Health investments and economic growth: a quantile regression approach
Abstract
Purpose
This study aims to analyse the relationship between health human capital and economic growth for a maximum sample of 92 countries over the period 1980-2010 taking into account countries’ heterogeneity by assessing how health variables affect different countries according to their position on the conditional growth distribution.
Design/methodology/approach
The paper estimates a growth regression applying the methodology proposed by Canay (2011) for regression by quantiles (Koenker, 1978, 2004, 2012a, 2012b) in a panel framework. Quantile regression analysis allows us to identify the growth determinants that present a non-linear relationship with growth and determine the policy implications specifically for underperforming versus over achieving countries in terms of output growth.
Findings
The authors’ findings indicate that better health is positively and robustly related to growth at all quantiles, but the quantitative importance of the respective coefficients differs across quantiles, in some cases, with the sign of the relationship greater for countries that recorded lower growth rates. These results apply to both positive (life expectancy) and negative (infant mortality rate, undernourishment) health status indicators.
Practical implications
Given the predominantly public nature of health funding, cuts in health expenditure should be carefully balanced even in times of public finances sustainability problems, particularly when growth slowdowns, as a decrease in the stock of health human capital could be particularly harmful for growth in under achievers. Additionally, the most effective interventions seem to be those affecting early childhood development that should receive from policymakers the necessary attention and resources.
Originality/value
This study contributes to the existing literature by answering the question of whether the growth effects of health human capital can differ in sign and/or magnitude depending on a country’s growth performance. The findings may help policymakers to design the most adequate growth promoting policies according to the behaviour of output growth.
Keywords
Citation
Rosendo Silva, F., Simões, M. and Sousa Andrade, J. (2018), "Health investments and economic growth: a quantile regression approach", International Journal of Development Issues, Vol. 17 No. 2, pp. 220-245. https://doi.org/10.1108/IJDI-12-2017-0200
Download as .RISPublisher
:Emerald Publishing Limited
Copyright © 2018, Emerald Publishing Limited
1. Introduction
Human capital is widely recognized as an important source of economic growth. As countries move towards knowledge-based economies, the existence of highly skilled human capital becomes increasingly important. It is thus not surprising that previous empirical research has focused on identifying the mechanisms of transmission from human capital accumulation to growth and assessing the respective magnitudes (Benhabib and Spiegel, 1994; Bloom et al., 2004; Howitt, 2005; Hanushek and Woessmann, 2011; Bleakley, 2010). However, most previous studies focus on formal education as the main source of human capital, while the impact on growth of health human capital has not attracted as much attention. Health influences labour productivity, the capacity to learn at school and to grow intellectually and physically (Jack and Lewis, 2009). Simultaneously, the decrease in mortality and morbidity allows for an increase in the proportion of the working age population, therefore contributing to raise per capita income. Higher longevity also creates a greater need for people to save for their retirement, Bloom and Canning (2000). It is thus not surprising that “several of the great” take-offs “in economic history […] were supported by important breakthroughs in public health, disease control, and improved nutritional intake […]” Sachs (2001, p. 22).
At the empirical level, studies usually assume homogeneity among countries by restricting the significance and magnitude of the relationship between health human capital and economic growth for the average economy. In fact, most previous studies use methodologies that estimate the average effect of health on output growth thus assuming that the parameters (slope coefficients) of the empirical model are country invariant, i.e. they assume parameter homogeneity. Allowing for parameter heterogeneity in the study of the health human capital-economic growth nexus in broad samples of countries can thus bring new insights. This is particularly true at a time when many countries are still under the effects of the 2007-08 financial crisis, with impressive growth slowdowns that demand a rigorous identification of the most effective sources of growth in periods of deceleration. Additionally, the identification of different growth impacts of health human capital across the distribution of the growth rate of output can be especially important for countries facing fiscal sustainability problems. Since investments in health are mainly publicly funded in many countries, a cut in public health expenditures, with the associated negative impact on health human capital, could be especially harmful for output growth in slow growing countries.
The main aim of this paper is to assess the importance of health human capital for economic growth in a broad sample of countries taking into account parameter heterogeneity. For this purpose, we allow for differentiated effects of health human capital on the output growth rate, conditioned by the location of the dependent variable at different parts of its distribution. Additionally, we will analyse the sensitivity of our results to the use of different proxies for health human capital, trying in this way to overcome, to some extent, measurement error problems. We apply a quantile regression approach to estimate output growth equations for a maximum sample of 92 countries over the period 1980-2010. According to Mello and Perelli (2003), quantile regression is a suitable estimation methodology in a growth context as it is possible to capture countries’ heterogeneity and assess how policy variables affect countries according to their position on the conditional growth distribution. In terms of policy implications, as suggested by Barreto and Hughes (2004), it may be the case that, due to the presence of other (un-modelled) determinants countries grow slower (or faster) relative to the conditions suggested by the variables that are included in the model. Quantile regression analysis allows us to identify the growth determinants that do not have the predicted effect on growth and determine specific policy implications for slow growers relative to fast growing countries in terms of output performance.
The remainder of this paper is organized as follows: Section 2 briefly reviews the theoretical predictions and empirical evidence on the nexus between health human capital and economic growth; Section 3 contains a data overview; Section 4 presents the empirical model and methodology; Section 5 presents and discusses the results; section 6 contains the main findings and suggestions for future research.
2. Is health correlated with economic growth? literature overview
The link between human capital and growth has been investigated mainly considering formal education (Benhabib and Spiegel, 1994; Benhabib and Spiegel, 2005; Miles, 2004). During the 20th century, impressive health improvements resulted in the extension of the concept of human capital to include, besides education “the general state of health of the working population”, Savvides and Stengos (2009, p. 4), notwithstanding the divergences at the empirical level on how to measure this concept. Health, just like education, differs across individuals and consists of a stock that can depreciate or increase over time (Grossman, 1972). Healthier workers are able to think better, are more focused and allocate more energy and higher effort to task performance. Health thus influences some of the workers’ characteristics that determine their productivity. Additionally, healthier workers are less likely to miss work due to sickness (Bloom and Canning, 2000).
The benchmark for modelling the relationship between health human capital and economic growth is the neoclassical growth model developed by Solow (1956) and augmented by Mankiw et al. (1992) to include human capital. In the neoclassical or exogenous growth theory, higher rates of accumulation of both human (health and education) and physical capital lead to permanently higher levels of income and increase the growth rate of output in the medium run. In AK-type endogenous growth models human capital contributes to economic growth because more educated/healthier workers increase not only their own productivity but also that of other individuals with whom they perform different tasks (Lucas, 1988). Human capital is also viewed by endogenous growth theory as the major determinant of innovation and technology diffusion activities (Romer, 1990; Nelson and Phelps, 1966; Barro and Sala-i-Martin, 1997). If health human capital increases there will be more ideas available and more innovation will take place in the technological leader countries. In the follower countries, a better health status will increase their absorption capacity in terms of adapting and implementing the technologies developed by the leaders.
Additionally, health human capital can have indirect growth impacts through its influence on other growth determinants such as demography, education, physical capital and income inequality and poverty (Bloom and Canning, 2000; Howitt, 2005; López-Casasnovas et al., 2005). If the health status of the population increases, school absence due to sickness is expected to decrease. Health allows enhancing learning capacity, since individuals will be better prepared, both physically and intellectually, to learn. In particular, better nourished children will have better cognitive skills (Alderman et al., 2006). Moreover, if health increases lead to a decrease in mortality or increase longevity, the higher will be the incentive to invest in education and acquire additional school qualifications (Miguel and Kremer, 2004). Furthermore, increasing longevity influences savings decisions. If people expect to live longer, they will save more for their retirement. Higher savings rates will in principle lead to higher investment rates and thus more physical capital accumulation, which in turn fosters growth. On the contrary, if people’s health is bad and they have “a short time horizon because they expect to die young, they have less reason to save and the economy fails to grow.” Lorentzen et al. (2008, p. 82).
Finally, promoting health can also reduce poverty (Sachs, 2001). Health improvements seem to have larger impacts on the standards of living of the poorer with weaker health. Poorer people which are better nourished see their education capabilities improve (Lorentzen et al., 2008) with positive consequences on their performance and economic growth. This is often the reason why improving the health status of the poorer is seen as a way to escape poverty traps (Sala-i-Martin, 2005).
