Food insecurity, environment, institutional quality, and health outcomes: evidence from South Asia | Globalization and Health

Conceptual framework

Prior literature has predominantly focused on the health production function (HPF), establishing a conceptual framework that emphasizes endogenous factors such as health expenditures, per capita income, employment, environmental quality, lifestyle, education, and genetics [10, 20,21,]– [22]. This approach traces back to the seminal work of Auster et al. [23], who explored the impact of environmental and healthcare indicators on mortality rates. However, despite the subsequent adoption of a similar pattern by substantial body of literature, most studies have overlooked exogenous factors. These factors include external social and economic shocks as well as institutions, which can either directly or indirectly influence health outcomes measured by mortality rate or life expectancy. In conceptualizing our study, we build upon this foundation, addressing the gap in literature. Figure 3 outlines the conceptual framework (extended HPF) designed for the present inquiry. Line (1) represents the conventional approach, as seen in studies like Onyimadu et al. [24] and Salgado et al. [25], offering for an extensive systematic review of studies examining how endogenous factors influence the subject. Line (2) illustrates how external shocks from global inflationary episodes, causing fluctuations in the general price level of food and non-food items, impact health outcomes. Line (3) outlines the direct effects of institutional quality on health outcomes, while Line (4) emphasizes the moderating effects of institutional quality on endogenous health factors.

Institutional quality

Institutional quality represents the overall efficiency, reliability, and effectiveness of institutions in an economy [26]. Fundamentally, institutions encompasses rules, policies, and practices that form and instruct the behavior of individuals and organizations in a society [27,28,29]. Institutional quality is a multifaceted concept that gauges that state’s power to govern its resources for the benefit of the nation. According to the World Health Organization [30], states are responsible for designing a country’s health system based on two key pillars: resource production and efficient service provision. This design aims to facilitate the achievement of three objectives, including institutions’ responsiveness, health system efficacy, and the availability of sufficient and just financial and physical resources [31, 32]. To assess the efficacy and quality of institutions, Kaufmann and Kraay [33] developed six indicators, measuring corruption control, the rule of law, government effectiveness, political stability, regulatory quality, and voice and accountability. In essence, the higher these measures, the higher the institutional quality, signifying that a country has a robust and transparent mechanism in place to ensure the fair and efficient utilization of its resources, resulting to positive outcomes for the nations [34].

Review of empirical studies

Our study aligns with prior empirical literature on several fronts, including the health consequences of economic growth, environmental degradation, demography, social factors, and institutional quality. For instance, researchers such as Dadgar and Norström [35], Gautam [36], Niu et al. [37], Spiteri and von Brockdorff [38], Erdoğan et al. [39], Salahuddin et al. [40], Knapp and Wang [41], and Mohapatra [42] have extensively examined the effects of economic growth, utilizing either per capita GDP or GDP growth rate, on health outcome indicators across diverse nations. They employed different statistical methods both at regional and country levels. The collective findings unanimously confirmed that economic growth plays a crucially positive role in influencing health outcomes. This positive influence operates through enhancement of individuals’ economic capacity, enabling them to afford better living conditions, accessing healthcare, and improve their living standards.

Furthermore, the review of literature reveals that numerous studies have delved into the relationship between environmental degradation and health outcome indicators, establishing a general concensus on the negative consequences of increased environmental degradation on population health. Noteworthy among these studies are Gasimli et al. [43], Mumtaz et al. [44], Omri et al. [45], Taghizadeh-Hesary et al. [46], Alimi et al. [47], Clark et al. [48], Li et al. [49], Emodi et al. [50], Zeeshan et al. [51], Murthy et al. [52], and Das and Debanth [53], which specifically explored the health consequences of CO₂ emissions, ecological footprint, non-renewable energy consomption, and climate change predictors on mortality rates, life expectancy, and mental health of populations across various countries. Their collective findings consistently suggest that environmental degradation is detrimental to public health. Comparatively, FI, a sensitive topic of policy discussions worldwide, has not recived extensive scrutiny in the exisitng literature. The available studies, conducted by Beyene [13], Benzekri et al. [11], Johnson et al. [54], Dean et al. [55], Seligman et al. [56], Militao et al. [57], Pengpid and Peltzer [58], Nagata et al. [59], and Nwosu et al. [60] have explored the effects of FI on different health outcome indicators, including mental health, life expectancy, infant mortality rates, and per capita health expenditures. Using diverse data sources, these empirical investigations span different countries and regions, excluding South Asia. Despite this diversity, their unanemous findings support the assertion that FI poses an early-stage threat to human well-being, acting as a harbinger for various diseases over time.

