German World War

7/30/2019 German World War

1/43

1

Alternative Estimates of the Effect of the Increase of Life

Expectancy on Economic Growth

Muhammad Jami Husain

Keele Management SchoolKeele University

Abstract

Until recently the literature has found evidence of a positive, significant, and sizable

influence of life expectancy on economic growth. This view has been challenged by

Acemoglu and Johnson (2007). They find no evidence that the large exogenous increase in

life expectancy led to a significant increase in per capita economic growth. This paper takes

up the modelling and estimation framework presented in Acemoglu and Johnson (2007),

and presents alternative estimates on the impact of life expectancy on population, GDP, and

GDP per capita by using alternative instruments, timeline, and country groups. The

findings suggest that the results differ significantly from that provided in Acemoglu and

Johnson (2007); that the generalization of the pessimistic outcomes with regards to the

impact of life expectancy on income per capita requires caution; and the increase of lifeexpectancy may have had a positive impact on income per capita growth.

JEL: I10, O40, J11

Keywords: Life Expectancy; Income; Economic Growth; Immunization; Instruments.


2/43

2

Alternative Estimates of the Effect of the Increase of Life

Expectancy on Economic Growth

1. ITRODUCTIO

Improvements of health and longevity are not simply viewed as a mere end- or by-product

of economic development but argued as one of the key determinants of economic growth,

and therefore provide means to achieve economic development and poverty reduction.

Hence, better health does not have to wait for an improved economy; rather, measures to

reduce the burden of disease, to give children healthy childhoods, to increase life

expectancy etc. will in themselves contribute to creating richer economies (see e.g. Suhrcke

et al, 2005). Until recently the literature has found evidence of a positive, significant, and

sizable influence of life expectancy (or some related health indicator) on economic

growth.1

This view has been challenged in a recent paper by Acemoglu and Johnson (2007). In the

face of critics scepticism that the cross-country regression studies may have established

merely a strong correlation between measures of health and both the level of economic

development and recent economic growth without being able to establish a causal effect of

health and disease environments on economic growth, they provided an empirical analysis

based on the international epidemiological transition, apparently led by the wave of

international health innovations and improvements that began in the 1940s, and found that

there was no evidence that the large exogenous increase in life expectancy led to a

significant increase in per capita economic growth. 2 An instrument was constructed,referred to as predicted mortality, based on the pre-intervention distribution of mortality

from 14 major diseases around the world and dates of global interventions. This instrument

appeared to have a large and robust effect on changes in life expectancy starting in 1940.

Then, it was shown that instrumented changes in life expectancy had a large effect on

population, a 1% increase in life expectancy leading to an increase in population of about

1.7-2%, but a much smaller effect on total GDP both initially and over a 40-year horizon.

Accordingly, the impact of an increase in life expectancy on per capita income was found

1A common empirical approach toward examining the impact of health on economic growth has been to

focus on data for a cross-section of countries and to regress the rate of growth of income per capita on theinitial level of health (typically measured by life expectancy, or survival rate), with controls for the initial

level of income and for other factors believed to influence steady-state income levels. These factors might

include, for example, policy variables such as openness to trade; measures of institutional quality, educational

attainment, and rate of population growth; and geographic characteristics. The application of growth

regression approach can be seen, inter alia, in Barro (1996); Barro and Lee (1994); Barro and Sala-I-Martin

(1995); Bhargava et al., (2001); Bloom, Canning, and Malaney (2000); Bloom and Malaney (1998); Bloomand Sachs (1998); Bloom and Williamson (1998); Caselli, Esquivel, and Lefort (1996); Gallup and Sachs

(2000); Hamoudi and Sachs (1999). The estimated effects of an increase in the life expectancy (expressed in

log terms) economic growth are the followings: 4.2% in Barro, 1996; 7.3% in Barro and Lee (1994); 5.8% in

Barro and Sala-I-Martin (1995); 6.3% Bloom, Canning, and Malaney (2000); 2.7% in Bloom and Malaney

(1998); 3.7% in Bloom and Sachs (1998); 4% in Bloom and Williamson (1998); 0.1% in Caselli, Esquivel,

and Lefort (1996); 3% in Gallup and Sachs (2000); and 7.2% in Hamoudi and Sachs (1999).2

The wave of international health innovations and the concomitant international epidemiological transitionled by the dramatic decrease in deaths from the infectious and parasitic diseases provide us with an empirical

strategy to isolate potentially-exogenous changes in health conditions.


3/43

3

to be insignificant or negative. The interpretation of the insignificant or negative impact of

life expectancy on GDP per capita is that, in the face of population growth, the other

factors in the production function, capital and land in particular, did not adjust. As

population increased, the capital-labour ratio decreased, which eventually led to a decrease

in the per capita income.

3

The political economy consequences of such findings bear critical implications in terms of

direct investments in health sectors. Such a finding is rather at odds with the advocates of

investing in health. Generalization of the pessimistic findings of Acemoglu and Johnson

(2007) should be subject to more scrutiny and this paper contributes towards this. In

particular, there is a need for considerable caution in interpreting the results obtained from

their framework of analysis for two reasons. First, the nature of the international

epidemiological transition that occurred around the 1940s and 1950s was unique and may

not be applicable to todays world. The changes in life expectancy that occurred mainly due

to the infectious diseases during that time may have different implications than the changes

that are being observed in recent times. For example, improvement in life expectancy in the

recent decades may not increase the population to the extent that it did during the

epidemiological transitions. Second, the nature of diseases has changed in todays world.

The diseases that affect the productive ages are probably becoming more important (e.g.

HIV/AIDS, heart diseases, cancer.) in recent times instead of the diseases that are more

fatal for children. Further study of the effects of the HIV/AIDS epidemic on economic

outcomes as well as more detailed analysis of different measures of health on human

capital investments and economic outcomes are major areas to be explored. Again,

Acemoglu and Johnson (2007) were unable to include Africa in their baseline analysis

owing to lack of data. Africa may be an important source of variation in the data and its

inclusion in the sample might have led them to different conclusions.4

This paper presents further analysis and estimates of the impact of life expectancy onincome by using alternative instruments, timelines, and country groups. The objective is to

investigate whether the findings of Acemoglu and Johnson (2007) prevail if different

instruments, time-lines, or country groups are used. Section 2 provides a description of

alternative sources of exogenous variations in life expectancy which can be used as

potential instruments. Section 3 presents the first stage estimates to highlight instrument

relevance and strengths. Section 4 presents the two-stage least square (2SLS) estimates of

the impact of life expectancy on the three major macro-variables, i.e. population, GDP, and

GDP per capita. What is shown below is that the results differ significantly from those

provided in Acemoglu and Johnson (2007); that the pessimistic outcome with regards to the

impact of life expectancy on income per capita requires caution; and that the positive

impact of life expectancy on income per capita is more apparent.

3However, the results of many micro-level studies (see e.g. Strauss and Thomas, 1998; Suhrcke et al., 2005;

Thomas and Frankenberg, 2002; Rugeret al., 2006; Behrman and Rosenzweig, 2001; Miguel and Kremer,

2004; Schultz, 2002) and cohort-level studies (e.g. Bleakley, 2006a; 2007) obtained large positive effects ofhealth on productivity. In that case, these latter effects, assuming that they exist and of partial equilibrium

nature, were fully offset by the crowding-out effect of population growth (Bleakley, 2006).4

Cervellati and Sunde (2009) suggest differing causal effects of life expectancy on income per capita due to

different phases of development. The negative impact of the increases in life expectancy on income is

observed for countries that did not go through the demographic transition, whereas in post-transitional

countries gains in life expectancy increase per capita income. By pooling the countries in their sample, they

argue, Acemoglu and Johnson (2007) cannot capture this nonlinear dynamic. Also, Bloom, Canning and Fink(2009) highlight that the healthiest nations in 1940 are those that benefited least from the health interventions

and also the ones that grew the most, giving a negative relationship between health interventions and growth.


4/43

4

2. THEORETICAL AD ESTIMATIO FRAMEWORK

The implication of the increase in the length of human life is modeled in a closed-economy

Solow type neoclassical growth model. The health variable is represented in terms of life

expectancy only. The aggregate production function of the economy i has constant returnsto scale.

