Estimating the covariates of childhood mortality from retrospective reports of mothers.

In this paper we compare various models for estimating the covariates of childhood mortality from the type of data commonly available in developing countries. Specifically, we examine how much precision is lost as various pieces of information, such as dates of birth and death for each child, are discarded. The conclusion which we reach is that even incomplete mortality data of the type collected in household surveys or censuses can yield estimates which are very close to those based on the much richer wealth of data collected in detailed maternity histories. Two substantive conclusions of interest are that in the two countries (Sri Lanka and Korea) we examined, the education of the father has a significant and pronounced effect on childhood mortality even when the mother's education is controlled, and once other covariates are controlled, there is no difference between urban and rural childhood mortality.