öMmföäflsäafaäsflassflassflas
ffffffffffffffffffffffffffffffffffff
Discussion Papers
Longevity Clubs
Ulla Lehmijoki
University of Helsinki and HECER
Discussion Paper No. 234
September 2008
ISSN 1795 0562
HECER – Helsinki Center of Economic Research, P.O. Box 17 (Arkadiankatu 7), FI 00014
University of Helsinki, FINLAND, Tel +358 9 191 28780, Fax +358 9 191 28781,
E mail info hecer@helsinki.fi, Internet www.hecer.fi
HECER
Discussion Paper No. 234
Longevity Clubs*
Abstract
This paper discusses the post war convergence of life expectancy in longevity clubs
derived from a global sample of countries by the regression tree method. Three clubs are
constructed by their literacy rates, whereas low income is typical of the fourth. The
convergence of life expectancy is analyzed both within and between the clubs. The results
show that because deficient literacy prevents diffusion of information, the average life 
expectancy growth in the club with the lowest literacy was sluggish relative to initial life 
expectancy. Furthermore, no convergence appeared within this club. By contrast, the club
with medium literacy proceeded steadily and exhibited convergence, presumably due to
externalities arising once a critical literacy level was reached. The slow progress in the
club with the highest literacy seems to be due the complexity of health information and
privatization of health. No convergence appeared within this club either. It also turns out
that, since AIDS is closely associated with poverty, most life expectancy laggards are
members of the low income club.
JEL Classification: J11, J24, I18.
Keywords: Life expectancy convergence, regression tree, longevity clubs.
Ulla Lehmijoki
Department of Economics
University of Helsinki
P.O. Box 17 (Arkadiankatu 7)
FI 00014 University of Helsinki
FINLAND
e mail: ulla.lehmijoki@helsinki.fi
* The author is grateful for the funding from the Yrjö Jahnsson Foundation.
1 Introduction
Since World War II there has been a remarkable growth in life-expectancy
worldwide. Given that this growth has typically been most rapid where the
initialvalueoflife-expectancywaslow, globalconvergenceoflife-expectancies
has taken place. This convergence, however, has not been unequivocal as life-
expectancy has increased only slowly in some countries and even decreased
in others.
Contradictoryflndingshavesubjectedlife-expectancyconvergencetoheated
debate currently. The optimists claim that most low-longevity countries ul-
timately catch up with the high-longevity ones (Ram 1998, Neumayer 2003,
Becker et al 2005, Riley 2005). The main argument is that the former can
adopt the latest technologies relatively easily to reduce infant mortality and
infectious diseases, whereas the advanced countries can derive progress only
from the slow flght against cardiovascular diseases and aging. Hence, coun-
tries with the greatest gaps to health leaders have the largest potential for
growth (Omran 1971, Deaton 2003, Vallin and Mesl¶e 2004, Edwards and
Tuljapurkar 2005, Cutler et al. 2006, Soares 2007). This optimism, however,
may be premature as convergence has stagnated in some cases. The so-called
club approach has thus gained acceptance (Mayer-Foulkes 2003). This ap-
proach assumes that countries can be classifled into longevity clubs with a
unique convergence rate in each. Rather than investigating convergence in a
broad sample of countries, it would thus be more appropriate to investigate
it in sub-samples, i.e., in clubs.
There are several reasons for the appearance of such clubs. Global in-
comes tend to diverge rather than converge as some countries are lagging
behind the generally rising trend (Pritchett 1997). Given that income is an
important determinant of life-expectancy (Preston 1975, Fogel 1994, 2004),
one expects divergence in life-expectancy as well. The difiusion of health
technology is not straightforward, either. The reason is that human capi-
tal, which is essential in absorbing of information, has been lowest in low
life-expectancy countries, thus making the realization of their high growth
potential cumbersome. Given the strong intergenerational heritage in hu-
man capital, the families and countries with low human capital tend to face
low life-expectancies for several generations, and low life-expectancy, in turn,
1
decreases the incentive to invest in human capital by increasing their risks.