= 1)( ititititit LKHAY (1)

Where 1+ ,Kit denotes capital,Litdenotes the supply of land, and Hitis the effective

units of labour given by Hit= hitit ; where it is the total population (henceforth,

representing those who are employed), while hit is the human capital per person. All labour

and land are inelastically supplied. The production function does not lose its generality if

the supply of land is normalized to unity for all countries (i.e.Lit = Li = 1). With regards to

the technological progress, the model maintains constant technology differences across

countries, but ignores differentiated technological progress (i.e iit AA = ). Also, cross-country differences in human capital per person and population are assumed to be constant.

Therefore, iit hh = and iit = .

Economies face depreciation of capital at the rate )1,0( and the savings (investment)

rate of country i is constant and equal to )1,0(is . This implies that the evolution of capital

stock in country i at time twill be ititiit KYsK )1(1 +=+ . Suppose that life expectancy

changes from0it

X to1it

X and remains at this level thereafter. After population and the

capital stock have adjusted, the steady-state capital stock level will be ii

i Ys

K

= .

Substituting into (1) and taking logs we obtain a simple relationship between income per

capita, the savings rate, human capital, technology, and population:

iiii

i

ii shA

Yy log

1

1log

1log

1log

1log

1log

+

+

=

(2)

Equation (2) shows that income per capita is affected positively by technology Ai, human

capital, hi, and the investment rate si. Population, i, has a negative effect on income per

capita. The term that captures the impact of population would drop out from the equation

if 01 . Acemoglu and Johnson (2007) suggests that for industrialized countries

land plays a small role in production due to only a small fraction of output being produced

in agriculture, so that 01 . Nevertheless, for many less-developed countries,

where agriculture is still important, 01 > and the direct effect of an increase in

population may be to reduce income per capita even in the steady state (i.e., even once the

capital stock has adjusted to the increase in population).5

5See Galor and Weil (2000), Hansen and Prescott (2002), and Galor (2005) for models in which at differentstages of development the relationship between population and income may change because of a change in

the composition of output or technology. In these models, during an early Malthusian phase, land plays an

important role as a factor of production and there are strong diminishing returns to capital. Later in the

development process, the role of land diminishes, allowing per capita income growth. Hansen and Prescott(2002), for example, assume a Cobb-Douglas production function during the Malthusian phase with a share

of land equal to 0.3.


5/43

5

The impact of increased life expectancy is assumed to be working initially through

increased population (both directly and also potentially indirectly by increasing total

births), so that: itiit X = (3)

whereXit

is life expectancy in country i at time t. Another important channel of influence

includes the impact that better health and longer life spans has in terms of increased

productivity. This increase in productivity may originate through a variety of channels,

including more rapid human capital accumulation (e.g Kalemli-Ozcan, Ryder, and Weil,

2000; Kalemni-Ozcan, 2002; Soares, 2005) or direct positive effects on (total factor)

productivity (e.g. Bloom and Sachs, 1998). Acemoglu and Johnson (2007) have used the

following isoelastic function in order to capture the beneficial effects of these variables on

productivity:

itiit XAA = and

itiit Xhh = (4)

Where iA and ih are some baseline differences across countries. Using the steady-state

value of capital stock together with (1), (3) and (4), we derive the long-run relationship

between log life expectancy and log per capita income below:

= iiitiitiitiit YsXXhXAY )(

it

iiii

it

it

X

shA

Y

log))1()((1

1

log1

1log

11log

1log

1)log(

+

+

+

+

=

Denoting itit Xx log is log life expectancy and

it

itit

Yy log , we obtain:

i

iiiiit

x

shAy

))1()((1

1

log1

1log

1log

1log

1log

1

+

+

+

+

=

(5)

The last term in equation (5) shows that an increase in life expectancy will lead to a

significant increase in long-run income per capita when there are limited diminishing

returns (i.e., 1 is small) and when life expectancy creates a substantial externality

on technology (high ) and/or encourages significant increases in human capital (high ).On the other hand, when and are small and 1 is large, an increase in life

expectancy would reduce income per capita even in the steady state.

Given the modeling framework, the empirical strategy followed by Acemoglu and Johnson

(2007) basically is to estimate equations similar to (5), and compare the estimates to the

parameters in these equations. More specifically, a fixed effects panel regression method is

used to capture the impact of life expectancy on the following four variables: population,

number of births, GDP, and GDP per capita. The fixed effects model examines country

differences in intercepts, assuming the same slopes and constant variance across groups.

Fixed effects models use least squares dummy variable (LSDV), within effect, and between

effect estimation methods. Thus, ordinary least squares (OLS) regressions with dummies,in fact, are fixed effects models. Such models assist in controlling for unobserved


6/43

6

heterogeneity, when this heterogeneity is constant over time. This paper uses panel data

where the unit of observations are the countries and time (decennial data since 1930

through 2000), and thus may have group (i.e. country specific) effects, time effects, or

both. This constant heterogeneity is the fixed effect for the individual countries. This

constant can be removed from the data, for example by subtracting each individual's meansfrom each of his observations before estimating the model.

Adding an error term, the generalized version of the estimating equation becomes:

ittiitit xy +++= (7)

where y is log income per capita, i is a fixed effect capturing potential technology

differences and other time-invariant omitted effects (i.e. iA , ih , i , and iK or is ), t

incorporates time-varying factors common across all countries, andx is log life expectancy

at birth. The coefficient is the parameter of interest equal to ))1()((1

1

+

when

equation (5) applies. Including a full set of country fixed effects, the is, is important, sincethe country characteristics iA , ih , i ,and iK or is would be correlated with life expectancy

(or health). Also, many country-specific factors will simultaneously affect health and

economic outcomes. Fixed effects at least remove the time-invariant components of these

factors.

Prior to investigating the effect of life expectancy on income per capita, its effects on

population, total births, and total income are reported. The equations for these outcome

variables are identical to (7), with the only difference being the dependent variable.

Equation (7), however, while estimated as it is, may be beset with the potential omitted

variable bias and reverse causality problems. In that case the causal effects of life

expectancy on income per capita or population are misleading. In particular, in equation(7), typically the population covariance term Cov(xit, it+k) is not equal to 0, because even

conditional on fixed effects, health could be endogenous.

In Acemoglu and Johnson (2007), the endogeneity problem has been addressed by

exploiting the potentially-exogenous source of variation in life expectancy attributable to

global health innovations and interventions. Specifically, the first-stage relationship is:

itti

I

itit uMx +++= ~~ , where IitM is the instrument, termed as predicted mortality,

derived from the worldwide variations in the death rates from different diseases, and due to

disease specific interventions at different points in time. The key exclusion restriction of

Cov(xit, it) equal to 0 is assumed, where itis the residual from equation (7).

Similarly in this paper, the first stage relationship using the alternative instruments is

presented in the equation below:

ittiitit uVx +++= ~~' (8)

Where 'itV includes alternative instruments like immunization coverage, occurrences of

conflicts and natural disaster dummies, used separately or in combinations. t~ represents

the time fixed effect controlling for unobserved omitted variables that changes over time

but are constant across entities; i~ captures entity fixed effect controlling for unobserved

omitted variables that differ across countries.


7/43

7

3. ALTERATIVE ISTRUMETS AD TIME-LIE

Acemoglu and Johnson (2007) use a predicted mortality instrument based on the death

rates from different diseases and the global health intervention dates that contributed to

rapid declines in mortality from different diseases. The analysis mainly refers to the period1940-1980 for which the predicted mortality instrument is deemed to be suitable. However,

the construction of the instrument raises several issues:

1. The assigned global intervention dummies assume uniform health interventions acrosscountries, which is clearly subject to doubt. The predicted mortality instrument assumes

zero values after the global intervention dates (i.e. from 1960 onwards). This may

render usage of this instrument artificial, or at least unique and applicable to analysis

pertaining to that timeline only. It is then worth investigating the notion that

instrumented life expectancy led to a much higher population increase and led to

insignificant or negative impact on per capita GDP with alternative instruments and/or

timelines.2. The notion that epidemiological transition occurred only during the post World War II

era requires caution. Riley (2001, 2007), using a vast bibliography6 that enabled him to

track the historical life expectancy situation for the majority of the countries around the

world, presents the approximate date (year) for each country from when the respective

countries registered sustained gains in life expectancy (termed health transition). Out

of more than 150 countries studied, 31 countries started their health transitions before

19007, 49 countries between 1900 and 19408, and 70 countries after 1940s9.

3. Due to scarcity of data African countries were excluded from the Acemoglu andJohnson (2007) analysis. Worldwide cross-country data for the period 1960 onwards

are relatively more accessible. One could then undertake the study with a similarmodeling approach to quantify the impact of life expectancy on economic growth for

the period 1960- to date, by which time the major health innovations to reduce

infectious diseases had occurred. In that case, the task is to find suitable instruments to

explain life expectancy during that period and its potential impact on growth through

the channels of human capital accumulation, total factor productivity and population

growth.

