Democracy Does Improve Health

Abstract

In this paper, we study the extent to which the spread of democracy affects country-level health outcomes in 115 countries, between 1960 and 2015. To do this, we use both the level and change measures of democracy in our regressions, concentrating on within-country variations. Our finding is that a one standard deviation increase of 0.35 in the level of democracy is associated with a 0.11 standard deviation increase in life expectancy, even after accounting for various country and time features. This corresponds to an increase in life expectancy of around 5 years for a country initially, with a mean life expectancy of 54 years. However, we do not find the change measure of democracy to be consistently influential. These results are robust to employing alternative model specifications, to using different subsamples of the data, and to alternative estimation techniques. We also find that these critical effects are retained when using other measures of health status. In particular, we observe that as the level of democracy rises, each of infant mortality, child mortality, and crude death decreases. We, therefore, conclude that the material role of democratic institutions in fostering population health is of first-order relevance.

1 Introduction

Sir Angus Deaton opens The Great Escape: Health, Wealth, and the Origins of Inequality by stating that: “Life is better now than at almost any time in history. More people are richer and fewer people live in dire poverty. Lives are longer and parents no longer routinely watch a quarter of their children die. Yet millions still experience the horrors of destruction and of premature death. The world is hugely unequal” (2013, p. 1). According to him: “The great escape in human history is the escape from poverty and death… You need a life to have a good life, and poor health and disability among the living can severely limit the capability to enjoy an other-wise good life” (pp. 23–24).¹

In the above, Deaton provides an apt summary of how poverty determines mortality, and why some societies continue to suffer calamitous mortality rates, that keep the life expectancy faced at birth by their citizens agonisingly low, which remains a topic of great debate amongst social scientists and policy makers. The United Nations (UN, 2015), in its 2030 Agenda for Sustainable Development, identified ensuring good health and promoting well-being at all ages as Goal 3 of 17 Sustainable Development Goals (SDGs). The destination is thus set, and the journey has begun. Undoubtedly, great strides forward have been made in recent times. For example, between the years 2000 and 2015, child mortality rate declined by 44%, and neonatal mortality rate fell by over 38%. Relative to the year 2000, the corresponding numbers in 2017 were reductions by 49% and 41%, respectively, for child and neonatal mortality rates (UN, 2019). This shows a considerable improvement globally in recent years, which appears to have been gaining speed.

Despite the monumental strides that have been made over the years in reducing the plights of the poorest citizens of the world,² the 2019 progress towards the SDGs report claims that “progress has stalled or is not happening fast enough…” (p. 9). However, the good news is that the existing literature has offered various explanations for why some countries live better, healthier, and longer than others. For instance, Cutler et al. (2006) have outlined a variety of socioeconomic factors as the fundamental causes of the differences in mortality rates between and within countries. Some of the factors they discuss are improvements in nutrition and water supply, availability of medications and vaccines for the treatment of sicknesses and prevention of others, and changes in income and education.³

Implicit in their discourse is the need for a political man or machine to organise the resources of a nation to the desired end of attaining prosperous health. Further, we know from the existing literature that access to more of the above-listed conditions alone is insufficient for explaining the cross-country differences in health and health policies, claiming that politics matter (Boone, 1996; Franco et al., 2004; Navarro et al., 2003; Ruger, 2005; Sen, 1981, 1999; Zweifel & Navia, 2000). Figure 1, which displays the unweighted averages of life expectancy (unlogged on the left y-axis) and the level of democracy (rescaled to lie between 0 and 1 on the right y-axis), underlines why one may want to study more the association between democracy and health systematically. For instance, the overall picture from the diagram suggests that there is some correlation between a country’s democratic level and its citizen’s life expectancy, especially since the 1980s. So, is this just a coincidence? Or does health follow where democracy leads?⁴ Our goal in this paper is to reassess whether political arrangements under democracy are superior to the ones under autocracy in providing better health outcomes for its population.