While demography, often measured by population growth and urbanization, is considered a health-endogenous factor, recent emprical studies yield mixed responses. For instance, Jemiluyi [61], Tripathi and Maiti [62], Perrott and Holland [63], and de Meijer et al. [64] concluded that growing population rate and rapid urbanization have negative impacts on public health. Within a given per capita income, these factors increase contemporary health expenditures and suppress the overall health outcomes. Conversely, studies conducted by Huang et al. [65] and Shao et al. [66] argue that urbanization is an effective means of increasing life expectancy and reducting infant mortality rates by facilitating people with swift access to better healthcare facilities. Additionally, the review of existing empirical literature reveals that studies conducted by Liao et al. [67], Gumus and Yurumez [68], Raghupathi and Raghupathi [69], and Gottfried and Sublett [70] explored the health effects of social factors, primarily proxied by school enrollment rate across various countries, using diverse statistical methods for their analysis. In consensus, their findings emphasize that the level of education and literacy have a positive impact on life expectancy and negative effects on mortality rates. Finally, the study delved into existing literature and discovered that recent works conducted by Socoliuc et al. [71], Rahman and Alam [72], Vian [73], Van De Bovenkamp et al. [74], De Luca [75], Onofrei et al. [76], Glynn [77], Farag et al. [78], Rosenberg [79], Koller et al. [80], and Hadipour et al. [81], mostly employing the rule of law or control of corruption as proxies for institutional quality, affirm that institutional quality is crucially in promoting positive health outcomes. Essentially, they highlight the importance of governance structure and anti-corruption acts in contributing to the efficacy of healthcare systems.

Research gaps

While recent empirical studies contribute significantly to existing literature, an added dimension would involve examining how contemporary health outcomes relate to externalities. Notably, there is a dearth of studies on the comprehensive effects of institutional quality on health outcomes, covering all aspects of the institutions both as a direct and moderating predictor. This gap is more tangible in the case of South Asian countries. Another gap is the absence, to our knowledge, of studies addressing global economic shocks, particularly global inflationary periods, which could significantly raise food prices, intensifying vulnerability to food security in South Asia. Moreover, apart from Gasimli et al. [43], who investigated the impact of environmental degradation on health outcomes, no other studies were found focusing on this aspect in South Asian countries. To address these gaps and align with our conceptual framework, we propose four key hypotheses: H₁: FI and environmental pollution have severe effects on health outcomes. H₂: Inflation uncertainty, as one of the key drivers of food price volatility, negatively impacts health outcomes. H₃: Institutional quality has a direct link with health outcome indicators. H₄: The interaction of institutional quality with inflation uncertainty increases or decreases the effects of endogenous health variables.

Data and variables

Our study covers the period from 2000 to 2021, incorporating the latest available data. The empirical investigation centers on South Asian countries, including Afghanistan, Bnagladesh, Bhutan, India, the Maldives, Nepal, Sri Lanka, and Pakistan. The selection of South Asia as the context of our study is guided by two compelling reasons. Firstly, despite the abundance of studies on the health implications of FI and environmental degradation, the bloc has not received extensive attention in the existing literature. Secondly, the region is at a precarious equilibrium characterized by low staple productivity, minimal returns to formers, supply shortages, highly volatile food prices, area diversification, and low per capita income. These factors collectively contribute to escalating threats of FI on health outcome indicators, yet there is insufficient number of studies to guide contemporary policy directions for South Asia. Therefore, addressing these challenges necessitates a comprehensive analysis of the current situation to inform effective policies and resource reallocation in South Asia.