6James C. Riley, Bibliography of Works Providing Estimates of Life Expectancy at Birth and of the

Beginning Period of Health Transitions in Countries with a Population in 2000 of at least 400,000 atwww.lifetable.de/RileyBib.htm (Last browsed on September 09, 2010) .7

Argentina, Australia, Austria, Belgium, Canada, Costa Rica, Cyprus, Czech Republic, Denmark, England

and Wales, Finland, France, Germany, Hungary, Iceland, Ireland, Italy, Japan, Luxembourg, Mexico,

Netherlands, New Zealand, Norway, Poland, Russia, Slovak Republic, Spain, Sweden, Switzerland,

Scotland, United States.8

Albania, Algeria, Bangladesh, Brazil, Bulgaria, Chile, China, Colombia, Cuba, Egypt, Estonia, Fiji, Ghana,Greece, Guatemala, Guyana, India, Indonesia, Jamaica, Jordan, Kenya, Latvia, Lithuania, Malaysia,

Mauritius, North Korea, Pakistan, Panama, Paraguay, Philippines, Portugal, Puerto Rico, Romania,

Singapore, South Africa, South Korea, Sri Lanka, Suriname, Syria, Taiwan, Trinidad and Tobago, Tunisia,

Turkey, Uganda, Ukraine, Uruguay, Venezuela, Vietnam, Yugoslavia.9 Eritrea, Ethiopia, Gabon, Gambia, Guinea, Guinea Bissau, Haiti, Honduras, Iran, Iraq, Kuwait, Laos,

Lebanon, Lesotho, Liberia, Libya, Madagascar, Malawi, Mali, Mauritania, Mongolia, Morocco,

Mozambique, Myanmar, Namibia, Nepal, Nicaragua, Niger, Nigeria, Oman, Papua New Guinea, Peru, Qatar,Rwanda, Saudi Arabia, Senegal, Sierra Leon, Solomon Islands, Somalia, Sudan, Swaziland, Tanzania,

Thailand, Togo, United Arab Emirates, Yemen, Zambia, Zimbabwe


8/43

8

The conjecture that the advent of antibiotics and vaccines in the 1940s or early 1950s (and

their subsequent worldwide application) allowed abruptly substantial gains in life

expectancy in general, and for low-income countries in particular, has been subject to

scrutiny. Widespread and effective dissemination of antibiotics and vaccines during the late

1940s until 1960 is doubtful, because this would have required developed health care andpublic health systems. These include physical infrastructure such as hospitals, health

centres, clinics and dispensaries as well as human capital in the forms of physicians and

nurses trained in Western medicine. Many of the countries that began sustained health

transitions in the 1940s and 1950s lacked such arrangements. According to Riley (2007)

if antibiotics and vaccines were in fact instrumental, one must imagine that they were

distributed outside the system, or mostly outside, such systems. In addition, ordinary

people must have learned quickly about these remedies, how to obtain them, and when and

how to use them, and they must also have been able to afford the new medications and

vaccines. Put that away, the proposition looks more doubtful (Riley, 2007, p.47).

The arrival of antibiotics (e.g. Penicillin and other antibiotics) in Latin America, Asia, and

Africa began in the late 1940s and early 1950s (Riley, 2007). Venezuela, for instance,

being a country with close contacts to the United States, started receiving antibiotics in

1950 (Rivora, 1998; cited in Riley, 2007). The question remains - did antibiotics actually

reach large proportions of the populations of Africa, Asia, and Latin America by 1950, or

even by 1960? Riley (2001, 2007) asserts that in most countries that began health

transitions between the 1890s and the 1920s or 1930s, mortality from diseases treatable

with antibiotics, such as tuberculosis and typhoid fever, was already much reduced by the

time the antibiotics became available.10 That was true also for the countries that began

health transitions before the 1890s.11 Thus the effects posited for antibiotics apply chiefly

to countries where bacterial infections remained leading cause of death, which would

include all the countries that initiated health transitions in the 1940s or later.

12

Antibiotics,and to much lesser degree sulfa drugs especially effective in treating wounds and sores, are

said to have allowed those countries to make rapid progress in controlling infectious

diseases.

In case of vaccines, widespread administering is slower. 13 Two of the most established

vaccines namely smallpox and typhoid were first introduced in 1796 and 1896 respectively.

However decades passed before usage was general in any large region in Africa, Asia, or

Latin America. The same delay applied to other vaccines. The date when a vaccine was

introduced in the West has little usefulness in determining when that vaccine was first used,

10

The list of countries in this category includes Hungary, Czech Republic, Slovak Republic, Spain, Argentina,Cyprus, Russia, Costa Rica, Estonia, Latvia, Lithuania, Ukraine, Cuba, Puerto Rico, Greece, Singapore,

Yugoslavia, Korea, Panama, Romania, Malaysia, Philippines, Bangladesh, Bulgaria, Chile, China, Fiji,

Ghana, Guyana, India, Indonesia, Jamaica, Mauritius, Pakistan, Portugal, Sri Lanka, Suriname, Taiwan,

Trinidad and Tobago, Tunisia, Uruguay, Albania, Brazil, Paraguay, South Africa, Turkey, Venezuela,

Vietnam, Algeria, Colombia, Egypt, Guatemala, Jordan, Kenya, Syria, Uganda (Riley, 2007).11

The list of countries in this category includes Denmark, France, Sweden, England, Norway, Canada,Belgium, Ireland, Australia, Netherlands, New Zealand. Mexico, Finland, Germany, Iceland, Switzerland,

Scotland, Italy, Luxembourg, Japan, Poland, United States, Austria (Riley, 2007).12

The list of countries in this category includes Bahrain, Lesotho, Papua New Guinea, Sierra Leon, Solomon

Islands, Afghanistan, Angola, Benin, Botswana, Burkina Faso, Cambodia, Cameroon, Central African

Republic, Chad, Comoros, Congo Republic, Cote d'Ivoire, Djibouti, Equatorial Guinea, Eretria, Ethiopia,

Gabon, Gambia, Guinea, Guinea Bissau, Haiti, Liberia, Madagascar, Malawi, Mali, Mauritania, Mongolia,

Nepal, Niger, Nigeria, Saudi Arabia, Somalia, Swaziland, Tanzania, Togo, Yemen, Zambia, Mozambique,Oman (Riley, 2007).13

The terms vaccination and immunization are used synonymously and interchangeably.


9/43

9

or first widely used, in the developing world, where, in any case, the medical infrastructure

was rarely developed enough to support mass immunizations. Vaccines with general or

limited usefulness against smallpox, typhoid fever, yellow fever, tuberculosis, tetanus,

plague, and whooping cough were introduced in the West before World War II, but only

smallpox vaccination seems to have been used widely in some parts of the developingworld before 1940s. After the war vaccines for polio (introduced in 1954), measles (1963),

mumps (1967), and rubella (1969) might have had substantial and immediate effects, but it

was not until the 1970s or even the 1980s before many of them were used widely. In the

absence of a detailed reconstruction of the timing and scale of vaccine use, it is difficult to

determine when they actually had a general effect. One exception was smallpox, which was

introduced in the 1970s and 1980s as part of the World Health Organizations Expanded

Program on Immunization (see, e.g Plotkin and Mortimer, 1994; Riley, 2007). In sum,

although antibiotics and vaccinations may have helped initiate declines in mortality rates in

many countries in the 1940s and 1950s; it seems likelier that their effect was felt somewhat

later and more weakly.

Nevertheless, there were significant survival gains across at least three decades, from the

1950s through the 1970s. While some of the improvements in health are the result of

overall social and economic gains, the major achievements were obtained by the specific

efforts to address major causes of disease and disability, such as providing better and more

accessible health services, introducing new medicines and other health technologies, and

fostering healthier behaviours through extensive campaigns and education. Nearly all of the

low-income countries that began health transitions between the 1890s and the 1920s or

1930s, before the availability of antibiotics or sulfa drugs, continued to add to those earlier

gains in the period 1970-2000 as well. But many countries that began transitions in the

1940s or later saw their gains falter in the 1980s in the face of HIV/AIDS, tuberculosis

(both drug resistant and non-resistant strains), resurgent malaria, unresolved problems withfecal disease, and other communicable diseases. The steps taken toward improved survival

among countries that initiated health transitions between the 1890s and the 1920s or 1930s

seem to have laid a better foundation for long-run gains than did the antibiotics and

vaccines deployed in the late 1940s and thereafter, or whatever other factors might be

posited as having mattered (Riley, 2007). Whatever the case, in the post epidemiological

transition period, one important element of the world-wide large scale health intervention

was the programme of immunization. Examples of effective public health programmes, not

necessarily hinging upon the national income level, exist to facilitate understanding the

determinants of the changes in population health (see for example, Levine et al., 2004;

Chandra, 2006). In this light, the cross-country vaccine adoption and implementation rates

by diseases may be a more relevant instrument in explaining exogenous variations in lifeexpectancy.