At the empirical level, researchers have yet to reach a consensus on the impact of health status and accumulation on growth. Early empirical studies find a positive relationship between health and economic growth in line with the pioneer work of Preston (1975). There is evidence pointing to health as an important growth determinant regardless of the period under analysis, type and number of countries included in the sample, health proxies used and model specification. However, negative and statistically significant impacts of health on output growth have also been found (Jack and Lewis, 2009). This motivated researchers to search for patterns across specific regions and among countries within the same income level group (Eggoh et al., 2015; Poças, 2012; Aghion et al., 2011; Bhargava et al., 2001). Overall, results appear sensitive to the health proxies used.
As far as panel data studies with wide samples of countries are concerned, Bloom et al. (2004) review some previous studies that use life expectancy to proxy for health status and conclude that the majority find a positive effect running from the initial level of health to output growth. They also estimate growth regressions with life expectancy as the main explanatory variable, over the period 1960-1990 for a sample of 104 countries. The results found point to a statistically significant and positive correlation, suggesting that health affects economic growth through its impact on labour productivity. Also using life expectancy to measure health human capital accumulation, Acemoglu and Johnson (2007) arrive at a negative correlation for a sample of 47 countries over the period(s) 1940-1980/1940-2000 implying that faster health accumulation is not beneficial for growth. According to the authors this is due to a Malthusian effect (the idea that population growth is expected to exceed resources growth) since for the period under analysis life expectancy grew at the same rate as population. In line with this work, Aghion et al. (2011) main aim is to study the importance of using different health proxies and growth theories, translated in the influence of the levels vs. the accumulation of health on output growth. They estimate a cross-country regression for a sample of 96 countries where they find a positive impact going from life expectancy to growth although health accumulation reveals to be less robust. In OECD countries the only health proxy with a positive correlation with growth is the reduction in the mortality rate below age forty. For the same time span, Lorentzen et al. (2008) explore other channels through which health might influence growth and they conclude that health can affect growth in a quantitatively more important way when countries simultaneously invest in physical capital and by influencing fertility rates, rather than by the human capital channel.
When assessing the relationship between health and economic growth, previous studies have mainly used linear regression techniques that estimate the average effect of health on output growth thus assuming that the slope coefficients of the empirical models are country invariant (parameter homogeneity). Soukiazis and Cravo (2007), Bhargava et al. (2001) and Cooray (2013) try to go beyond assessing the average effect by dividing their samples into different income groups (low, middle and high-income countries). They also proxy health with life expectancy but Cooray (2013) uses a sample of 210 countries, while Soukiazis and Cravo (2007) consider 77 countries (for the periods 1990-2008 and 1980-2000, respectively). Increasing health, according to the results in Soukiazis and Cravo (2007) is growth enhancing for low income countries whereas it has no statistically significant impact in high income countries. When Cooray (2013) uses adult survival rates as a proxy for health, the results point to a positive growth influence in upper-middle and high-income countries, while the influence is negative in low- and lower-middle income countries. The study by Bhargava et al. (2001) proxies the health status with adult survival rates considering a sample of 92 countries and obtains results similar to those of Soukiazis and Cravo (2007) although their results indicate that the relationship becomes insignificant after a certain GDP per capita threshold, providing a possible explanation for the differences in the results from other studies. Other studies that tried in some way to deal with the issue of parameter heterogeneity are those that restrict the sample to specific countries within a geographical region or an institutional group thus focusing on more homogeneous groups of countries, as in Eggoh et al. (2015) for African Countries and Poças (2012) for OECD countries. Poças (2012) finds evidence that health boosts growth in OECD countries, especially when considering the proxy health care quality and the mortality rates associated with specific diseases. On the other hand, Eggoh et al. (2015) conclude that increasing health expenditures may have a negative growth influence, even when the level of health expenditures for the countries is low, if education expenditures are below a certain threshold.
However, the previous studies cannot provide information on whether the health-growth nexus differs across countries that are under or over-performers in terms of growth, i.e. they fail to account for the entire conditional distribution of the output growth rate. As stated by Canarella and Pollard (2004, p. 3), “(…) finding the magnitude of the effects of the explanatory variables at the tails of the conditional growth distribution is likely to be more interesting and useful than finding the magnitude of such effects at the conditional mean.” Wang (2011) and Miles (2004) are examples of studies that investigate heterogeneous effects of different growth determinants. Miles (2004) focus on educational human capital. This author considers a sample of 77 countries for the period 1970-1998 and applies a pooled quantile regression approach. He finds different marginal effects of human capital between slow growers and fast growers. In line with this reasoning and focusing on the relationship between health and growth, Wang (2011) investigates the impact of health expenditures on growth for a sample of 31 countries over the period 1986-2007 applying a quantile regression approach. The results obtained indicate that there is a positive influence among middle and high performers, while for low performers the influence is negative. Our research differs from this on three main respects. First, Wang (2011) focus on the role of health expenditure (proxied by total expenditure on health; total expenditure on health care; and expenditure per capita) in the explanation of output behaviour, while the present study examines also the role of different health status indicators in the explanation of economic growth. For comparison purposes, given that Wang (2011) is the only study, to our knowledge, that applies quantile regression to the analysis of the health-growth nexus, we also include in our analysis a proxy for health expenditures although this implies shortening the period under analysis to 1995-2010. Second, in the tradition of growth empirics we estimate a growth regression with per capita output growth as our dependent variable and health proxies as the main explanatory variables, taken alongside other control variables considered relevant growth determinants in the theoretical and empirical growth literature. We estimate what is known in the literature as an ad hoc growth regression that encompasses neoclassical and endogenous growth models explanations of output behaviour in the long run (Barro, 1991; Ciccone and Jarocinski, 2010; Crespo-Cuaresma et al., 2011; Moral-Benito, 2012). In Wang (2011) the author estimates bivariate regressions with total output and, alternatively, one of the three proxies for total health expenditures he considers in his analysis. Third, the quantile regression approach used by Wang (2011) does not control for fixed effects (characteristics intrinsic to each country that influence the respective growth behaviour and that are different across countries, while remaining constant over time) and thus for omitted variables bias as is our case using the quantile regression proposed by Canay (2011) for a panel data framework.
3. Data overview
Our broadest sample includes a balanced panel data set for 92 countries (see Table AI in the Appendix[1]) from 1980 to 2010. The data needed for the estimation of our growth regressions were computed with information obtained mainly from the Penn World Table (PWT) (Feenstra et al., 2015), version 8.0 and the World Development Indicators (WDI). Using data originally from the PWT 8.0 we computed data for real GDP per capita at constant PPP by dividing output at constant international PPP at 2005 prices by total population. From the WDI we extracted several health proxies based on the following criteria: the availability of data for extended periods of time and a large number of countries. The choice of the period under analysis was thus dictated by data availability and econometric reasons. The paper applies a quantile regression approach [with the estimation methodology proposed by Canay (2011)], which is demanding in terms of the structure of the data since it requires a balanced panel data set, i.e. the number of time periods T must be the same for all countries i. We analyse a wide sample of countries, with quite different health proxies’ data availability, especially for the less developed countries. Extending the period under analysis beyond 2010 would imply missing data for a number of countries in our sample and thus an unbalanced panel data set, making it impossible to apply our selected estimation methodology, maintaining the analysis of the same health proxies. As more data becomes available, it would be interesting to update our data set. Although the health status of a population and its accumulation are difficult concepts to measure, the classical procedure for their evaluation is based on 5 D’s: death, disease, disability, discomfort and dissatisfaction (Lohr, 1988). These provide negative outcome indicators and therefore the doubt remains whether it is more accurate and conceptually correct to measure the lack of health (negative indicators) rather than its existence (positive indicators). The latter refers to wellness and quality of life, which involves a lot of subjectivity in its measurement. Thus, we also considered both positive and negative health indicators. The positive health indicators used are life expectancy, female and male survival rates to the age of 65 and consumption of kilocalories per day per person. The negative health indicators are female and male adult mortality rates, infant mortality rate and the prevalence of undernourishment. Finally, as mentioned at the end of Section 2, for comparison purposes with Wang (2011), although at the cost of half of the time period under analysis that gets reduced to 1995-2010, we also proxy for health using public health expenditures per capita. As suggested by an anonymous reviewer, it would also be interesting to control for the efficiency of these expenditures using proxies for institutional quality, such as the level of corruption or governance. However, since the estimation methodology used requires a balanced panel data set, controlling for the efficiency of health expenditure with these proxies would also imply missing data for a number of countries due to the limited data availability of corruption and governance indicators and thus an unbalanced panel. We acknowledge that previous literature such as Bergh and Henrekson (2011) and Oto-Peralías and Romero-Ávila (2013) show that there is a nonlinear relationship between public spending and economic growth that depends on public sector institutional quality. On the other hand, evidence from studies that examine the impact of public spending on health controlling for good governance is mixed. For instance, Wagstaff and Claeson (2004) find that health spending decreases under-five mortality as long as the quality of governance is high, while Lewis (2006) finds no association between the effectiveness of health spending and measures of the effectiveness of institutions in the health sector (government effectiveness or corruption measures). We leave the empirical analysis of this interesting topic for future research, as more data becomes available. For a summary of the variables used and respective sources see Table AII in the Appendix. Table I presents some descriptive statistics for the main variables.
