Introduction

Pollution is among the many environmental challenges the world is facing today. It is a major concern in much of the developing countries as the impact of pollution is more severe with levels that often cause poor health, deaths, and disabilities for many people, which consequently leads to higher mortality rates. Air pollution was linked with 1 out of every 8 deaths or around 7 million people globally (Rees 2016). According to Pena and Rollins (2017), the effect from disclosure about environmental air pollution is much greater in developing regions, where population growth and rapid industrialisation have resulted in low air quality. The emissions from substantial industrial activities and high vehicle emissions sustained from heavy pollutants greatly contribute to air pollution and are a significant cause to global climate change. According to Marchwinska-Wyrwal et al. (2011), about half the pressure on human health from air pollution is endured by people in developing countries. This unfortunate association can be particularly relevant in developing countries, as raised by Di et al. (2017) and Deguen et al. (2015), where people with a low socioeconomic status are more vulnerable to elevated rates of pollutants, and hence, the adverse effect rate is greater than people with a higher socioeconomic status.

However, in many developing countries, the environmental cost of pollution is unavoidable as the industrial sector is a crucial sector that generates the primary engine of economic growth and development. Thus, the problem with linking mortality to economic growth and environmental pollution is crucial for developing nations, particularly for the countries that have lack of environmental policies (Tanaka 2015; Aliyu and Ismail 2016; Zhao 2019), inappropriate health infrastructures (Rasoulinezhad et al. 2020; Kiross et al. 2020), and fewer contributions of renewable energy resources in economic sectors (Martins et al. 2018; Van Tran 2020). Moreover, the region’s economy still relies on fossil fuel consumption due to the cheapest and energy resource available, leading to higher economic growth, but eventually comes with more environmental pollution that causes social negative effects related with health problems such as increased mortality rates.

Theoretically, a sizeable number of epidemiological studies have established a positive association between air pollutant exposure and mortality incidence rate (Fang et al. 2016; Garcia et al. 2016; Yitshak-Sade et al. 2019; Fan and Wang 2020; Oluleye and Adabale 2020; Wu et al. 2020). Air pollution coming from a variety of sources has a range of chemical compositions and toxicity that can harm human health when exposed to it. Various pollutants were used in the literature, along with particulate matter (PM), sulphur dioxide (SO2), nitrogen dioxide (NO2), and carbon dioxide (CO2), among others, to evaluate environmental pollution. However, this study emphasizes on particulate matter (PM2.5) and CO2 emissions as air pollution measurement. PM2.5 is categorized under national pollutants, which is likely to be addressed by regulation at the national level, whereas CO2 is classified as an international pollutant and has a global impact (Frankel and Rose 2005). This study is similar with Aliyu and Ismail (2016), albeit they use PM10 and CO2 to proxy air pollution. Contrastingly, our study uses PM2.5, which is suggested to be more toxic compared to huge particles (Zanobetti and Schwartz 2009) and presents the greatest health hazards because the particles may reach deep through the lungs and the bloodstream, which leads to cardiovascular and respiratory diseases that contribute to higher incidence of mortality (Anenberg et al. 2010; Zeger et al. 2008; Marchwinska-Wyrwal et al. 2011; Cesaroni et al. 2013; Carey et al. 2013; Garcia et al. 2016). Moreover, many of countries in developing region are among the highest exposed population to ambient pollution of PM2.5 (World Bank 2019). On the other hand, earlier research indicates that mortality rates of a country are a comprehensive measure of the health status of that country, resulting from several economic factors that may consist of income and public health expenditure, or social factors including education level of the population, and environmental factors (Halicioglu 2011). In addition, governance quality may also be a crucial factor in understanding the health outcomes (Pautrel 2009).

