The groupwise decomposition: Estimating group-specific contributions to differences between demographic measures

Eugenio Paglino

doi:10.4054/DemRes.2026.54.14

Volume 54 - Article 14 | Pages 441–470

The groupwise decomposition: Estimating group-specific contributions to differences between demographic measures

By Eugenio Paglino

Date received:

11 Nov 2024

Date published:

6 Mar 2026

Word count:

4674

Keywords:

decomposition, estimation, life table functions

DOI:

10.4054/DemRes.2026.54.14

Replicable Material:

demographic-research.54-14 (zip file, 165 kB)
readme.54-14 (text file, 4 kB)

Download PDF
Submit a Response Letter

Abstract

Background: Several general decomposition algorithms have been developed to decompose the difference between the values of demographic measures in two populations. However, fewer techniques are available to researchers wishing to understand how different subpopulations are contributing to the difference in the value of a demographic measure between the total population and a reference subpopulation.

Methods: This paper proposes a general decomposition to answer these ‘groupwise’ questions. As for other general decomposition techniques, the method assumes that the aggregate measure is computed from matrices of discrete demographic data but is agnostic to the specific function used to compute the measure of interest. After proving some of the decomposition’s properties, its relationships with other methods are discussed, and extensions are considered.

Results: The proposed method is used to investigate the contribution of United States counties by metropolitan/nonmetropolitan code to the life expectancy gap between large metropolitan areas and the US as a whole. Results from this new approach are compared to those obtained by adapting other general decomposition algorithms, and their computational efficiency is evaluated.

Contribution: This paper provides a new decomposition method to investigate how different subpopula-tions contribute to the difference between a demographic measure for the total population and for a reference population. This setup is common in demography, and the decomposition method developed in this paper adds one more tool to improve our understanding of how different groups contribute to the demographic profile of a population.