DEMOGRAPHIC RESEARCH
A peer-reviewed, open-access journal of population sciences
DEMOGRAPHIC RESEARCH
VOLUME 29, ARTICLE 23, PAGES 617-640
PUBLISHED 27 SEPTEMBER 2013
http://www.demographic-research.org/Volumes/Vol29/23/
DOI: 10.4054/DemRes.2013.29.23
Research Article
Age groups and the measure of population
aging
Hippolyte d’Albis
Fabrice Collard
c 2013 Hippolyte d’Albis & Fabrice Collard.
This open-access work is published under the terms of the Creative
Commons Attribution NonCommercial License 2.0 Germany, which permits
use, reproduction & distribution in any medium for non-commercial
purposes, provided the original author(s) and source are given credit.
See http://creativecommons.org/licenses/by-nc/2.0/de/
Table of Contents
1 Introduction 618
2 Related literature 619
3 Endogenous age groups and the measurement of population aging 621
3.1 An example 622
3.2 Methodology 625
4 The dynamics of population aging in industrialized countries 628
4.1 Aging in the US 628
4.2 On the number of groups 631
4.3 An international perspective 633
5 Conclusion 635
6 Acknowledgements 636
References 637
Appendix 639
Demographic Research: Volume 29, Article 23
Research Article
Age groups and the measure of population aging
Hippolyte d’Albis1
Fabrice Collard2
Abstract
BACKGROUND
Measures of population aging are important because they shape our perception of demographic
trends. Indicators of aging based on fixed ages contributed to a dramatic portrayal
of demographic evolutions, some of which were associated with the myth of decline.
OBJECTIVE
We propose a new measure of population aging, based on the relative age of each individual
in the population. Our approach builds on previous work by Aghevli and Mehran
(1981) and relies on optimal grouping techniques that are used to determine the various
age groups within a population. The cutoff ages for these groups, such as the age from
which an individual is considered to be an older person, are then endogenous variables
that depend on the entire population age distribution at any given moment.
METHODS
We show how to apply optimal grouping techniques to age distributions and how to calculate
various indicators of aging, which are invariant with respect to proportional rescaling
of distributions. We compute these indicators for the US, and a sample of 13 other industrialized
countries.
RESULTS
We find that, contrary to common arguments for an aging population, the share of elderly
individuals within the total population has not increased much, and has remained stable in
these countries. These results complement and reinforce the earlier findings of Sanderson
and Scherbov (2005, 2007) who also reassessed the aging phenomenon.
1 Correspondending author. Paris School of Economics, University Paris 1/ Centre d’Economie de la Sorbonne,
106 boulevard de l’Hôpital, 75013 Paris, France. E-mail: dalbis@univ-paris1.fr.
2 Department of Economics, University of Bern.
http://www.demographic-research.org 617
d’Albis & Collard: Age groups and the measure of population aging
“The first part of life is childhood. The second is your child’s childhood. And
then the third, old age.” Barbara Kingsolver, The Lacuna
1. Introduction
Population aging is often perceived as a very widespread phenomenon. According to
the last United Nations “Population Aging Report” (2009), the proportion of the global
population aged over 60 years was 8% in 1950, 10% in 2000, and is expected to reach
21% in 2050. In this report, the United Nations have used a very specific, albeit very
common, type of measurement for assessing the population aging phenomenon, namely
the proportion of population aged over 60. And yet, it is evident that today’s 60-year-olds
are often very different from their parents at the same age and have absolutely nothing
in common with their grandparents at the same age. The age at which one becomes
an older person is a notion that changes over time; thus, calculating the proportion of
older persons based on a fixed age only provides us with biased information. The use
of such an indicator is often justified on the ground that these fixed ages (60, 65 or 80,
depending on the study) correspond to the eligibility ages of certain social programs,
most notably the pay–as–you–go pension system. However, recent events, for example
in Europe, show that these ages also undergo changes (see notably Fenge et al., 2008,
and references therein). Indicators, though simple, are not neutral. While studying the
history of social representation that defines old age as starting from 60 years, Bourdelais
(1994, 1999) showed that indicators of aging based on fixed ages contributed to a dramatic
portrayal of demographic evolutions, some of which were associated with the myth of
decline. The aim of our paper is to propose a new means of determining the various age
groups in a population and to recalculate new indicators of aging based on the cutoff ages
of these groups.
