Estimating Subnational Fertility Rates with a Principal Component-Based Bayesian Spatiotemporal Model

Ameer Dharamshi, University Of Washington
Celeste Winant, University of California, Berkeley
Magali Barbieri, Institut National d'Études Démographiques (INED)
Monica Alexander, University of Toronto

Understanding the economic, social, and cultural consequences of declining fertility rates requires high-quality fertility data over long periods of time and at fine geographic scales. This is a challenging task as at subnational levels, small population counts lead to high stochasticity; consequently, zero birth observations are a frequent occurrence. In this paper, we propose a principal component-based Bayesian spatiotemporal model to estimate age-specific fertility rates in small subnational areas. The model exploits structural patterns in fertility through the principal components, and stabilizes estimation by pooling information on the local manifestation of these patterns in space and time. We apply our model to county-level fertility data from California for 1982-2022 to estimate age-specific fertility rates and downstream fertility indicators. The model appears to perform well in these initial experiments and we identify several interesting patterns in Californian fertility for further study.

Keywords: Bayesian methods , Spatial Demography, Small area estimation, Fertility

See extended abstract.