Therefore, unfavorable initial conditions are di–cult to overcome and long-
lasting vicious cycles may follow; some countries may even be caught in a
low life-expectancy trap (Becker et al. 1990, Soares 2005). Furthermore, low
human capital is associated with high fertility, which tends to decrease life-
expectancy through maternal and infant deaths, especially in poor countries.
Caldwell (2000) also associates high fertility rates with extramarital sexual
activities, increasing the risk of AIDS.
In this paper, I concentrate on the need to classify the data into longevity
clubs. Both the number and boundaries of these clubs are derived from the
data by the regression tree method suggested by Breiman et al. (1984) and
Durlauf and Johnson (1995), a data-sorting method which partitions the
sample into sub-samples to flnd the best piecewise linear model. The main
advantage of this method is that instead of applying some pre-determined
partitioning criterion, it chooses both the partitioning variable and its split
value for each split to generate the homogenous sub-samples possible. Suc-
cessive splits grow up a tree, starting from the root (the full sample) to
the leaves (clubs). By supplying several theoretically plausible candidates
to serve as the classifying variables one can learn from their importance in
terms of the longevity convergence in the sub-samples or clubs.
The regression tree analysis suggests that a global sample of countries
should be partitioned into four convergence clubs according to human cap-
ital and income. The role of human capital is important, since three clubs
are characterized by their initial literacy rates, whereas low initial income
is typical of the fourth. A rich convergence analysis is now viable since one
can investigate the life-expectancy convergence both within and between the
clubs. The results show that because deflcient literacy prevents difiusion of
information, the average life-expectancy growth in the club with the lowest
literacy was slower than expected from its initial life-expectancy. No con-
vergence arose within this club either. By contrast, the club with medium
literacy proceeded rapidly, also exhibiting convergence within the club, pre-
sumably because of health externalities arising once some critical literacy
level was reached. An explanation for the slow progress in the club with the
highest literacy may in turn be in the complexity of health information, mak-
2
ing the absorbtion of this information less easy. Since privatization of health
eliminates externalities, no convergence has appeared within this club. The
analysis here also implies that because the AIDS epidemic was so strongly
fueled by poverty, an unwanted re-appearance of the negative association
between income and life-expectancy occurs. Given that poverty is still a
considerable problem, this paper thus pays attention to the risk of escalation
of this epidemic in the future.
The outline of the paper is the following: Section 2 performs a specifl-
cation test and partitions the data into longevity clubs. Sections 4 and 3
analyze convergence within and between the clubs, and discuss the role of
literacy, incomes, and fertility. Section 5 ofiers conclusions and closes the
paper.
2 Regression Tree Analysis
Consider the post-war growth of life-expectancy.1 In 1960, the average life-
expectancy in a sample of 125 countries was only 51:72 years but this in-
creased to 64:63 years by 2001. Since countries with the lowest initial values
exhibited the most rapid growth, convergence of life-expectancy appeared
in this sample. Figure 1 illustrates this by showing a clear negative associ-
ation between the initial life-expectancy and its growth, but Figure 1 also
shows that convergence was far from universal as many low life-expectancy
countries fell behind and, in the most unfortunate cases, life-expectancy even
decreased. Examples are Zimbabwe, Zambia, and Botswana with decreases
of 10:46, 6:13, and 2:98 years, respectively. It is thus possible that the full
sample should be partitioned into sub-samples to derive unbiased estimates
for convergence. Hence, we perform a speciflcation test to see whether any
sub-samples exist in the data.
The theoretical discussion above suggests that unequal life-expectancy
growth appears where considerable disparities exist in initial income, human
capital, and fertility so that we take these variables as candidates to classify
the data. The income is measured by (the log of) the real per capita GDP
and human capital by the literacy rate. Both variables were measured in
1life-expectancy at birth, both sexes, denoted by life-expectancy and longevity in the
text.
3
30 50 70 90-20
-10
0
10
20
30
40
 
Life-expectancy growth (years)
Initial life-expectancy (years)
Figure1: life-expectancygrowthregressedagainstinitiallife-expectancy(125
countries, period 1960-2001).