Despite increasing interests regarding adverse health effects from air pollution across developing nations, the majority of accessible studies on the topic focus on children (Siddique et al. 2011; Cesur et al. 2013; Greenstone and Hanna 2011; Tanaka 2015; Arceo et al. 2016). The behaviour and exposure to health risks towards adults are crucial as they are within the most economically productive age span that contribute towards economic growth. With limited studies on this, the adult mortality rate is becoming a crucial factor for a comprehensive evaluation of population mortality trends. Exposure to toxins from ambient air pollution substances in their working areas, arduous physical work, and physical activity may have long-term health implications. This presents a problem for adults, who are proportionately exposed to environmental pollutants while working. Air pollution–related health studies on adult populations showed they are a subpopulation considered vulnerable to air pollution exposure (Barreca et al. 2017; Sanyal et al. 2018; Achilleos et al. 2019). Numerous studies have supported growing evidence that the harm of PM2.5 is expanding gradually and becomes as one of the leading risk factors for death. In various populations, it has been reported that increased concentrations of PM2.5 cause increased susceptibility to chronic diseases such as respiratory diseases and lung cancer, thereby increasing the probability of death rates (Fang et al. 2016; Garcia et al. 2016; Li et al. 2018; Wang et al. 2020). Yu et al. (2020) found an increased risk of mortality due to the long-term exposure of PM2.5 in areas reported with consistently low PM2.5 concentrations in Australia. Relatedly, Tian et al. (2019) found that short-term acquaintance to PM2.5 was positively correlated towards adult hospital admittances in China, even at rates lower than the present Chinese ambient air quality requirements. These studies show that PM2.5 can contribute in lowering life spans even in places where the levels of concentration are considered safe, suggesting that the adverse impact could be more prevalent as air pollution increases. The results of Taghizadeh-Hesary and Taghizadeh-Hesary (2020) also confirm that PM2.5 and CO2 are major risk factors for lung cancer that contributes to higher incidence of mortality in Southeast Asia. In another study, Cao et al. (2018) demonstrate that in 31 provinces in China, a significant association exists between PM2.5 and mortality based on geographically weighted regression model.

The study of the impacts of energy use on mortality in many studies also revealed that any increase in carbon emissions has a considerable effect on mortality (Erdoğan et al. 2019; Li et al. 2019; Matthew et al. 2019; Anser et al. 2020). Aliyu and Ismail (2016) employed an air pollution and mortality dataset in 35 African countries and found that particulate matter (PM10) and carbon dioxide (CO2) at higher levels has a major impact on increased mortality in adults. It is discovered that CO2 emissions in the Commonwealth of Independent States region had a positive relationship with mortality from cardiovascular disease, diabetes mellitus, cancer, and chronic respiratory disease among adults (Rasoulinezhad et al. 2020). Similarly, a growing industrial sulphur dioxide (SO2) emissions by ten thousand tonnes in some cities would lead to increased local mortality (Chen et al. 2017). On the other hand, air pollution has done harm not only to health but also economic losses. A recent study in China reported that by 2030, PM2.5 emissions could lead to a loss of 2.0% in gross domestic production, a loss of 210 billion Chinese yuan in health spending, and a loss of about 10,000 billion lives, if no control policies undertaken (Xie et al. 2019).

However, there is a growing argument that more active governmental roles addressing this issue can help to control the adverse health effects of pollution in better-governed nations (Siddiqi et al. 2009; Lin et al. 2014; Aliyu and Ismail 2016). In some studies, the governance effect was direct and positive (Glatman-Freedman et al. 2010; Holmberg and Rothstein 2011; Burchi 2011; Olafsdottir et al. 2011). For example, according to Rosenberg (2018), if Paraguay improves the quality of governance to its fullest extent, the country can achieve almost a 16% reduction in mortality. Ahmad and Hasan (2016) found that high rate of corruption (bad governance) prevents the improvement of health status in the long run. Similarly, good governance in the country is found to play a significant role for improving the health state along with other socio-demographic factors in Ethiopia (Biadgilign et al. 2019). On the other hand, the governance effect on health was found to be indirect or modified by contextual factors (Farag et al. 2013; Hu and Mendoza 2013). Bollyky et al. (2019) suggest that democracies contribute to health benefits for mortality causes when a nation is followed by free and fair elections. Doucouliagos et al. (2019) shows that the effectiveness of health aid in reducing mortality is conditional on good governance (measured either as government effectiveness or control of corruption). A study conducted by Makuta and O’Hare (2015) found that an efficiently managed government correlates positively with better health outcomes. Its effect is modified by the governance level, by having a greater health effect in nations with a better governance level and less influence in nations with lower government performance. This can be attributed to the improved efficiency and better allocation of resources available as the quality of governance increases. This result is similar to Sirag et al. (2017). In addition, according to Kim and Wang (2019), governance quality and quantity had major impacts on public health. They measured the quantity of government performance by public expenditure on health while regulating for control, efficiency of government, regulatory quality, voice and accountability, and the rule of law for government quality.

