The Wealth-Health Nexus: New Global Evidence
Authors
- First Online:
DOI: 10.1007/s11293-012-9347-x
- Cite this article as:
- Jaunky, V.C. Atl Econ J (2013) 41: 115. doi:10.1007/s11293-012-9347-x
- 208 Views
Abstract
Using a world sample of countries, this paper re-examines the U-shaped relationship between per capita GDP (wealth) and life expectancy at birth (health). Since cross-sectional dependence across countries is detected, second-generation panel unit root and cointegration tests are employed. All the variables are found to be integrated in one order as well as cointegrated. Various quadratic specifications are also employed and the hypothesis is confirmed.
Keywords
Per capita GDPLife expectancyCross-sectional dependencePanel FMOLSJEL
I15C33O11Introduction
Healthcare improvements constitute a major moral prerogative to any nation. Healthy citizens, whether unskilled or skilled, enhance an economy’s productive capacity by being both physically and mentally apt. As such, the question of whether better health care can stimulate economic growth intrinsically arises. There is an on-going debate in literature on the impact of increased life expectancy on the wealth of nations and diverse results have been obtained.
During infant stages of development, any improvements on the level of health lead to a fall in wealth because of the occurrence of the Malthusian population effect. The latter postulates an exponential expansion in population growth. Yet, this effect is only temporary as fertility is apt to drop. As further health improvements occur beyond the turning point α, human capital accumulations and economic development are stimulated. This eventually causes population growth rate to fall (Hansen 2012).
Various studies have investigated the impact of life expectancy on economic growth. Acemoglu and Johnson (2007) find a negative but statistically insignificant impact whereas Zhang and Zhang (2005), Bloom et al. (2009), Turan (2009) and Aghion et al. (2010) uncover a significantly positive one. Hansen (2012) is among the first to provide evidence of a U-shaped relationship using a world panel of 119 countries throughout the period of 1940–1980. Yet, while the models used in literature rely heavily on cross-sectional studies which may not adequately capture the effects of health, some have employed panel data of little importance given to non-stationary series, which can lead to spurious inferences.
Descriptive statistics for the period 1970–2009
Data | Mean | Standard deviation | Minimum | Maximum |
---|---|---|---|---|
Real GDP per Capita ($) | 6105.71 | 8738.81 | 57.78 | 56388.99 |
Life Expectancy (years) | 63.68 | 11.46 | 26.82 | 82.93 |
Results
Prior to estimating the above equation, some preliminary tests are conducted. According to the Hausman’s (1978) specification test, the null hypothesis (H_{0}) of no systematic difference in coefficients between the fixed-effects (FE) and random-effects (RE) panel data models is rejected with the test statistics equal to χ^{2}(2) = 245.68 [0.000]*. The FE panel data model best fits the data. The Greene (1993) groupwise heteroskedasticity test statistics are χ^{2}(106) = 15955.74 [0.000]* and χ^{2}(106) = 15998.47 [0.000]* for the FE and RE panel data models respectively. Thus, the H_{0} of homoskedasticity is rejected. Next, the Wooldridge (2002) no first-order autocorrelation (AR(1)) test statistic is F(1,106) = 289.25 [0.000]*. The H_{0} is rejected. The Pesaran (2004) test of H_{0} no cross-sectional dependence (CSD) is equal to 38.02 [0.000]* for the FE panel data model. The p-value is in square brackets. The presence of CSD is found. In addition to the FE and RE panel data models, the Prais and Winsten (1954) heteroskedastic panel corrected standard error model which can control for AR(1) specific to each panel is applied (StataCorp 2007).
As exposed in Table 4, the U-shaped hypothesis is supported. Since β_{1} and β_{2} are statistically significant at the 1 % level, a precise and meaningful value of α can be obtained from the models. For instance, a 99 % confidence interval for α is reported for the FE, RE, and PW panel data models. All α’s are found to lie within this estimated interval. Their p-values are also computed to be 0.000, implying they are statistically significant from zero. Nevertheless, these three models are based on the stationarity assumption and ignore any non-stationary process of the series which can result in spurious inferences. Efficient models such as the fully modified ordinary least squares (FMOLS) model need be considered. Preliminary tests such as panel unit root and cointegration tests are accordingly required. When performing panel unit root tests, two distinct specifications are utilized. One test includes a constant term only while the other contains both a constant term and a time trend. Macroeconomic data tends to display a trend over time. It is more fitting to consider a regression with a constant and a trend at level form. Since first-differencing tends to extract any deterministic trends, inferences will be carried out by considering a constant term only.