Another source of exogenous variation in life expectancy may be caused by epidemics,

wars, famines and other disastrous events. In this light, the cross-country occurrences of

major conflicts may potentially have contributed to the health outcomes. Therefore, for the

period 1960-2005 the alternative instruments to explain life expectancy may be divided

into two categories: public health interventions measured in terms of (or proxied by)

percentage of population covered under several types of vaccines (e.g. DPT, BCG,

Measles.); and natural disasters (e.g. occurrence of major conflicts, drought, and floods).

Post-war remarkable achievements in the world also include the agricultural (green)

revolution, perceived to be the outcome of the extensive research and development of high

yielding and highly resistant varieties of seeds, and leading to the increased availability of


10/43

10

food and nutrition across the world in general. This paper also highlights the link between

increased energy intake and life expectancy; and investigates how life expectancy

instrumented by calories per capita affects economic outcomes. The influences of these

factors will be discussed in detail below.

3.1. IMMUIZATIO PROGRAMMES

Immunization is a proven tool for controlling and eliminating life-threatening infectious

diseases and is estimated to avert over 2.5 million deaths each year. It has clearly defined

target groups; it can be delivered effectively through outreach activities; and vaccination

does not require any major lifestyle change (WHO, 2003). The Expanded Programme on

Immunization (EPI) launched by the World Health Organization (WHO) in 1974 increased

immunisation from five percent of all children to 80 percent in a span of thirty years

(Tangermann, 2007). This was mainly possible due to coordinated efforts from a coalition

of partners: governments, the United Nations Development Programme, UNICEF,

development agencies, the World Bank, the Rockefeller Foundation, Medecins sans

Frontires, and Rotary International.14. Since 2000, the GAVIhas been very successful at

re-focusing immunization activities globally. 15 For people in developing countries,

successful immunisation programmes save thousands of lives, and organisations including

UNICEF and the WHO are committed to making vaccines against measles, polio and other

serious diseases available to as many children as possible.

Immunization is one of the most cost-effective public health interventions, with

demonstrated strategies that make it accessible to even the most hard-to-reach and

vulnerable populations. It has eradicated smallpox, lowered the global incidence of polio

by 99% since 1988 and achieved dramatic reductions in diseases such as measles,

diphtheria, whooping cough (pertussis), tetanus and hepatitis B. The cost of fullyimmunizing a child with the six traditional EPI vaccines through routine health services

were estimated to be approximately 15 USD per child in the 1980s and approximately 17

USD per child in the 1990s. Thus, with an annual birth cohort of approximately 91.4

million in low-income countries, estimates of total immunization costs in 1998 were 1.123

billion USD (GAVI, 2001). Yet in almost 50 nations, 60 percent of the children are not

immunised. The result is three million children dying every year from diseases that are

entirely preventable, 30 million infants having no access to basic immunisation each year,

and a child in the developing world ten times more likely to die of a vaccine-preventable

death than a child in an industrialised nation. National income levels may have played a

minor role in this regard.

The most widely used vaccination in the world isBacille Calmette Guerin (BCG), which is

made of a live, weakened strain of mycobacterium bovis (a cousin of mycobacterium

tuberculosis, the TB bacteria). BCG is effective in reducing the likelihood and severity of

TB in infants and young children. Developed in the 1930s it still remains the only

vaccination available against tuberculosis. This vaccine is particularly important in areas of

14However, the programs could not have been successful without the involvement of political, religious and

community leaders, all of whose contribution amounted to what has been described as the greatest social

mobilisation effort in peace-time. (http://www.immunisation.nhs.uk/About_Immunisation/Around_the_world/The_Expanded_Program_on_Immunization_EPI; browsed in December, 2008)15

GAVI partners include governments in industrialized and developing countries, UNICEF, WHO, the Billand Melinda Gates Foundation, the World Bank (WB), NGOs, foundations, vaccine manufacturers, andtechnical agencies such as the US Centers for Disease Control and Prevention (CDC).


11/43


12/43


13/43


14/43

14

by 12 per thousand. Ghobarah, Huth, and Russett (2003) conduct a cross-national analysis

on the impact of conflict on public health and highlight the significant burden of death and

disability. Iqbal (2006) assesses this relationship between conflicts and population health

by analyzing cross-country data on summary measures of public health between 1999 and

2001 and asserts that interstate and intrastate conflict negatively influences the healthachievement of states and, therefore, the human security of their populations. Moreover his

analysis suggests that the negative effect of war on health is particularly intense in the short

term following the onset of a conflict. Neumayer and Plmper (2006) emphasise the gender

dimension of health impacts by providing the first rigorous analysis of the impact of armed

conflict on female relative to male life expectancy. On average, they find that over the

entire conflict period of interstate and civil wars women are more adversely affected than

men.

Glei et al. (2005) show how the national mortality and life expectancy estimates can be

significantly underestimated for countries that experience substantial war losses in a given

time period. They conducted a case study of Italy and their results indicate that estimates

currently available from the Human Mortality Database (HMD) greatly underestimate

period mortality during wartime among all Italian males, and may even underestimate

mortality among civilian males. Their work also highlights how failing to account for war

mortality presents problems in making inter-country mortality comparisons.

Ghobarah et al. (2003, 2004) discusses different channels through which public health is

affected by major conflicts in general, and civil wars in particular. These conflicts affect

the public health of a population at several levels. Firstly, the most observable direct effect

of conflict is the number of people killed and wounded. Conflicts render a large part of the

population exposed to hazardous conditions caused by events such as refugee flows and

movements of soldiers, giving rise to epidemics as the mobile groups act as vectors for

disease. The ability of war-torn societies to deal with new threats to public health isweakened by conflict and, consequently, the negative effect on health outcomes continues

to grow. During periods of conflict, resources get diverted toward military purposes, and

public health spending goes down, even as the public health needs of the population

escalate. This diversion of resources away from public health is often accompanied by a

decline in infrastructure that further impedes the ability of a society to handle public health

issues. Damage to the general infrastructure combines with damage to, or possible

destruction of, the health care infrastructure to make a society incapable of facing its

increasing public health challenges. Public health is affected not only by direct damage to

hospitals and other medical facilities, but also by damage to elements of the larger

infrastructure such as transportation, the water supply, and power grids. In addition to

infrastructural damage, food shortages and possibly famines often ensue during andimmediately after wars (Iqbal, 2006; Ghobarah et al., 2004). Additionally, major conflicts

often induce out-migration of highly trained medical professionals, and this loss of human

capital may not be reversed by their return or replacement until long after the wars end.

Armed conflicts often inflict displacement of population, either internally or as refugees;

and induce them to stay in crowded makeshift camps for years.17 Unhealthy food, polluted

water, improper sanitation, and dismal housing turn these sites into new vectors for

infectious disease (e.g. measles, acute respiratory disease, and acute diarrheal disease).

Peoples immune systems are drastically weakened due to malnutrition and stress. Children

17The Rwanda civil war generated 1.4 million internally displaced persons and another 1.5 million refugees

into neighboring Zaire, Tanzania, and Burundi (Ghobarah et al., 2004).


15/43

15

become especially vulnerable to infections.18 Tooles (2000) analysis of civil wars asserts

that the crude mortality rates among newly arrived refugees had been five to twelve times

above the normal rate. The spread of diseases from the refugee camps to the larger segment

of the population puts non-displaced populations at greater risk. Prevention and treatment

programmes already weakened by the destruction of health care infrastructure during warsbecome overwhelmed, especially if new strains of infectious disease bloom. For example,

efforts to eradicate Guinea worm, river blindness, and poliosuccessful in most countries-

-have been severely disrupted in states experiencing the most intense civil wars. Drug

resistant strains of tuberculosis develop and in turn weaken resistance to other diseases.

