Presented to the School of Economics

De La Salle University - Manila

1st Term, A.Y. 2019-2020

In partial fulfillment
of the course
In ECONMET V24

The Determinants of Life Expectancy: A Cross-Country Multiple Linear Regression Analysis

Submitted by:
Carlos, Eugenio III B.
Reyes, Juan Alfonso T.
Tiongco, Liam R.

Submitted to:
Dr. Kris Francisco

Submitted on:
December 16, 2019
 Table of Contents

I. Introduction

A. Background of the Study

B. Research question

C. Motivation

D. Significance of the Study

II. Literature Review

III. Methodology

A. Framework

B. Estimation

C. Data Sources

IV. Results and Discussion

V. Conclusion

VI. References

VII. Appendix
 INTRODUCTION
Background
For many years now, people of different fields and disciplines have been studying the variables that
affect a person’s life expectancy to address man’s desire for a long, healthy life. In the OECD Glossary of
Statistical Terms (2007), life expectancy is defined as the average number of years that a person at that age
can be expected to live. The most commonly used indicator for life expectancy is life expectancy at birth
(LEB), which specifically measures the average age of death expected for people who were born in a certain
year, assuming that the mortality patterns during that year are constant (Suzuki, Khokhar, & Fantom, 2019).
With the presence of more advanced technology, medical resources, and more stable political environments,
we evaluate regressors that could affect life expectancy in modern times.
Research Question
Life expectancy varies from year to year, country to country. The researchers seek to find a
common denominator among the discrepancies. With that, we aim to answer the question “What are the
economic and socio-economic determinants of life expectancy?”
Motivation
The objective of the study is to determine the variables that affect a country's life expectancy rate,
both positively and negatively, and if the relationships we hypothesized are valid. The primary motivation
of our analysis is to create an unbiased model of life expectancy that will validate and contribute to what is
currently known about life expectancy and the factors that affect it using the classical regression model.
Significance of the Study
This research can help extend the life expectancy in a country. Given that the findings in our model
are significant, we can determine variables that we should invest in to increase life expectancy, and variables
that should be avoided which decrease life expectancy. Furthermore, this research could be of significant
contribution to government institutions and policy makers which are trying to maximize the life expectancy
and welfare of the country.

LITERATURE REVIEW
The Preston Curve
The Preston Curve was created by Samuel H. Preston in 1975, signifying the relationship between
income and life expectancy. The basic hypothesis is that there is a causal relationship between income and
life expectancy. With that, we can expect higher-income countries to have a higher average life expectancy
age (Dhongde, 2016).
Education and Life Expectancy at Birth
The relationship between Education and Life expectancy has caught the attention of researchers
in recent years. According to Rogot, Sorlie, and Johnsons (1992), there is a strong relation between
Education and Life Expectancy. Another research mentioned that there is a causal relationship between
the characteristics of a person who desires more education and a longer life expectancy (Kaplan, Spittel,
Zeno, n.d.).
Health Expenditure Per-Capita and Life Expectancy at Birth
Health Expenditure Per-Capita is the amount spent by a country's government on public health
services per person and for collective medical services (OECD, n.d.).. Intuitively, higher health
expenditure per capita means more money is allocated for health, which in return improves the average
health of a person in a given country. The better health condition, the higher the life expectancy.
Unemployment Rate and Life Expectancy at Birth
The unemployment rate is defined as the portion of the labor force that is without a job, expressed
as a percentage (Chappelow, 2019). In 2016, Singh and Siahpush conducted a study that looked into the
association between the rate of unemployment and life expectancy, stating that unemployment is a strong
determinant for various health outcomes. Since employment is an indication of income, then it can be
assumed that when the unemployment rate is low, more people have access to basic healthcare necessities,
which may increase life expectancy at birth.
Primary Care Physician Supply and Life Expectancy
Some studies found that the supply of physicians has a positive effect on life expectancy. In 2019,
Basu et al. conducted a study on the physician supply in the United States and how it affected life
expectancy. They found that every ten additional primary care physicians per 100,000 population was
associated with a 51.5-day increase in life expectancy.
Using Multiple Linear Regression Model in Modelling Life Expectancy
Since life expectancy is a measure of how long an individual is expected to live—and given that
many factors affect the well-being of people—a multiple linear regression that would explain life
expectancy through explanatory variables will produce an accurate model. In 2000, Chen and Ching
constructed a model that explained life expectancy at birth. After refining the model, they ended up with
nine regressors, with an adjusted 𝑅2 value of 0.8690. It was found that their model predicted life expectancy
accurately for countries not included in their data set.

