Health and Income Variation – A Panel Data Study on the Developed and Less Developed Economies

Abstract

In this paper, human capital in the form of ‘health status’ is introduced into a neoclassical economic growth model as one of the main factors differentiating rich and poor countries. Various panel data models are used to examine how health and other growth factors affect average income in different countries. Our main empirical finding indicates that a one-year increase in life expectancy (the health status measure) raises GDP per capita by 0.5–0.9%. Based on this result, a baseline health status can be established to help poor countries achieve a targeted economic growth rate.

Keywords:

• Health
• GDP per capita
• neoclassical economic growth model
• panel data models

JEL Classifications :

• I1
• O1
• O4

Acknowledgements

The authors would like to thank the Editors, Dr Chinan Chiang, and an anonymous referee for precious comments and suggestions on this paper. I-Ming Chiu is grateful to the Lindback Foundation for financial support through Rutgers University. Any remaining errors are the authors’ own.

Notes

¹Economic well-being is represented by ‘average income’, which is measured by ‘GDP per capita’ using international PPP price. We use these two terms interchangeably in this paper.

²In addition to education and health, working experience is another potential candidate for representing human capital (Bloom et al., 2004). Experience data can be collected via survey questions and used in micro studies. However, such data often do not exist in macro (across-country) studies.

³In practice, it also includes other factors such as culture, political stability, government policies, etc.

⁴In section 4, we show that education and health are linearly related to the log of average income.

⁵Using the World Bank's Gross National Income (GNI) per capita criteria in 2008, these 52 economies can be divided into ‘High & Upper Middle Income’ and ‘Lower Middle & Low Income’ countries based on average GDP per capita data calculated from our sample. The former countries include Austria, Canada, Chile, Colombia, Cyprus, Denmark, the Dominican Republic, Finland, Greece, Iceland, Italy, Israel, Jamaica, Japan, Malaysia, Mauritius, the Netherlands, Portugal, Singapore, Sweden, Switzerland, Trinidad and Tobago, Turkey, the UK, the US, and Uruguay, Venezuela (27 countries in total). The latter countries include Benin, Bolivia, Cameroon, Central African Republic, Ecuador, El Salvador, Guatemala, Honduras, India, Indonesia, Kenya, Mozambique, Nicaragua, Nigeria, Pakistan, Paraguay, Philippines, Rwanda, Sri Lanka, Syria, Thailand, Togo, Tunisia, Uganda, Zimbabwe (25 countries in total).

⁶The idiosyncratic component (ϵ _it ) is both heteroskedastic and serially correlated.

⁷The estimation equation after differencing is Δ y _it =Δ l _it +α* Δ k _it +δ₁* Δ ED _it +δ₂* Δ H _it +Δ ϵ _it , where ‘Δ’ is the difference symbol. For example, Δ y _it =y _it −y _it−1. If the estimated coefficients (α, δ₁, δ₂) are quite different after applying the differencing method, it is likely the results we get from estimating equation (10) are spurious.

⁸The test is not actually a selection test. It is to compare whether the fixed effect model (estimator) is different from the random effect model (estimator). If the null hypothesis is true, the difference between these two estimators approaches zero asymptotically (see Verbeek, 2004, 10.2.3).

⁹All of the regressors except for the labor–population ratio are replaced by their first lagged counterparts. The labor–population ratio in each country has been steady over the sample period and so it is unlikely to be affected by average income.

¹⁰Although the random effect model is rejected by the Hausman test, we don't refute its usefulness for inference purposes. The individual specific effect a _i in equation (9) can be ‘random’ in nature, considering that our data on 52 countries are a mere sample drawn from the population of more than 200 countries. Frees (2004) explains why both fixed and random effect models are important in practice.