Thus it is suggested that countries with stronger governance would certainly have better health policies that can safeguard their countries from unintended pollution adverse effects and thus benefit the population’s health outcomes. Therefore, this study argues whether there is a possibility the role of governance in mitigating the effects of air pollution on mortality is feasible. With this background, the objective of this study is to investigate the air pollution effects towards adult mortality.

Recognizing the enormity of current literature about these issues in high-income settings, this study’s contribution is derived from the lack of studies that focus on developing settings. The study follows the generalized method of moments (GMM) technique, which are to be discussed later, with a panel of 72 developing countries. The rest of the article is set out as follows. In the next section, the details of the estimation models, method, and data are provided. The outcomes of the estimations are then addressed in Section 3 followed by a discussion. Finally, Section 4 provides the conclusion with a few summary remarks.

Methodology

Following Coneus and Spiess (2012) and Aliyu and Ismail (2016), Eq. 1 is empirically modelled to estimate the effects of air pollution on adult mortality rate:

$$ {lAM}_{it}={\alpha}_0+{\alpha}_1{lAM}_{it-1}+{\alpha}_2{lAP}_{it}+{\alpha}_3l{Y}_{it}+{\alpha}_4{lPHS}_{it}+{\alpha}_5{lEDU}_{it}+{\mu}_i+{\varepsilon}_{it}\kern0.5em $$
(1)

where lAM is the dependent variable proxy by adult mortality rates, which refers to the probability that those who are 15 years old will die before they reach age 60. lAP is the variable of interest, namely, air pollution which is a proxy for PM2.5 levels and CO2 emissions. lY denotes real GDP per capita, lPHS is the public health expenditure, and lEDU represents the education level of the population; the subscripts i and t, respectively, represent country and time period; μt and ɛ denote country-specific effects and the error term, respectively. The lagged dependent variable (lAMit − 1) refers to the dynamic effect, where the existence of the probability of mortality relies on itself in the preceding year, and the coefficient must be less than 1 because of the persistency of the variable to be statistically significant.

The main variables of interest are air pollution proxy by using PM2.5 and CO2, which are expected to have a negative relationship with mortality rate. PM2.5 can reach the lower respiratory tract due to its small size and thus have a greater potential for causing lung and heart diseases that can eventually lead to a reduced mortality expectancy (Ghorani-Azam et al. 2016). Additionally, carbon dioxide can cause severe health issues that have been shown to be significantly linked to higher mortality rates in humans (Satish et al. 2012). Thus, higher mortality rates in adults are expected throughout developing countries due to air pollution exposure.

As in the standard literature, relevant explanatory variables that may affect adult mortality are included in the regressions. Income is measured as real GDP per capita to reflect the level of economic growth for a nation. Government health spending is a measure of public investment in the health sector and human capital, which includes the provision of health facilities, family planning programs, nutrition initiatives, and emergency health assistance that include access to health care and facilities. Education level is considered to be another influential determinant of mortality where it increases labour market productivity as well as income growth. Education also raises people’s health awareness and hence they tend to take precautions to create a better state of health for themselves. Therefore, a positive effect is expected from these explanatory variables on adult mortality.

Previous studies indicated that countries with stronger governance would implement health policies to favour the social variables of health (Siddiqi et al. 2009; Lin et al. 2014; Kim and Wang 2019). Thus, the need for better governance to improve health outcomes has been growing. Hence, the specified estimation below includes governance quality index (GQI) as an additional explanatory variable:

$$ {lAM}_{it}={\alpha}_0+{\alpha}_1{lAM}_{it-1}+{\alpha}_2{lAP}_{it}+{\alpha}_3l{Y}_{it}+{\alpha}_4l{PHS}_{it}+{\alpha}_5{lEDU}_{it}+{\alpha}_6{GQI}_{it}+{\mu}_i+{\varepsilon}_{it}\kern3.25em $$
(2)

where GQI represents six components of the Worldwide Governance Indicators (WGI), namely, voice and accountability, political stability, government effectiveness, regulatory quality, rule of law, and control of corruption. All of these components are derived on the basis of a set of variables and eventually converted into standardized indices that range approximately between − 2.5 and + 2.5. The principal component analysis (PCA) method is employed to construct GQI from the six components of governance quality to specifically study the impact of aggregate quality of governance rather than focus on individual components, as their role is equally important.

