Death clustering, birth spacing and infant mortality in Brazil

Many studies have demonstrated a detrimental effect of short birth intervals on infant survival. These studies have used a variety of statistical techniques which assume independence between observations but this assumption is violated if individual women contribute more than one birth to the sample because the survival chances of siblings are correlated. This thesis investigates the consequences of this familial correlation for the relationship between birth spacing and infant mortality in Brazil through the use of random-effects logistic models which control for familial correlation. The data used are from the 1986 Brazil Demographic and Health Survey (DHS). Initial analyses of the relationship between birth spacing and infant mortality confirm that short birth intervals are an important risk factor for neonatal and postneonatal mortality in Brazil, and that the magnitude of the effects are similar to those observed elsewhere in South America. The analysis is then repeated using random-effects logistic models and the results compared with those obtained earlier using logistic regression. Family effects are very highly significant but the relationship between birth spacing and infant mortality is preserved. The final part of the thesis attempts to explain the causes of familial association by utilising the additional information available in the Brazil DHS on health care use, household environment and infant feeding. In the neonatal period none of the variables considered help to explain family effects but in the postneonatal period breastfeeding behaviour appears to play an important role.