The Quantity and Quality of Life and the Evolution

of World Inequality
By GARY S. BECKER, TOMAS J. PHILIPSON, AND RODRIGO R. SOARES*

GDP per capita is usually used to proxy for the quality of life of individuals living in
different countries. Welfare is also affected by quantity of life, however, as represented
by longevity. This paper incorporates longevity into an overall assessment of the
evolution of cross-country inequality and shows that it is quantitatively important. The
absence of reduction in cross-country inequality up to the 1990s documented in previous
work is in stark contrast to the reduction in inequality after incorporating gains in
longevity. Throughout the post–World War II period, health contributed to reduce
significantly welfare inequality across countries. This paper derives valuation formulas
for infra-marginal changes in longevity and computes a “full” growth rate that
incorporates the gains in health experienced by 96 countries for the period between
1960 and 2000. Incorporating longevity gains changes traditional results; countries
starting with lower income tended to grow faster than countries starting with higher
income. We estimate an average yearly growth in “full income” of 4.1 percent for the
poorest 50 percent of countries in 1960, of which 1.7 percentage points are due to
health, as opposed to a growth of 2.6 percent for the richest 50 percent of countries, of
which only 0.4 percentage points are due to health. Additionally, we decompose
changes in life expectancy into changes attributable to 13 broad groups of causes of death
and three age groups. We show that mortality from infectious, respiratory, and digestive
diseases, congenital, perinatal, and “ill-defined” conditions, mostly concentrated before age
20 and between ages 20 and 50, is responsible for most of the reduction in life expectancy
inequality. At the same time, the recent effect of AIDS, together with reductions in mortality
after age 50—due to nervous system, senses organs, heart and circulatory diseases—
contributed to increase health inequality across countries. (JEL I10, I31, J17, O57)

Although GDP per capita is usually used as a tries, material gain is obviously only one of
proxy for the quality of life in different coun- many aspects of life that enhance economic
well-being. Overall economic welfare depends
on both the quality and the quantity of life:
*
Becker: Department of Economics and Sociology, Uni- yearly income and the number of years over
versity of Chicago, 1126 East 59th Street, Chicago, IL
60637, and Hoover Institution (e-mail: gbecker@midway.
which this income is enjoyed. Both past and
uchicago.edu); Philipson: Irving B. Harris Graduate School recent estimates suggest that longevity has been
of Public Policy Studies, University of Chicago, 1155 East a quantitatively important component of the
60th Street, Chicago, IL 60637 (e-mail: t-philipson@ overall gain in welfare in the United States
uchicago.edu); Soares: Department of Economics, Univer- during the twentieth century (Dan Usher, 1973;
sity of Maryland, College Park, MD 20742, and Catholic
University of Rio de Janeiro (e-mail: soares@econ. David Cutler and Elizabeth Richardson, 1997;
umd.edu). Earlier versions of this paper circulated under the William D. Nordhaus, 2003; Kevin M. Murphy
title “Growth and Mortality in Less Developed Nations” and and Robert Topel, 2003; similar evidence for
as NBER Working Paper No. 9765. We benefited from Sweden is presented by Kristina Burström et al.,
comments from Magnus Johannesson, Chad Jones, Casey
Mulligan, Eric Neumayer, Richard Rogerson (the editor), an
2003). This paper analyzes whether gains in life
anonymous referee, and seminar participants at Brown Uni- expectancy have also been an important deter-
versity, University of Arkansas, University of Chicago, minant of the evolution of welfare inequality
University of Wisconsin-Madison, Virginia Tech, World across countries.
Bank, the 2002 AEA Meetings, and the UN-WIDER Con- A vast literature has investigated the evolu-
ference “Sharing Global Prosperity.” Becker and Philipson
received support from the George J. Stigler Center for the tion of the cross-country dispersion in income
Study of the Economy and the State, and Becker from the per capita, and whether poor countries tend to
John M. Olin Foundation and the NICHD (Grant No. 5401). grow faster than rich ones (see, for example,
277
278 THE AMERICAN ECONOMIC REVIEW MARCH 2005

TABLE 1—EVOLUTION OF CROSS-COUNTRY INEQUALITY IN INCOME AND LIFE EXPECTANCY, 1960 –2000

Income per capita Life expectancy

1960 1990 2000 1960 1990 2000
Relative mean dev. 0.4751 0.4733 0.4215 0.1179 0.0507 0.0509
Coeff. of variation 1.2344 1.2529 1.1662 0.2629 0.1245 0.1361
Std. dev. of logs 1.0178 1.0300 0.9620 0.2552 0.1326 0.1513
Gini coeff. 0.5104 0.5187 0.4873 0.1293 0.0690 0.0730
Regression to the mean ⫺0.0069 ⫺0.0741 ⫺0.6133 0.0364
over previous date (p-value ⫽ 0.86) (p-value ⫽ 0.00) (p-value ⫽ 0.00) (p-value ⫽ 0.31)

Notes: Income per capita is GDP per capita in 1996 international prices, adjusted for terms of trade (Penn World Tables 6.1).
Life expectancy is life expectancy at birth (World Development Indicators, World Bank). Inequality measures weighted by
country population (abstracting from within country inequality). Sample includes 96 countries, comprising more than 82
percent of the world population. Regression to the mean is the coefficient of a regression of the change in the variable over
the period on its initial level (natural logs used in the income regressions; weighted regressions).

Robert J. Barro and Xavier Sala-i-Martin, 1995; change in the variable on its initial level (the
Angel de la Fuente, 1997; Gregory Mankiw et natural logarithm, in the case of income per
al., 1992; Danny T. Quah, 1996; Stephen L. capita).
Parente and Edward C. Prescott, 1993). All These data, as previous research has found,
these studies give virtually the same results, show that regressions to the mean, Gini coeffi-
namely, that income inequality across countries cients, coefficients of variation, and other mea-
was not reduced during most of the postwar sures of inequality do not show evidence of
period. reduction in income inequality across countries
In contrast, evidence suggests that cross- up to the 1990s.1 During the 1990s, the eco-
country inequality in several different dimen- nomic success of China and India, together
sions of health was substantially reduced with their huge populations, reduced the cross-
throughout the twentieth century (see, for ex- country dispersion in income per capita. (With-
ample, Francois Bourguignon and Christian out these two countries, the sample shows
Morrison, 2002; Brian Goesling and Glenn increasing inequality between 1960 and 1990,
Firebaugh, 2004; Eric Neumayer, 2003; Randa and stable inequality between 1990 and 2000;
Sap and Stephen C. Smith, 2002; Stephen this is also in good part behind the results ob-
Younger, 2001). So the trends in income and tained in Sala-i-Martin, 2002.)
health inequality have indeed been strikingly Table 1 also presents the same dispersion
different in the recent past, and, therefore, trust- measures usually used for income for the case
ing on income alone to infer the evolution of of life expectancy at birth. In this case, the
welfare inequality across countries may lead to evidence is diametrically opposite. By any mea-
misleading results. sure, life expectancy inequality declines sub-
We illustrate the cross-country trends in stantially over the entire period. Countries
health and income inequality between 1960 and starting with low longevity tended to gain more
2000 in Table 1. Income is measured with the in life expectancy than countries starting with
gross domestic product per capita adjusted for high longevity.2 The regression to the mean
terms of trade (Penn World Tables 6.1), and health
is measured by life expectancy at birth (World
1
Bank World Development Indicators). The sam- As pointed out by Milton Friedman (1992), zero-mean
measurement error in the initial period income tends to
ple includes 96 countries, comprising more than generate spurious negative correlation, artificially increas-
82 percent of the world population in 1960. ing the degree of regression to the mean. Even with this
The table shows several dispersion measures bias, it is not uncommon for one to obtain a positive
(weighted by country populations) for income coefficient in these regressions.
2
per capita and life expectancy at birth, for the As opposed to the negative impact of measurement
error on regression to the mean in income, there are reasons
years 1960, 1990, and 2000. The last row pre- to believe measurement error has a positive impact in the
sents the regression to the mean over the previ- case of longevity. An upward bias in poor-country life
ous period: the coefficient of a regression of the expectancies is commonly believed to occur due to incom-
VOL. 95 NO. 1 BECKER ET AL.: QUANTITY AND QUALITY OF LIFE AND WORLD INEQUALITY 279