The main difficulty in characterizing the relative size of older populations lies in the
determination of the age at which an individual becomes an older person. We propose
to use all the statistical information contained in the population age distribution to define
this age. We proceed in the following manner: we predefine a certain number of age
groups, then “optimally” divide single age-classes among these different groups. The
optimal grouping rule, as proposed by Aghevli and Mehran (1981), consists in selecting
cutoff ages for groups such that age differences are a minimum within each group and a
maximum between groups. The resulting age group–based representation is then optimal
as it gives the best portrayal of the initial distribution. Information loss arising from
the grouping of data is therefore minimal. This procedure leads to a clear endogenous
definition of an elderly individual: one classifies as elderly any individual whose age is
closer to the average age of the elderly group than to the average of any other group. A
618 http://www.demographic-research.org
Demographic Research: Volume 29, Article 23
direct implication of this grouping is that the definition of old age fundamentally relies
on the relative position of each cohort in total population and thus depends on the entire
shape of the age distribution. Our concept of the stages of life is a relative one: the
“age” of individuals within a given cohort depends on the size of the other cohorts. This
is a statistical interpretation of the what is nicely characterized by B. Kingsolver in the
quotation we reproduced above.
Optimal grouping techniques were initially used by Aghevli and Mehran (1981) and
Davies and Shorrocks (1989) for income distribution issues, and applied by Esteban et
al. (2007) to polarization measurements. In this paper, we demonstrate in a formalized
manner how to apply these techniques to age distributions in order to calculate cutoff
ages. The latter then allow us to calculate various indicators of aging, which are invariant
with respect to proportional rescaling of distributions. These calculations are no more
complicated than those proposed in related contributions, which will be described below.
Most notably, in the extreme case where only two age groups are considered, our indicator
of aging becomes the proportion of individuals whose age is greater than the mean age.
Applying this technique to total US population, we find that the age at which one becomes
an older person has dramatically increased over the last century. In our benchmark
experiment involving 4 age groups, we find that the entry age into oldness was 48.7 years
in 1933 and skyrocketed to 57.6 years in 2005. Most industrialized countries exhibit the
same behavior of the entry age into oldness. We then find that the share of the so-defined
elderly in total population remained stable over time and does not display a pronounced
upward sloping trend. We then compute the elder-child ratio and find that its time average
increased over the last 50 years by less than 6.5% in the US, and by less than 8% on average
in a sample of 13 industrialized countries. These findings then suggest that aging is
less pronounced when a measure that takes evolutions in the age distribution into account
is used.
The remainder of the paper is organized as follows. Section 2 compares our method to
recent contributions in the field. Section 3 describes our approach to defining endogenous
age groups and defines our aging indicators. Section 4 revisits aging in the US and in 12
other industrialized countries in light of our new indicators. A last section offers some
concluding remarks.
2. Related literature
Our work is part of continuing efforts in the latest research on the demographics of aging.
Two distinct bodies of work have led to the proposal of indicators of aging that are not
based on the constancy of the age at which one becomes an older person. The first of
these is founded on a simple idea, initially developed by Ryder (1975), that defines an
http://www.demographic-research.org 619
d’Albis & Collard: Age groups and the measure of population aging
individual’s age not according to the number of years lived since birth, but according to
the remaining number of years that he or she is expected to live. Thus, Ryder proposes
considering an individual as an older person when his or her remaining life expectancy
is less than ten years. This type of characterization, which may be used to define the
proportion of older persons in a population, constitutes a major advancement, as it enables
the distinction between individual and population aging. This idea has been pursued by
Sanderson and Scherbov (2005, 2010), who establish the mean age of an age pyramid that
is recalculated based on the life expectancy at each age.
However, such approaches have two drawbacks. First, at a given date, a cohort’s life
expectancy is unknown and its estimation using a period life table is imperfect (Goldstein
and Wachter, 200