A Bayesian Approach to Combining Population and Survey Data for Male Fertility Estimation

Michael Rendall, RAND
Ryan Admiraal, University of Washington
Mark S. Handcock, University of Washington
Kara Joyner, Cornell University
Elizabeth Peters, Cornell University
Felicia T. Yang, Cornell University

Where survey data are used for the analysis of male fertility, data quality is a concern. Survey data are known to be worse for men than for women. Birth registration data, especially for non-marital births, are also available in less detail for men, and often with lower degrees of completeness. To address these estimation problems, male fertility rates estimated from survey data in the 2002 National Survey of Family Growth are first evaluated against age-specific rates calculated from birth registrations and population estimates. Male birth reports by age and other variables reported commonly between men and women are then evaluated against female birth reports in the survey data. The information from these evaluations is used in Bayesian estimation allowing for successive relaxation of assumptions about the accuracy of the population data from birth registrations and the survey data. Reductions in variance and bias about the estimates are demonstrated.