Fig. 1

Democracy and life expectancy, Notes: The figure shows the evolution of the level of democracy and life expectancy over the sample period. The base sample is an unbalanced panel of 115 countries from 1960 to 2015, with data averaged over five-year windows. Variable definitions and data sources are provided in the text; see Sect. 2

There exists an abundance of literature examining this link between democracy and health. On the one hand, a large cluster of papers holding the traditional view passed down from Aristotle have documented a positive connection between democracy and health outcomes (e.g., Lake & Baum, 2001; Navarro et al., 2003; Franco et al., 2004; Ghobarah et al., 2004; Ruger, 2005; Besley & Kudamatsu, 2006; Safaei, 2006; Klomp & Haan, 2009; Wigley & Akkoyunlu-Wigley, 2011; Gerring et al., 2012; Kudamatsu, 2012; Garcia, 2014; Wang et al., 2019). Meanwhile, there is a growing body of work that have reports little, nonexistent, or negative impact of democracy on population health (e.g., Burroway, 2016; Gauri & Khaleghian, 2002; Norris, 2012; Ross, 2006; Rothstein, 2011; Shandra et al., 2004). The research endeavours mentioned so far have been mainly concerned with establishing a nonneutral (positive or negative) relationship between democracy and health.

What may be more interesting than identifying the correlation between democracy and health, however, is determining whether the association is causal (Gauri & Khaleghian, 2002; Shandra et al., 2004; Norris, 2012). In this paper, we provide one of the first attempts to fill this gap in the existing literature. We are aware of only one other paper that explicitly reports causal effects of democracy on health (Batinti et al., 2022). It considers the effect of transitions to democracy on the average height of adults, finding that the spread of democracy raised average height by 0.7 cm. Although this study covers an extended period from mid-nineteenth century to the 1970s, it includes only 15 European countries. In this paper, we re-examine this difficult issue using a larger panel data covering 115 countries from 1960 to 2015. Our paper can thus be viewed as extending the literature by showing not only a robust and strong effect of democracy on the health outcomes of a nation but, more importantly, submitting evidence of a causal effect that runs from democracy to health.

In line with the bulk of previous research, our finding is that democracy improves health.⁵ More specifically, we find that, after controlling for various country and time features, a one standard deviation increase of 0.35 in the level of democracy is associated with a 0.11 standard deviation increase in life expectancy in the baseline analysis that uses fixed effect estimator. This is equivalent to an increase in life expectancy of around 5 years for a country, initially, with a mean life expectancy of 54 years. However, we find that changes in democracy, whether an increase or a decrease, exert no consistently significant effect on health outcomes. We confirm that the results are robust to a wide array of econometric tests, including (i) employing alternative model specifications, such as controlling for the dynamics of the dependent variable and possible nonlinear effect of income; (ii) using different subsamples of the data, such as regressing the model on observations for the least democratic countries, the most democratic countries, and excluding countries from each continent (Africa, Americas, Asia, Europe, and Oceania) at a time; and (iii) estimation by two-stage least squares (2SLS), providing a result which—given our identification strategy—we believe is novel to the literature on democracy-life expectancy nexus, and the generalised method of moments (GMM) approaches.

Besides, we utilise alternative health measures, such as infant mortality, child mortality, and crude death, and find that they are all negatively and statistically significantly associated with democracy. More specifically, a one standard deviation increase in the level of democracy of 0.35 will, in these instances, reduce the standard deviation of infant mortality, child mortality, and crude death by 0.05, 0.06, and 0.11, respectively. These values lead to a decline in (i) infant mortality to around 67 per thousand live births for a country initially with a mean level of around 86 per thousand live births; (ii) child mortality to approximately 95 per thousand from approximately 132 per thousand; and (iii) crude death to about 12 per thousand from 13 per thousand. While democracy reduces the probability of all types of mortality more than nondemocracies, we observe that it is more effective in combating infant mortality and child mortality, whereas it is less potent in dealing with crude death.

The rest of the paper is organised as follows. Section 2 provides the definitions of variables, data sources, and some exploratory evidence. Section 3 provides the econometric specification and estimation strategies. Section 4 provides our main results and robustness exercises. Section 5 provides the conclusion.

2 Data Description

2.1 Sample Size and Data Sources

The study sample is composed of an unbalanced panel of 115 countries spanning every continent.⁶ For these countries, and to study the effects of democracy on health outcomes, we collect the relevant outcome measures, independent variables, and additional controls for the two halves of each decade from 1960 to 2015. Unless noted otherwise, all data are extracted from the Quality of Government (QoG) Institute’s database at the University of Gothenburg, but the primary sources are acknowledged below. Depending on the model specification in use, the total number of observations ranges from 617 to 916.