Conflicts reduce the pool of available resources for expenditures on the health care system,

and deplete the human and fixed capital of the health care system. For example, heavy

fighting in urban areas is likely to damage or destroy clinics, hospitals, laboratories, and

health care centres, as well as water treatment and electrical systems. Rebuilding this

infrastructure is unlikely to be completed quickly in the post-war period (Ghobarah et al.,

2004).

The Liberian civil war in 2003 may be a classic example of the way population health is

affected by war conditions. Cholera, among other diseases, spread at alarming rates as

internally displaced people with the disease headed towards the refugee camps outside the

city. The conflict situation made it impossible for either Liberian authorities or

international agencies to carry out the extensive process of water chlorination that would

halt further spread of cholera (Iqbal, 2006). Furthermore, afflicted people are unable to

access medical facilities because of the security situation. In September 2003, the World

Health Organization reported that only 32% of the Liberian population had access to clean

water, no more than 30% of the population had access to latrines, and there had been no

regular garbage collection in Monrovia since 1996. The SKD Stadium, the largest camp for

internally displaced people in Monrovia, housed about 45,000 people who cook and sleepin any sheltered spot they can find, in hallways and in tiny slots under the stadium seats,

and there were six nurses in the health centre for 400 daily patients (WHO 2003). In the

Sudan, two decades of conflict have exposed the population to diseases (such as yellow

fever), malnutrition, displacement of large groups, poverty, and famine. The Iraqi

population experienced near destruction of their health care system, previously one of the

best in the Middle East, during the first Gulf War. Public health in Iraq continued on a path

of steady decline during a decade of sanctions and internal repression, before the general

and health infrastructures were subjected to a second war. In 1993, Iraqs water supply was

estimated at 50% of pre-war levels (Hoskins 1997), and war-related post-war civilian

deaths numbered about 100,000 (Garfield and Neugut 1997).

The Conflict Variable (Dummy)

To assess the presence of conflict, this paper uses data from the Peace Research Institute of

Oslo (PRIO) Dataset on Armed Conflict (Gleditsch et al., 2002). These data measure

conflict according to both its intensity and its type including domestic and international

conflict. In the PRIO dataset the conflicts are categorised in terms of two levels of

intensity: Minor: between 25 and 999 battle-related deaths in a given year and War: at

least 1,000 battle-related deaths in a given year. This paper creates dummies for the

conflict variable only using the second category. This gives 467 occurrences of conflicts in

59 countries during 1950-2007 periods.

18For more please see e.g. Smallman-Raynor and Cliff (2004).


16/43

16

3.3. ATURAL DISASTER

Another type of exogenous variation in life expectancy is provided by the occurrences of

natural disasters (e.g. drought, floods, cyclones), which often led to famine and severe

epidemics, and nutritional shocks to a large segment of the population. The channelsthrough which these disastrous events impart a negative impact on life expectancy

intuitively resemble those depicted earlier for the major conflicts, which include the direct

effects in terms of deaths, malnourishment, reduced immune systems, crop destruction, loss

of productive land and water recourses, and increased morbidity; and indirect effects of

resource constraints, infrastructure destruction, stresses and social tensions. However, the

potential long run impacts are also highlighted in many studies. For instance, van den Berg,

Lindeboom and Portait (2007) use historical data for the period of 1845-48, which includes

the Dutch potato famine, and investigate whether exposure to nutritional shocks early in

life affects later-life mortality. During this period, all potato and grain crops in Europe

failed due to Potato Blight and bad weather conditions. They found that men who were

exposed to severe famine at least four months before birth and directly thereafter had aresidual life expectancy at age 50 that was significantly lower (a few years) than otherwise,

but that the mortality rate at earlier ages was not affected. Studies based on the Dutch

hunger winter under German occupation at the end of World War II (Ravelli et al., 1998;

Roseboom et al., 2001) and on Chinas great famine (Meng and Qian, 2006; Chen and

Zhou, 2007) indicated significant long-run effects on adult morbidity.

3.4. CALORIE PER CAPITA AD YIELD19

The world since the second half of the twentieth century has witnessed decades of high

productivity growth in crop agriculture led mainly by crop genetic improvement, includingboth the diffusion of existing high yielding varieties and the development of new varieties.

In the face of unprecedented population increases and limited natural resources per capita,

food production in most developing countries has increased over the decades, which

enabled them to avert large scale hunger and malnutrition leading to premature death. The

process was facilitated by the establishment of the extensive research infrastructure the

system of international agricultural research centres (IARCs) that were organized under the

rubric of the Consultative Group for International Agricultural Research (CGIAR). IARCs,

in collaboration of the national agricultural research systems (NARS) around the world,

developed and maintained genetic resource collections (gene banks) and fostered free

exchange of genetic resources between NARS and IARC programmes. Most IARC

programmes developed strong breeding programmes where advanced breeding lines andfinished varieties were developed. These materials were made available to NARS through

international testing and exchange programmes. The initial apparent success in crop

productivity was observed in wheat and rice, with shorter, early-maturing, and less

photoperiod sensitive plant types. These high yielding and highly resistant new plant types

led to the incipient popular concept of Green Revolution.

The achievements of the international and national collaborative research initiatives may be

measured in terms of the releases of new crop varieties, adoption of modern varieties, and

subsequent crop yield increases. There was a steady increase of varietal releases in almost

all of the crops through the 1960s into the late 1980s and 1990s. For instance, average

19The content of this section is largely based on Evenson and Gollin (2003).


17/43


18/43


19/43

19

The elasticity values for the BCG vaccine coverage range between 0.018 and 0.043 with an

average of about 0.030. When the regression is run for the group of countries that began

health transition before 1900, the coefficient is very low (0.009) and imprecise with the t-

value of only 1.21. This group includes countries are the economically developed nations

mostly from the West, where the BCG vaccination coverage matters less to influence lifeexpectancy. The elasticity values are higher for the low income groups (e.g. 0.043, 0.039,

and 0.042 in columns 1, 2, and 3). In all cases the F-statistics for the log of the BCG

coverage coefficient are more than 8 (except in column 6) suggesting this variable to be a

potential instrument for explaining life expectancy.

Relatively higher elasticity values are observed when the log of DPT coverage is used. The

elasticity values range from 0.021 to 0.045 with an average value of 0.031. The coefficient

is again low and imprecisely estimated from the countries that began the transition of life

expectancy before the twentieth century. The coefficients in general are more significant

and the corresponding F-statistics qualifies the variable to be a potential instrument. In the

bottom panel, similar and even higher elasticity values are observed when the log of

measles coverage is used to explain life expectancy. This variable appears to be the

strongest instrument of the three used.

The relatively higher elasticity values for the measles coverage is probably due to the fact

that this vaccine is usually administered to most children over 9 months and that most of

the children may have already been covered with DPT and BCG vaccines. Therefore, this

variable captures a wider impact of immunization. This is further clarified with the

estimates presented in Table 2. The first panel uses all the vaccine coverage as explanatory

variables in the regression specification. The resulting coefficients become statistically

insignificant for most samples, despite the coefficients jointly being significant in terms of

the model fit (F-Statistics). Also, the elasticity values appear to be unpredictable. This

indicates the collinear relationship between the variables. The bottom panel uses only DPTand measles as explanatory variables, which then provide better estimates, although in

many cases the elasticity values are statistically insignificant. Therefore, in the two-stage

least squares estimates used to estimate the impact of life expectancy on the major macro-

variables (i.e. population, GDP, and GDP per capita), the immunization variables are used

individually as instruments.

4.2. MAJOR COFLICTS AD LIFE EXPECTACY

Table 3 presents the first stage relationship between life expectancy and the occurrence of

major conflicts between 1950 and 2005. There are three panels in the table: the first paneluses yearly cross-country data on the life expectancy from the World Development

Indicators by the World Bank; the second (middle) panel uses data on the life expectancy

as found in the Demographic Year Book (2005). The reason for this was primarily due to

the fact that data frequency and data points differ between these two sources. The bottom

panel uses 5-yearly interval data on life expectancy from the World Development

Indicators (2007) and the conflict dummy. The conflict dummy in this later case is different

from the one used in the yearly data in this case a country would carry a value of 1 if it

faced at least one conflict in the preceding 4 years. For instance, the conflict occurrence

dummy for country X will have a value of 1 in time t if it suffers from at least one

conflict during t-4 to t. In doing so it assumes a lagged effect of wars and conflicts along

with the contemporary impacts. In each country group, the regression is applied only forthe countries that faced at least one major conflict during 1950-2005.


