Subnational probabilistic projections fertility: rethinking form Latin America

ANDREOZZI LUCIA

Journal of Demography

College of Population Studies at Chulalongkorn University

Año: 2024 vol. 39

2730-3934

General trends in fertility, mortality, and migration can be discerned andprojected into the future with reasonable results. However, considerableuncertainty is attached to each specific trend in a particular country or region.Thus, subnational projections represent a special chapter within demographicprojections. After introducing subnational projections and its close relationshipwith fertility, this work proposes as its main objective to project fertility rates atthe subnational level for Argentina using a probabilistic method, i.e., the Bayesianhierarchical model (BHM), and then compare the results with the point estimatesof deterministic projections published by its national agency for statistics.Bayesian hierarchical modelling is a statistical model written at multiple levelsthat estimates the parameters of the posterior distribution using the Bayesianmethod. In the second stage, forecasts were obtained from two models, oneincluding all the countries available in the World Population Prospects (WPP)and a second model based only on a subgroup of countries (Argentina, Colombia,Chile, Cuba and Uruguay), both of which are based on data from 1980 to 2010.This set of countries was selected using the transition to identify a subset ofcountries with similar patterns that includes Argentina, but showing differentpatterns compared with the rest of the region. A time period is selected to includeyears for which these countries have achieved adequate data quality. BHM forsubnational projections is an extremely useful and flexible method, presentingmany advantages over classical methods. The Bayesian framework is a powerfulscheme to generate national and subnational projections for mortality, fertility,and population. This research aims to address the limitations and uncertainties inexisting subnational population projections and reinforce the use of probabilisticmodels: specifically, BHM, which respects data and avoids forcing it to getcaught up in a mathematical assumption.