Table 1 Descriptive statistics

2.2 Health Indicators

Our core indicator of a country’s health status is life expectancy from the World Bank’s World Development Indicators (WDI, 2016). Life expectancy measures the number of years a newborn infant is expected to live if the prevailing mortality patterns at the time of its birth persisted throughout its life. Meanwhile, we have also considered three alternative health indicators (also from WDI) as dependent variables: infant mortality, child mortality, and crude death. Infant mortality is the number of infants per thousand live births dying before reaching the age of one; child mortality is the number of infants per thousand that will die before reaching the age of five; crude death is the number of deaths per thousand population.

2.3 Democracy Measures

We represent the average level of democracy using the revised combined polity score from the Polity IV database of the Center for Systemic Peace. This dataset is compiled based on the conceptual framework of Eckstein and Gurr’s (1975) patterns of authority, and the Polity2 scores are formulated around three assessed sub-scores. The first is how institutionalised, competitive and open the process of executive recruitment is. The second is the degree of institutional constraints on the decision-making powers of the executive arm of the government. The third is the degree of institutionalisation (or regulation) of, and the extent of, government restriction on political competition (Marshall et al., 2018).

The Polity2 scores are computed by subtracting Polity IV’s institutionalised autocracy measure from its institutionalised democracy measure. Given this, the Polity2 scores range from $-$ 10 to + 10 because both autocracy and democracy indicators are characterised on an additive eleven-point (0–10) scale. The lower values (closer to $-$ 10) suggest more coherent autocratic regimes, while the higher values (closer to + 10) indicate more enduring democratic polities.⁷ This 21-point measure is then normalised and rescaled to range from 0 (full dictatorship) to 1 (full democracy).

As there have been debates about the potential gains from having democratic capital, some insightful research outputs have, in addition to a proxy for the level, or the contemporaneous measure, of democracy, also included measures of a country’s democratic experience as part of their controls in evaluating the influence of democracy on the health outcomes of a country’s population (see, for example, Besley & Kudamatsu, 2006; Ross, 2006; Gerring et al., 2012). Another way to think about a country’s constitutional history is to consider its political regime durability, or lack thereof, as in Minier (1998). The innovation this author brought to bear was to examine the extent to which it was the level of, or the change in, democracy that was crucial for economic growth. We adopt such a strategy here, and utilise both the level and change measures of democracy in our attempt to identify whether (or not) there is a causal effect of democracy on health.

This approach has enabled us to empirically implement both the substantive and minimalist conceptualisations of democracy (Cheibub et al., 2010). Besides, Cheibub et al. (2010) have criticised the use of continuous measures of political regimes.⁸ Thus, we borrow the approach of Minier (1998), supplementing our polychotomous measure of democracy by constructing two additional measures, which are dichotomous, to capture increases and decreases in the levels of democracies around the world (our change measures of democracy). While Minier (1998) used these change measures of democracy to re-visit the link between economic growth and democracy, to our best knowledge, this is the first paper to adopt it when (re-)examining the connection between health and democracy.⁹

We have coded our two 0–1 binary indicators for monitoring the extent of democratic engagements and practices based on the Democracy-Dictatorship database of Cheibub et al. (2010), appended by updates from Bormann and Golder (2013). The qualifications for codifying countries as becoming more democratic/dictatorial over a five-year window are: (i) a change in classification by types of democracies (parliamentary, mixed or presidential) towards or away from types of dictatorships (monarchic, military or civilian dictatorships); (ii) the fraction of years the change is observed must be greater than or equal to 0.5; and (iii) turn off the binary indicator from 1 to 0 after ten periods.

Our objective in creating binary variables for democracy increases and decreases is to capture changes, such that the third condition, just stated, is needed and justifiable under the assumption that countries that fall into this category have settled down into the new democracy/dictatorship status.¹⁰ To clarify the construction of our change measures of democracy, we consider Argentina for illustration, which is represented in Fig. 2. We begin in the early sixties, when Argentina changed from presidential democracy, in 1961, under Arturo Frondizi, to military dictatorship in 1962, when Jose Maria Guido was acting president. The country, however, immediately reverted to presidential democracy the following year with the election of Arturo Umberto Illia. This new transition only lasted from 1963 to 1965, when, once again, the country succumbed to Juan Carlos Ongania, a military dictator. These reversals between democratic and dictatorial rules continued for another few years, and it was not until the election and installation of Raul Alfonsin in 1983 that this pattern ceased.