According to Figure A1 in the Appendix that contains a scatter plot relating countries’ average growth rates for two sub-periods, 1980-1995 and 1995-2010, most of the observations are located in the first quadrant. This implies that the countries that were slow (fast) growers in the first sub-period remained slow (fast) growers in the second sub-period. This supports the importance of further investigating whether the explanatory variables we consider in our growth regressions have different growth impacts across different parts of the distribution of the output growth rate. Countries can reap additional growth benefits from the identification of the determinants that have higher impacts according to the respective growth performance and in this way eventually overcome poverty or middle-income traps.
Table AIII in the Appendix contains the correlation matrix between health indicators. The negative indicators (such as mortality rates and population undernourished) are negatively correlated to the positive ones (life expectancy, survival rates), as expected. In spite of being a positive health indicator, the variable measuring the consumption of calories presents a negative correlation with the other positive indicators, suggesting that in some countries the daily calories consumption is beyond that which is beneficial for health. The correlation between health expenditures and the other health indicators is positive relative to the positive health status indicators and negative relative to the negative health status indicators, as expected.
4. Empirical growth models and quantile regressions
To assess the importance of health and its different proxies for growth we estimate what is known in the literature as an ad hoc growth regression since it is not directly derived from a particular growth model but incorporates growth determinants highlighted by both the exogenous and the endogenous growth literature. We consider each health proxy alternatively (to avoid multicollinearity) together with a set of control variables identified as important growth determinants in the empirical and theoretical economic growth literature (Sala-I-Martin et al., 2004; Moral-Benito, 2012).
Our baseline growth regression is thus given by equation (1):
As far as the expected sign of the estimated coefficients is concerned, we assume that a better health status plays a key role in fostering workers’ productivity. Thereby, real GDP per capita growth rates are believed to move in line with life expectancy, survival to age of 65, kilocalories per day and health expenditure per capita; and to vary inversely with adult and infant mortality rates and the prevalence of undernourishment. Investing in physical capital or education increases input availability and thus fosters growth. The initial level of output controls for the convergence hypothesis, through diminishing returns or technological catch up. Higher levels of output, closer to the steady state equilibrium or the technological frontier, lead to slower growth. In the neoclassical framework, the population growth rate has a negative growth impact. Additionally, a country’s GDP per capita can grow faster as countries open their market to foreign countries and allow more goods to be traded with the rest of the world, due to scale effects, increased competitiveness and/or technological diffusion. Finally, Barro (1990) considers the share of government expenditures as a powerful determinant of growth. He argues that increasing the share of non-productive government expenditures can lower growth.
We have estimated equation (1) using quantile regression with the main aim of identifying health growth effects beyond those allowed by conventional estimation procedures. This method, first proposed by Koenker and Bassett (1978), estimates models for conditional quantile functions, Q_{τ}(Y/X): the influence of a set of variables X on Y is estimated for univariate quantiles τ ∈ (0,1) of the distribution of Y rather than focusing on the expected value of the response variable as do least squares estimation, E(Y/X). The mean effects reflect only a specific part of the distribution (the central part). Similarly, univariate quantiles of the empirical distribution also correspond to a particular location of the distribution with value y such that P(Y ≤ y) = τ. Thinking of quantiles as a central part of a particular location of the distribution (like the median or the mean) makes it possible to solve the minimization problem in the same way as that for the conditional mean[2]. In summary, quantile regression minimizes a weighted sum of absolute deviations given by:
By applying the quantile regression procedure, it is possible to generate estimates of the influence of the covariates on the dependent variable for each quantile τ of the distribution of the response variable. The estimations were carried out using the linear programming procedure available for R studio (Koenker, 2012).
The quantile regression approach presents several advantages when compared to conventional estimation methods such as ordinary least squares (OLS). It provides summary statistics on both the central part and the tails of the distribution of the response variable allowing for a more complete investigation of the influence of specific covariates[3]. Quantile regression is also a more robust estimation procedure when the errors are not iid, as it is more robust to non-normal errors and outliers (as it minimizes asymmetrically absolute deviations), while OLS can be inefficient if the errors are highly non-normal. Furthermore, we can easily compare regression coefficients of specific quantiles to least squares estimates. The interpretation is very similar: a one-unit increase in the predictor variable associated to the estimated coefficient produces a change in the dependent variable expressed by the coefficient obtained for the specific quantile of the response variable.
Additionally, panel data enables us to control for unobserved fixed effects. To address the over parametrization resulting from parameter heterogeneity (Koenker, 2004) we eliminate the fixed effects applying the method proposed by Canay (2011). This corresponds to a two-step estimator that is consistent and asymptotically normal as both the number of units and periods grow. Assuming that fixed effects affect all quantiles in the same way, the effect on the conditional mean will also be the same. Therefore, we first estimate the conditional mean within the model and then purge this model from the individual effects. Afterwards it is thus possible to run a simple quantile regression subtracting the individual effects from the dependent variable. In the next section, we first present the estimation results of a panel fixed effects model for a better discussion and comparison of the results with those obtained when applying the fixed effects quantile regression proposed by Canay (2011) (for the 0.05, 0.25, 0.5, 0.75 and 0.95 quantiles).
5. Results
In this section, we present the results of estimating our baseline growth regression [see equation (1)] with the different health proxies for our sample of a maximum of 92 countries over the period 1980-2010. First, we applied unit root tests[4] to the variables and residuals of the regressions that proved to be stationary. We thus eliminated the possibility of obtaining spurious relationships. To estimate the panel least squares model we applied the Hausman (1978) test[5] to confirm fixed effects consistency that would otherwise jeopardize the panel methodology suggested by Canay (2011).
We present the results from the quantile regression estimations in two different ways to facilitate the interpretation of our findings. On the one hand, we plot (Figure 1) the evolution of the marginal effects of the different health proxies across output growth quantiles, at 90 per cent confidence intervals, together with the marginal effect of the least squares estimation, also at a 90 per cent confidence interval (straight lines). Furthermore, to allow for a clearer interpretation of the graphical analysis, Tables II and III provide an overview of the estimated coefficients for the different health proxies, for positive health indicators and health expenditures in Table II and for negative health indicators in Table III. In the estimations, we considered five quantiles, τ’s (τ = 0.05, 0.25, 0.5, 0.75, 0.95) to get a better understanding of the potential changes in coefficients across the conditional distribution. Additionally, for each variable we present the results of the test of coefficient homogeneity[6] across quantiles along the output growth rate conditional distribution. The null hypothesis corresponds to slope equality across quantiles so if the test rejects the null hypothesis we are in the presence of statistically significant differences in slope coefficients for the explanatory variables.
Starting with the analysis of the results for the positive health indicators (Figure 1 (a) to (d) and Table II), as we can see in Figure 1, part (a), the inverse of the life expectancy logarithm[7] (ille) shows a negative estimated coefficient, as expected, that decreases in intensity from low (slow growers) to high (fast growers) quantiles. These findings thus suggest that an increase in life expectancy (that corresponds to a decrease in ille, the variable considered in the estimations) has a positive impact on growth, stronger for slow growers. A higher life expectancy is the result of better early life health and thus includes a specific channel from childhood health to faster growth: healthier children go on to invest more in human capital, perform better at school and work and have less absences due to illness (Bleakley, 2010), which in turns promotes growth, as confirmed by our results. Life expectancy is also a proxy for adult health/mortality, which is positively correlated with growth because:
healthier adults are more able to work (illness reduces the ability to work); and
early-life investments in human capital should increase if adults are healthier since human capital will be used more intensely than if it is idled because of disease (Bleakley, 2010).