coefficient implies that, on average, each addi- come per year and the value of years enjoyed,
tional ten years of life expectancy in 1960 were changes the traditional results regarding cross-
associated with a reduction of roughly six years country inequality. Our main result is that in-
in life expectancy gains in the following 40 corporating longevity changes the conclusions
years. But the decline in health inequality is related to the evolution of welfare inequality
concentrated entirely between 1960 and 1990. over time: countries became significantly more
After that, the effects of AIDS in Africa are felt, equal between 1960 and 2000. In particular, we
and life expectancy inequality increases slightly estimate an average yearly growth in “full in-
between 1990 and 2000. come” of 2.8 percent, of which roughly three-
These two patterns suggest that incorporating quarters are due to income per capita and one-
longevity into an overall assessment of the quarter to longevity. For the poorest 50 percent
changes in cross-country inequality may be im- of countries in 1960, however, there is an aver-
portant, as the extent of changes in income age yearly growth of 4.1 percent, of which 1.7
inequality is small compared to the changes in percentage points are due to health, as opposed
life expectancy inequality. In addition, changes to the richest 50 percent of countries, for which
in income and health inequality have followed the average yearly growth in “full income” is
completely different patterns, suggesting that a 2.6 percent, and only 0.4 percentage points are
large component of the recent changes in health due to health.
is orthogonal to income. We also disaggregate mortality data by age
This paper tries to account for the impact of groups and causes of death to try to understand
longevity on the evolution of welfare across the determinants of the cross-country reduction
countries during the last few decades. The use in life expectancy inequality, and the diseases
of per capita income to evaluate welfare im- responsible for the observed gains in welfare.
provements assumes that it reflects the level of For each age group and cause of death, we
economic welfare enjoyed by the average per- compute a counterfactual measure of the mor-
son, but it has well-known shortcomings such as tality rate that would be observed in the 1990s
not measuring nonmarket goods and home- had mortality rates by all causes and at all ages
production. We try to fill in one of those gaps by but the ones in question remained at their 1960
incorporating survival rates throughout a per- values. This approach allows us to estimate the
son’s life. In particular, we consider the evolu- life expectancy gain attributable to reductions in
tion of welfare for a “hypothetical life-cycle mortality at each age and by each specific cause
individual” (HLCI). A HLCI for a given year of death. We show that changes in mortality due
and country is defined to have the income per to infectious, respiratory, and digestive dis-
capita of the country in every year of life and to eases, and congenital and perinatal conditions,
face throughout life the survival probabilities mostly concentrated at early ages, are the most
determined by the country’s cross-sectional sur- important factors determining the reduction in
vival curve. Our results refer to inequality life expectancy inequality. In other words, mor-
across different societies as measured by differ- tality at early ages by these causes of death fell
ences in welfare of this hypothetical individual. more rapidly in poor than in rich countries. At
We estimate the monetary value of longevity the same time, reductions in mortality due to
gains and add it to the observed gains in income nervous system, senses organs, heart, and cir-
per capita. This gives the change in income that culatory diseases worked toward increasing in-
would have been observed if all the welfare gain equality, as mortality among the elderly by
in the period had taken the form of income these causes fell more rapidly in rich than in
growth. We then analyze how the growth in this poor countries. The large changes in mortality
“full income,” including both changes in in- observed in the developing world are consistent
with the interpretation that poor countries ab-
sorbed technology and knowledge previously
available in rich countries, at relatively low
plete data from rural areas. This upward bias has been costs, while most of the changes in mortality in
reduced, however, by improved collection of mortality sta-
tistics in poor countries in the last decades. The reduction in developed countries took advantage of recent de-
upward bias for poor countries would induce a positive, as velopments on the frontier of medical technology.
opposed to negative, bias on the regression toward the mean. Our paper relates to the original work of Dan
280 THE AMERICAN ECONOMIC REVIEW MARCH 2005

Usher (1973), which was developed further by where S(t) is the probability of survival to age t,
Sherwin Rosen (1988) and Murphy and Topel y(t) is income at t, c(t) consumption at t, and r ⫽
(2003). Our work also relates to existing mea- ␳ the assumed interest rate. This budget con-
sures created by the United Nations that attempt straint assumes full annuity insurance, or the
to incorporate nonmaterial aspects into broader existence of a complete contingent claims
measures of well-being (UNDP, 2002). How- market.
ever, as discussed at length in Philipson and Now consider a given country at two points
Soares (2002), our methods differ in that pref- in time, with lifetime income and survival func-
erences revealed by market behavior, and not by tions denoted by Y and S, and Y⬘ and S⬘ respec-
arbitrary assumptions from government or in- tively. We are interested in the infra-marginal
ternational agencies, dictate the relative impor- income W(S,S⬘) that would give a person in this
tance of nonmaterial aspects in the overall country the same utility level observed in the
evaluation of welfare. first period, but with the mortality rates ob-
The structure of the paper is outlined as fol- served in the second:
lows. Section I discusses the methodology used
to value longevity gains, extending previous (3) V共Y⬘ ⫹ W共S, S⬘兲, S兲 ⫽ V共Y⬘, S⬘兲.
analysis of marginal changes in longevity to
consider valuations of infra-marginal changes.
Section II contains our basic results on the re- The growth rate in the “full” lifetime income
duction in inequality induced by considering that values both the gains in material income
longevity gains. Section III considers age- and and the gains in longevity then corresponds to
cause-specific contributions to this reduction in G ⫽ [Y⬘ ⫹ W(S,S⬘)]/Y ⫺ 1. This “full” growth
inequality. Section IV concludes the paper. rate is thus the standard growth rate added with
the growth rate in income measuring the gains
in longevity.
I. Valuing Infra-Marginal Changes in Longevity Income can be used to measure material im-
provements only with a set of assumptions that
A. Theory justifies using a single number to portray
changes in a country’s welfare. Similar simpli-
Previous work of Usher (1973), Rosen fying assumptions are needed here to measure
(1988), and Murphy and Topel (2003) derive the material value equivalent to the life expect-
formulas to value marginal changes in survival ancy gains. More precisely, to calibrate the
rates. Here we extend this work by providing model for commonly available national income
the corresponding formulas for infra-marginal and mortality statistics for a given country and
changes. Consider the indirect utility function year, we consider a HLCI who receives the
V(Y,S) of an individual with survival function S country’s income per capita in all years of life
and lifetime income Y: and faces throughout life the country’s cross-
sectional survival function. Under the main-