Fig. 2

Time series plot of political regime changes—the case of Argentina, Notes: The figure shows the increases and decreases in the level of democracy for Argentina over the sample period. The base sample is an unbalanced panel of 115 countries from 1960 to 2015, with data averaged over five-year windows. Variable definitions and data sources are provided in the text; see Sect. 2

At this point (1983), we turn on the binary indicator for increases in democracy for the next ten years, after which it is turned off. As it happens, Argentina has remained under democratic governance since then. The implementation in practice, though is that an increase in a democracy is equal to unity in 1985 and zero in all subsequent years. Our conceptualisation invokes a form of jostling for supremacy between democracy and dictatorship per given period (generally over a five-year window). Continuing with Argentina’s illustration, we proceed as follows. Using information from 1960 to 1964, we code 1965 as an increase in democracy because it lasted longer than the dictatorial disruption (4/5 years which is higher than our lower bound in condition (ii)).

Next, 1970 is coded as a decrease in democracy because the military administered the country during this period, which was interrupted in 1973 by a stint of democratic rule. We nonetheless still coded 1975 as a decrease in democracy because 2/5 = 0.4 < 0.5.¹¹ Conditions (i)–(iii) imply that 1980 is zero for both an increase and a decrease in democracy. The next and last change for Argentina happened in 1985 when an increase in democracy took a unit value because of the country’s return to presidential democracy in 1983. It is this approach that we utilise to code the binary indicators for increases and decreases in democracy for each of the 115 countries in the sample.

2.4 Other Control Variables

Our analysis considers two sets of potential determinants of health outcomes around the world. For the first set of controls, we follow an established large body of work that centres on socioeconomic-health nexus (see, for example, Hadley & Osei, 1982; Duleep, 1986; Marmot et al., 1987; Kunst & Mackenbach, 1994; Ettner, 1996; Pritchett & Summers, 1996; Wildman, 2001, 2003; Meer et al., 2003). Whatever measure is used to proxy socioeconomic status (whether wealth, education, occupation, social class, or self-help), this literature has found evidence of a positive link between socioeconomic position and the health outcomes of a population (Safaei, 2006).

As a result, we have included time-varying factors, such as GDP pc (gross domestic product divided by total population, measured in constant 2010 $US from WDI), population density (people per square km of land area from WDI), years of schooling (average total number of years of education in the population above 25 years of age from Barro & Lee, 2013), and growth of GDP pc (percentage change on previous year’s GDP pc as defined above from WDI). These controls were decided upon based on existing literature (see, for example, Pritchett & Summers, 1996, Franco et al., 2004, Besley & Kudamatsu, 2006, Ross, 2006, Gerring et al., 2012, and Wang et al., 2019).

Then, building on insights provided by Gerring et al. (2012), we entered as the second set of controls largely time-invariant covariates. These consist of ethnic diversity (the probability that two randomly selected persons belong to different ethnic groups from Alesina et al., 2003), colonial roots (dummy variables taking the value of 1 for former British, French, Portuguese, Spanish, and other European (Belgian, Dutch, and Italian) colonies, respectively, and 0 otherwise from Teorell & Hadenius, 2007), legal origins (dummy variables taking the value of 1 when a country is recognised as having British common law, French civil law, German civil law, Scandinavian law, and Socialist law, respectively, and 0 otherwise from La Porta et al., 1999), religious affiliations (fraction of each country’s population that is Roman Catholic, Muslim, and Protestant in 1980, with the residual organised into “other religions” from La Porta et al., 1999), and latitude (the absolute value of the latitude of a country’s capital city divided by 90, to lie between 0 and 1 from La Porta et al., 1999).

