N=5,000).