20/43

20

A negative relationship between conflicts and life expectancy is evident from Table 3. The

mean coefficients in the three panels are -1.28, -1.69, and -1.37 respectively, which

indicates that occurrence of conflicts are associated with the reduction of life expectancy

(in units of year). The coefficients for five of the country groups are insignificant at the 5%

level, but the coefficients are statistically significant when the regression is run with all thecountries and with WB low income countries. Using the life expectancy data from the

Demographic Year Book provided more statistically significant coefficients for these two

groups. The associated F-statistics indicate that occurrence of major conflict may not be a

strong instrument, since the values lie below 10 (see, Stock and Watson, 2006).

Nevertheless, this variable is used in the 2SLS estimates as a robustness check.

4.3. ATURAL DISASTERS AD LIFE EXPECTACY

Table 4 reports the impact of natural disasters on life expectancy for different samples of

countries. After controlling for country and time fixed effects none of the coefficients is

statistically significant, suggesting no impact of floods and drought occurrence on the

outcome of national level life expectancy. Therefore, natural disaster variables are dropped

off the instrument list in the 2SLS estimates.

4.4. LIFE EXPECTACY ISTRUMETED BY CALORIE PER CAPITA AD YIELD

Table 5 shows the first stage calorie per capita - life expectancy and yield life expectancy

relationship. Controlling for the fixed country and time effects, the estimates show a very

significant and robust relationship. A 1 percent increase in calories per capita increases life

expectancy in the range of 0.16 0.28 percent, with a mean elasticity value of 0.24. The t-

values indicate that the coefficients are significant at the 1% level even when the standarderrors are robust, and that corresponding F-statistics are large enough to safely consider this

variable as an instrument. The bottom panel in Table 5 shows the impact of increases in

yield on life expectancy. The yield variable here is the average of the yields (kg/hectare) of

the ten major crops weighted by the area harvested for the corresponding crops. The

elasticity values range from 0.03 to 0.07 and half of them could not be estimated precisely

as evident from the robust standard errors and the t-values. Therefore, the subsequent

instrumental variable estimates capturing the impact of life expectancy increases on

population, GDP, and per capita income use calories per capita as the instrument, along

with the immunization and conflict variables.

4.5. LIFE EXPECTACY ISTRUMETED BY CALORIE PER CAPITA AD

IMMUIZATIO COVERAGE

Table 6 reports the first stage relationship while more than one instrument is used to

explain life expectancy. The upper panel uses DPT coverage and calorie per capita as

instruments and the lower panel uses measles and calorie per capita as instruments. The

associated t-values, joint tests of significance (F-statistics) indicate that these instruments

may perform very well for many of the country groups even when used in combinations.


21/43

Table 1: First Stage Estimates: Life Expectancy and Immunization Coverage

1 2 3 4 5

All Countries WB Low

incomecountries

WB Lower

MiddleIncome

countries

Countries that

began healthtransition after

1940

Countries that

began healthtransition between

1900 and 1940

Co

bet

be

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 1

Dependent Variable: Log Life Expectancy

Log BCG 0.029 0.043 0.039 0.042 0.02

Standard error (Robust) 0.007 0.013 0.013 0.015 0.007

t-value 4.29 3.36 2.97 2.85 2.93

F-statistics 18.40 11.29 8.83 8.14 8.57

Number of observation 776 262 272 351 232

Number of countries 158 53 55 67 44


Log DPT 0.028 0.045 0.04 0.039 0.021


t-value 5.24 3.69 2.82 3.33 3.08

F-statistics 27.44 13.62 7.98 11.10 9.48




Log Measles 0.03 0.046 0.04 0.058 0.019

Standard error (Robust) 0.006 0.014 0.015 0.016 0.005 t-value 5.28 3.38 2.74 3.60 4.10

F-statistics 27.83 11.44 7.51 12.97 16.77



Note: Regressions with a full set of year and country fixed effects. Robust standard errors, adjusted for clustering by country are


22/43

Table 2: First Stage Estimates: Life Expectancy and Immunization Coverage (Combined Effect)

1 2 3 4 5


incomecountries

WB Lower

MiddleIncome

countries

Countries that


1940

Countries that

began healthtransition between

1900 and 1940

Coun

begtra

bef

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 198


Log BCG 0.001 -0.035 0.03 0.013 -0.019 0

t-value 0.13 -0.95 2.35 0.86 -0.57

Log DPT 0.02 0.05 0.02 0.02 0.02

t-value 2.27 2.62 0.94 1.87 0.91

Log Measles 0.02 0.03 0.22 0.04 0.02

t-value 2.08 1.33 1.74 2.28 0.99

Joint test of significance 9.07 4.67 4.12 4.41 4.56




Log DPT 0.018 0.037 0.028 0.021 0.011 0

t-value 2.19 2.44 1.69 1.95 0.75

Log Measles 0.02 0.02 0.02 0.04 0.01

t-value 2.50 1.41 1.74 2.76 1.14

Joint test 16.28 7.48 4.46 7.23 7.98





23/43

Table 3: First Stage Estimates: Life Expectancy and Major Conflicts

1 2 3 4

All Countries WB Low income

countries

WB Lower

Middle Incomecountries

Countries that


1940

Countri

healthbetwee

1950-2005 1950-2005 1950-2005 1950-2005 19

Log Life Expectancy (World Development Indic

Conflict Dummy -1.43 -2.196 -0.496 -1.165 -

Standard error (Robust) 0.608 1.075 0.595 0.69

t-value -2.35 -2.04 -0.83 -1.69

F-statistics 5.53 4.17 0.69 2.85

Number of observation 1295 527 488 617

Number of countries 64 28 26 33

Log Life Expectancy (Demographic Yearbook

Conflict Dummy -2.182 -2.886 -0.618 -1.396 -


t-value -3.55 -2.74 -1.11 -1.70

F-statistics 12.61 7.50 1.22 2.90



Log Life Expectancy (5-year interval pan

Conflict Dummy -1.606 -2.157 -0.943 -0.99 -


t-value -2.50 -2.82 -0.83 -1.96

F-statistics 6.26 7.94 0.68 3.85





24/43

Table 4: First Stage Estimates: Life Expectancy and atural Disasters

All Countries WB Lowincome

countries

WB LowerMiddle

Income

countries

Countriesthat began

health

transition

after 1940

Countriesthat began

health

transition

between

1900 and

1940

1980-2000 1980-2000 1980-2000 1980-2000 1980-2000

Dependent Variable: Log Life Expectan

Dummy for Flood Occurrence -0.346 -0.456 0.01 -0.163 -0.363


t-value -1.43 -1.07 0.02 -0.31 -0.89 Number of observation 2711 574 721 819 665


1970-2000 1970-2000 1970-2000 1970-2000 1970-2000

Dependent Variable: Log Life Expectan

Dummy for Drought Occurrence -0.086 -1.139 -0.187 0.364 -1.308


t-value -0.21 -1.56 -0.45 0.56 -1.82

Number of observation 1366 236 390 417 329 Number of countries 108 29 36 42 26



25/43

Table 5: First Stage Estimates: Life Expectancy Agricultural Revolution

1 2 3 4 5


incomecountries

WB Lower

MiddleIncome

countries

Countries that


1940

Countries that

began healthtransition

between 1900 and

1940

Count

begantran

befor

1960-2000 1960-2000 1960-2000 1960-2000 1960-2000 1960


Log Calorie Per Capita 0.245 0.239 0.25 0.213 0.281 0.

Standard error (Robust) 0.043 0.07 0.062 0.06 0.068 0.

t-value 5.72 3.43 4.04 3.56 4.13 3

F-statistics 32.72 11.77 16.30 12.68 17.03 11

Number of observation 934 273 268 378 285 2Number of countries 129 39 38 54 40 2


Log Yield 0.032 0.03 0.037 0.038 0.063 0.