冕 ⬁ tained assumptions, this implies that optimal

(1) V共Y, S兲 ⫽ max exp共⫺␳t兲S共t兲u共c共t兲兲 dt consumption equals the constant income, c(t) ⫽
0
c ⫽ y, so that the indirect utility function can be
expressed in terms of the yearly income y, as in:

subject to
(4) V共y, S兲 ⫽ u共y兲A共S兲

(2) Y⫽ 冕
0
⬁
exp共⫺rt兲S共t兲y共t兲 dt where A(S) ⫽ 兰⬁0 exp(⫺rt)S(t) dt is the value of
an annuity based on the survival function S. If
w(S,S⬘) is the yearly, as opposed to lifetime,

冕
income that measures the gain in longevity in a
⬁
manner similar to before, w satisfies
⫽ exp共⫺rt兲S共t兲c共t兲 dt
0 (3⬘) u共y⬘ ⫹ w共S, S⬘兲兲A共S兲 ⫽ u共y⬘兲A共S⬘兲.
VOL. 95 NO. 1 BECKER ET AL.: QUANTITY AND QUALITY OF LIFE AND WORLD INEQUALITY 281

with the corresponding annual “full-income” expression (5) and this functional form, we ob-
growth rate g ⫽ [y⬘ ⫹ w(S,S⬘)]/y ⫺ 1. The value tain closed form solutions for w and g.3
of longevity gains measured in yearly income is The set of parameters (␣,␥,r) needed to com-
related to the value measured in lifetime income pute these values can be calibrated from other
according to W(S,S⬘) ⫽ A(S)w(S,S⬘). The usual parameters more commonly estimated in the
critiques of GDP as a measure of “full in- “value of life” and consumption literatures.
come”—that it does not incorporate value of More precisely, we have that ␣ ⫽ c1 ⫺ 1/␥(1/␧ ⫺
leisure, household production, and nonmarket (1/(1 ⫺ 1/␥))), where ␧ ⫽ (u⬘/(c)c)/u(c) is the
goods—also apply to our methodology. In fact, elasticity of the instantaneous utility function,
we attempt to fill in one of these gaps, as it often discussed and estimated in empirical stud-
relates to home-produced or nonmarket health. ies of compensating differentials for occupa-
tional mortality risks. In particular, Murphy and
B. Calibration Topel (2003, p .23) estimate ␧ to equal 0.346.
Martin Browning et al. (1999, p. 614), after
Here we discuss our methods for calibrating exhaustively reviewing the estimates from the
the parameters used throughout the empirical empirical literature on the inter-temporal elas-
analysis to estimate the infra-marginal willing- ticity of substitution (␥), suggest that it is
ness to pay for gains in survival rates w(S,S⬘). slightly above unity. Using ␥ ⫽ 1.250, ␧ ⫽
The monetary value of the gains in longevity 0.346, and c ⫽ $26,365, we calibrate the value
measured in annual income and the growth rate of ␣ to equal –16.2.4,5 We assume the annual
in “full-income” are determined implicitly from interest rate r to be 0.03.
expression (3⬘). If we can invert the instanta-
neous utility function u(.), they can be written
as:

(5) w ⫽ u ⫺1 冋 u共y⬘兲A共S⬘兲
A共S兲 册
⫺ y⬘, and
3
The closed form expression for w is:

冋 y⬘ 1 ⫺ 1/ ␥
A共S⬘兲
冉 冊
⫹␣ 1⫺
1

冋 册
w⫽
u共y⬘兲A共S⬘兲 1 A共S兲 ␥
⫺1
g⫽u ⫺1.
A共S兲 y
4
⫻ 冉 冊册
A共S⬘兲 ⫺ A共S兲 ␥/共␥ ⫺ 1兲
A共S兲
⫺ y⬘.
The value of consumption is the value of U.S. per
As stressed by Rosen (1988), two dimensions capita income in 1990 in the PWT 6.1 dataset, matching the
of the instantaneous utility function affect the year in which Murphy and Topel (2003) estimate ␧ using
willingness to pay for extensions in life expect- U.S. data. With the calibrated utility parameters, an indi-
ancy. The first is the substitutability of con- vidual with annual income equal to $353 would be indif-
ferent between being alive or dead. The only values of the
sumption in different periods of life, i.e., the GDP per capita variable (adjusted for terms of trade, rgdptt)
inter-temporal elasticity of substitution, and the in the PWT 6.1 dataset below $353 are the ones for the
second is the value of being alive relative to Democratic Republic of Congo between 1994 and 1997.
5
being dead. We calibrate the following func- Notice that the functional form adopted is flexible
tional form for the instantaneous utility function enough to accommodate an income-elasticity of the will-
ingness to pay for changes in life expectancy that actually
to capture these two different dimensions: changes with income. So the calibration using U.S. data is
not limiting in the sense of imposing a willingness to pay
c 1 ⫺ 1/ ␥ that does not belong to the less-developed countries we
(6) u共c兲 ⫽ ⫹␣ want to analyze. If we look at the income-elasticity of the
1 ⫺ 1/ ␥ marginal willingness to pay for life extensions, it varies
from 1.2 for average levels of income per capita (around
$10,000 in 2000), to 1.9 and 3.8 for, respectively, $1,000
The parameter ␣ determines the level of annual and $500 of income per capita. Therefore, the functional
consumption at which the individual would be form adopted is flexible enough to identify underlying pref-
indifferent between being alive or dead, arising erence parameters that, in principle, can be used irrespective
of the income level. Viscusi and Aldy (2003) make an
from the normalization of the utility of death to extensive review of estimates of the “value of a statistical
zero. If the inter-temporal elasticity of substitu- life” around the world. For the developed countries included
tion ␥ is larger than 1, then ␣ is negative. With in their review, our parameterization implies “values of a
282 THE AMERICAN ECONOMIC REVIEW MARCH 2005

TABLE 2—VALUE OF LIFE EXPECTANCY GAINS BY REGION OF THE WORLD AND GROUPS OF COUNTRIES, 1960 –2000

1960 2000 Value of life Yearly growth

exp. gains in Lifetime present rate of full
Life GDP Life GDP annual value of life income
exp. p.c. exp. p.c. income exp. gains (percentage)
Europe & Central Asia 68 6,810 76 18,281 1,809 51,706 2.7
East Asia & Pacific 42 1,317 71 5,866 2,600 60,957 4.8
Latin Am. & the Carib. 56 3,459 70 7,161 1,365 36,935 2.3
Middle East & N. Africa 48 1,935 69 5,525 1,817 46,076 3.4
North America 70 12,380 77 32,880 2,804 81,993 2.7
South Asia 44 892 63 2,346 635 15,504 3.1
Sub-Saharan Africa 41 1,470 46 1,573 72 1,612 0.3
Poorest 50% countries in 1960 41 896 64 3,092 1,456 33,673 4.1
Richest 50% countries in 1960 65 7,195 74 18,162 2,076 58,957 2.6
World 49 2,983 67 7,236 1,627 40,626 2.8

Notes: Income per capita is GDP per capita in 1996 international prices, adjusted for terms of trade (Penn World Tables 6.1).
Life expectancy is life expectancy at birth (World Development Indicators, World Bank). Regional averages weighted by
country population. Sample includes 96 countries, comprising more than 82 percent of the world population. Value of life
expectancy gains based on the authors’ calculations.