2.5 More Democracy, Better Health? Some Exploratory Evidence

2.5.1 Descriptive Statistics

Table 1 documents the descriptive statistics for our variables, reporting on the mean, standard deviation, minimum, and maximum values. The table presents these statistics for all countries, and also separately for high- and low-democracy countries. We designate countries into high-democracy if their mean level of democracy over the sample period is greater than the median value for all countries; otherwise, they are classified as low-democracy countries.

Looking at these descriptive statistics across the three sample groups (All countries, High-democracy countries, and Low-democracy countries), and focussing on health outcomes and democracy measures, we observe that the mean life expectancy in the sample of all countries (4.17 in logs) is lower than in the high-democracy countries (4.29 in logs) but higher than in the low-democracy countries (4.1 in logs). These statistics are reversed for the remaining three health indicators: the mean values of infant mortality, child mortality, and crude death (all in logs) in the full sample are, respectively, 3.37, 3.66, and 2.18. Further, these values are found to be larger than the corresponding estimates for high-democracy countries (2.48, 2.68, and 2.12) and smaller than those for the low-democracy countries (3.86, 4.19, and 2.21). We also show that the mean level of democracy in the world sample (0.61) is lower than in the high-democracy sample (0.93), but higher than in the low-democracy sample (0.43). Unsurprisingly, most of the changes in democracy (whether an increase or a decrease) are taking place amongst the low-democracy countries.

Further, life expectancy ranges from 3.38 to 4.42 (in logs) in the panel data; see Table 1. The countries with the shortest life expectancy in the sample are Rwanda (3.38 in 1990), Mali (3.54 in 1970), and Sierra Leone (3.59 in 1990). The country with the longest life expectancy is Japan (4.42 in 2010), trailed very closely by Switzerland (4.4171 in 2010) and Spain (4.4165 in 2010).¹² In terms of infant mortality, the countries with the highest values are Mali (5.2296 in 1970), Sierra Leone (5.2017 in 1970), and Liberia (5.1971 in 1970), while the countries with the lowest values are Singapore (0.7605 in 2010), Japan (0.7779 in 2010), and Finland (0.7864 in 2010).

Child mortality ranges from 0.9683 to 5.9472 in logs. The countries with the highest values are Mali (5.9472 in 1970), Niger (5.8068 in 1970), and Sierra Leone (5.7574 in 1970), while the countries with lowest values are Finland (0.9683 in 2010), Singapore (1.0078 in 2010), and Luxembourg (1.0221 in 2010). With respect to crude death, Qatar (0.4056 in 2010), United Arab Emirates (0.4097 in 2010), and Bahrain (0.8475 in 2010) are countries with the lowest values, whereas Rwanda (3.5898 in 1990), Mali (3.3955 in 1970), and Niger (3.3304 in 1970) have the highest values.

Two general patterns emerge from Table 1 and the above discussions. First, it appears that high-democracy countries have better health status than low-democracy countries. High-democracy countries live longer, their babies’ survival rates are higher, and they suffer lower maximum crude death. Interestingly, however, the data shows that the minimum value of crude death is experienced in a low-democracy country: 0.41 (in logs) compared to 1.38 (in logs) for high-democracy countries.¹³ We also observe that high-democracy countries tend to have more socioeconomic and culture-historic requisites for better health performance. More specifically, relative to the low-democracy countries, high-democracy countries, on average, have higher income per capita, are more educated, have lower population heterogeneity, are likely to be the coloniser rather than the colonised, and have higher fractions of their population professing to be Protestants rather than Muslims. Second, the best (worst) health outcomes are reported for later (earlier) periods in the sample, which could be as a result of global health trends (Ross, 2006), but which we argue correlates with rising level of democracy across the world.

2.5.2 Univariate Correlations

In Table 2, we report the Pearson correlation matrix amongst the main variables of interest along with the corresponding number of observations in brackets and p-values in parentheses for our full sample of 115 countries. It is clear from the documented values in column (1) that the level of democracy is highly correlated with the health measures. Take for example, the correlation between democracy and life expectancy—our primary indicator of a country’s health—is 0.5227, which is statistically different from zero with a p-value = 0.0000.