METHODOLOGY
Framework
For this study, the dependent variable used is the life expectancy at birth (lifexp). The regressors
we used are the natural logarithm of real GDP per capita measured in constant 2011 international dollars,
ln(rgdppc); the unemployment rate as estimated by the International Labor Organization,𝑅𝑅𝑅𝑅𝑅; the
natural logarithm of current health expenditure per capita measured in current international dollars
ln(𝒉𝒉𝒉𝒉𝒉); the number of physicians per 1,000 people, 𝑅ℎ𝑅𝑅; and the primary completion rate
measured as a percentage of the relevant age group, pri. The hypothesized relationships between the
regressand and the regressors can be found in table 1. The hypothesized population regression function is
expressed as follows: 𝑅𝑅𝑅𝑅𝑅𝑅 = 𝑅0 + 𝑅1 𝑅𝑅(𝑅𝑅𝑅𝑅𝑅𝑅) + 𝑅2 𝑅𝑅(ℎ𝑅𝑅𝑅𝑅𝑅) + 𝑅3 𝑅𝑅𝑅𝑅𝑅 +
𝑅4 𝑅ℎ𝑅𝑅 + 𝑅5 𝑅𝑅𝑅 + 𝑅.
Estimation
This study utilizes the ordinary least squares method of estimating the parameter slope coefficients
in a multiple regression model. In order for this method to provide the most efficient, unbiased estimates,
Data Sources
The study makes use of cross sectional data manually searched and compiled from databases of the
World Bank Databank. The data being used in this study is from 2015, which was the year chosen as it is
the most recent year with the most observations overall for the variables. Data on 217 countries was
collected for this study but due to a number of countries having missing observations for the specified
variables for that year, many countries had to be dropped from the study. Hence, this study will be analyzing
data from the 54 countries that had data available for 2015.

RESULTS AND DISCUSSION

Results of the Regression
Using the statistical software STATA and the data presented in the appendix, the following
estimated regression equation was obtained:
̂
𝒉𝒉𝒉𝒉𝒉𝒉 = 𝒉𝒉. 𝒉𝒉𝒉𝒉 − 𝒉. 𝒉𝒉𝒉𝒉𝒉𝒉(𝒉𝒉𝒉𝒉𝒉𝒉) + 𝒉. 𝒉𝒉𝒉𝒉𝒉𝒉(𝒉𝒉𝒉𝒉𝒉)
− 𝒉. 𝒉𝒉𝒉𝒉𝒉𝒉𝒉𝒉𝒉 − 𝒉. 𝒉𝒉𝒉𝒉𝒉𝒉𝒉𝒉 + 𝒉. 𝒉𝒉𝒉𝒉𝒉𝒉𝒉
Based on the intercept, the predicted life expectancy at birth of an individual from a country when
𝒉𝒉(𝒉𝒉𝒉𝒉𝒉𝒉) = 𝒉, 𝒉𝒉(𝒉𝒉𝒉𝒉𝒉) = 𝒉, 𝒉𝒉𝒉𝒉𝒉 = 𝒉, 𝒉𝒉𝒉𝒉 = 𝒉, and 𝒉𝒉𝒉=0 is approximately
39 years, which is somewhat low. However, it is not sensible to interpret the intercept this way given the
model, especially since there is no real values for 𝒉𝒉𝒉𝒉𝒉 and 𝒉𝒉𝒉𝒉𝒉𝒉that result in their natural
logarithms being equal to zero. The slope coefficient of 𝒉𝒉(𝒉𝒉𝒉𝒉𝒉𝒉) implies that for a country with a
real GDP per capita that is higher by 1%, predicted life expectancy at birth is decreased by approximately
0.003 years, which is unexpected given our hypothesized relationship between real GDP per capita and life
expectancy at birth. The slope coefficient for 𝒉𝒉(𝒉𝒉𝒉𝒉𝒉)implies that a 1% increase in a country’s health
expenditure per capita is associated with a life expectancy at birth that is higher by 0.045 years, following
the hypothesized relationship. The slope coefficient for unemployment is -0.0818, which implies that for
every unit increase in unemployment, life expectancy at birth is decreased by -0.0818. This satisfies the
hypothesized relationship, as unemployment is associated with poverty and low output. For phys, which
measures the number of physicians there are for every 1000 population, the slope coefficient is -0.2157,
which is interpreted as for every unit increase in phys, life expectancy at birth decreases by -0.2157. This
is unexpected, as the a priori expectation was a positive effect based on the related literature. As for the
primary completion rate pri the slope coefficient is 0.0941, which indicates that for every unit increase in
pri, life expectancy at birth increases by 0.0941, following the hypothesized relationship.
The resulting 𝑅2 and adjusted 𝑅2 was calculated to be 0.8237 and 0.8035, respectively. This
implies that the estimated model explains approximately 80% of the variation in life expectancy at birth.
To test the individual and joint significance of the regressors, the p-value of the F-statistic and the
coefficients’ t-statistics in the regression were examined. The STATA output is shown in Appendix D
shows that the variables are jointly significant as 𝑅 > 𝑅 = 0.0000. Individually, however, only lnhexpc
and pri were statistically significant, with a p-value less than 0.05, while the rest were higher. The p-value
for lnrgdppc was 0.830, which may be an indication that health expenditure may have been a better
prediction for life expectancy rather than real GDP per capita.