Additionally, a higher life expectancy is expected to increase savings (Bloom et al., 2004) and thus capital accumulation, allowing in this way for faster growth. The estimated coefficients (Table II) are statistically significant at 0.1 per cent for most quantiles with the exception of the coefficient for the 0.95 quantile that presents no statistical significance and the coefficient for the 0.75 quantile that is only significant at the 1 per cent level. Besides these results being statistically significant and corresponding to different slopes for the estimated quantiles, also the p-value of the slope equality test indicates that in this case we can reject the null hypothesis of equal slopes for life expectancy in the growth regression across quantiles. For higher quantiles, the results suggest that the accumulation of health human capital, proxied by a higher life expectancy, becomes a less relevant growth determinant (the same applies to the other health proxies as shown in the next paragraphs). Fast growers increase at a higher rhythm their output levels, so it might be the case that, as they approach the technological frontier, input accumulation becomes less important for growth. Growth becomes driven by technological progress/innovation activities and thus other more direct measures of innovation, such as R&D expenditures or patents, would probably be more relevant growth determinants (Benhabib and Spiegel, 1994, 2005).
Kilocalories consumption per day per person (ilkcal) is also a positive health indicator thus we expect a positive correlation with output behaviour. This positive sign might be explained using the adult and childhood health arguments above, specifically good nutrition early in life might affect adult productivity because lack of nutrients as a child can lead to cognitive or physical deficits as an adult, while well-nourished adults are more productive (Bleakley, 2010). Kilocalories consumption is also introduced in the regression as its inverse so we expect that the estimated coefficient has a negative sign (just like the inverse of life expectancy). According to the results presented in Figure 1(b) and Table II, the estimated coefficients for this variable confirm the expected negative sign across all quantiles and so a positive association between kilocalories consumption and output per capita growth. Figure 1, part (b), suggests that the influence of calorie consumption across the quantiles of the growth rate distribution is quite similar to the results obtained when the regression is estimated by least squares until around the 0.8 quantile, when the magnitude becomes lower. However, the estimated coefficients for the lowest (0.05) and highest (0.95) quantiles are not statistically significant and the coefficients for the other quantiles are very similar, which is confirmed by the result of the slope equality test that does not allow us to reject the null hypothesis. The lack of statistical significance at the lower quantiles could be the result of excessive calorie consumption when growth slows down and so thus income, with individuals increasing the consumption of less expensive food choices such as fast food, which in turn does not improve the health status of the population. Recall also that in Table AIII kcal presents a negative correlation with the other positive indicators, suggesting that in some countries the daily calories consumption is beyond that which is beneficial for health. For τ = 0.95 a similar argument to that used for life expectancy applies.
The link from adult survival to growth occurs, for instance, though the individuals’ provisions for the future that depend on the time horizon to which those decisions are expected to apply. If there is a lower risk of premature death, individuals’ will save more and accumulate more human capital (Lorentzen et al., 2008). Additionally, (Lorentzen et al., 2008, p. 85) argue that:
The premature death of an adult means the total loss of any human capital investments and the inability of that adult to personally enjoy the fruits of other investments. The death of an infant, while tragic and costly in its own right, has less severe economic consequences.
The adult survival rates (see Table II) were disaggregated by gender (asr.m, and asr.f, m for males and f for females). The results for males present a negative sign, for most quantiles, contrary to the theoretical predictions, although only statistically significant after quantile 0.75. Overall, for this indicator the results of the slope equality test do not allow us to reject the null hypothesis of parameter homogeneity. A possible explanation for the lack of statistical significance at most of the quantiles analysed could be that the overall adult male survival rate, as opposed to age specific rates, is not a good proxy to capture the effect of health on growth. This is because it is better health at a young age (the prime productive years) that has longer-term consequences in terms of workers productivity and thus growth. For instance, Aghion et al. (2011) find that only the reduction in mortality below age forty generates productivity gains for OECD countries and Lorentzen, McMillan and Wacziarg (2008) conclude that a greater risk of death during the prime productive years is associated with less growth. The negative sign might be explained by the following mechanism: lower survival rates (higher mortality) improve per capita GDP because with a smaller male workforce, female labour becomes more valuable and this leads to a reduction in fertility that implies higher living standards for the remaining population/survivors (Young, 2005). By contrast, the estimators for the female variable have the expected positive sign and present higher statistical significance and at more locations of the growth rate distribution. The difference in sign relative to the variable for males might be explained by the more important role females play in the accumulation of human capital of their offspring or by the increasing share of females in the labour market over the recent past that enhanced the contribution of women’s health human capital for growth. In this case, the estimated coefficient for quantile 0.95 presents the lowest value but it is not significant, similar to the estimators for ille and ilkal. Furthermore, there appears to be no pattern of change for the asr.f estimated coefficients across quantiles, which can be confirmed by looking at Figure 1(d). The estimated coefficient increases from the 0.05 to the 0.25 quantiles, it then decreases when we move to the 0.5 quantile and increases again for quantile 0.75, when it reaches its highest value (see Table II). However, the results of the slope equality test for female adult survival rates do not allow us to reject the null hypothesis of parameter homogeneity.
The results using the negative health indicators can be found in Figure 1 parts (e) to (h) and Table III. For all the proxies used we expect a negative correlation with growth, based on similar arguments to those discussed in the previous paragraphs for the positive health indicators. Ill health, as reflected in higher mortality of infants and adults or undernourishment, implies working less hours, with less effort, reduces cognitive ability and capital accumulation (if parents die, they cannot invest in their children; people save less because the length of retirement shortens). It can also suggest high-risk environments and deter capital flows through FDI or tourism (Jack and Lewis, 2009). All these different mechanisms in turn hamper growth.
In Table III and Figure 1(e) and (f), the adult mortality rates were also disaggregated by gender (amr.m and amr.f, m for males and f for females). Relative to adult survival rates, mortality rates are more sensitive to child mortality rates (Bhargava et al., 2001) thus including both adult and childhood health effects on growth. The estimated coefficients are negative as expected for the female adult mortality rate, while the variable for males presents positive estimated coefficients for some quantiles, contrary to what expected, although in most cases not statistically significant. The results of the slope equality test for the male adult survival rates do not allow us to reject the null hypothesis of parameter homogeneity and the estimated coefficients are only statistically significant after quantile 0.75. On the contrary and similar again to the results with female survival rates, the estimated coefficients for female adult mortality rates present higher statistical significance and at more locations of the growth rate distribution, which can be confirmed by looking at Figure 1(f). The main difference relative to the female survival rate pertains to the significance for quantile 0.05 where female adult mortality rates are not statistically significant. Similar also is the fact that the results of the slope equality test for the female adult survival rates do not allow us to reject the null hypothesis of parameter homogeneity. Given the close association between survival and mortality rates (see also Table AIII), it is not surprising that the results with adult mortality rates are basically a mirror image of the results with adult survival rates. The previous economic interpretation of the results for the adult survival rates thus applies here, with the necessary adaptions to the change in sign.
Infant mortality rate is a proxy for childhood (lack of) health, which might have specific effects on output behaviour because most of an individual’s human capital and physiological development occurs early in life and thus an unhealthy childhood can have effects that persist through life and negatively affect income (Bleakley, 2010). Our findings suggest that an increase in infant mortality rate indeed has a negative impact on growth. The pattern of behaviour across quantiles of the estimated coefficient for the infant mortality rate (imr) [Figure 1 (g)] is similar to that obtained for life expectancy [Figure 1(a)]. The statistical significance is identical too: no significance is found for the 0.95 quantile and for the 0.75 quantile the estimated coefficient increases the respective statistical significance (from 1 to 0.1 per cent). The results from the slope equality test indicate that it is possible to reject the null hypothesis of parameter homogeneity at the 10 per cent significance level. Similar to the results obtained with life expectancy, imr is a more relevant growth determinant for slow growers, suggesting a more important role for input availability in countries more distant from the technological frontier.
As for the share of undernourishment in total population (under), the estimated coefficients are negative as expected and, for the lowest quantiles, they are also statistically significant. The analysis of the results presented in Figure 1(h) indicates that moving from low to higher quantiles of the output growth rate distribution increases the value of the estimated coefficients (they become less negative), thus the association becomes less intense. Additionally, the results of the slope equality test indicate that it is possible to reject the null hypothesis of parameter homogeneity at the 1 per cent significance level. However, for the 0.75 and 0.95 quantiles the estimated coefficient for undernourishment is not statistically significant. As with the results with life expectancy and the infant mortality rate, the effects of undernourishment are stronger for under achievers in terms of output per capita growth and a similar reasoning applies related to the distance to the technological frontier.
