Bayesian Multivariate Spatial Estimation of Small-Area Fertility Schedules

Renato M. Assuncao, Universidade Federal de Minas Gerais
Carl P. Schmertmann, Florida State University
Suzana M. Cavenaghi, Universidade Estadual de Campinas

Accurate estimation of localized rate schedules and age patterns is essential in many demographic analyses, and has become more important with greater access to geocoded data. In this paper we explore applications of Bayesian spatial methods to estimation of rate schedules for small areas. The main novelty is treating each rate schedule as a vector of parameters to be estimated simultaneously, rather than element-by-element. We extend Bayesian spatial methods to estimate vectors of local rates –- specifically, fertility schedules for a set of over 3500 Brazilian municipalities in 1991. The main idea is to estimate local ASFRs by “borrowing strength” not only from women of the same age in neighboring areas, but also from women in other age groups within the area, and from observed regularities in ASFR schedules across locations. This method promises significant improvements in local-area rate estimates when both spatial and age-related patterns in rates are strong.