II. The Effect of Life Expectancy on World sification, and for the groups of poorest and
Inequality richest countries in 1960 (population-weighted
averages).
We use expression (5) to calculate the value The average value of longevity gains in terms
of the longevity gains observed between 1960 of annual income for the entire sample is
and 2000 and to evaluate the impact of the $1,627. The value is somewhat higher for the
changes in longevity on cross-country inequal- richest countries: $2,076 against $1,456 (in in-
ity. Per capita income figures (adjusted for ternational prices). But the relation between
terms of trade) are taken from the Penn World these values and the initial income is much
Tables version 6.1 (variable rgdptt). In order to higher for poor countries, where it reaches 163
include a sample that is as representative as percent, as opposed to 29 percent for the richest
possible, we use life expectancy at birth num- half of the sample. This tendency is also re-
bers from the World Bank World Development flected in the growth rate of “full income.” In
Indicators and apply the deterministic version this case, since the initial income level is lower
of the methodology discussed in the previous for developing countries, the difference be-
section. tween the richest and poorest countries is re-
Table 2 presents the results for the value of versed: the average yearly growth for the top
longevity gains and the growth rate of “full half of the sample is 2.6 percent, against 4.1per-
income,” together with income and life expect- cent for the bottom half.
ancy statistics, using the value of the parameters The regional profile of the value of longevity
derived in the previous section. The value of changes also reflects this trend. In terms of the
longevity gains is presented in two forms: yearly growth in “full income,” East Asia and
yearly income (w) and total discounted lifetime the Pacific, the Middle East and North Africa,
value (W). Results are presented for the regions and South Asia emerge as the top performers.
of the world according to the World Bank clas- Apart from the well-known development suc-
cess of the Southeast Asian countries, this also
reflects to a great extent the more recent success
of China and India, in both the economic and
statistical life” between $1.5 and $2.5 million. These are health arenas. But perhaps most striking is the
typically in the lower range of estimates discussed in Vis-
cusi and Aldy (2003). If anything, our parameterization will dismal performance of sub-Saharan Africa,
tend to underestimate the value of reductions in mortality which displays the lowest growth rate of “full-
rates. income” in the sample. As we discussed before,
VOL. 95 NO. 1 BECKER ET AL.: QUANTITY AND QUALITY OF LIFE AND WORLD INEQUALITY 283

TABLE 3—EVOLUTION OF CROSS-COUNTRY INEQUALITY IN FULL INCOME, 1960 –2000

Income per capita Full income

1960 1990 2000 1990 2000
Relative mean dev. 0.4751 0.4733 0.4215 0.4397 0.3760
Coeff. of variation 1.2344 1.2529 1.1662 1.1664 1.0463
Std. dev. of logs 1.0178 1.0300 0.9620 0.9758 0.9476
Gini coeff. 0.5104 0.5187 0.4873 0.4935 0.4561
Regression to the mean ⫺0.0069 ⫺0.1338 ⫺0.1006 ⫺0.2638
over 1960 (p-value ⫽ 0.86) (p-value ⫽ 0.01) (p-value ⫽ 0.02) (p-value ⫽ 0.00)

Notes: Income per capita is GDP per capita in 1996 international prices, adjusted for terms of trade (Penn World Tables 6.1).
Full income calculated by the authors with 1960 as base year, incorporating gains in life expectancy at birth (World
Development Indicators, World Bank). Inequality measures weighted by country population (abstracting from within country
inequality). Sample includes 96 countries, comprising more than 82 percent of the world population. Regression to the mean
is the coefficient of a regression of the change in the natural log of income over the period on its initial level (weighted
regressions).

this is partly due to the reversal of the gains in As long as the income elasticity of the value of
life expectancy that was brought about by the life is not much below unity, using some mea-
outbreak of the AIDS epidemic in the 1990s. sure of “full income” in this regression would
Nevertheless, the size of the sub-Saharan pop- unambiguously increase the degree of conver-
ulation is relatively small when compared to the gence, since richer countries in 1960 also had
other developing regions, and the net effect of higher life expectancy.
health is to reduce overall inequality. These results indicate reduction in welfare
This indicates that, unlike income changes, inequality throughout most of the postwar pe-
longevity changes since 1960 reduced the dis- riod, in the sense that countries with higher
parity in welfare across countries. Table 3 ex- initial income tended to have significantly lower
plores this point further by repeating the same subsequent welfare gains. Incomes 100 percent
income dispersion measures presented in table higher in 1960 were associated, on average,
1, and by additionally calculating the same sta- with “full-income” growth rates 10 percent
tistics for “full income” in 1990 and 2000 (tak- lower in the following 30 years, and 26 percent
ing 1960 as the base period). As the table lower in the following 40 years. This result is
shows, by any statistic, the inclusion of life not surprising, given the negative correlation
expectancy in the measure of “full income” between life expectancy gains and income. As
generates significant reductions in inequality long as the income elasticity of value of life is
between 1960 and 1990, and also a significant not much above unity, any value attached to
increase in the rate of reduction in inequality longevity would work toward increasing con-
between 1960 and 2000. The coefficient on vergence. W. Kip Viscusi and Joseph E. Aldy
ln (income in 1960) in the “full-income” regres- (2003) conclude, from various types of evi-
sion to the mean equations is negative and sta- dence, that this elasticity is less than unity, but
tistically significant. Higher income in 1960 is their results for countries are greatly affected by
consistently associated with lower growth in a couple of extreme observations for India.
“full income” in the 30-year period between Without these observations, Becker and Julio J.
1960 and 1990 and in the 40-year period be- Elias (2003) get an elasticity of about unity.
tween 1960 and 2000. The evidence presented here also indicates
The ideal independent variable in the right- that the relative importance of health improve-
hand side of this regression should be a measure ments, when compared to income gains, was
of “full income in 1960.” Since the approach systematically higher for the developing world.
discussed in Section I does not allow us to The share of the welfare improvements ob-
calculate the value of given levels of life ex- served between 1960 and 2000 due to mortality
pectancy, but only the value of changes in life reductions— calculated as value of longevity
expectancy, we are forced to use the 1960 value gains in annual income/(value of longevity
of income per capita rather than “full income.” gains in annual income ⫹ increase in annual
284 THE AMERICAN ECONOMIC REVIEW MARCH 2005