Finally, Table II and Figure 1(i) contain the results with health expenditures. The assumption here is that efficient and strategic public spending leads to better health outcomes (Nixon and Ulmann, 2006) and, in turn, improved population health promotes economic growth through the various channels described in the previous paragraphs. Also confirming theoretical predictions, the estimated coefficients when using the proxy corresponding to the inverse of the initial level of public health expenditures per capita (ilgh) are negative and statistically significant across all quantiles (see Table II), indicating in turn a positive correlation between higher spending in health and growth. Regarding the magnitude of the estimated coefficients, the results for the slope equality test do not reject the null hypothesis of parameter homogeneity, although the estimated coefficients are slightly higher for the median and around the 0.8 quantile of the growth rate distribution. From the inspection of Figure 1(i) that contains the estimated coefficients for the variable ilgh it is possible to see that the black line that represents the coefficients across quantiles does not cross the red dashed line that represents the least squares confidence intervals. This also supports the finding that the estimated coefficients are not very different from the one obtained with least squares estimation. The overall results suggest in any case that when countries (either slow or fast growers) face public finances sustainability crisis they should cut public expenditure carefully not to jeopardize long-term output growth, especially in countries where health expenditures are mainly publicly funded. This result is in line with the findings of Wang (2011) for the period 1986-2007, who found a positive effect of health expenditures on growth, although only for medium and fast growers.
Regarding the control variables considered, the results are very similar for the regressions with the different health proxies. Table AIV in the Appendix contains the estimated coefficients with quantile regression as well as the within least squares model considering life expectancy as the proxy for health human capital. The estimated coefficient for education (educ) is statistically significant and positive as expected and the slope equality test reveals parameter heterogeneity, with a higher impact for the lowest quantile 0.05, a result in line with previous studies, Miles (2004). The variable controlling for convergence (ly) also presents the expected sign (negative) and is statistically significant across quantiles. Also, statistically significant and with the expected sign (positive) across all quantiles is the variable which measures countries’ openness to trade (lopen). Population growth (n) is statistically significant at the 1 per cent significance level with fixed effects but it is only significant with quantile regression for the 0.05 and 0.25 quantiles (at the 10 per cent significance level). Government consumption (g) presents an estimated negative coefficient but it is not statistically significant for any quantile. As for gross fixed capital formation (gfcf), the sign is negative, contrary to theoretical predictions. A possible explanation for this result is related to non-productive investments over the period under analysis that could have resulted in a crowding-out effect relative to productive investments thus hampering growth.
Overall, the results indicate that increasing health human capital is growth enhancing. However, for the higher growth quantiles, in particular the 0.95 quantile, the estimated coefficients are not statistically significant (except for public expenditures on health). The growth benefits from health human capital accumulation are bound to be exhausted as countries grow faster and close the technological gap and so technological progress and associated proxies such as R&D expenditures, R&D personnel, patents, number of graduates in STEM, etc. become the relevant growth determinants. Furthermore, the results for the various health proxies present some interesting differences. For instance, when we consider health proxies according to gender, female health proxies are positively related with growth (in particular at the lower quantiles) and the contrary applies to male proxies (although for most quantiles these are not statistically significant). This result could be related to the increasing share of females in the labour market over the last decade that enhanced the contribution of women’s health human capital for growth. It could also be the case that the economic growth gains from female’s health are higher than those from males’ since there are potential externalities from better females’ health associated with their children[8]. Nevertheless, the magnitude of this impact does not seem to vary across quantiles.
The most robust results in terms of differentiated growth impacts across quantiles are obtained when using the proxies’ life expectancy, infant mortality rates and undernourishment. The results from the slope equality test and the estimated coefficients across quantiles reveal differences in the magnitude of the influence of these health proxies. Improvements in the health status have a greater impact on slow growers (located in the 0.05 and the 0.25 quantiles) compared to those that performed above the median growth rate (located in the 0.75 quantile). Therefore, slow growers benefit more from an increase in health relative to fast growers, highlighting the adverse growth effects of not investing in health during periods of growth slowdowns.
An important policy implication of the former results is the potential more important contribution of a healthier young population to the improvement of economic growth for under achievers in terms of per capita output growth. This conclusion is derived from the fact that the difference in results across quantiles is only statistically significant (slope equality test) for three health variables that measure (are strongly influenced by) early-life health: the infant mortality rate, life expectancy and undernourishment. For the latter measures, for instance Lorentzen et al. (2008) and Jack and Lewis (2009) show that infant mortality is a major source of variation in life expectancy. Additionally, infant mortality is closely associated with undernourishment through e.g. maternal malnutrition (Lechtig, 1980) and lack of nutrients in childhood itself (Wegman, 2001) (see also the high correlations in Table AIII in the Appendix). Since our results suggest that the effect of child health is stronger for low-growth economies, policymakers in these countries should divert their attention and policies to invest more in prevention and treatment in the early stages of life, reducing the effect of diseases on the physical and cognitive abilities of children. According to Jack and Lewis (2009), one of the most robust results from microeconomic studies on the economic consequences of improved health is that, among other type of early-life interventions, prenatal care, food supplements for malnourished children and micronutrients for disadvantaged children, have long-term implications for enhanced health status. They help to raise the potential for long-term academic achievement and become successful professionals and improve well-being throughout life, increasing in this way productivity and output. Previous literature also indicates that ill health in the early years is likely to impair children over the course of their life: safeguarding health during childhood is thus more important than at any other age (Belli et al., 2005). As for adult health, our findings suggest that policies aimed at improving health in later stages of life are not as relevant (see the results with adult mortality and survival rates), especially in the case of males. However, health opportunities for females should be increased based on the spillovers conveyed by the following African proverb, adapted to a health context: “If we educate a boy, we educate one person. If we educate a girl, we educate a family and a whole nation” (taken from Knowles et al., 2002, p. 118). As for the control variables, the estimated coefficients tend to be fairly similar across quantiles, with the exception of education, indicating again a stronger impact for slow growers. The latter empirical results reiterate the importance of fostering education human capital and removing barriers to trade to promote growth (see Table AIV).
6. Conclusion
In recent decades, we have witnessed huge improvements in vaccination, infectious diseases treatments and access to medical care throughout the world. Nowadays, people are expected to live longer than ever before and with better quality of life. In this study, we revisit the role of health human capital on economic growth by applying a quantile regression approach to identify different signs and magnitudes for the influence of various health proxies across the distribution of the output per capita growth rate. This can lead to more specific policy implications regarding health determinants of economic growth according to growth performance. For this purpose, we considered a (maximum) sample of 92 countries over the period 1980-2010 and applied the quantile approach proposed by Canay (2011) that allows us to extend quantile regression to a panel data framework.
The results obtained endorse investing in health as a means of improving growth performance in our sample. Additionally, our findings suggest that the location on the output per capita growth rate distribution matters in terms of the magnitude of the relationship between health and macroeconomic performance. Countries will benefit more from investments that improve the health status of the respective populations when they are experiencing growth slowdowns (higher estimated coefficients for the lower quantiles of the growth rate distribution). This is true for the health proxies’ life expectancy, infant mortality rate and the prevalence of undernourishment, for which we obtained statistically significant coefficients, with the expected sign that changed across quantiles. Additionally, the results indicate that both the health status of infants as well as that of (female) adults have an important role in explaining per capita income growth: we found evidence supporting different growth contributions from females (mothers) health relative to that of males, positive and statistically significant in the first case.
The former results lead to different policy implications for under-performing countries in terms of actions that can foster output growth. For under achievers (those located at the lower growth quantiles) it is especially important not to overlook health improvements because this can have more important negative repercussions on long run growth rates as well as further depressing growth in the medium-run. Additionally, the most effective interventions seem to be those affecting early childhood development that should receive from policymakers the necessary attention and resources. As countries grow faster and close the technological gap, the association between health human capital and growth becomes less intense, an indication that the accumulation of inputs becomes a less important source of growth, which in turn becomes driven by innovation activities. In these cases, policymakers should probably redirect their attention to promoting science and technology.
While we have shown that there is evidence of parameter heterogeneity in the health-growth relationship in our sample over the period under analysis, further research is needed to understand why such parameter heterogeneity exists. Additionally, future research should explore other mechanisms of transmission from health to economic growth to provide a more complete picture in terms of direct and indirect effects and policy implications across the distribution of the growth rate of output.