Selection of variables

Dependent variables

The selection of the variables aligns with study’s objectives and is consistent with prior empirical literature [82,83,84,85,86]. We employ life expectancy at birth (LE), representing the number of years a newborn kid would survive if the prevailing mortality patterns at the time of its birth persist throughout its life. Additionally, we incorporate infant mortality rates (MR), indicating the number of kids who die before reaching one year of age per 1,000 live births per year. In this study, LE and MR are used as dependent variables. While LE represents a broader overview of a nation’s health outcomes, MR is considered as a micro-predictor. It is essential to examine the response of both macro- and micro-health outcome predictors to the explanatory variables.

Explanatory variables

In addition to two innovatively constructed variables for inflationary shocks and the institutional quality index (details in the next section), the study builds upon previous studies [13, 87,88,89,90,91,]– [92] and employs three indicators, namely, prevalence of undernourishment (PN), per capita kilocalorie supply (KS), and inequality in per capita calorie intake (CI), as explanatory variables to measure FI in South Asia. PN is expressed as the percentage of people with insufficient regular food intake to maintain a typical, active life; a data value of 2.5 indicates a malnutrition rate lower than 2.5%. Moreover, KS represents the amount of all types of daily food supplies, measuring the available quantity of food for consumption. CI is expressed as the coefficient of variation of energy intakes, with a higher coefficient indicating greater inequality. These indicators are widely used in literature and are considered as best-fit proxies for measuring FI.

Control variables

To control for the effects of various social, economic, demographic, and environmental factors, the study gauges social factors through the school enrollment rate (SE), expressed as the gross percentage of enrollment in primary schooling to the total enrollment [83, 93]. SE is employed to capture the effects of education and literacy on the subject. Moreover, to account for macro-level economic variations and their effects on the dependent variables, per capita GDP growth (PG, annual %) is employed as a control variable [94, 95]. Per capita health expenditure (HE, constant 2015 US$) is utilized, following [96] and [97], to control for their effects on LE and MR. In this context, HE enables the assessment of the effects of out-of-pocket spending on the subject. Additionally, per capita CO₂ emissions (CO₂e), resulting from the use of fossil fuels and industry, serve as an environmental degradation variable [53]. Lastly, the study incorporates urbanization (UR, % of population) as a control variable for its effects on LE and MR [98, 99].

Construction of new variables

This part addresses the construction of the inflationary shock variable and institutional quality index. The persistent growth in the general price level of food items, especially when it is unpredictable in the future, cannot be overruled, Considering the previous period of inflationary episodes in South Asia that hindered general food prices, we innovatively construct a predictor of inflation uncertainty (InF). This allows for a more precise evaluation of the effects of economic variability on both LE and MR. In doing so, we use the datapoints of the annual inflation rate and the generalized autoregressive conditional heteroskedasticity (GARCH) model of Bollerslev [100] as follows:

$$VAR\left( {{\varepsilon _{INF,t}}} \right)=\sigma _{t}^{2}+{\vartheta _0}+{\vartheta _1}\varepsilon _{{INF,t – 1}}^{2}+\zeta \sigma _{{t – 1}}^{2}$$

(1)

In Eq. (1), $ VAR\left({\epsilon }_{INF,t}\right)$ is the conditional variance of error term of the annual inflation rate, $ {\vartheta }_{0}$ and $ {\vartheta }_{1}$ are the intercept and autoregressive conditional heteroskedasticity parameter, respectively, and $ \zeta {\sigma }_{t-1}^{2}$ represents the GARCH parameter. Additionally, since the 1980s, political instability, ineffective government, and, most importantly, corruption have been serious issues in South Asian economies that have brought local and international concerns to the fore [101]. This has been an alarming concern to most of the financial aid to uplift poverty, basic healthcare services, and FI. However, South Asian governments adopted programs of anti-corruption, they only remained as populist mottos. Based on the Worldwide Governance Indicators (see Fig. 4), though all institutional indicators are comparatively lower than other regions, political stability stands at 26.53 percentile rank, followed by regulatory quality at 30.65 percentile rank, voice and accountability at 33.29 percentile rank, rule of law at 37 percentile rank, and government effectiveness at 38.91 percentile rank. Values below the 50-percentile rank are alarming signs of poor institutional quality.

