Estimating the total fertility rate in developing countries

Leontine Alkema, National University of Singapore
Adrian Raftery, University of Washington
Samuel J. Clark, University of Washington

The objective is to estimate the total fertility rate in countries with issues with data quantity and quality, and assess its uncertainty. Biases and differences in error variance between observations are estimated using linear regression. Based on a data set of seven countries in western Africa we find that retrospective surveys with a recall period of more than five years overestimate the total fertility rate, while direct estimates and observations from longer ago underestimate fertility. The error variance is larger for observations with a one-year time span, observations that were collected before the mid 1990s, and one Demographic Health Survey. The total fertility rate is estimated using a local smoother, and the uncertainty assessment is carried out using a weighted likelihood bootstrap. Model validation shows that taking data quality into account gives more calibrated and sharper confidence intervals; on average uncertainty is reduced by about 40%.