Figures
Descriptive statistics for the main variables
Variable | Obs. | Mean | SD | Median | 1st quantile (0.25) | 3rd quantile (0.75) |
---|---|---|---|---|---|---|
Δly | 552 | 0.0166 | 0.0392 | 0.0175 | −0.0043 | 0.0369 |
le | 552 | 64.77 | 11.1228 | 67.94 | 56.08 | 74.11 |
asr.m | 552 | 0.6095 | 0.1618 | 0.6333 | 0.4864 | 0.7448 |
asr.f | 552 | 0.6969 | 0.1785 | 0.7478 | 0.5577 | 0.8520 |
imr | 534 | 0.0494 | 0.0419 | 0.0371 | 0.0114 | 0.0798 |
amr.m | 540 | 0.2595 | 0.1240 | 0.2376 | 0.1629 | 0.3237 |
amr.f | 540 | 0.1926 | 0.1302 | 0.1491 | 0.0886 | 0.2689 |
gh | 270 | 820.93 | 1,062.645 | 343.13 | 90.32 | 1,248.04 |
under | 252 | 0.2111 | 0.1331 | 0.1945 | 0.1003 | 0.3050 |
kcal | 252 | 154.400 | 105.1652 | 142.344 | 70.125 | 229.000 |
Δly–average annual growth rate of real GDP per capita; le–life expectancy; asr.m–adult male survival rate; asr.f–adult female survival rate; imr–infant mortality rate; amr.f–adult female mortality rate; amr.m–adult male mortality rate; gh–public health expenditures per capita; prevalence of undernourishment; kcal –consumption of calories per day per person. Δly, le, asr.m, asr.f, imr, amr.m, amr.f relate to 1980-2010. gh relates to 1995-2010 and under and kcal relate to 1990-2010
Source: Authors’ calculations with R
Estimates of the quantile panel model and fixed effects model for positive health indicators and health expenditure
ille | ilkcal | asr.m | asr.f | ilgh | |
---|---|---|---|---|---|
Fixed effects | −1.3308*** (0.3532) | −0.0333 (0.0981) | −0.8320*** (0.1370) | 0.1359 (0.0878) | −0.0566 (0.0915) |
Quantile | |||||
τ = 0.05 | −1.7878*** (0.3737) | −0.0432 (0.0315) | 0.0087 (0.0700) | 0.1458* (0.0633) | −0.8442*** (0.1139) |
τ = 0.25 | −1.9006*** (0.3693) | −0.0628* (0.0277) | −0.0176 (0.0385) | 0.1675*** (0.0362) | −0.7918*** (0.0775) |
τ = 0.50 | −1.2211*** (0.3199) | −0.05219. (0.0281) | −0.0410 (0.0417) | 0.1373** (0.0457) | −0.7583*** (0.0568) |
τ = 0.75 | −0.9630** (0.3047) | −0.0688* (0.0343) | −0.0924* (0.0392) | 0.1717*** (0.0411) | −0.8599*** (0.0662) |
τ = 0.95 | −0.8711 (0.7011) | −0.0383 (0.0417) | 0.0158 (0.0817) | 0.0646 (0.0926) | −0.8858*** (0.1465) |
Slope equality test | 2.9263 (0.0198)* | 1.5782 (0.1773) | 0.9327 (0.4440) | 1.6671 (0.1548) | 1.2765 (0.2772) |
No. countries | 92 | 63 | 92 | 92 | 90 |
Time period | 1980-2010 | 1990-2010 | 1980-2010 | 1980-2010 | 1995-2010 |
ille – inverse of life expectancy; ilkcal – inverse of consumption of calories per day per person; asr.m – adult male survival rate; asr.f – adult female survival rate; ilgh – inverse of public health expenditures per capita. Standard errors in parenthesis. The slope equality test refers to the test statistic with the p-value in parenthesis;
,
,
and “.” denote the statistical significance at the 0.1, 1, 5 and 10 % levels, respectively
Source: Authors’ calculations using R
Estimates of the quantile panel model and fixed effects model for negative health indicators
amr.m | amr.f | imr | under | |
---|---|---|---|---|
Fixed effects | 0.0297 (0.0901) | −0.1055 (0.0901) | −0.4293** (0.1305) | −0.0400 (0.0589) |
Quantile | ||||
τ = 0.05 | −0.0230 (0.0771) | −0.1114 (0.0721) | −0.6179*** (0.1761) | −0.1143** (0.0427) |
τ = 0.25 | 0.0144 (0.0428) | −0.1427** (0.0485) | −0.5761*** (0.0973) | −0.0909*** (0.0232) |
τ = 0.50 | 0.0312 (0.0465) | −0.1058* (0.0503) | −0.4637*** (0.1001) | −0.0538* (0.0246) |
τ = 0.75 | 0.0964* (0.0422) | −0.1457** (0.0503) | −0.3436*** (0.0999) | 0.0029 (0.0304) |
τ = 0.95 | −0.0150 (0.0867) | −0.0150 (0.1093) | −0.1878 (1.1664) | 0.0227 (0.0597) |
Slope equality test | 1.9437 (0.1005) | 0.9001 (0.4629) | 2.1549 (0.0716)^{§} | 3.7563 (0.0048)** |
No. countries | 90 | 90 | 89 | 63 |
Time period | 1980-2010 | 1980-2010 | 1980-2010 | 1990-2010 |
amr.f–adult female mortality rate; amr.m–adult male mortality rate; imr–infant mortality rate; under–prevalence of undernourishment. Standard errors in parenthesis. The slope equality test refers to the test statistic ant with the p-value in parenthesis;
,
,
and ^{§} denote the statistical significance at the 0.1, 1, 5 and 10 % levels, respectively
Source: Authors’ calculations with R software
List of countries
Albania | Costa Rica | Ireland | Namibia | Sudan |
Argentina | Cote d’Ivoire | Israel | Nepal | Swaziland |
Australia | Denmark | Italy | Netherlands | Sweden |
Austria | Dominican Republic | Jamaica | New Zealand | Switzerland |
Bahrain | Egypt | Japan | Niger | Tanzania |
Bangladesh | El Salvador | Jordan | Norway | Thailand |
Belgium | Finland | Kenya | Pakistan | Togo |
Benin | France | Laos | Panama | Trinidad and Tobago |
Bolivia | Gabon | Lesotho | Paraguay | Tunisia |
Botswana | Gambia | Liberia | Peru | Turkey |
Brazil | Germany | Malawi | Philippines | Uganda |
Bulgaria | Ghana | Malaysia | Poland | United Kingdom |
Burundi | Greece | Mali | Portugal | United States |
Cambodia | Guatemala | Mauritania | Rwanda | Uruguay |
Cameroon | Honduras | Mauritius | Senegal | Vietnam |
Canada | Hungary | Mexico | Sierra Leone | Zambia |
Chile | India | Mongolia | Singapore | Zimbabwe |
China | Indonesia | Morocco | South Africa | |
Colombia | Mozambique | Spain |
Source: Authors’ own compilation
List of variables
Notation | Description | Source | Source’s notation |
---|---|---|---|
Δly | Real GDP per capita annual average growth rate (calculated as the annual average growth rate of real GDP at chained PPPs divided by total population) | PWT 8.0 | rgdpo; pop |
ly | Logarithm of the initial level of real GDP per capita in PPP’s (calculated dividing real GDP in 2005 international USD by total population) | PWT 8.0 | rgdpo; pop |
gfcf | Average investment share | PWT 8.0 | csh_i |
n | Population growth rate | World Bank | SP.POP.GROW |
lopen | Trade (exports plus imports) as a percentage of GDP | PWT 7.1 | openk |
g | Share of government consumption in GDP at current PPP | PWT 8.0 | csh_g |
educ | Barro and Lee (2013) average years of total schooling of people aged 15 and over | Barro and Lee (2013) | yr_sch |
ilgh | Inverse of public health expenditures per capita (calculated multiplying total health expenditure per capita by public health expenditures as a share of total health) | WDI | SH.XPD.PCAP.PP.KD SH.XPD.PUBL |
ille | Inverse of the logarithm of life expectancy at birth in total years’ | WDI | SP.DYN.LE00.IN |
imr | Share of infants dying before reaching one year of age | WDI | SP.DYN.IMRT.IN |
asr.m | Share of male new-born infants that would survive to age 65 | WDI | SP.DYN.TO65.MA.ZS |