The regression specification is extended by allowing the interaction of the air pollution measure with the governance quality index. According to Lin et al. (2014), sound policymaking and effective policy execution by the government are viewed to be an important determinant of the population’s health. Good governance, therefore, can impose better environmental policies to improve the environmental conditions (Tamazian and Rao 2010; Greenstone and Hanna 2011; Tanaka 2015; Ibrahim and Law 2016). Accordingly, the equation for regression with an interaction term is as follows:

$$ {lAM}_{it}={\alpha}_0+{\alpha}_1{lAM}_{it-1}+{\alpha}_2{lAP}_{it}+{\alpha}_3l{Y}_{it}+{\alpha}_4l{PHS}_{it}+{\alpha}_5{lEDU}_{it}+{\alpha}_6{GQI}_{it}+\kern0.75em {\alpha}_7\left({lAP}_{it}\ast {GQI}_{it}\right)+{\mu}_i+{\varepsilon}_{it} $$
(3)

where (lAPit ∗ GQIit) measures the interaction between PM2.5 and governance quality index. The coefficients of these interaction terms are employed to test whether the negative effect of air pollution on adult mortality is modified by the quality of governance. A negative and significant β7 in Eq. 3 confirms that governance quality plays a role as moderator on reducing air pollution risk to reduce adult mortality. Understandably, this interaction suggests that in order to achieve an environmental goal consistent with the public’s interest, there is a need for specific task allocation to the appropriate governance to ensure its role is effective. The environmental benefits from this could enhance population health and would contribute to greater productivity in the economy.

Given the panel nature of the data, Eqs. 1, 2, and 3 were assessed by employing the panel generalized method of moments (GMM) estimator, designed for dynamic panel data models that were proposed by Holtz-Eakin et al. (1988), Arellano and Bond (1991), and Arellano and Bover (1995). GMM is employed because it controls the country-specific effects, which cannot be used with country-specific dummies because of the dynamic structure of the equation (Abdul Bahri et al. 2019). In addition, GMM controls over the simultaneity bias, since few from the explanatory variables are possibly endogenous.

Consequently, the traditional panel models (such as pooled OLS, fixed-effect panel model, and random-effect panel model) are not suitable due to the presence of country-specific effects and the lagged dependent variable or potential endogeneity of explanatory variables. Thus, to tackle these problems, the GMM estimator is employed. There are two types of GMM estimators, which are the difference and system GMM. Precisely, by compelling first the differences, the GMM method removes the country-specific effects or whichever time-invariant country-specific variable. Later, to overcome the resultant correlation of lagged dependent variable and disturbance terms following first differencing, Arellano and Bond (1991) propose the applying of instrumental variables to deal with the regressors’ endogeneity problem.

In the first difference GMM estimation, the difference lagged dependent variables and the endogenous variables may be instrumented accordingly by their lags in levels, whereas the exogenous variables may become their own instruments. While the above difference estimator is capable of controlling the country-specific effects and simultaneity bias, it still has a considerable shortcoming. Alonso-Borrego and Arellano (1999) and Blundell and Bond (1998) indicate that while the explanatory variables are persistent, the variables’ lagged levels turn out to be weak instruments and thus can result in biased parameter estimates within small samples and a larger variance asymptotically. Blundell and Bond (1998) display that the system GMM estimator is capable of lessening the biases and imprecision related to the difference estimator. This estimator is used as it increases efficiency by utilizing both lagged levels and lagged differences. Further, alternatively, GMM estimators deemed in their one-step and two-step forms. Theoretically, the two-step estimator is farther efficient in regard to the one-step estimator due to the use of optimal weighting matrices. Hence, a two-step system GMM estimator is adopted as it provides a better result, and for this cause, this study also employs this estimation technique.