income between 1960 and 2000)— has an aver- technological improvements (see Samuel H.
age of 28 percent for the entire world. But this Preston, 1980; Soares, 2005). Stable concave
number is above 30 percent for East Asia and returns to investments in health cannot account
the Pacific, the Middle East and North Africa, for the changing cross-sectional relationship be-
and South Asia, and below 14 percent for Eu- tween income and life expectancy. Moreover,
rope and Central Asia and for North America. since investments in health are much larger
Overall, the evidence shows that longevity for developed than for developing countries—
changes in the period between 1960 and 2000 measured either in absolute terms or as shares of
worked toward reducing the disparity in welfare income—a stable health production function
across countries. The actual reduction in dispar- could not explain the convergence in life ex-
ity depends on the specific values of the param- pectancy, unless returns to investments in health
eters ␣ and ␥; that is, on the relative importance were implausibly higher for the less-developed
of quantity and quality of life. Nevertheless, the world.
qualitative role played by mortality reductions,
and the fact that their impact on welfare was A. Data
quantitatively important, should be obvious.
These results would be even stronger if we To understand the nature of the changes in
accounted for expenditures on health and R&D, mortality in the developing world, we decom-
because part of the gains in life expectancy is pose the gains in life expectancy into different
driven by these expenditures. Since most of age groups and causes of death. The World
these expenditures are undertaken by the devel- Health Organization Mortality Database con-
oped world, the share of truly exogenous reduc- tains number of deaths by age and cause of
tions in mortality is certainly higher for the death for the years between 1960 and 2000.
less-developed countries.6 Therefore, conver- Causes of death in the different years are clas-
gence in welfare would be higher if the endog- sified according to the current International
enous part of longevity gains were netted out. Classification of Diseases (ICD) codes, so data
for different periods have to be made compati-
III. The Causes of Changes in Mortality ble by matching codes of the different versions
Inequality of the ICD. As we will be dealing with rather
broad groups of causes of death, this will not be
Cross-country life expectancy convergence a problem.
would follow if countries shared the same con- We define the following 13 groups of causes
cave health production technology, as the inter- of death: R01: infectious diseases; R02: neo-
national evidence suggests, and spent the same plasms; R03: endocrine, metabolic, and blood
on health. Countries with higher initial mortal- diseases, and nutritional deficiencies; R04:
ity would then have larger mortality reductions mental disorders; R05: diseases of the nervous
because they have much higher returns on in- system and senses organs; R06: heart and cir-
vestments in health than do countries with lower culatory diseases; R07: respiratory and diges-
mortality. tive diseases; R08: urinary and genital diseases;
The shift in the income–life expectancy pro- R09: abortion and obstetric causes; R10: skin
file noticed elsewhere suggests, however, that and musculoskeletal diseases; R11: congenital
this is not the whole story, and that a consider- anomalies and perinatal period conditions; R12:
able part of the changes in longevity is related to ill-defined conditions; and R13: accidents, sui-
cides, and homicides. The grouping of the codes
6
from the ICD-6/7 and ICD-9 into these 13 cat-
For example: in 2000, health expenditures per capita in egories is described in the Appendix.
the richest OECD countries were typically above $2,000,
reaching $4,252 for the United States; in the same year, Mortality data by age groups and causes of
these expenditures were below $300 for countries like death for earlier years are available only for a
China, India, Mexico, Poland, and Turkey. In terms of restricted sample of countries. In order to in-
shares of GDP, these values corresponded to around 10 crease cross-country comparability in our data-
percent for the richest countries, as opposed to 5 percent for
the developing countries cited above (data from the World set, we use the mortality data as ten-year
Bank World Development Indicators). Expenditures on averages centered in the reference years: 1965
R&D are usually even more concentrated than that. corresponds to the average for the period be-
VOL. 95 NO. 1 BECKER ET AL.: QUANTITY AND QUALITY OF LIFE AND WORLD INEQUALITY 285

tween 1960 and 1969, and 1995 corresponds to B. Age- and Cause-Specific Changes in Life
the average between 1990 and 1999 (or years Expectancy
available in these intervals). Even after this, the
sample includes only 49 countries.7 The most To consider age-specific changes in longev-
critical problem with this dataset is that it con- ity, define the survival function conditional on
tains only one sub-Saharan African country age a according to S(t,a)⫽S(t)/S(a), for t ⱖ a.
(Mauritius). So we cannot possibly expect this Any change in the survival function from S to S⬘
exercise to reveal the effects of AIDS mortality can be decomposed into changes attributable to
in the later part of the twentieth century. But, as different age groups. Without loss of generality,
the previous discussion pointed out, the behav- consider the age group between ages a and a ⫹
ior of life expectancy in sub-Saharan Africa i, where a ⱕ a ⫹ i ⱕ t. The survival function
after 1990 immediately reveals the overwhelm- that would be observed if only changes in mor-
ing effects of AIDS on the recent evolution of tality between ages a and a ⫹ i had taken place
mortality across countries. So the real puzzle would be S⬘a ⫽ S(t,a ⫹ i)S⬘(a ⫹ i,a)S(a). This
lies in the reduction on life expectancy inequal- counterfactual survival function gives the prob-
ity prior to 1990, not in the effects of AIDS after ability of survival up to age t according to the
that. probabilities between ages a and a ⫹ i observed
In this respect, apart from the absence of in the second period, and the probabilities in
sub-Saharan Africa, there does not seem to be other age groups observed in the first period.
any specific bias induced by the reduced sam- Now consider cause-specific changes in mor-
ple. In fact, the experience of the countries tality. Let there be K competing independent
included in the sample seems to be quite typical causes of mortality, inducing the overall sur-
of the cross-country changes in life expectancy vival function S ⫽ 兿kk ⫽ 1 Sk, where Sk denotes
observed prior to the arrival of AIDS. The co- the survival function of cause k. Define the
efficient of regression to the mean in life ex- counterfactual survival function S⬘k ⫽ S⬘k 兿i ⫽ k
pectancy between 1965 and 1995 in this sample Si. Similar to the case of age-specific mortality
is equal to – 0.55, as compared to a coefficient changes, this expression gives the survival func-
equal to – 0.61 observed between 1960 and tion that would be observed if only changes in
1990 in the larger sample used in Section I. mortality from cause of death k had taken place.
In any case, this is a limitation imposed by Both these decompositions can be applied to
data availability, and we can do nothing about any given survival function, so that applying
it. When interpreting the results related to age- them sequentially one obtains a counterfactual
and cause-specific mortality, it is important to survival function S⬘k,a, which simulates the sur-
keep in mind that they are probably an accurate vival function that would be observed if only
description of the pre-AIDS experience of tech- changes in mortality by one specific cause of
nological diffusion from developed to develop- death (k) and in one specific age group (between
ing countries, but not of the recent experiences ages a and a ⫹ i) had actually taken place.
of sub-Saharan Africa and of the extremely poor With the age- and cause-specific survival
countries of the world. functions S⬘k,a, we can immediately construct
corresponding counterfactual measures of life
expectancy, each one defined as L⬘k,a ⫽ 兰⬁ 0 S⬘k,a(t)
dt. L⬘k,a is the exact analog of S⬘k,a in terms of
7
The countries included in this sample are the following: life expectancy. For our purposes, it gives the
Argentina; Australia; Austria; Barbados; Belgium; Belize; life expectancy that would be observed in 1995
Bulgaria; Canada; Chile; Colombia; Costa Rica; Cuba; if only mortality rates between ages a and a ⫹
Former Czechoslovakia; Ecuador; Egypt; El Salvador; Fin-
land; France; Former Federal Republic of Germany;
i, and due to the kth cause of death, had actually
Greece; Hong Kong; Hungary; Iceland; Ireland; Italy; Ja- changed between 1965 and 1995.
pan; Luxembourg; Malta; Mauritius; Mexico; Netherlands; Now consider three age groups— between
New Zealand; Norway; Philippines; Poland; Portugal; ages 0 and 19, 20 and 49, and above 49 —and
Puerto Rico; Romania; Singapore; Spain; Sweden; Trinidad the 13 causes of death defined above. This strat-
and Tobago; United Kingdom, England and Wales; United
Kingdom, Northern Ireland; United Kingdom, Scotland; egy allows the decomposition of the gains in life
United States of America; Uruguay; Venezuela; Former expectancy observed in the period into changes
Yugoslavia. attributable to each age group and cause of
286 THE AMERICAN ECONOMIC REVIEW MARCH 2005

death, plus a higher-order term.8 Let L␶ denote is given by a linear regression of ⌬L on a

the vector of life expectancy at birth for differ- constant plus L65. Define X65 ⫽ [1 L65], a
ent countries in year ␶, and index the age groups matrix containing a column of ones, and a col-
by the initial age. Then umn with the life expectancy at birth for the
different countries in 1965. The convergence
coefficient is given by the second term in ␤ ⫽
冘 冘 ⌬L
13