asr.f | Share of female new-born that would survive to age 65 | WDI | SP.DYN.TO65.FE.ZS |
amr.m | Probability of a 15-year old male dying before reaching age 60 at the beginning of the period | WDI | SP.DYN.AMRT.MA |
amr.f | Probability of a 15-year old female dying before reaching age 60 at the beginning of the period | WDI | SP.DYN.AMRT.FE |
ilkcal | Inverse of the logarithm of number of calories consumed per day per person | WDI | SN.ITK.DFCT |
under | Prevalence of undernourishment in the population | WDI | SN.ITK.DEFC.ZS |
Source: Authors’ own compilation
Health variables’ correlation matrix (58 countries 1995-2010)
le | kcal | asr.m | asr.f | amr.m | amr.f | imr | under | gh | |
---|---|---|---|---|---|---|---|---|---|
le | 1.00 | −0.58 | 0.97 | 0.99 | −0.91 | −0.97 | −0.93 | −0.61 | 0.68 |
kcal | −0.58 | 1.00 | −0.58 | −0.59 | 0.56 | 0.58 | 0.60 | 0.99 | −0.62 |
asr.m | 0.97 | −0.58 | 1.00 | 0.96 | −0.97 | −0.96 | −0.86 | −0.61 | 0.59 |
asr.f | 0.99 | −0.59 | 0.96 | 1.00 | −0.91 | −0.99 | −0.90 | −0.61 | 0.65 |
amr.m | −0.91 | 0.56 | −0.97 | −0.91 | 1.00 | 0.93 | 0.75 | 0.58 | −0.49 |
amr.f | −0.97 | 0.58 | −0.96 | −0.99 | 0.93 | 1.00 | 0.86 | 0.61 | −0.60 |
imr | −0.93 | 0.60 | −0.86 | −0.90 | 0.75 | 0.86 | 1.00 | 0.64 | −0.80 |
under | −0.61 | 0.99 | −0.61 | −0.61 | 0.58 | 0.61 | 0.64 | 1.00 | −0.63 |
gh | 0.68 | −0.62 | 0.59 | 0.65 | −0.49 | −0.60 | −0.80 | −0.63 | 1.00 |
See Table AII for a description of the variables
Source: Authors’ calculations with R
Quantile regression estimation results with life expectancy (1980-2010 for 92 countries)
Fixed Effects | Quantile | ||||||
---|---|---|---|---|---|---|---|
τ = 0.05 | τ = 0.25 | τ = 0.5 | τ = 0.75 | τ = 0.95 | Equality test | ||
Int | 1.0694*** (0.1203) | 1.1161*** (0.1116) | 0.9658*** (0.0893) | 0.8796*** (0.0876) | 0.9196*** (0.1967) | ||
ille | 0.1179*** (0.0312) | −1.7878*** (0.3737) | −1.9006*** (0.3693) | −1.2211*** (0.3199) | −0.9630** (0.3047) | −0.8711 (0.7011) | 2.9263 (0.0198)* |
ly | −0.0947*** (0.0067) | −0.0975*** (0.0362) | −0.0947*** (0.0030) | −0.0953*** (0.0017) | −0.0921*** (0.0027) | −0.0997*** (0.0061) | 1.0509 (0.3793) |
educ | 0.035*** (0.0019) | 0.0157*** (0.0018) | 0.0123*** (0.0010) | 0.0135*** (0.0008) | 0.0118*** (0.0009) | 0.0132*** (0.0021) | 3.6312 (0.0058)** |
n | 0.4719** (0.1791) | 0.5397^{§} (0.2770) | 0.2919^{§} (0.1507) | 0.2255 (0.1561) | 0.1770 (0.1971) | 0.0774 (0.5229) | 0.3536 (0.8416) |
gfcf | −0.0090*** (0.0020) | −0.0065 (0.0049) | −0.0093*** (0.0015) | −0.0108*** (0.0015) | −0.0096*** (0.0017) | −0.0085* (0.0034) | 0.7577 (0.5537) |
lopen | 0.0225*** (0.0057) | 0.0196*** (0.0048) | 0.0207*** (0.0026) | 0.0218*** (0.0017) | 0.0258*** (0.0027) | 0.0281*** (0.0048) | 1.6778 (0.1522) |
g | 0.0110 (0.0110) | −0.0283 (0.0362) | −0.0121 (0.0150) | −0.0195 (0.0213) | 0.0111 (0.0251) | 0.0603 (0.0611) | 1.0396 (0.3851) |
See Table AII for a description of the variables. The slope equality test refers to the test’s statistic with the p-value in parenthesis;
,***
,**
* and ^{§}denote the statistical significance at the 0.1 %, 1 %, 5 % and 10 % levels, respectively
Source: Authors’ calculations with R
References
Acemoglu, D. and Johnson, S. (2007), “Disease and development: the effect of life expectancy on economic growth”, Journal of Political Economy, Vol. 115 No. 6, pp. 925-985.
Aghion, P., Howitt, P. and Murtin, F. (2011), “The relationship between health and growth: when Lucas meets Nelson-Phelps”, Review of Economics and Institutions, Vol. 2 No. 1, pp. 1-24.
Alderman, H., Hoddinott, J. and Kinsey, B. (2006), “Long term consequences of early childhood malnutrition”, Oxford Economic Papers, Vol. 58 No. 3, pp. 450-474.
Barreto, R.A. and Hughes, A.W. (2004), “Under performers and over achievers: a quantile regression analysis of growth”, Economic Record, Vol. 80, pp. 17-35.
Barro, R. (1990), “Government spending in a simple model of endogenous growth”, Journal of Political Economy, Vol. 98 No. 5, pp. S103-S125.
Barro, R. (1991), “Economic growth in a cross section of countries”, Quarterly Journal of Economics, Vol. 106 No. 2, pp. 407-443.
Barro, R. and Lee, J.W. (2013), “A new data set of educational attainment in the world, 1950-2010”, Journal of Development Economics, Vol. 104, pp. 184-198.
Barro, R. and Sala-I-Martin, X. (1997), “Technological diffusion, convergence, and growth”, Journal of Economic Growth, Vol. 2 No. 1, pp. 1-27.
Belli, P., Bustreo, F. and Preker, A. (2005), “Economic benefits of investing in child health”, Bulletin of the World Health Organization, Vol. 83 No. 10, pp. 721-800.
Benhabib, J. and Spiegel, M.M. (1994), “The role of human capital in economic development: evidence from aggregate cross-country data”, Journal of Monetary Economics, Vol. 34 No. 2, pp. 143-173.
Benhabib, J. and Spiegel, M.M. (2005), “Human capital and technology diffusion”, in Aghion, P. and Durlauf, S. (Eds), Handbook of Economic Growth, North Holland, Amsterdam.
Bergh, A. and Henrekson, M. (2011), “Government size and growth: a survey and interpretation of the evidence”, Journal of Economic Surveys, Vol. 25 No. 5, pp. 872-897.
Bhargava, A., Jamison, D., Lau, L. and Murray, C. (2001), “Modeling the effects of health on economic growth”, Journal of Health Economics, Vol. 20 No. 3, pp. 423-440.
Bleakley, H. (2010), “Health, human capital, and development”, Annual Review of Economics, Vol. 2, pp. 283-310.
Bloom, D.E. and Canning, D. (2000), “The health and wealth of nations”, Science, Vol. 287 No. 5456, pp. 1207-1209.
Bloom, D.E., Canning, D. and Sevilla, J. (2004), “The effect of health on economic growth: a production function approach”, World Development, Vol. 32 No. 1, pp. 1-13.
Canarella, G. and Pollard, S.K. (2004), “Parameter heterogeneity in the neoclassical growth model: a quantile regression approach”, Journal of Economic Development, Vol. 29 No. 1, pp. 1-31.
Canay, I. (2011), “A simple approach to quantile regression for panel data”, The Econometrics Journal, Vol. 14 No. 3, pp. 368-386.
Ciccone, A. and Jarocinski, M. (2010), “Determinants of economic growth: will data tell?”, American Economic Journal: Macroeconomics, Vol. 2 No. 4, pp. 222-246.
Cooray, A. (2013), “Does health capital have differential effects on economic growth?”, Applied Economics Letters, Vol. 20 No. 3, pp. 244-249.
Crespo-Cuaresma, J., Foster, N. and Stehrer, R. (2011), “Determinants of regional economic growth by quantile”, Regional Studies, Vol. 45 No. 6, pp. 809-826.