GMM estimator consistency leans on two validity tests. First, the Hansen overidentification restrictions shall be used to check the instruments’ validity. From the test, the instruments are shown to be valid and endogeneity bias can be overcome when the null of overidentification hypothesis is failed to be rejected. Secondly, in regard to the serial correlation checking, the null hypothesis of no serial correlation is not rejected in AR (2). The failure to reject the Hansen J-test and AR (2) test’s null hypothesis suggests that the GMM estimator is reliable and efficient and has a lack of bias, and the instruments are valid (Arellano and Bond 1991; Arellano and Bover 1995; Blundell and Bond 1998).

This study employed data from a panel of 72 developing countries covering the 2010–2017 period. The countries and time period were chosen dependent on the data’s availability. Mortality rates for adults were collected from the World Development Indicators of the World Bank. Particulate matter (PM2.5) was estimated via micrograms per cubic meter, while carbon dioxide (CO2) was estimated via metric tons per capita, real GDP per capita shows the level of income or economic progress, public health spending is the amount of public and private health spending as a percentage of total GDP, and education level is measured by the school enrolment rate. These data likewise were extracted from the World Bank. Nevertheless, the government quality indicator comprises of voice and accountability, political stability, government effectiveness, regulatory quality, rule of law, and control of corruption measures which were taken from the World Governance Indicator of the World Bank.

Results and discussion

In this section, the results are reported following the two-step system GMM estimator. The results of Hansen J test are stated to report the validity of the dynamic panel model. Since the p value of the Hansen test in any of the models is statistically not significant, the null hypothesis cannot be rejected. The AR (2) revealed that the moment conditions of the models are satisfied since the p values are not statistically significant. Therefore, the Arellano-Bond test confirms the second-order serial correlation and the Hansen test indicates that the over-instrumentation problem is minimized in all regressions, justifying that the estimated model met the diagnostic tests.

The datasets of descriptive statistics are presented in Table 1. The table shows the means, standard deviations (within countries, between countries, and overall countries), as well as minimum and maximum values. According to the table, the difference of minimum and maximum values for adult mortality rates is large, which is 490 (per 1000 adults). It is shown that Lebanon and Lesotho have the lowest and highest rate of mortality at 57 and 548, respectively. In the PM2.5 tabulation, the same discrepancy is shown. The maximum concentration of PM2.5 from the samples is 97.6, whereas the lowest concentration is 10.5 in micrograms per cubic meter as reported in India and Kiribati. This may indicate plausibly that air pollution could be a significant factor in the rising adult mortality in developing countries. In addition, the governance quality index is low (indicated by negative values) for most countries, as shown by the mean in the table.

Table 1 Descriptive statistics for adult mortality rates and its determinants
Full size table

The correlation results show that there is a positive correlation between adult mortality and pollutants, i.e. the correlation coefficient is 0.071 for PM2.5 and 0.155 for CO2. On another hand, there is a negative correlation between adult mortality and the rest of the variables, including the interactive terms, i.e. the correlation coefficient is 0.158 for PM2.5 and 0.037 for CO2. The overall results come to the conclusion that a negative association of interactive terms emphasized the need for government role to moderate the adverse effect of air pollution on health (Table 2).

Table 2 Correlation matrix
Full size table

Regression results are given in Tables 3 and 4. Table 3 displays the results of the basic regression (column 1) and the regressions results, where governance quality index is added (column 2). In addition, the expanded regression models with the added interaction between air pollution (PM2.5) and governance quality index are shown in column 3. The robustness of the estimated results was then tested by re-estimating the models with CO2 as the air pollution predictor as presented in Table 3.

Table 3 GMM estimation results: PM2.5 and governance quality index effects on adult mortality rates
Full size table
Table 4 GMM estimation results: CO2 and governance quality index effects on adult mortality rates
Full size table

The basic regression indicates consistent findings with the past empirical studies. PM2.5 carries a significantly positive coefficient, indicating that the increasing level of particulate air pollution contributes to an increase in adult mortality rates in the developing countries. The analysis shows that a 10% rise in the PM2.5 level causes adult mortality rates to rise between 0.04% and 0.06%. Therefore, the indication that higher adult mortality is associated with PM2.5 seems to confirm the epidemiologic finding that air-polluting particles can lead to higher mortality rates (Yitshak-Sade et al. 2019; Oluleye and Adabale 2020; Wu et al. 2020). These findings are compatible with He et al. (2016) and Aliyu and Ismail (2016) and de Keijzer et al. (2017) and Cao et al. (2018). Moreover, the proxies of real GDP per capita, government health spending, and education level have negative and significant coefficients, denoting their beneficial impact on reducing the adult mortality rates as expected. This result confirms the findings of Zhang et al. (2014) and Grigoli and Kapsoli (2017) that these explanatory variables are associated with lower mortality.