(7) ⌬L ⫽ k,a ⫹ ⌬LH (X65⬘X65)⫺1X65⬘⌬L. By substituting ⌬L from

a ⫽ 兵0,20,50其 k ⫽ 1 expression (7), we can write

where ⌬L is the change in life expectancy ob- ␤ ⫽ 共X65⬘X65兲⫺1X65⬘

served between 1965 and 1995. ⌬Lk,a, for k ⫽

冋 册
1,. . . ,13, and a ⫽ 0, 20, 50, is the change in life
冘 冘 ⌬L
13
expectancy attributable to the kth cause of death
and to the age group starting at age a, defined as ⫻ k,a ⫹ ⌬LH .
a ⫽ 兵0,20,50其 k ⫽ 1
⌬Lk,a ⫽ Lk,a 95
–L65, and ⌬LH is the change in
life expectancy due to the interaction among
mortality changes in the different groups of ages This expression gives a natural decomposition
and causes of death (higher-order terms). for the convergence coefficient:
With this decomposition, we can see what
types of mortality were responsible for the gains
冘 冘␤
13
in life expectancy observed in the last few de-
(8) ␤⫽ k,a ⫹ ␤H
cades. In addition, we can examine the reasons
a ⫽ 兵0,20,50其 k ⫽ 1
behind the differential behavior of life expect-
ancy across developing and developed coun-
tries, and shed some light on the reasons behind where ␤k,a is the vector of coefficients of the
the reduction in cross-country health inequality OLS regression of ⌬Lk,a on X65.
observed up to the beginning of the 1990s. In words, the coefficient of the regression of
In this respect, we concentrate the analysis on changes in life expectancy on initial life expect-
the regression to the mean in life expectancy. ancy levels can be decomposed into coefficients
The approach outlined above gives a direct de- of regressions of cause- and age-specific
composition of the overall regression to the changes in life expectancy on initial life expect-
mean in life expectancy into the regression to ancy levels, plus a residual term (␤H). That is,
the mean attributable to each cause and age. By regression to the mean in life expectancy is
definition, the regression to the mean coefficient decomposed into regression to the mean attrib-
utable to the 13 underlying causes of death and
three age groups, plus a residual term. This
8 allows us to evaluate the role of different causes
Given that changes in mortality from different causes
and ages interact with each other in generating the final of death in generating the observed reduction in
survival function, this decomposition does not explain ex- life expectancy inequality, and also to analyze
actly 100 percent of the shift in this function when infra- in what ages this reduction in inequality was
marginal changes in mortality are being considered. (This is concentrated.
the competing risks nature of mortality rates, as discussed
by William H. Dow et al., 1999.) Formally, this is a first-
order decomposition of changes in the survival function. For
marginal changes in S, it would indeed generate an exact C. Results from the Decomposition of Life
decomposition. There are decomposition strategies in the Expectancy Changes
demographic literature that explain 100 percent of the
changes in life expectancy, but they rely on arbitrarily
assigning changes in life expectancy resulting from the Age- and cause-specific survival rates are
interaction between different causes or ages to one specific constructed using death and population data
component (see Preston et al., 2001). With infra-marginal from the World Health Organization Mortality
changes, the interaction among higher-order terms is rele- Database. Mortality rates are assumed to be
vant, and it is impossible to attribute their effects to any
particular cause or age group. In any case, the decomposi- constant inside the age intervals for which data
tion suggested here accounts for more than 80 percent of the are tabulated. Table 4 presents the life expect-
changes in life expectancy in our sample. ancy changes that can be attributed to each
VOL. 95 NO. 1 BECKER ET AL.: QUANTITY AND QUALITY OF LIFE AND WORLD INEQUALITY 287

TABLE 4—DECOMPOSITION OF LIFE EXPECTANCY GAINS BY CAUSE OF DEATH AND AGE GROUP, REGIONS OF THE WORLD,
1965–1995

Europe & E. Asia & Latin Am. & Middle East North Whole
Change in life expectancy Cent. Asia Pacific the Carib. & N. Africa America sample
Total 8.6 5.4 10.2 18.0 5.8 7.1
By cause of death:
R01: Infectious 0.8 0.4 1.1 ⫺0.1 0.0 0.4
R02: Neoplasms 0.2 0.1 0.1 0.0 0.1 0.1
R03: Endocrine, metabolic, and blood 0.3 0.0 0.0 0.4 ⫺0.2 0.0
diseases, nutritional deficiencies
R04: Mental disorders 0.0 0.0 0.0 0.0 0.0 0.0
R05: Nervous system and senses organs 2.3 1.4 0.9 0.1 1.2 1.4
R06: Heart and circulatory ⫺0.2 0.0 ⫺0.4 ⫺1.6 2.3 0.4
R07: Respiratory and digestive 1.3 1.1 3.5 12.1 0.5 1.6
R08: Urinary and genital 0.2 0.2 0.1 0.0 0.1 0.1
R09: Abortion and obstetric causes 0.1 0.0 0.1 0.1 0.0 0.0
R10: Skin and musculoskeletal 0.0 0.0 0.0 0.1 0.0 0.0
R11: Congenital anomalies and perinatal 0.7 1.0 1.1 2.6 1.0 1.0
period conditions
R12: Ill-defined 1.2 0.5 2.8 2.9 0.0 0.9
R13: Accidents, suicides, and homicides 0.5 0.3 ⫺0.1 0.4 0.4 0.3
By age group:
Between 0 and 19 2.5 2.2 5.9 16.2 1.5 3.0
Between 20 and 49 1.2 0.4 1.4 0.8 0.6 0.7
50 and above 4.6 2.7 2.3 0.5 3.5 3.1

Notes: Decomposition of life expectancy calculated by the authors based on age- and cause-specific mortality data from the
World Health Organization. Regional averages weighted by country population. Sample includes 49 countries. The total life
expectancy change for East Asia & Pacific is very different from the one presented in Table 2 because the sample used here
excludes some of the main beneficiaries of the life expectancy gains in the region, such as China, Indonesia, Korea, and
Thailand.

cause of death and each age group, by regions rope, and Asia. At the same time, there are
of the world and for the whole sample. certain diseases that had a relatively small im-
The table shows that, overall, the most im- pact on the overall change in life expectancy,
portant changes in life expectancy came from but were very important in one particular
diseases of the nervous system and senses or- region. This is the case of heart and circula-
gans, respiratory and digestive conditions, con- tory diseases, which had a relatively small
genital anomalies, perinatal period conditions, overall impact, but were the most important
and ill-defined causes. Also, most of these re- factors determining changes in life expect-
ductions in mortality were concentrated before ancy in North America. This is also the case
age 20 or after age 50. of infectious diseases, which had a significant
But the composition of these mortality impact in life expectancy in Latin America
changes in terms of causes of death and age and the Caribbean.
groups was very different across the different These differences are also clear in the age
regions. Mortality reductions in nervous system profile of life expectancy gains across the dif-
and senses organs diseases were very important ferent regions. The most expressive gains in the
in Europe and Central Asia, but almost irrele- age group under 20 are observed in the Middle
vant in the Middle East and North Africa. On East and North Africa, and in Latin America
the other hand, respiratory and digestive condi- and the Caribbean, while most of the gains
tions, together with ill-defined causes, were the above 50 are observed in Europe and Central
main factors in determining life expectancy Asia, and in North America.
gains for the Middle East and North Africa and In order to further understand the differential
for Latin America and the Caribbean, but were impact of mortality by different causes and
much less important for North America, Eu- age groups on the evolution of cross-country
288 THE AMERICAN ECONOMIC REVIEW MARCH 2005