Standard error (Robust) 0.012 0.027 0.019 0.019 0.021 0.

t-value 2.59 1.09 1.90 2.02 3.05 2

F-statistics 6.72 1.19 3.61 4.09 9.31 7

Number of observation 1154 334 319 458 313 2

Number of countries 181 51 54 68 47 2



26/43

Table 6: First Stage Estimates: Instruments Used in Combinations

1 2 3 4 5 6

All

Countries

WB Low

incomecountries

WB Lower

MiddleIncome

countries

Countries that


1940

Countries that

began healthtransition

between 1900and 1940

Countrie

began htransitio

before 1

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 1980-20


Log DPT 0.020 0.028 0.035 0.030 0.015 0.008 Standard Error 0.050 0.011 0.013 0.012 0.007 0.005

t-value 4.050 2.490 2.610 2.490 2.080 1.640

Log Calorie per capita 0.167 0.241 0.127 0.243 0.082 -0.026 Standard Error 0.052 0.085 0.660 0.081 0.040 0.030

t-value 3.220 2.840 1.910 3.000 2.030 -0.800



Joint test (F-statistics) for log

DPT & log Calorie/capita

14.82 7.50 7.75 7.65 4.66 1.36


Log Measles 0.025 0.033 0.054 0.057 0.014 0.015 Standard Error 0.005 0.010 0.017 0.014 0.004 0.006

t-value 4.990 3.250 3.190 4.150 3.510 2.370

Log Calorie per capita 0.172 0.253 0.126 0.271 0.074 -0.031 Standard Error 0.053 0.082 0.060 0.081 0.039 0.030

t-value 3.23 3.100 2.030 3.250 1.900 -1.040



Joint test (F-statistics) for log

Measles & log Calorie/capita

16.79 11.76 9.14 12.10 8.59 3.00



27/43

27

5. MAI RESULTS: 2SLS ESTIMATES

This section presents the two stage least square estimates of the impact of life expectancy

on the three major macro variables: Population, GDP, and per capita GDP. The life

expectancy variable is instrumented using different instruments separately and incombination. All the resulting elasticities are reported to assess the potential impact of life

expectancy on the variable of interest. Table 7, Table 8, and Table 9 report the results,

where each table consists of eight panels and therefore continues in multiple pages to

accommodate large volume of results. When more than one instrument is used (i.e. panel 7

and panel 8 in Table 7, Table 8, and Table 9) four additional statistics are reported:

Kleinbergen-Paap LM Statistics of under-identification test; Kleinbergen-Paap Wald F-

statistic of weak identification test; Hansen J-statistics; and p-value associated with the

Hansen J-Statistic.20

5.1. LIFE EXPECTACY AD POPULATIO

The instrumented life expectancy by the predicted mortality instrument in Acemoglu and

Johnson (2007) suggests that there is a large, and relatively precise and robust effect of life

expectancy on population. However, we do not observe similar result using alternative

instruments, timelines, and country classifications reported in Table 7. Large elasticity

values are observed in column 1 where the regression is run on all the countries for which

data are available. One percent increase in life expectancy leads to the increase of

population in the range of 1.06 to 2.53 while using immunization coverage and calorie per

capita, individually and in combinations, as instruments. But this is not the case when we

use major conflict as the instrument for life expectancy. The impact of life expectancy on

population cannot be measured precisely. The list of countries in this case includes only 60countries that faced at least one major conflict. Although in column 3 (WB Lower Middle

Income countries), column 5 (countries that began health transition between 1900 and

1940), column 6 (countries that began health transition before 1900, column 7 (panel of 59

countries), and column 8 (initial middle and poor countries) the elasticity values are high

while using immunization and calorie per capita as instruments, most of them are

statistically insignificant. Strikingly, for the WB low income countries (column 2) and the

countries that health began health transition began after 1940, no discernible impact of life

expectancy on population is observed whatever instrument is used. To summarize, elastic

population response due to increase in life expectancy is observed when immunization

programme or calorie intake is used as instruments, with most of them imprecisely

measured; the population response is absent while running the regression for low income

countries; and using the conflict dummy as instrument tends to produce negative impact of

life expectancy on population, but all the coefficients are statistically insignificant.

Therefore, the elastic response of population to the increase of life expectancy is not

obvious, as found in Acemoglu and Johnson (2007).

20The Hansen J-statistics are reported by partialling out the country-clusters due to the existence of a rank-

deficient estimate of the covariance-matrix of orthogonality conditions. This is common when the cluster

option is used, as well as, in the presence of singleton dummy variables. However, according to the Frisch

WaughLovell (FWL) theorem (Frisch and Waugh, 1933; Lovell, 1963) the coefficients estimated for a

regression in which some exogenous regressors are partialled out from the dependent variable, theendogenous regressors, the other exogenous regressors, and the excluded instruments will be the same as the

coefficients estimated for the original model (Baum et al., 2007).


28/43

Table 7: 2SLS Estimates: Impact of Life Expectancy on Population

1 2 3 4 5


income

countries

WB Lower

Middle

Income

countries

Countries

that began

health

transition

after 1940

Countries

that began

health

transition

between 1900

and 1940

Cou

that

hea

tran

befo

Page 1 of 3 for Table 7Dependent Variable: Log Population

Panel 1: Life expectancy instrumented by BCG Vaccin

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 198

Log of life expectancy 2.467 -0.156 1.995 0.772 2.813 4.86

Standard error (Robust) 1.153 0.440 1.205 0.841 2.209 3.10

t-value 2.140 -0.360 1.660 0.920 1.270 1.56



Dependent Variable: Log Population

Panel 2: Life expectancy instrumented by DPT Vaccin

Log of life expectancy 2.536 0.137 2.586 -0.083 3.351 3.44


t-value 3.330 0.380 1.830 -0.190 1.620 0.67



Dependent Variable: Log PopulationPanel 3: Life expectancy instrumented by Measles Vacc

Log of life expectancy 1.815 -0.001 1.651 0.221 1.969 0.70


t-value 2.790 0.000 1.360 0.890 1.720 0.20




29/43

1 2 3 4 5


income

countries

WB Lower

Middle

Income

countries

Countries

that began

health

transition

after 1940

Countries

that began

health

transition


Cou

that

hea

tran

befo


Panel 4: Life expectancy instrumented by Major Conf

1950-2004 1950-2004 1950-2004 1950-2004 1950-2004 195

Log of life expectancy 0.058 -0.242 1.224 0.069 0.530


t-value 0.100 -0.350 0.550 0.060 0.920




Panel 5: Life expectancy instrumented by Major Conflict Dummy

Log of life expectancy -0.549 0.835 -1.893 0.690 -1.206


t-value -0.560 1.170 -0.570 0.500 -0.900




Panel 6: Life expectancy instrumented by Calorie p

1960-2000 1960-2000 1960-2000 1960-2000 1960-2000 196

Log of life expectancy 0.678 0.218 0.057 0.179 0.558 4.2


t-value 1.520 0.450 0.060 0.290 0.740 2.80




30/43

1 2 3 4 5


income

countries

WB Lower

Middle

Income

countries

Countries

that began

health

transition

after 1940

Countries

that began

health

transition


Cou

that

hea

tran

befo


Panel 7: Life expectancy instrumented by DPT vaccine coverage

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 198

Log of life expectancy 1.327 0.249 1.652 -0.102 2.750 0.98


t-value 1.980 0.720 1.280 -0.290 1.570 0.25



Under Identification LM Test* 17.79 8.73 6.494 10.19 6.967 3.93Weak Identification test** 9.36 4.49 5.98 4.72 3.473 1.15

Hansen J-test Statistics *** 5.064 1.91 0.00 0.095 0.625 0.87

p-value, Hansen J-test 0.024 0.17 0.995 0.758 0.43 0.35


Panel 8: Life expectancy instrumented by Measles vaccine coverag

Log of life expectancy 1.057 0.146 1.243 0.054 1.944 -0.1


t-value 1.790 0.550 1.050 0.230 1.410 -0.0



Under Identification LM Test* 19.69 8.92 6.13 12.36 7.46 6.15

Weak Identification test** 11.36 6.097 6.92 7.11 6.53 2.37

Hansen J-test Statistics *** 1.12 6.465 0.06 0.024 0.002 0.80

p-value, Hansen J-test 0.29 0.011 0.81 0.87 0.967 0.37


* Kleinbergen-Paap LM statistic; ** Kleinbergen-Paap Wald F statistic; ***Partialling out the country dummies


31/43

31

5.2. LIFE EXPECTACY AD GDP

The 2SLS estimates for the impact of life expectancy on GDP are remarkably different

from those that we observe in Acemoglu and Johnson (2007). The general finding in

Acemoglu and Johnson (2007) suggests an inelastic or negative impact of life expectancyon GDP, or at least a smaller impact of life expectancy on GDP than on population.

Furthermore most of the coefficients are statistically insignificant. This scenario is reversed

if alternative instruments are used. Table 8 presents the impact of life expectancy on GDP

using different instruments, different country groups, and varying timelines. In most of the

cases the GDP response is highly elastic and in many cases the coefficients are precisely

estimated with small robust standard errors. The exceptions are observed only when the

regressions are run on country groups comprised of mostly developed nations.