With the addition of governance quality index in the model, the coefficient tends to be not statistically significant in affecting adult mortality. However, when PM2.5 and governance quality index in the equivalent model are examined, the result shows that adult mortality has negative signs and is statistically significant at 5% significant level. With a one-unit improvement in governance quality index, it decreases mortality among adults in the developing countries by 0.001%. This suggests that with government action initiatives, the negative impact of pollution can be lessened. The fact that government indicator is only significant in evaluating adult mortality rates with this interaction effect suggests that countries with better governance quality can act as a plausible channel to reduce air pollution risk on health outcomes (lower adult mortality rates). Along that narrative, Biadgilign et al. (2019) and Kim and Wang (2019) conducted studies that provide evidence that good governance could correspond to better health outcomes. Nevertheless, the reported results here are slightly lower, perhaps because this study focuses on developing sample where the scores of governance level are particularly lesser.

The interactive terms’ result implies that the quality of governance influences health outcomes more effectively when it is directly targeted towards air pollution. The other study found to examine the effect of governance quality on air pollution is a study by Aliyu and Ismail (2016), while other studies interact public health spending with the quality of government to measure the effects on health outcomes (Makuta and O’hare 2015; Sirag et al. 2017; Doucouliagos et al. 2019; Kim and Wang 2019). The findings of Aliyu and Ismail (2016) were, however, unclear where there is no general conclusion that can be drawn by interacting the only component of government effectiveness of governance quality with PM10 and CO2.

Table 4 presents robustness findings by re-estimating the estimation models in Table 3 by replacing the air pollution indicator with CO2. An association between air pollution and adult mortality remains robust across all specifications. In basic estimations (column 1), carbon dioxide emission positively affects adult mortality but is not statistically significant. This result indicates that despite the wide usage of CO2 as the indicator of air pollutants present, PM2.5 could be a better measurement of an air pollutant as it is more harmful to health (significantly affects mortality). The estimation analyses in Table 3 also consistently show that the real GDP per capita, government health spending, and education level reduce levels of mortality for adults (except for education level in column 3).

However, in all regressions (columns 2 and 3), the inclusion of governance quality index as well as with CO2 shows that there is a negative association on adult mortality rates, but the variables are statistically not significant. This result is compatible with empirical findings in other developing country studies made by Delavari et al. (2008), Monsef and Mehrjardi (2015), and Amuka et al. (2018). Compared to CO2, which is considered to be a global pollutant, there is obvious public policy imperative to mitigate local pollutants (PM2.5) made by the government as global pollution requires participation from other countries altogether.

Conclusion

This study investigates the effects of air pollution on adult mortality levels in developing countries. From the air pollution and adult mortality analyses, the increasing particulate air pollution levels (PM2.5) are likely to have a significant effect on the rising rates of adult mortality rates. Additionally, the effects of air pollution appear to be more noticeable in particulate air pollution (PM2.5) than carbon dioxide emissions. The indication on the interaction effects of air pollution (PM2.5) and government quality on adult mortality rates is clear, implying that the government does not independently affect adult health, but only significant when measured together in the same model in the respective developing nations. The findings suggest that the adverse effect of air pollution could be mitigated when the government plays an effective role.

Consequently, our analysis suggests that a country should concentrate its efforts to improve air quality and creating healthy physical and social living environments with finer governance. Effective governance can lead to better policies as effective environmental regulations can be implemented, thus affecting all other functions of the health system, leading to enhanced health system performance and, therefore, better health outcomes. Increased health status may not only an outcome in its own right but also a requirement for the development process, as good health is an essential source of human capital. With better health costs for healthcare, services can be reduced and consequently improve labour productivity of a country (Dormont et al. 2010). Therefore, the interplay of governance levels and population health can be accomplished with higher governance quality in light of the movement towards economic growth for preferable living and health status in addition, government should focus on fostering local institution to monitor and promoting the use of technological developments in economic sectors and extrapolating these to develop effective policies for international sustainability.