1960. The third candidate is fertility. Fertility was already relatively low in
developed countries in 1960, but in developing countries it was universally
high, discriminating them only poorly. Since the flrst family planning pro-
grams, however, had started to run with the outcome that fertility decreased
in some developing countries but was resistant in some others, we choose
the fertility change from 1950 to 1960, instead of choosing the initial fertil-
ity rate as an indicator of (inherited) fertility behavior,. The demographic
data comes from the United Nations (2007), the income data from Maddison
(2001) (GPD per capita in constant international dollars), and the literacy
data from the World Bank (2000). All variables are presumably exogenous
to life-expectancy growth from 1960 to 2001.
To perform the speciflcation test, we divide the full sample into two sub-
samples by means of the variables above.2 Let T stand for the period length
(41 years) and Ei;t for the life-expectancy in country i at time t. We estimate
the standard convergence equation
ln(Ei;t=Ei;0)=T = fi+fl ln(Ei;0)+†i; (1)
by least squares for the full sample and for the sub-samples separately. The
null is that the full sample obeys the common linear model. The Wald test
2The flrst (second) sub-sample consists of countries with higher (lower) than the mean
ofpercapitaGDP andliteracy, andfaster(slower)thanthemeandecreaseintotalfertility.
4
of similarity of fi and fl in the sub-samples yields F¡statistics of 3:21 which
with (2;121) degrees of freedom refers to a p¡value of 0:04, so that the null
is rejected in favor of sub-samples. Hence, the club approach is justifled here.
Even though the speciflcation test suggests that at least some sub-samples
exist in the data, their number and boundaries cannot be properly revealed
by mechanical splits. We therefore discover the clubs by the regression tree
method. This method partitions the sample to flnd the best piecewise linear
model. For each split, the algorithm chooses the splitting variable and its
splitvaluetominimizethesumofsquaredresiduals(SSR)inthesub-samples.
Only one-step look ahead and binary splits are used. A detailed description of
the regression tree method is available in Breiman et al. (1984) and Durlauf
and Johnson (1995).
Figure 2 reports the result of the regression tree analysis. The flrst split
comes in terms of the initial literacy rate, the split value being 90:45%.
Countries with an initial literacy rate higher than this value constitute the
flrst longevity club, which we refer to as the \High-Literacy" club (symbol h).
Figure 2 indicates that this club has 26 members with average life-expectancy
growth of 7:62 years. The second split comes in terms of literacy again,
separating the countries with an initial literacy rate lower than (or equal
to) 35:75% to the \Low-Literacy" club (46 members, average life-expectancy
growth 14:98 years, symbol l). The initial literacy rate in all remaining
countries is between 35:75% and 90:45%, but some of these countries had
markedly low incomes initially. The remaining countries are thus partitioned
in terms of their incomes, the split value being 6:90. Countries with income
below(orequal to)this valueconstitute the\Low-Income"club (14members,
average life-expectancy growth 7:56 years, symbol g), while the remaining
countries constitute the \Medium-Literacy" club (39 members, average life-
expectancy growth 15:90 years, symbol m). For a complete list of countries
in these four clubs, see Appendix A.
One can continue the partitioning until the algorithm is unable to flnd
new splits to reduce SSR, or all degrees of freedom are used up (the number
of countries in the sub-samples is equal to or smaller than twice the num-
ber of regressors). Even though the model flt always increases, every split
makes its interpretation more di–cult so that an excessive number of clubs
5
    K68
Literacy
Literacy
90.45% 
35.75%     26High Literacy7.62
K6C
46
Low Literacy14.98
K67 K6D
Income
   6.90 
14  
 
39
Medium LiteracyLow Income 15.907.56
  Split variable
Split value
  Club symbol
  Number of members Life-expectancy growth 
    Name of the club
Figure 2: The regression tree. The right arrow indicates the countries for
which the split variable takes values larger than the split value. The left
arrow indicates the countries for which the split variable takes values smaller
than or equal to the split value.
is not desirable. Some \over fltting" may also appear in the tree since it is
adapted according to every detail of the present sample, thus decreasing its
predictive power in the \population". Therefore, a cross validation exercise
is performed for SSR to see, what number of clubs generates the best flt in
the \population", with the outcome that the best number of clubs turns out
to be four as already depicted in Figure 2.3 For details in tree pruning, see
3In the ten-fold cross validation performed here, the full sample is partitioned into ten
approximately equal pieces stochastically. Nine of them are used to develope a tree which
is then fltted to the tenth to calculate SSR. By repeating this and averaging, one can
calculatea populationestimatefor SSR for each number of clubs. The club number which
minimizes this estimate refers to the best tree size and to the best piecewise linear model.
