Probabilistic Indirect Estimation of under-Five Mortality Rates

Carl P. Schmertmann , Florida State University

I introduce and test a new, probabilistic approach to the classic Brass problem of indirect estimation of under-five mortality (q5) from aggregate data on children ever born and children dead by age of women. A Bayesian model can exploit the same regularities as classic Brass methods, but without restrictive assumptions about demographic constants. A Bayesian approach does not require fixed temporal patterns in rates, and it can use data from multiple maternal age groups to estimate current q5. It can also produce uncertainty measures that incorporate both sampling error and potential errors in demographic assumptions. The proposed approach uses simple parametric models to describe fertility and mortality levels and patterns over the period covering approximately 30 years before the survey, and treats the parameters in these models as unknown quantities with prior distributions. Tests use DHS data and compare model results to current UN estimates.

See paper

 Presented in Session 29. Estimation Methods