Results of Diagnostic Tests

To test for multicollinearity, the researchers computed the correlation coefficients between the
regressors and the regressors’ variance inflation factors. The STATA output shown in Appendix E indicates
that there is a high level of multicollinearity present in the model as the variance inflation factors for
𝑅𝑅(𝑅𝑅𝑅𝑅𝑅𝑅) and 𝑅𝑅(ℎ𝑅𝑅𝑅𝑅)are well above 10. As shown by the correlation coefficient of 0.9619,
there is an almost perfect linear relationship between 𝑅𝑅(𝑅𝑅𝑅𝑅𝑅𝑅)and 𝑅𝑅(ℎ𝑅𝑅𝑅𝑅), making the
slope coefficients in the model less precise. Hence, it is likely that dropping one of these would allow the
OLS method to provide better estimates. Since 𝑅𝑅(𝑅𝑅𝑅𝑅𝑅𝑅)was found to be individually insignificant
and 𝑅𝑅(ℎ𝑅𝑅𝑅𝑅) was found to be individually significant at the 1% level, dropping
𝑅𝑅(𝑅𝑅𝑅𝑅𝑅𝑅)from the regression could potentially allow for better estimates. Overall, this implies that
the CLRM assumption of no two regressors having a perfect linear relationship was almost violated in the
model.
To test for heteroskedasticity, the Breusch-Pagan and White tests for heteroskedasticity were used.
The STATA outputs for these tests are shown in Appendices E and F. Both tests assume that the variance
of the errors is constant or that the errors are homoskedastic. Since the resulting p-values from the tests are
both significant at the 1% level, there is strong evidence against the null hypothesis that the errors are not
homoskedastic, hence we reject the null hypothesis. The consequence of violating the CLRM assumption
that the error term has a constant variance is that the slope coefficients in the model are no longer efficient.
 To test for misspecification, the Ramsey RESET test and a link test were used. The STATA outputs
for these tests are provided in Appendices G and H. The results of the RESET test imply that the null
hypothesis of no omitted variables could not be rejected at the 10% significance level as shown by the p-
value of 0.1228. Hence, it can be said that the model does not have omitted relevant variables. The linktest
shows that the predicted value is a significant predictor, as expected, and that the squared prediction is not
a significant predictor of the dependent variable even at the 20% level, which should be the case when the
model is properly specified. Overall, it can be said that the model was properly specified.

CONCLUSION AND RECOMMENDATIONS

In summary, this study shows that health expenditure per capita and the primary completion rate
are appropriate predictors of a country’s life expectancy at birth. In addition, it is possible that health
expenditure per capita may be a more apt predictor life expectancy at birth than real GDP per capita.
However, due to violation of the homoskedasticity and no multicollinearity assumption, the values for the
slope coefficients obtained in our multiple linear regression are not efficient and are unlikely to provide
accurate predictions for countries not included in the data set. The model overall, though, is able to explain
much of the variation in life expectancy at birth, approximately 80%.
 REFERENCES