6
Breiman et al. (1980) and Durlauf and Johnson (1995).
LIF60 LIF01 GROWTH LIT60 DTFR GDP60 GDP01
Full Sample 51.72 64.63 12.91 54.89 0.168 2644 6380
High-Literacy 69.72 77.34 7.62 98.23 0.089 6611 17264
Medium-Literacy 55.16 71.06 15.90 68.62 0.194 2555 6293
Low-Literacy 40.80 55.79 14.98 20.10 0.150 1075 1539
Low-Income 44.58 52.15 7.57 50.47 0.299 677 2319
Table 1: Longevity club statistics. The notations are: initial and flnal life-
expectancies and life-expectancy growth (LIF60, LIF01, GROWTH), initial literacy
rates (LIT60), fertility changes (DTFR), initial and flnal incomes (GDP60, GDP01).
All numbers are averages.
The order by which the variables enter to partition the data provides
information about their importance. Given that the flrst two splits come
in terms of the literacy rate, the analysis here suggests that literacy (human
capital) is the variable which most e–ciently discriminates between the clubs.
Naturally, it is not the only one. Table 1 shows that the clubs have difierent
initial incomes and fertility changes as well. Since the High-Literacy club, for
example, had the slowest fertility change and highest initial income, whereas
the Medium-Literacy club had medium values for these variables as well, not
all life-expectancy disparities can be explained by literacy alone. Its role is
the most decisive, however, thus supporting the idea that the efiects of the
initial human capital are intergenerational and long-lasting. The importance
of literacy also suggests that health technology difiusion has been the driving
force in life-expectancy growth, since the capacity to absorb information
seems to be the critical condition for this growth.4
4Note that the algorithm does not choose fertility to partition the data, even though its
association with infant and maternal mortality seems indisputable, and Caldwell (2000)
convincingly discusses its role in explaining AIDS contamination rates. A possible ex-
planation for this mismatch is that the fertility change before the research period applied
here as a proxy did not adequately predict the fertility rate at the time of the break-out
of AIDS, which occurred much later and seems to explain most life-expectancy failures,
as is discussed in Section 4. An alternative explanation is that literacy explains fertility
to such a degree that fertility itself loses its explanatory power.
7
3 Convergence Within the Clubs
The conventional result is that convergence of life-expectancy appears in
broad samples of countries (Ram 1998, Neumayer 2003, Becker et al. 2005,
Riley 2005). Table 2 reports that the sample of 125 countries here also ex-
hibits highly signiflcant fl-convergence as calculated by Equation (1). The
speciflcation test above suggests, however, that the derived estimate may
be biased and that one should merely concentrate on clubs. Table 2 shows
that statistically signiflcant fl-convergence only appeared within the Medium-
Literacy and Low-Income clubs, while less than signiflcant convergence ap-
peared in the High-Literacy club. In the Low-Literacy club, the association
between initial life-expectancy and its growth was negative but not signifl-
cant at any conventional level. Figure 3 illustrates these results by showing a
clear downward sloping trend in the Medium-Literacy and Low-Income clubs
(symbols m and g), whereas no such trend is seen in the High and Low-
Literacy clubs (symbols h and l). Life-expectancy convergence thus seems
to be a much more complicated process than is implied by the full-sample
analysis alone.
fl t-test prob.
Full Sample -0.784 -6.814 0.000
High-Literacy -0.661 -1.352 0.094
Medium-Literacy -1.450 -7.877 0.000
Low-Literacy -0.239 -0.607 0.273
Low-Income -2.723 -1.868 0.043
Table 2: Convergence statistics.