For example, in column 1 of Table 8, when the regression is run using all country data the

elasticity values are 1.06, 2.02, and 2.22 when immunization coverage is used as an

instrument; 4.72, 8.44, and 4.67 when major conflict is used as an instrument (and appliedonly to countries affected by major conflict); 4.67 with calories per capita as an instrument

for life expectancy; 3.51 and 3.72 when life expectancy is instrumented by both calorie per

capita and immunization coverage. All of the coefficients are significant except the first

one. The estimated impact of life expectancy on GDP is also robust and significant for the

WB low-income countries. Column wise, the highest elasticity values are observed for

countries that began health transitions between 1900 and 1940. These are the countries,

according to Riley (2007) that continued to perform better during the second half of the

twentieth century into the 1980s and 1990s. The countries that began health transition after

the 1940s lost some of the life expectancy gains of the 1980s and 1990s; and lower

elasticity values are observed for these countries when immunization coverage is used as an

instrument.


32/43

Table 8: 2SLS Estimates: Impact of Life Expectancy on GDP

1 2 3 4 5


incomecountries

WB Lower

MiddleIncome

countries

Countries

that beganhealth

transition

after 1940

Countries

that beganhealth

transition

between 1900

and 1940

C

thh

tr

b

Page 1 of 3 for Table 8Dependent Variable: Log GDP

Panel 1: Life expectancy instrumented by BCG Vacc

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 19

Log of life expectancy 1.062 2.311 2.590 0.493 3.905 -0

Standard error (Robust) 1.404 1.015 1.749 1.754 3.282 5

t-value 0.760 2.280 1.480 0.280 1.190 -0



Dependent Variable: Log GDP

Panel 2: Life expectancy instrumented by DPT Vacc

Log of life expectancy 2.019 2.436 0.239 1.186 5.360 6


t-value 2.010 2.250 0.140 1.060 1.730 0



Dependent Variable: Log GDPPanel 3: Life expectancy instrumented by Measles Va



t-value 2.120 2.690 0.640 0.440 2.280 -0




33/43

1 2 3 4 5


income

countries

WB Lower

Middle

Income

countries

Countries

that began

health

transition

after 1940

Countries

that began

health

transition


C

th

h

tr

b


Panel 4: Life expectancy instrumented by Major Con

1950-2004 1950-2004 1950-2004 1950-2004 1950-2004

Log of life expectancy 4.722 3.734 8.774 4.483 3.598


t-value 1.880 2.090 1.000 1.740 0.860




Panel 5: Life expectancy instrumented by Major Conflict Dumm

Log of life expectancy 8.436 6.159 7.187 7.243 4.426


t-value 1.970 2.160 1.100 1.440 1.630




Panel 6: Life expectancy instrumented by Calorie

1960-2000 1960-2000 1960-2000 1960-2000 1960-2000 19



t-value 4.570 2.270 2.650 2.500 3.560 2




34/43

1 2 3 4 5


income

countries

WB Lower

Middle

Income

countries

Countries

that began

health

transition

after 1940

Countries

that began

health

transition


C

th

h

tr

b


Panel 7: Life expectancy instrumented by DPT vaccine coverag

1980-2000 1980-2000 1980-2000 1980-2000 1980-2000 1



t-value 2.950 2.300 1.130 1.460 2.460 0



Under Identification LM Test* 17.79 8.73 6.49 10.19 6.97 3

Weak Identification test** 9.37 4.49 5.98 4.72 3.47 1

Hansen J-test Statistics 5.06 0.497 2.20 4.57 1.16 4

Hansen J-test p-value 0.03 0.48 0.14 0.033 0.28 0


Panel 8: Life expectancy instrumented by Measles vaccine covera



t-value 3.210 2.480 1.740 1.330 2.730 -0


Number of countries 116 38 35 53 38 2Under Identification LM Test* 19.69 8.92 6.13 12.36 7.46 6

Weak Identification test** 11.36 6.097 6.92 7.11 6.53 2

Hansen J-test Statistics *** 1.681 0.347 1.14 3.69 1.15 3

p-value, Hansen J-test 0.195 0.556 0.29 0.06 0.28 0


* Kleinbergen-Paap LM statistic; ** Kleinbergen-Paap Wald F statistic; ***Partialling out the country dummies.


35/43

35

5.3. LIFE EXPECTACY AD GDP PER CAPITA

The estimated impact of life expectancy on GDP per capita is summarized in Table 9

below, and presented in detail in Table 10. The figures are the elasticity values, i.e. thecoefficients of log life expectancy, after controlling for country and time fixed effects,

using instrumental variables, and different country groups. The statistical significance of

the coefficients (as indicated by asterisk) is based on robust standard errors.

Table 9: Impact of Life Expectancy on GDP per Capita: Summary of 2SLS estimates

All

Countries

WB Low

income

countries

WBLower

Middle

Incomecountries

Countriesthat began

health

transitionafter 1940

Countriesthat beganhealth

transition

between1900 and1940

Countriesthat beganhealth

transition

before1900

Panel of

59

countries

InitialMiddle

Income

and Poorcountries

Life expectancy

Instrumented by

Impact of Life Expectancy on GDP per Capita

(the elasticity values)

BCG coverage -1.11 2.41* 1.29 -0.26 1.22 -5.16 0.46 -0.51

DPT coverage -0.33 2.23 * -1.62 1.28 2.03 3.93 -2.08 -1.55

Measles coverage 0.70 2.13* -0.04 0.21 4.37* -2.80 1.47 3.80

Conflict Dummy (1) 4.59* 3.80* 7.30 4.24 3.22 2.30 7.51

Conflict Dummy (2) 8.93* 5.28* 8.24 6.56 5.28 8.17 11.20

Calorie per capita 3.99* 3.11* 4.31* 3.91* 4.40* 4.71 2.11* 4.36*

DPT coverage andCalorie per capita

2.18* 2.53* 0.63 1.18 6.50* 2.68 0.04 0.16

Measles coverage &calorie per capita

2.67* 2.39* 1.99 0.75 7.44* -2.03 4.65* 6.69

Note: (*) beside the values indicates that the coefficient of log life expectancy is significant at 5% level.

The estimated coefficients closely follow the pattern of the effects of life expectancy on

population and GDP. We find a positive impact on GDP per capita in cases where theimpact on GDP is larger (more positive) than on population. A total of 62 coefficients are

reported in Table 9, out of which 44 take a value of more than one, seven coefficients lie

between 0 and 1, and 11 coefficients are negative indicating that an increase in life

expectancy reduces GDP per capita. However, all the inelastic and negative coefficients are

statistically insignificant, whereas 22 out of 44 coefficients with an elasticity value larger

than one are statistically significant at the 5% significance level. All the coefficients for the

WB low income countries are significant and show that a 1% increase in life expectancy

increases GDP per capita by 2.13 to 5.28%. While the pattern of impact varies across

country groups and instruments, the reported coefficients suggest a large positive impact of

life expectancy on GDP per capita in a large number of specifications, especially for the

group of countries at the lower range of income.


36/43

Table 10: 2SLS Estimates: Impact of Life Expectancy on GDP Per Capita

1 2 3 4 5


incomecountries

WB Lower

MiddleIncome

countries

Countries

that beganhealth

transition

after 1940

Countries

that beganhealth

transition

between 1900

and 1940

Cou

thathea

tran

befo

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 198

Page 1 of 3 for Table 10 Dependent Variable: Log GDP Per CapitaPanel 1: Life expectancy instrumented by BCG Vaccin

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 198

Log of life expectancy -1.105 2.408 1.294 -0.255 1.220 -5.1


t-value -0.570 2.340 0.940 -0.120 0.570 -0.9



Dependent Variable: Log GDP Per Capita

Panel 2: Life expectancy instrumented by DPT Vaccin

Log of life expectancy -0.332 2.228 -1.618 1.281 2.030 3.92


t-value -0.310 2.300 -0.960 1.010 0.930 0.41



Dependent Variable: Log GDP Per CapitaPanel 3: Life expectancy instrumented by Measles Vacc

Log of life expectancy 0.703 2.134 -0.041 0.210 4.367 -2.7


t-value 0.670 2.570 -0.020 0.230 1.940 -1.2




37/43

1 2 3 4 5


income

countries

WB Lower

Middle

Income

countries

Countries

that began

health

transition

after 1940

Countries

that began

health

transition


Cou

that

hea

tran

befo

1980-2004 1980-2004 1980-2004 1980-2004 1980-2004 198

Page 2 of 3 for Table 10 Dependent Variable: Log GDP Per

German World War

Documents

Transcript of German World War