Basu, S., Berkowitz, S. A., Phillips, R. L., et al. (2019). Association of primary care physician supply
with population mortality in the United States, 2005-2015. Jama Internal Medicine. Retrieved
from https://jamanetwork.com/journals/jamainternalmedicine/article-
abstract/2724393?fbclid=IwAR0zHG7ZsU5IH8KxVPBj46fwx2oZY_JJg2Wi2lTIn0KxVDOT7--
wZBWYUNw

Chappelow, J. (2019) Unemployment Rate. Retrieved from

https://www.investopedia.com/terms/u/unemploymentrate.asp

Chen, M. & Ching, M. (2000). A statistical analysis of life expectancy across countries using multiple
regressors. Retrieved from https://www.seas.upenn.edu/~ese302/Projects/Project_2.pdf

Garcia, J. et al. (2016) Modern Day Evaluation of the Preston Curve: The Relationship Between Life
Expectancy and Income. Retrieved from
https://pdfs.semanticscholar.org/0ed0/5451d2875b8f974192d435d474118c9a98e3.pdf

Kaplan, R. et al (2014) Education Attainment and Life Expectancy. Retrieved from

https://journals.sagepub.com/doi/pdf/10.1177/2372732214549754

Organization for Economic Cooperation and Development. (2007). Life Expectancy. Retrieved
from OECD Glossary of Statistical terms website:
https://stats.oecd.org/glossary/detail.asp?ID=1530

Singh, G. and Siahpush, M. (2016) Inequalities in US Life Expectancy by Area Unemployment

Level, 1990–2010. Retrieved from https://www.hindawi.com/journals/scientifica/2016/8290435/

World Bank Data (n.d.). GDP per capita, PPP (constant 2011 international $). Retrieved from
https://data.worldbank.org/indicator/NY.GDP.PCAP.PP.KD

World Bank Data (n.d.). GDP per capita, PPP (constant 2011 international $). Retrieved from
https://data.worldbank.org/indicator/NY.GDP.PCAP.PP.KD

World Bank Data (n.d.). Life expectancy at birth, total (years) - Sub-Saharan Africa. Retrieved
from Roser, M. (2019). Life Expectancy. Retrieved from https://ourworldindata.org/life-
expectancy
 APPENDICES

A. Data

Country lifexp rgdppc hexpc phys pri unemp

Luxembourg 82.29268293 93727.3206 6420.317415 2.9222 79.07543 6.669000149

Kuwait 75.13 70265.23663 2927.440047 2.5789 100.579 2.200000048

Norway 82.30487805 64028.39153 6268.112398 4.394 100.8554 4.296000004

Switzerland 82.89756098 57324.67907 7611.309849 4.1782 96.82687 4.801000118

Denmark 80.70243902 45787.98395 5021.768695 3.6743 103.8827 6.168000221

Sweden 82.20487805 45698.49506 5299.966773 4.2861 105.2293 7.43200016

Austria 81.1902439 44253.75425 5194.065652 5.0701 99.65672 5.72300005

Germany 80.64146341 44043.17794 5328.899379 4.1383 97.9372 4.624000072

Bahrain 76.762 43926.60212 2326.821277 0.9257 99.51365 1.075999975

Oman 76.887 39243.18166 1815.504829 2.1225 103.6481 3.542000055

United Kingdom 80.95609756 39009.73489 4087.728534 2.778 100.8826 5.301000118

Malta 81.89756098 34484.53079 3389.006631 3.826 101.8052 5.390999794

Italy 82.54390244 34292.53482 3288.745489 3.9174 97.74971 11.89599991

Korea, Rep. 82.02439024 34177.65447 2502.25111 2.2596 98.01557 3.599999905

Spain 82.83170732 32357.47404 3175.133883 3.8493 101.2097 22.05699921

Israel 82.05121951 32104.41912 2750.48431 3.1906 103.5788 5.250999928

Slovenia 80.77560976 29042.72523 2676.208687 2.8099 97.90015 8.961999893

Slovak Republic 76.56341463 28397.41072 2028.355297 3.4414 92.44306 11.48099995

Estonia 77.5902439 27610.24195 1860.325048 3.4152 96.44263 6.186999798

Lithuania 74.32195122 27041.96812 1853.699283 4.2993 101.89 9.119999886

Malaysia 75.461 25390.43526 1037.658054 1.5132 100.0088 3.099999905

Poland 77.45121951 25307.24394 1686.15661 2.3112 101.7504 7.500999928

Hungary 75.56829268 25077.23063 1859.518082 3.1159 98.59476 6.813000202

Russian Federation 71.18341463 24516.54995 1328.253037 3.9632 96.14293 5.571000099