TABLE 5—CONTRIBUTION OF AGE- AND CAUSE-SPECIFIC MORTALITY CHANGES TO

REGRESSION TO THE MEAN IN LIFE EXPECTANCY, 1965–1995
(Percentage)

Cause of death/age group 0–20 20–50 Above 50 All ages

R01: Infectious 3.7* 3.0* 1.0 7.8*
R02: Neoplasms ⫺0.7* ⫺0.7 ⫺0.4 ⫺1.8
R03: Endocrine, metabolic, and blood diseases, 1.3* 1.6* ⫺0.5 2.4
nutritional deficiencies
R04: Mental disorders ⫺0.1* 0.2 0.2 0.3
R05: Nervous system and senses organs 0.0 ⫺0.2 ⫺11.4* ⫺11.5*
R06: Heart and circulatory ⫺0.8* ⫺2.0* ⫺20.0* ⫺22.7*
R07: Respiratory and digestive 77.8* 3.0* ⫺0.5 80.7*
R08: Urinary and genital 0.3* ⫺0.4* ⫺1.3* ⫺1.3*
R09: Abortion and obstetric causes 0.1* 1.1* 0.0* 1.2*
R10: Skin and musculoskeletal 0.1* 0.1* 0.3* 0.5*
R11: Congenital anomalies and perinatal 10.5* ⫺0.1* 0.0* 10.4*
period conditions
R12: Ill-defined 8.5* 4.1* 20.4* 33.7*
R13: Accidents, suicides, and homicides ⫺0.7 ⫺1.2 ⫺1.6* ⫺3.4*
All causes 109.6* 8.9* ⫺23.7* 100.0*

Notes: Calculations based on coefficients from (population-weighted) regressions of the

changes in life expectancy attributable to each specific cause of death and age group on the
life expectancy at birth in 1965. * denotes statistical significance at 5 percent of the
coefficients in these regressions. Decomposition of the life expectancy changes based on the
authors’ calculations using World Health Organization data (49 countries).

inequalities in health, we apply the decomposi- tually no impact on overall health inequality,
tion strategy described in the previous section to but two played a considerable role in increasing
the regression to the mean in life expectancy. inequality: mortality by nervous system and
As mentioned before, regression to the mean in senses organ diseases, and heart and circulatory
life expectancy can be decomposed into regres- diseases reduced convergence by more than 34
sion to the mean in changes in life expectancy percent of its actual value. In the case of ner-
attributable to each cause of death and age vous system and senses organs diseases, mor-
group. The coefficient of regression to the mean tality reductions were experienced by both
in this restricted sample is equal to – 0.55 (sta- developed and developing countries, but the
tistically significant at any standard significance extent of these reductions was considerably
level). We run 56 regressions of age- and cause- larger for developed countries. In terms of heart
specific changes in life expectancy (13 causes of and circulatory diseases, mortality reductions
death, three age groups, plus all ages and all were considerable for North America, but basi-
causes of death together) on the initial life ex- cally nonexistent— or even negative—for the
pectancy level (each one giving one of the ␤k,a rest of the world.
coefficients from the previous section). We then In relation to the causes of death that worked
calculate the contribution of the specific age toward reducing health inequality, the action is
group and cause of death to the overall regres- concentrated in a handful of cases: infectious,
sion to the mean in life expectancy (ignoring the respiratory, and digestive diseases, congenital
constant, ␤k,a/␤). The results are presented in anomalies, perinatal period conditions, and “ill-
Table 5. defined” conditions accounted for roughly 133
Of the 13 causes of death, five contributed to percent of the observed regression to the mean.
increased dispersion in life expectancy across Among these, respiratory and digestive diseases
countries, meaning that the behavior of mortal- were by far the most important, accounting for
ity due to these causes worked against regres- 81 percent of the regression to the mean. Note
sion to the mean in life expectancy. Most of that this group also includes infectious diseases
these five “divergent” causes of death had vir- related to the respiratory tract, such as pneumo-
VOL. 95 NO. 1 BECKER ET AL.: QUANTITY AND QUALITY OF LIFE AND WORLD INEQUALITY 289

nia and influenza, and digestive tract diseases diseases). The concept is of a developed center
such as appendicitis and cirrhosis. The second that generates health and medical knowledge to
most important contribution to convergence be absorbed eventually by the underdeveloped
comes from “ill-defined” causes and conditions. periphery.
This most likely reflects the relative improve-
ment of medical practice and record keeping IV. Concluding Remarks
behavior in developing countries.9
The age-specific pattern of these causes of This paper shows that life expectancy gains
death is also obvious from the decomposition. in the 40 years between 1960 and 2000 have
Almost all of the inequality enhancing effect of been an important component of improvements
nervous system and senses organs diseases, and in welfare throughout the world. We estimate
heart and circulatory diseases, took place via the value of the gains in health during this
reductions in mortality after age 50. At the same period to be of the same order of magnitude as
time, the reduction in inequality via respiratory the gains in income and, for the poorest half of
and digestive diseases was almost entirely con- the world, to represent 40 percent of the overall
centrated before age 20, with some effects still welfare gains. The effects of health are suffi-
being felt between ages 20 and 50; infectious cient to revert the results regarding the evolu-
diseases had similar effects across these two tion of cross-country inequality up to the 1990s.
age groups. Overall, reductions in mortality Once health is accounted for, there is a signif-
up to age 20 were responsible for most of the icant reduction in inequality throughout the
regression to the mean in life expectancy ob- world up to 1990 and, even with the AIDS
served in the period, with some additional epidemic, a much more significant reduction in
contribution from changes in life expectancy inequality between 1960 and 2000 than can be
between ages 20 and 50. In contrast, changes perceived from income alone.
in mortality after age 50 significantly contrib- The decline in life expectancy inequality can
uted to increase health inequality across be attributed, to a great extent, to reductions in
countries. mortality due to infectious, respiratory, and di-
These results support the view that recent gestive diseases in developing countries. Nev-
reductions in mortality in the developing world ertheless, some causes of death have actually
have been due in part to the absorption of pre- contributed toward increased health inequality.
viously available technologies (for arguments in This is obviously true for AIDS, but it is also
this direction, see Preston, 1980; Soares, 2005). true for the cases of nervous system, senses
The group of infectious, respiratory, and diges- organs, heart, and circulatory diseases, for
tive diseases, congenital anomalies, and perina- which developed countries took advantage of
tal period conditions includes the types of recent advances on the frontier of medical tech-
diseases for which educational health programs nology. The ongoing AIDS epidemic in Africa
and simple interventions can have large benefi- and rising infection rates in Asia, coupled with
cial effects. On the other side of the spectrum, recent advances in medical technology that are
developed countries benefited relatively more unlikely soon to become available in the devel-
from reductions in mortality that required new oping world, raise the possibility that the post-
technological developments, relatively costly war trends in health inequality may be reversed
change of habits, and expensive surgical inter- in the near future.
ventions (heart, circulatory, and nervous system
APPENDIX: CLASSIFICATION OF ICD CODES INTO
CAUSE OF DEATH GROUPS
9
The fact that “ill-defined” conditions were relatively
more common in developing countries in 1965 tends to R01: infectious diseases: icd-6/7 a: a001-
underestimate the actual convergence in the other causes of a043; icd-6/7 b: b001-b017; icd-9: b01-b07.
death. This is so because a larger share of the reduction in R02: neoplasms: icd-6/7 a: a044-a060; icd-6/7
mortality in developing countries is being attributed to b: b018-b019; icd-9: b08-b17. R03: endocrine,
“ill-defined” causes and conditions. Which causes of death
experience the biggest underestimation depends on the cor- metabolic and blood diseases, nutritional defi-
relation between cause of death and misreporting (“ill- ciencies: icd-6/7 a: a061-a066; icd-6/7 b: b020-
defined”). We do not deal with this problem. b021; icd-9: b18-b20. R04: mental disorders:
290 THE AMERICAN ECONOMIC REVIEW MARCH 2005