To flnd an intuitive explanation for the club-speciflc convergence results
aboveweturntotheroleofliteracyagainasitsimportanceisindicatedbythe
regression analysis. Consider flrst the Low-Literacy club. By construction,
the literacy rate in all countries of this club was initially lower than 35:75%,
the lowest value being 2:00%. Furthermore, since low (initial) literacy was
closely associated with low (initial) life-expectancy (correlation 0:69), the
least advanced countries with the highest potential for growth also faced the
most serious constraints in realizing this potential, as low literacy seriously
limited absorbtion of new knowledge in these countries. Thus, in spite of
8
their good prospects, these countries only proceeded at the average rate, i.e.,
convergence failed within this club.
30 50 70 90
-20
-10
0
10
20
30
40
h hhh
h
h
h h
h
h
h
h
hhh h
hh
g
g
g
g
g
g
g
g
g
g
gg
g
m
mm
m
m
m
m m m
mmm
m
mm
m
mm
m m
m
m m
m
m
m
m
m
m
m m
m
m
m
m
ml
l
l
l
l
l
l
l
l
l
l
lll
l
l
l
l ll l
l
l
l
l
l
l l
l
l
l
l
l
l
l
l
l
l l
l l
ll
l
l
Life-e xpectancy g rowth (y ears)
Initial lif e-e xpectancy (y ears)
Figure 3: The Longevity Clubs.
Consider then the Medium-Literacy club. By construction, its initial lit-
eracy rate varied from 35:75% to 90:45%, being closely associated with initial
life-expectancy again (correlation 0:68). The least advanced countries should
therefore have faced the most serious constraints in this club as well. De-
spite this, highly signiflcant convergence appeared, so that one wonders why
the above reasoning is not valid here. Presumably an explanation can be be
found in external efiects typically associated with human capital and literacy
(Lucas 1988). Externalities arise because literate members of society inform
the illiterate ones, either orally or by showing things in practice. Once a
critical literacy level is reached, most people achieve new information, and
further gains in literacy are less important as they induce little improve-
ment in absorption marginally.5 To put this difierently, one can say that the
literacy constraint becomes less binding and the potential for growth dic-
5It is possible that the this critical value was close to 35:75%, identifled here to distin-
guish the Medium-Literacy club from the Low-Literacy club.
9
tates the life-expectancy growth, thus inducing convergence within this club.
This conclusion is conflrmed by facts from the Low-Income club, in which
club, the initial literacy rate was less closely related to initial life-expectancy
(correlation only 0:22) so that the worst absorbtion constraints were not so
typical of the least advanced countries, and these could utilize their poten-
tials for growth e–ciently. Hence most rapid convergence was seen in this
club (Figure 3, Table 2).
Table 2 also shows that even though some convergence appeared in the
High-Literacy club, this convergence was not signiflcant at the 5% level.
To comprehend this, note that the High-Literacy club had already reached
the highest stage of health transition with cardiovascular diseases and aging
as the most serious obstacles to life-expectancy growth. Since the causes
of these intricate problems are not completely understood thus far, and no
simple treatment is available yet, health information tends to be markedly
complex in this club. Complicated instructions are given to patients and
proflt-targeted misinformation is common. Not just literacy but critical in-
formation absorbtion becomes important, and it is far from self-evident that
adequate health information reaches everyone. Privatization of health is
typical, so that positive health externalities are less likely (Vallin and Mesl¶e
2004, Soares 2007). Given that the correlation between initial literacy and
life-expectancy is still high (0:60), the most backward countries tend to have
the greatest absorbtion constraints in this club. In spite of high literacy
rates, the complexity of health information and missing externalities make
these constraints binding, thus explaining why no (statistically signiflcant)
convergence appeared in this club.
4 Convergence Between the Clubs
Theconvergencehypothesismaintainsthatthequickestlife-expectancygrowth
is seen in the initially most backward countries, and one expects this principle
to hold for club-speciflc average life-expectancies as well. Figure 4 illustrates
the beginning and end-of-period values. Consider flrst the High-Literacy club
where the average life-expectancy was initially as high as 69:72 years and in-
creased further by 7:62 years (see also Table 1). However, as compared with
10
the Medium-Literacy club, with an initial value of 55:16 years and an increase
of 15:90 years, the High-Literacy club proceeded slowly. This is in line with
the convergence hypothesis: life-expectancy gains in the High-Literacy club
were di–cult to achieve, whereas the less advanced Medium-Literacy club
was able to absorb new knowledge easily (Omran 1971, Deaton 2003, Vallin
and Mesl¶e 2004, Edwards and Tuljapurkar 2005, Cutler et al. 2006, Soares
2007). As a consequence, the average life-expectancy in the Medium-Literacy
club approached that in the High-Literacy club.
