Reconstructing Long Term Fertility Trends by Pooling Birth Histories

Bruno D. Schoumaker, Université Catholique de Louvain

In developing countries - where civil registration systems are deficient - fertility trends are often inferred from a few estimates from censuses and surveys. Such an approach tends to mask changes that occurred between two data points. Fertility trends computed in this way may also be largely influenced by data quality issues. In this paper, we present a method for reconstructing and smoothing fertility trends by combining birth histories from multiple surveys, using Poisson regression and restricted cubic splines. Data used in the paper come from World fertility surveys and Demographic surveys. The method is illustrated with data from a variety of countries, with varying numbers of surveys and with different data quality issues. The method is also validated using microsimulations methods.