icd-6/7 a: a067-a069; icd-9: b21. R05: diseases ties under Competing Risks.” American
of the nervous system and senses organs: icd- Economic Review, 1999, 89(5), pp. 1358 –71.
6/7 a: a070-a078; icd-6/7 b: b022-b023; icd-9: Friedman, Milton. “Do Old Fallacies Ever Die?”
b22-b24. R06: heart and circulatory diseases: Journal of Economic Literature, 1992, 30(4),
icd-6/7 a: a079-a086; icd-6/7 b: b024-b029; pp. 2129 –32.
icd-9: b25-b30. R07: respiratory and digestive Goesling, Brian and Firebaugh, Glenn. “The
diseases: icd-6/7 a: a087-a107; icd-6/7 b: b030- Trend in International Health Inequality.”
b037; icd-9: b31-b34. R08: urinary and genital Population and Development Review, 2004,
diseases: icd-6/7 a: a108-a114; icd-6/7 b: b038- 30(1), pp. 131– 46.
b039; icd-9: b35-b37. R09: abortion and obstet- Mankiw, N. Gregory; Romer, David and Weil,
ric causes: icd-6/7 a: a115-a120; icd-6/7 b: David N. “A Contribution to the Empirics of
b040; icd-9: b38-b41. R10: skin and musculo- Economic Growth.” Quarterly Journal of
skeletal diseases: icd-6/7 a: a121-a126; icd-9: Economics, 1992, 107(2), pp. 407–37.
b42-b43. R11: congenital anomalies and peri- Murphy, Kevin M. and Topel, Robert H. “The
natal period conditions: icd-6/7 a: a127-a135; Economic Value of Medical Research,” in
icd-6/7 b: b041-b044; icd-9: b44-b45. R12: ill- Kevin M. Murphy and Robert H. Topel, eds.,
defined: icd-6/7 a: a136-a137; icd-6/7 b: b045- Measuring the gains from medical research:
b046; icd-9: b46. R13: accidents, suicides, and An economic approach. Chicago: University
homicides: icd-6/7 a: a138-a150; icd-6/7 b: of Chicago Press, 2003.
b047-b050; icd-9: b47-b56. Neumayer, Eric. “Beyond Income: Convergence
in Living Standards, Big Time.” Structural
REFERENCES Change and Economic Dynamics, 2003,
14(3), pp. 275–96.
Barro, Robert J. and Sala-i-Martin, Xavier. Eco- Nordhaus, William D. “The Health of Nations:
nomic growth. Advanced Series in Econom- The Contribution of Improved Health to Liv-
ics. New York: McGraw-Hill, 1995. ing Standards,” in Kevin M. Murphy and
Becker, Gary S. and Elias, Julio J. “Introducing Robert H. Topel, eds., Measuring the gains
Incentives in the Market for Live and Cadav- from medical research. Chicago: University
eric Organs.” Unpublished Paper, 2003. of Chicago Press, 1999.
Bourguignon, Francois and Morrison, Christian. Parente, Stephen L. and Prescott, Edward C.
“Inequality among World Citizens: 1820 – “Changes in the Wealth of Nations.” Federal
1992.” American Economic Review, 2002, Reserve Bank of Minneapolis Quarterly Re-
92(4), pp. 727– 44. view, 1993, 17(2), pp. 3–16.
Browning, Martin; Hansen, Lars Peter and Heck- Philipson, Tomas J. and Soares, Rodrigo R.
man, James J. “Micro Data and General Equi- “World Inequality and the Rise in Longev-
librium Models,” in John B. Taylor and ity,” in Boris Pleskovic and Nicholas Stern,
Michael Woodford, eds., Handbook of mac- eds., Annual World Bank Conference on De-
roeconomics, Vol. 1A. Amsterdam: Elsevier velopment Economics 2001/2002. Washing-
North-Holland, 1999, pp. 543– 633. ton, DC: World Bank, 2002, pp. 245–59.
Burström, Kristina; Johannesson, Magnus and Preston, Samuel H. “Causes and Consequences
Diderichsen, Finn. “The Value of the Change of Mortality Declines in Less Developed
in Health in Sweden 1980/81 to 1996/97.” Countries During the Twentieth Century,” in
Health Economics, 2003, 12(8), pp. 637–54. Richard S. Easterlin, ed., Population and
Cutler, David M. and Richardson, Elizabeth. economic change in developing countries.
“Measuring the Health of the U.S. Popula- Chicago: University of Chicago Press, 1998,
tion.” Brookings Papers on Economic Activ- pp. 289 –341.
ity, 1997, 0(0), pp. 217–71. Preston, Samuel H.; Heuveline, Patrick and Guil-
de la Fuente, Angel. “The Empirics of Growth lot, Michel. Demography–Measuring and
and Convergence: A Selective Review.” modeling population processes. Oxford:
Journal of Economic Dynamics and Control, Blackwell Publishing, Ltd., 2001.
1997, 21(1), pp. 23–73. Quah, Danny T. “Empirics for Economic
Dow, William H.; Philipson, Tomas J. and Sala-i- Growth and Convergence.” European Eco-
Martin, Xavier. “Longevity Complementari- nomic Review, 1996, 40(6), pp. 1353–75.
VOL. 95 NO. 1 BECKER ET AL.: QUANTITY AND QUALITY OF LIFE AND WORLD INEQUALITY 291

Rosen, Sherwin. “The Value of Changes in Life pectancy,” in Milton Moss, ed., The measure-
Expectancy.” Journal of Risk and Uncer- ment of economic and social performance,
tainty, 1988, 1(3), pp. 285–304. Studies in Income and Wealth, Vol. 38.
Sala-i-Martin, Xavier. “The World Distribution NBER Conference of Research in Income
of Income (Estimated from Individual Coun- and Wealth. New York: Columbia University
try Distributions),” in National Bureau of Press, 1973, pp. 193–225.
Economic Research, Inc., NBER Working Viscusi, W. Kip. “The Value of Risks to Life and
Papers: No. 8933, 2002. Health.” Journal of Economic Literature,
Sap, Randa and Smith, Stephen C. “Human Cap- 1993, 31(4), pp. 1912– 46.
ital Convergence: A Joint Estimation Ap- Viscusi, W. Kip and Aldy, Joseph E. “The Value
proach.” IMF Staff Papers, 2002, 49(2), pp. of a Statistical Life: A Critical Review of
200 –11. Market Estimates throughout the World,”
Soares, Rodrigo R. “Mortality Reductions, Edu- in National Bureau of Economic Research,
cational Attainment, and Fertility Choice.” Inc., NBER Working Papers: No. 9487,
American Economic Review (forthcoming). 2003.
United Nations Development Programme. Human Younger, Stephen D. “Cross-Country Determi-
development report 2002, Oxford: Oxford nants of Declines in Infant Mortality: A
University Press, 2002. Growth Regression Approach.” Cornell Uni-
Usher, Dan. “An Imputation to the Measure of versity, Cornell Food and Nutrition Policy
Economic Growth for Changes in Life Ex- Program: Working Paper No. 130, 2001.