0 
20 
40 
60 
80 
1960 2001
K68K20K20K20K20K20K20K20K20K20K20K20K20K20K20K20K6DK20K20K20K20K20K20K20K20K20K20K20K20K20K20K67K20K20K20K20K20K20K20K20K20K20K20K20K20K20K6C
Figure 4: Beginning and end-of-period average life-expectancies.
Consider then the Low-Literacy club, the average life-expectancy of which
was 40:80 years initially and then increased by 14:98 years (Figure 4, Table
1). At flrst sight, this seems to be a satisfactory result. But it is not.
Comparison with the Medium-Literacy club shows this clearly. In 1960, the
life-expectancy gap between the Medium and High-Literacy clubs was 14:56
years, while that between the Low and High-Literacy clubs was 28:91 years.
Hence, the potential for growth was much higher in the Low-Literacy club
which, however, utilized it only to a limited extent by closing some 52% of
the gap, while the Medium-Literacy club closed more than 100% of its gap.
Presumably, literacy is an explanation since its average initial value was only
20:10% in the Low-Literacy club but 68:62% in the Medium-Literacy club,
implying that these clubs had difierent opportunities to utilize their growth
potential. Health externalities further exacerbated the disparity since they
are likely to exist only in the latter club.
Soares (2007) provides further insight into the role of literacy by maintain-
11
ing that the efiects of some health measures, such as large-scale immuniza-
tions and improvements in public health infrastructure, are independent of
the personal characters of the agents. But since some gains related to prepa-
ration of food, treatment of water, and nursing of infants, for example, derive
from personal practices, their efiects depend more on the absorbtion of knowl-
edge on the agent’s side (Soares 2007). Thus the observed life-expectancy
growth in the Low-Literacy club was probably caused by the former type
of measure alone, whereas both types were e–cient in the Medium-Literacy
club.
Now turn to the Low-Income club, in which the average life-expectancy
increased by 7:57 years, implying that only 30% of its initial life-expectancy
gap was closed (Figure 4, Table 1). Given that this club was constructed on
the basis of its low (initial) incomes, malnutrition and many income-related
diseases must have increased mortality in this club. It has been claimed,
however, that these outcomes of poverty have decreased, and the association
between life-expectancy and income has become less tight (Soares 2007). It
does indeed seem that a great deal of the longevity disaster in the Low-
Income club was caused by the AIDS epidemic. This is clearly illustrated
in Figure 5, which shows how the promising convergence trend of this club
was interrupted in the 1980s, implying that a great deal of the regression
was due to the break-out of AIDS. The HIV prevalence numbers of 2:67%,
0:26%, 1:48%, 2:45%, and 11:14% in the full sample, High, Medium, and
Low-Literacy clubs, and the Low-Income club in 2005 support this, even
though the association between income and AIDS is not one-to one, since
the Low-Income club consists of four low-AIDS countries (Burma, China,
The Comorros, and Vietnam), whereas four countries with two-digit HIV
prevalence rates (Central Africa, Mozambique, Namibia, and South Africa)
are classifled into other clubs.
Since poverty exacerbates mortality from AIDS by imposing serious lim-
its on treatment and prevention (UNAIDS and WHO 2007), it seems that
AIDS has established the poverty-mortality connection again. In many
cases, one can flnd historical reasons for poverty. The Sub-Saharan coun-
tries are good examples; being seriously impaired by their colonial roots,
they were markedly vulnerable when the HI virus started to spread and the
12
30 
40 
50 
60 
70 
80 
90 
Lowh Income Low Literacy 
Medium Literacy 
High Literacy 
1960  1980  2000 
     Year
Life-expectancy (years)
Figure 5: Trends in life-expectancy.
life-expectancy toll was dramatic. History explains why African countries
were the flrst to confront the exacerbated efiects of AIDS, but if we take the
role of poverty seriously we see it as a future risk as well. Table 1 shows that
the average income is now the lowest in the Low-Literacy club. Therefore,
the analysis here raises concerns about the escalation of AIDS in this club.
