Statistical Modeling for Age-Specific Fertility Rate: A Finite Mixture Model Approach

• Activity Number: 237
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Government Statistics Section
• Author(s): Seongyong Kim*+ and Daeyoung Kim and Yousung Park
• Companies: Korea University and University of Massachusetts, Amherst and Korea University
• Keywords: age specific fertility rate ; population projection ; finite mixture model ; EM algorithm ; statistical inference

Abstract

The pattern of age-specific fertility rate has been typically bell-shaped, but recently, it has been changed for developed countries. For example, a second peak has appeared in young age, thus age specific fertility rates has bi-modal shape. To reflect these changes, many researchers suggested using mixture models. However, their approach is based on least-squares method and thus is quite limited for detailed statistical inference such as interval estimation and investigation of features of subpopulations.

In this paper, we propose to employ finite mixture models using maximum likelihood (ML) approach, and illustrate this approach for six countries. For the component density in the model, we compare several commonly used distributions in Demography, and for the selection of proper number of components, we take into account not only model selection criteria but also stability of component estimates over the data period.

To better understand the structure of the fertility behavior in the population, we also investigate the resulting clusters (subpopulations) by exploring covariates (such as marital status, birth order, race) for members of each cluster.