(a): Pesaran panel unit root test statistics (b): Hadri and Kurozumi panel unit root test statistics
Variables | Deterministics | Level form | First-difference | ||
t-bar | Z | t-bar | Z | ||
LGDP_{it} | Constant | −1.399 | 4.192 [1.000] | −2.161 | −4.191 [0.000]* |
Constant + Trend | −2.001 | 3.987 [1.000] | −2.487 | −1.729 [0.042]^{+} | |
LEX_{it} | Constant | −1.779 | 0.011 [0.504] | −2.346 | −6.233 [0.000]* |
Constant + Trend | −2.317 | 0.271 [0.607] | −2.462 | −1.432 [0.076]^{‡} | |
\( LEX_{it}^2 \) | Constant | −1.798 | −0.198 [0.422] | −2.342 | −6.186 [0.000]* |
Constant + Trend | −2.308 | 0.372 [0.645] | −2.468 | −1.507 [0.066]^{‡} | |
Variables | Deterministics | Level form | First-difference | ||
ZA_{spc} | ZA_{la} | ZA_{spc} | ZA_{la} | ||
LGDP_{it} | Constant | −0.098 | −0.063 | 0.092 | 0.208 |
Constant + Trend | 3.935* | 3.730* | 1.820^{+} | 2.144* | |
LEX_{it} | Constant | −1.995 | −1.982 | −1.119 | −1.212 |
Constant + Trend | 4.264* | 4.334* | 3.906* | 3.772* | |
\( LEX_{it}^2 \) | Constant | −1.980 | −1.969 | −1.120 | −1.197 |
Constant + Trend | 4.585* | 4.644* | 4.159* | 4.066* |
Westerlund panel cointegration test statistics
Statistics | Without trend | With trend | ||||||
---|---|---|---|---|---|---|---|---|
Value | Z | P-value | Robust P-value | Value | Z | P-value | Robust P-value | |
Gt | −2.737 | −13.378 | 0.000* | 0.000* | −3.729 | −14.769 | 0.000* | 0.000* |
Ga | −4.076 | 3.306 | 1.000 | 0.037^{+} | −4.489 | 12.901 | 1.000 | 1.000 |
Pt | −16.084 | −4.810 | 0.000* | 0.000* | −25.647 | −2.199 | 0.014^{+} | 0.027^{+} |
Pa | −2.988 | −1.033 | 0.151 | 0.001* | −5.428 | 7.743 | 1.000 | 0.970 |
Regressions
Statistics | FE | RE | PW | FMOLS | ||||
---|---|---|---|---|---|---|---|---|
β_{0} | 1.19 | 72.98 | 0.71 | 74.33 | −6.13 | 63.57 | – | – |
(0.29)* | (16.40)* | (0.31) ^{+} | (16.33)* | (0.53)* | (5.62)* | – | – | |
β_{1} | 1.55 | −34.26 | 1.66 | −35.06 | 3.32 | −32.27 | −0.57 | −39.88 |
(0.07)* | (8.09)* | (0.07)* | (8.05)* | (0.13)* | (2.80)* | (−1.41)^{‡} | (1.75)* | |
β_{2} | – | 4.45 | – | 4.56 | – | 4.53 | – | 4.99 |
– | (1.00)* | – | (0.99)* | – | (0.35)* | – | (0.22)* | |
R^{2} | 0.15 | 0.31 | 0.68 | 0.74 | 0.95 | 0.96 | – | – |
Greene (1993) | 14333.29* | 5955.74* | 14321.66* | 15998.47* | – | – | – | – |
Pesaran (2004) | 63.36* | 38.02* | 58.51* | 31.928* | – | – | – | – |
Observations | 4280 | 4280 | 4280 | 4280 | 4280 | 4280 | 4280 | 4280 |
Countries | 107 | 107 | 107 | 107 | 107 | 107 | 107 | 107 |
α | – | 46.91 | – | 46.53 | – | 35.15 | – | 54.38 |
C.I. | – | [40.09,53.73] | – | [39.89,53.18] | – | [31.77,38.52] | – | – |
Conclusions
The paper investigates the relationship between wealth and health status using a world panel data for the period 1970–2009. A U-shaped relationship is uncovered. However, conventional estimators reveal a turning point of about 35 and 47 years, while a much greater value of 54 years is obtained when applying an efficient estimator such as the panel FMOLS. In sum, health effects on wealth are inclined to vary over different stages of economic development. Policymakers should therefore take for this non-monotonic relationship into account when designing healthcare schemes.