To reiterate, starting from the importance of human capital (literacy)
and income indicated by the regression tree analysis above, I have explained
the convergence of life-expectancy by these factors alone. Naturally, several
other explanations are possible. One is the role of trade because knowledge is
often embodied in goods. Another is institutional, as good institutions, legal
governments, and political stability are factors promoting difiusion. Cultural
factors are important as well, but data availability limits the analysis. Never-
theless, the intergenerational nature of human capital and its critical role in
absorbtion of information are intuitively appealing in explaining long-lasting
disparities in life-expectancy growth. Only the recent AIDS epidemic turns
out to be an exception, since its roots seem not to be in in low literacy but
in poverty.
13
5 Conclusions
This paper discusses the post-war convergence of life-expectancy in longevity
clubs which are identifled in the data by the regression tree method. Four
clubs appear. Given that three clubs are constructed by their literacy, this
analysis suggests that literacy or, more broadly, human capital, has played
an essential role in convergence of life-expectancy.
Convergence of life-expectancy is analyzed both within and between the
clubs. The results show that because deflcient literacy prevents difiusion of
information, the growth in the club with the lowest literacy rate was slug-
gish. No convergence appeared within this club. By contrast, the club with
literacy at the medium level proceeded rapidly and exhibited convergence,
presumably the result of high literacy and health externalities rising once
critical literacy is achieved. The slow progress in the club with the highest
literacy was probably caused by the complexity of health information and
privatization of health. No convergence appeared within this club either.
Most life-expectancy laggards are members of the Low-Income club because
the mortality efiects of AIDS are exacerbated by poverty.
The results of this paper suggest that continuous efiorts are needed to
remove the constraints on life-expectancy growth in the less advanced coun-
tries. Educational programs are important, but more frank measures are
also needed because the health efiects of educational programs appear slowly.
Profound poverty needs attention because it increases the risk of AIDS, and
we should also flnd policies to improve the absorbtion of complex health
information in the most advanced countries; measures which increase exter-
nalities and decrease privatization could be successful here. But we should
also reconsider the roles of biological and man-made obstacles in the slow-
down of life-expectancy growth in the most advanced countries, since it is
possible that we have accepted it as an intrinsic element of convergence too
easily. One of the future challenges thus is to investigate the extent to which
this slowdown can be alleviated by well-constituted health policies.
14
A Appendix: Countries and Clubs
High Literacy Club:
Argentina, Australia, Austria, Belgium, Bulgaria, Canada, The Czech Re-
public, Denmark, Finland, France, Germany, Hungary, Ireland, Italy, Japan,
The Netherlands, New Zealand, Norway, Poland, Romania, Spain, Sweden,
Switzerland, The United Kingdom, The United States of America, Uruguay.
Medium Literacy Club:
Albania, Bahrain, Bolivia, Brazil, Chile, Colombia, Costa Rica, Cuba, The
Dominican Republic, Ecuador, El Salvador, Greece, Guatemala, Honduras,
Indonesia, Israel, Jamaica, Jordan, Lebanon, Madagascar, Malaysia, Mauri-
tius, Mexico, Namibia, Nicaragua, Panama, Paraguay, Peru, ThePhilippines,
Portugal, Singapore, South Africa, South Korea, Sri Lanka, The Syrian Arab
Republic, Thailand, Trinidad and Tobago, Turkey, Venezuela.
Low Literacy Club:
Afghanistan, Algeria, Angola, Bangladesh, Benin, Burkina Faso, Burundi,
Cambodia, Cameroon, The Central African Republic, Chad, Congo, C^ote
d’Ivoire, Djibouti, Egypt, Eritrea, Gabon, Gambia, Ghana, Guinea, Guinea-
Bissau, Haiti, India, Iran, Iraq, Laos, Libya, Mali, Mauritania, Mongolia,
Morocco, Mozambique, Nepal, Niger, Nigeria, Oman, Pakistan, Rwanda,
Senegal, Sierra Leone, Somalia, Sudan, Togo, Tunisia, Yemen, Zaire.