Eggoh, J., Houeninvo, H. and Sossou, G.A. (2015), “Education, health and economic growth in african countries”, Journal of Economic Development, Vol. 40 No. 1, pp. 93-111.
Feenstra, R.C., Inklaar, R. and Timmer, M.P. (2015), “The next generation of the Penn world table”, American Economic Review, Vol. 105 No. 10, pp. 3150-3182.
Grossman, M. (1972), “On the concept of health capital and the demand for health”, Journal of Political Economy, Vol. 80 No. 2, pp. 223-255.
Hanushek, E.A. and Woessmann, L. (2011), “How much do educational outcomes matter in OECD countries?”, Economic Policy, Vol. 26 No. 67, pp. 427-491.
Hausman, J.A. (1978), “Specification tests in econometrics”, Econometrica, Vol. 46 No. 6, pp. 1251-1271.
Howitt, P. (2005), “The role health plays in economic growth”, in Lopez-Casasnovas, G., Rivera, B. and Currais, L. (Eds), Health and Economic Growth: Findings and Policy Implications, MIT Press, Cambridge.
Jack, W. and LEWIS, M. (2009), “Health investments and economic growth: macroeconomic evidence and microeconomic foundations”, The World Bank, Policy Research Working Paper Series, 4877.
Knowles, S., Lorgelly, P. and Owen, P.D. (2002), “Are educational gender gaps a break on economic development? Some cross-country empirical evidence”, Oxford Economic Papers, Vol. 54 No. 1, pp. 119-149.
Koenker, R. (2012), Quantreg: Quantile Regression.
Koenker, R. (2004), “Quantile regression for longitudinal data”, Journal of Multivariate Analysis, Vol. 91 No. 1, pp. 74-89.
Koenker, R. and Bassett, G. (1978), “Regression quantiles”, Econometrica, Vol. 46 No. 1, pp. 33-50.
Koenker, R. and Bassett, G. (1982), “Robust tests for heteroscedasticity based on regression quantiles”, Econometrica, Vol. 50 No. 1, pp. 43-61.
Lechtig, A. (1980), “Relationship between maternal nutrition and infant mortality”, in Santos W., Lopes, N., Barbosa, J.J., Chaves, D. and Valente, J.C. (Eds), Nutritional Biochemistry and Pathology. Nutrition and Food Science (Present Knowledge and Utilization), Springer, Boston, MA, Vol 3.
Levin, A., Lin, C.-F. and Chu, J. (2002), “Unit root tests in panel data: asymptotic and finite-sample properties”, Journal of Econometrics, Vol. 108 No. 1, pp. 1-24.
Lewis, M. (2006), “Governance and corruption in public health care systems”, Center for Global Development Working Paper 78, Center for Global Development.
Lohr, K.N. (1988), “Outcomes measures: concepts and questions”, Inquiry: a Journal of Medical Care Organization, Provision and Financing, Vol. 25 No. 1, pp. 37-50.
López-Casasnovas, G., Rivera, B. and Currais, L. (2005), “Health and economic growth”, in López-Casasnovas, G.., Rivera, B. and Currais, L. (Eds), Health and Economic Growth: Findings and Policy Implications, MIT Press, Cambridge, Mass.
Lorentzen, P., Mcmillan, J. and Wacziarg, R. (2008), “Death and development”, Journal of Economic Growth, Vol. 13 No. 2, pp. 81-124.
Lucas, R. (1988), “On the mechanics of economic development”, Journal of Monetary Economics, Vol. 22 No. 1, pp. 3-42.
Mankiw, G., Romer, D. and Weil, D.N. (1992), “A contribution to the empirics of economic growth”, Quarterly Journal of Economics, Vol. 107 No. 2, pp. 407-437.
Mello, M. and Perelli, R. (2003), “Growth equations: a quantile regression exploration”, Quarterly Review of Economics and Finance, Vol. 43 No. 4, pp. 43-667.
Miguel, E. and Kremer, M. (2004), “Worms: identifying impacts on education and health in the presence of treatment externalities”, Econometrica, Vol. 72 No. 1, pp. 159-217.
Miles, W. (2004), “Human capital and economic growth: a quantile regression approach”, Applied Econometrics and International Development, Vol. 4 No. 2, pp. 23-52.
Moral-Benito, E. (2012), “Determinants of economic growth: a Bayesian panel data approach”, The Review of Economics and Statistics, Vol. 94 No. 2, pp. 566-579.
Nelson, R. and Phelps, E. (1966), “Investment in humans, technology diffusion and economic growth”, American Economic Review, Vol. 56 Nos 1/2, pp. 69-75.
Nixon, J. and Ulmann, P. (2006), “The relationship between health care expenditure and health outcomes”, The European Journal of Health Economics, Vol. 7 No. 1, pp. 7-18.
Oto-Peralías, D. and Romero-Ávila, D. (2013), “Tracing the link between government size and growth: the role of public sector quality”, Kyklos, Vol. 66 No. 2, pp. 229-255.
Poças, A. (2012), “The Interrelations between Health, Human Capital and Economic Growth. Empirical Evidence for OECD countries and Portugal”, FACULTY OF ECONOMICS, U. O. C. (ed.) PhD in Economics dissertation.
Preston, S. (1975), “The changing relation between mortality and level of economic development”, Population Studies, Vol. 29 No. 2, pp. 231-248.
Romer, P.M. (1990), “Endogenous technological change”, Journal of Political Economy, Vol. 98 No. 5, pp. 71-102.
Sachs, J.D. (2001), “Macroeconomics and health: investing in health for economic development”, Report of the Commission on Macroeconomics and Health Geneva, World Health Organization, Geneva.
Sala-I-Martin, X. (2005), “On the health-poverty trap”, lopez-Casasnovas, G., Rivera, B. and Currais, L.. (Eds), Health and Economic Growth: Findings and Policy Implications, The MIT Press, Cambridge, Mass.
Sala-I-Martin, X., Doppelhofer, G. and Miller, R. (2004), “Determinants of long-term growth: a Bayesian averaging of classical estimates (BACE) approach”, American Economic Review, Vol. 94 No. 4, pp. 813-835.
Savvides, A. and Stengos, T. (2009), Human Capital and Economic Growth, Stanford University Press, Stanford.
Solow, R. (1956), “A contribution to the theory of economic growth”, Quarterly Journal of Economics, Vol. 70 No. 1, pp. 65-94.
Soukiazis, E. and Cravo, T. (2007), “The interaction between health, human capital and economic growth: empirical evidence”, Mimeo, Faculty of Economics, University of Coimbra.
Wagstaff, A. and Claeson, M. (2004), Rising to the Challenge: The Millennium Development Goals for Health, World Bank, Washington, DC.
Wang, K. (2011), “Health care expenditure and economic growth: quantile panel-type analysis”, Economic Modelling, Vol. 28 No. 4, pp. 1536-1549.
Wegman, M. (2001), “Infant mortality in the 20th century, dramatic but uneven progress”, The Journal of Nutrition, Vol. 131 No. 2, pp. 401S-408S.
Young, A. (2005), “The gift of the dying: the tragedy of AIDS and the welfare of future African generations”, Quarterly Journal of Economics, Vol. 120 No. 2, pp. 423-466.
Acknowledgements
The authors would like to thank an anonymous reviewer for the insightful suggestions and comments that greatly contributed to improving the final version of the paper.
Corresponding author
About the authors
Francisca Rosendo Silva is an Economist. She earned her Master in Economics degree from the Faculty of Economics, University of Coimbra, Portugal, with a specialization in growth economics.
Marta Simões is an Assistant Professor at the Faculty of Economics, University of Coimbra and a Researcher affiliated with the Centre for Business and Economics Research (CeBER), University of Coimbra. Her research interests include economic growth, inequality, human capital, innovation and applied econometrics. She is the author or co-author of several publications in international scientific peer reviewed journals, book chapters and conference proceedings. She holds a PhD degree in Economics from the Faculty of Economics, University of Coimbra.
João Sousa Andrade holds a Doctorat d’État en Sciences Économiques, University of Poitiers. He is a Full Professor at the Faculty of Economics, University of Coimbra and a Researcher affiliated with the Centre for Business and Economics Research (CeBER), University of Coimbra. His research interests include macroeconomics, monetary policy and applied econometrics. He is the author and co-author of several publications in international scientific peer reviewed journals, book chapters and conference proceedings.