Thus, to account for both the direct and moderating effects of institutional quality on the subject, we innovatively construct a comprehensive institutional quality index (InQ) following the distance-based approach proposed by Sarma [102]. This technique has recently gained prominence in the literature and has several advantages over common methods [103,104,105,106]. Figure 5 displays the constructed institutional quality index (InQ).

Initially, the study compiled relevant data at the country level and subsequently constructed a comprehensive panel for South Asia, encompassing 8 countries. The datasets for LE, MR, SE, PG, HE, UR, and annual CPI-based inflation rate were sourced from the World Development Indicators [107]. Additionally, the datasets for PN, KS, and CI were obtained from UN-FAO [108]. The data for per capital CO₂e was sourced from the Global Carbon Budget, available in [109]. Finally, datasets for constructing the InQ have been compiled from Worldwide Governance Indicators [110] sources.

Estimation methods

Our primary objectives are to investigate how both endogenous and exogenous predictors influence health outcomes in South Asia. To that end, we modify the existing health production function using the lines of direction shown in our conceptual framework. First, to test the effects of FI, environmental factors, and other economic and social indicators in the presence of InQ and InF on health outcomes, we specify the following multivariate long-run equation:

$$\begin{gathered} H{O_{it}}=\delta +{\eta _1}P{N_{it}}+{\eta _2}K{S_{it}}+{\eta _3}C{I_{it}}+{\eta _4}S{E_{it}}+{\eta _5}P{G_{it}}+{\eta _6}H{E_{it}} \hfill \\ \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,+{\eta _7}C{O_2}{e_{it}}+{\eta _8}U{R_{it}}+{\eta _9}In{Q_{it}}+{\eta _{10}}In{F_{it}}+{\wp _t}+{\varepsilon _{it}} \hfill \\ \end{gathered} $$

(2)

where all variables are defined before, HO refers to health outcome proxied by LE and MR, $ \delta $ is the intercept, and $ {\eta }_{1}$ to $ {\eta }_{10}$ are the long-run coefficients. Subscripts $ i$ represents the countries and $ t$ denotes time dimension. Equation (1) and the subsequent regressions account for country-specific unobserved fixed effects represented by $ \wp $. Finally, $ {\epsilon }_{it}$ presents the error term. To examine the moderating effects of InF and InQ on the relationship between health outcomes and the endogenous variables, we specify the following equation:

$$H{O_{it}}=\delta +\sum\limits_{{j=1}}^{{10}} {{\eta _j}{X_{it}}+} \,\theta \left( {{Z_{it}} \times {X_{it}}} \right)+\,{\wp _t}+{\varepsilon _{it}}$$

(3)

where $ {\eta }_{j}$ refers to the long-run coefficients of the explanatory variables $ {X}_{it}$ and $ \theta $ represents the long-run coefficient of the interaction term of the $ {Z}_{it}$ (InF or InQ) with the explanatory variables, using separate regressions for each interaction models. In order to incorporate the interaction terms into Eq. (3), we follow the same methodology as proposed by Abaidoo and Agyapong [111] and Dada and Ajide [112]. In doing so, we differentiate the health outcome indicators (LE and MR) with respect the explanatory variables as follows:

$$\frac{{\partial H{O_{it}}}}{{\partial {X_{it}}}}={\eta _j}+\theta {Z_{it}}$$

(4)