Latvia 74.4804878 23018.7929 1393.638929 3.1736 98.0601 9.873000145

Mauritius 74.35317073 18879.49485 1147.854245 2.0246 101.0687 7.409999847

Mexico 74.904 17495.16111 974.4987939 2.2513 103.9694 4.313000202

Iran, Islamic Rep. 75.796 16683.43161 1329.564222 1.14 100.2644 11.06000042

Colombia 76.531 13114.57363 832.6895424 1.9379 101.7113 8.298999786

Maldives 77.691 12683.75875 1298.722179 3.1381 106.9581 5.677999973

South Africa 62.649 12345.9288 1079.639021 0.7827 87.25533 25.15600014

Mongolia 69.111 11329.87797 471.8075028 3.2426 92.99312 4.861000061

Sri Lanka 76.316 11077.59739 463.7359234 0.9746 101.1093 4.670000076

Tunisia 75.922 10860.51981 802.0708307 1.2999 99.01363 15.21800041

Ecuador 76.143 10704.27326 983.0623625 2.0734 103.5557 3.615999937

Egypt, Arab Rep. 71.302 10242.53436 496.9782118 0.81 96.90645 13.05200005

Georgia 72.973 9005.338678 717.1269015 5.0976 115.434 14.07999992

Fiji 67.103 8775.920925 320.0539774 0.8374 107.0184 4.547999859

Jordan 74.078 8407.679672 568.2245546 2.8407 72.06589 13.07499981

Bhutan 70.419 8380.123518 287.3170207 0.3188 100.6622 2.450000048

Jamaica 74.098 8047.090058 487.7806742 0.4175 85.9205 13.51099968

Timor-Leste 68.459 7657.278335 138.0459775 0.6696 99.2806 3.342999935

Cabo Verde 72.117 6007.21829 339.5523941 0.7694 91.03706 12.11400032

Moldova 71.478 5695.630725 514.257167 3.2002 89.78202 3.694000006

Vietnam 75.11 5608.471859 334.3622215 0.7887 105.964 1.858999968

Pakistan 66.577 4459.146517 134.3635841 0.9753 64.95687 3.565999985

Sudan 64.429 4262.233622 288.4135835 0.41 59.83708 13.15499973

Senegal 66.747 3001.819107 136.6757348 0.2018 59.85945 6.756999969

Sao Tome and 69.377 2889.580149 167.6038376 0.32 77.48247 13.60499954

Principe

Rwanda 67.45 1774.476218 119.3613243 0.1408 69.06014 1.098999977

Uganda 61.373 1749.602349 119.9585244 0.0908 54.74301 1.870000005

Gambia, The 60.91 1481.4762 79.50438266 0.1077 69.27261 9.538999557

Togo 59.927 1467.245159 91.80855556 0.0487 92.71515 1.692000031

Central African 50.881 702.9853244 42.08366354 0.0629 40.87417 6.754000187

Republic

B. Methodology

Table 1. Hypothesized relationships between the regressand and the regressors

Relationship
Regressor with Intuition
Regressand

rgdppc rgppc measures real wealth per person. Higher wealth allows better
access for healthcare so we can expect it to have a positive relation
+ to life expectancy.

unemp Unemp measures the percentage of individuals with no formal

work. High unemployment is related to poverty and low output.
- Lower output and poverty means less access to private health care.
We expect this to have a negative relation to life expectancy

pri Pri is the measure of the rate of primary school completion.

+ Characteristics of those pursuing more education are linked to
having higher life expectancy. We expect this to have a positive
relationship with life expectancy.

hexpc Hexpc is the amount the government spends on public and

collective health services. The higher the expenditure, the more
+ health care is available. We expect this to have a positive
relationship with life expectancy.

phys Phys is a measure of the number of physicians there are for every
1000 population. A greater ratio signifies more physicians per
+ patients, therefore better attention to their health.We expect this to
have a positive relationship with life expectancy.
C. STATA Regression Results

D. STATA Results for Regressors’ Correlation Coefficients and Variance Inflation Factors
E. White Test for Heteroskedasticity

F. Breusch-Pagan Test for Heteroskedasticity

G. Ramsey RESET Test

 H. Link Test

I. Plagiarism results