Low Income Club:
Botswana, Burma, China, TheComorros, EquatorialGuinea, Kenya, Lesotho,
Malawi, Swaziland, Vietnam, Tanzania, Uganda, Zambia, Zimbabwe.
15
References
Becker Gary S., Kevin Murphy, and Robert Tamura. 1990. \Human capital,
fertility, and economic growth," Journal of Political Economy 98(5): S12-
S37.
Becker Gary S., Tomas J. Phillipson, and Rodrico R. Soares. 2005. \The
quantity and quality of life and the evolution of world inequality," American
Economic Review 95(1): 277{291.
Breiman, Leo, Jerome L. Friedman, R. A. Olshen, and Charles J. Stone.
1984. Classiflcation and regression trees. Belmont, CA. Wadsworth.
Caldwell, John. 2000. \Rethinking the African AIDS epidemic," Population
and Development Review 26(1): 117{135.
Cigno, Alexander. 1998. \Fertility decisions when infant survival is endoge-
nous," Journal of Population Economics 11: 21{28.
Cutler, David, Agnus Deaton, and Adriana Lljeras-Muney. 2006. \The
determinants of mortality," Journal of Economic Perspectives 20(3): 97{120.
Deaton, Agnus. 2003. \Health, inequality, and economic development,"
Journal of Economic Literature 49(1): 113{158.
Durlauf, Steven, and Paul A. Johnson. 1995. \Multiple regimes and cross-
country behaviour," Journal of Applied Econometrics 10: 365-384.
Edwards, Ryan D., and Shripad Tuljapurkar. 2005. \Inequality in life-
spans and a new perspective on mortality convergence across industrialized
countries," Population and Development Review 31(4): 645{674.
Fogel, Robert W. 1994. \Economic growth, population theory, and physiol-
ogy: the bearing of long-term processes on the making of economic policy,"
American Economic Review 84(3): 369{395.
Fogel, Robert W. 2004. The Escape from Hunger and Premature Death,
1700-2100 { Europe, America, and the Third World. Cambridge, Cambridge
University Press.
Lucas, Robert E. 1988. \On the Mechanics of Economic Development,"
Journal of Monetary Economics 22: 3^u-42.
Maddison, Angus. 2001. The World Economy: Historical Statistics. Paris,
OECD, CD-rom.
Mayer-Foulkes, David. 2003. \Convergenceclubsincross-countrylife-expectancy
dynamics," Chapter 5 in: Perspectives on Growth and Poverty. Van der Ho-
even, Rolph, and Anthony Shorrocks (eds.) Tokyo, United Nations Univer-
sity Press.
16
Neumayer, Eric. 2003. \Beyond income: convergence in living standards,
big time," Structural Change and Economic Dynamics 14: 275 A =296.
Omran, Abdel. 1971. \The epidemiological transition: a theory of the
epidemiology of population change," Milbank Memorial Fund Quarterly 49:
509{538.
Preston, Samuel H. 1975. \The changing relation between mortality and
level of economic development," Population Studies 29(2): 231{248.
Pritchett, Lant. 1997. \Divergence, big time," Journal of Economic Per-
spectives 11(3): 3{17.
Ram, Rati. 1998. \Forty years of the life span revolution: an exploration of
the roles of \convergence," income, and policy," Economic Development and
Cultural Change 46(4): 849{857.
Riley, James R. 2005. \The timing and pace of health transition around the
world," Population and Development Review 31(4): 741{764.
Soares, Rodrigo R. 2005. \Mortality reductions, educational attainment, and
fertility choice," American Economic Review 95(3): 580{601.
Soares, Rodrigo R. 2007. \On the determinants of mortality reductions in
the developing world," NBER Working Paper No. W12837.
Vallin, Jacques, and France Mesl¶e. 2004. \Convergences and divergences
in mortality. a new approach to heath transition," Demographic Research,
Special Collection 2, Article 2, 11{44.
World Bank. 2000. World development indicators 2000. Washington CD.
CD-rom.
UNAIDS and WHO. 2007. AIDS epidemic update. Geneva, Joint United
Nations Programme on HIV/AIDS and World Health.
United Nations. 2007. World population prospects. The 2006 revision. New
York .
17