where the sign of $ \theta $ is a priori-indeterminant due to the expected effects (positive or negative) of the explanatory variables on HO. For example, we expect $ \theta $ to be positive in the relationship between LE and KS and negative in reducing the effects of CO₂e on MR. To estimates Eqs. 2 and 3, we first consider the fixed effects (FE) model, where $ \wp $ is considered as the country-specific effects. The estimation of FE model is based on the assumption that $ {\epsilon }_{it}$ is correlated with $ {X}_{it}$ and uncorrelated with $ \wp $ [113]. Nonetheless, random effects (RE) model is an alternative to FE technique. It assumes that $ \wp $ is a random variable and uncorrelated with $ {X}_{it}$. If this assumption holds, then RE estimators would be more reliable than FE model [114]. This hypothesis can be tested using Hausman’s [115] specification approach. Additionally, the instrumental variables approach (IV) is another empirical competitor, which considers that there might be some exogenous variables, such as InQ, InF, and CO₂e, as in our case, correlated with $ {\epsilon }_{it}$. It offers a mechanism to still estimate accurate and consistent coefficients. The IV regression takes the following form:

$$\begin{gathered} {Y_{it}}=\mu +{\vartheta _1}X_{{it}}^{1}+{\vartheta _1}X_{{it}}^{2}+\varepsilon _{{it}}^{1} \hfill \\ X_{{it}}^{1}=\mu +{\theta _2}X_{{it}}^{1}+{\theta _3}X_{{it}}^{2}+{\theta _4}X_{{it}}^{3}+\varepsilon _{{it}}^{2} \hfill \\ \end{gathered} $$

(5)

where $ {X}_{it}^{1}$, $ {X}_{it}^{2}$ and $ {X}_{it}^{3}$ refer to the endogenous variables, exogenous variables, and instrumental variables, respectively, $ \theta $ refers to the vector of reduced from coefficients, and $ {\epsilon }_{it}^{1}$ and $ {\epsilon }_{it}^{2}$ present the normal multivariate variance-covariance matrix. If the homoscedastic assumption holds true, then IV regression would be a good substitution. Nevertheless, in the presence of autocorrelation, cross-sectional dependence, heteroskedasticity, and endogeneity issues, neither of the above-cited models would be reliable. Therefore, to account for these issues, we estimate the IV-generalized method of moment (IV-GMM) model of Blundell and Bond [116]. It provides unbiased and consistent coefficients and has gained statistical prominence in prior literature. The IV-GMM model is also suitable for small samples like ours (t = 176), whether balanced or unbalanced [117]. Furthermore, unlike pooled OLS, FE, and RE techniques, the GMM model does not require the sample to hold normality assumptions [118]. For brevity, the moment conditions (MMs) of the GMM model, which were conducive to its development, take the following form:

$$E\left[ {{Z_{it}}{\varepsilon _{it}}\left( \vartheta \right)} \right]=E\left[ {{Z_{it}}\left( {{Y_{it}} – \left( \vartheta \right)X^{\prime}_{{it}}} \right)} \right]=0$$

(6)

where $ {X}_{i,t-1}$, $ {X}_{i,t-2},$ and $ {X}_{i,t-i}$ are the instruments used. The model can be estimated using system- or difference-GMM. The system-GMM model simultaneously includes two MMs for differenced and level equations, and it is evidently more accurate than difference-GMM [119, 120]. The difference-GMM, however, removes the fixed effects by differencing the employed data [121]. Further, the system-GMM is estimated using one-step or two-step estimators. Based on conventional asymptotics theory, however, estimators would be asymptotically normal in both approaches, but the two-step system-GMM estimator yields a comparatively smaller variance [122]. Therefore, the present study adopts the two-step system-GMM model. For estimation of the system-GMM, we used STATA/BE-17, in which the “xtabond2” command comes with a built-in diagnostic check for the first- and second-order autocorrelation, testing the well-being of the instruments used in the model under the presence of the first- and absence of the second-order autocorrelation [123]. Nevertheless, this inquiry does not aim to discuss the preference and technicality of the GMM model; the above-cited studies can be found highly informative.

Source link