Table 2 Correlation coefficients

In addition, the level of democracy is found to be strongly correlated with the remaining health indicators—infant mortality ($-$ 0.5678), child mortality ($-$ 0.5704), and crude death ($-$ 0.1719)—again with a p-value = 0.0000 in all three cases. We also observe that the correlations are quite high and strong between the health variables; see columns (4)-(6). Life expectancy is negatively correlated with infant mortality ($-$ 0.8771, p-value = 0.0000), child mortality ($-$ 0.9026, p-value = 0.0000), and crude death ($-$ 0.7534, p-value = 0.0000). Moreover, and as one might expect, infant mortality, child mortality, and crude death all show positive and strong correlation coefficients between them.

3 Estimation Strategy

To characterise the effects that democracy has on life expectancy and other key indicators of a healthy society, we run regressions of the following baseline linear model:

$${H}_{ct}=\alpha {D}_{c,t-1}+\beta {D}_{c,t-1}^{R}+\gamma {D}_{c,t-1}^{F}+\Theta {{\varvec{Y}}}_{c,t-1}+\Gamma {{\varvec{X}}}_{c}+{\delta }_{c}+{\eta }_{t}+{\epsilon }_{ct}$$

(1)

where c = country, t = time, H = life expectancy, infant mortality, child mortality, or crude death, D = level of democracy, D^R = increase in democracy, D^F = decrease in democracy, Y = GDP pc, years of schooling, population density, and growth of GDP pc, X = ethnic diversity, colonial roots, legal origins, religious affiliations, and latitude, $\delta$ = country dummies, $\eta$ = time dummies, and $\epsilon$ = error term.

Following the existing literature (e.g., Besley & Kudamatsu, 2006; Gerring et al., 2012; Ross, 2006; Wang et al., 2019), our baseline estimation of Eq. (1) uses RE and FE techniques, although our emphasis is on the fixed effect (FE) results. As argued by Acemoglu et al., (2008, pp. 809–810), “fixed effect regressions… are well suited to the investigation of the relationship between income and democracy, especially in the postwar era. The major source of potential bias in a regression is country-specific, historical factors influencing both political and economic development.” Following this strand of literature, we estimate the within-country effect of democracy on the health of a country’s population. Essentially, this approach stresses the importance of probing the association between democracy and health within a nation over time, rather than just across nations. For example, instead of comparing how a country, say Nigeria, is health-wise relative to the rest of the world now that it has returned to democratic rule for more than 20 years, with regular elections and peaceful transfers of political powers. The emphasis, however, should be on scrutinising whether Nigeria’s chances of becoming relatively healthier is improved as it is becoming relatively more democratic. This is our motivation for mostly providing evidence using the FE estimator.

However, while the FE estimation technique allows us to control for potential confounding factors that may otherwise exacerbate the problem of omitted variable bias, it is essential to remember that it is not the magic bullet. We have, therefore, also pursued estimation by two alternative methods. First, we employ two-stage least squares (2SLS) regressions that use plausibly exogenous variations in a country’s regional democratisation wave (an external instrument made available by Acemoglu et al., 2019) to instrument for within-country variations in a country’s level of democracy. Second, we utilise both the difference and system GMM estimators that use relevant lags of appropriate moment conditions (internal instruments) to instrument for all variables of interest.¹⁴

4 Empirical Results

4.1 Baseline Estimates

We begin by documenting results using the RE estimator, which helps to obtain the unique effects of the included observed country-specific time-invariant controls. We then report estimates from the FE estimator in which case consideration is given to unobserved heterogeneity, although we are no longer able to produce the unique effects of our preferred time-invariant factors.

Table 3 documents our baseline estimates for the relationship between the level and change measures of democracy and life expectancy, which is arranged into two panels. Panel (a) displays the results using the RE estimator and Panel (b) shows the results using the FE estimator. Whilst we have included the estimated coefficients of the core independent variables, we have not reported estimates for the four time-varying controls and the time-invariant covariates to save space. Additionally, we have only documented estimates from our most structured model specification and indeed focussed on discussions around the effects of the level and change measures of democracy on our measures of health performance.¹⁵

Table 3 Democracy and life expectancy

Beginning with the results in Panel (a), columns (1)–(3) utilise our world sample. Column (1) reports that the level of democracy is positively related to life expectancy, and this coefficient of 0.064 (standard error = 0.001) is significantly different from zero at the 99% confidence level. Surprisingly, both of our change measures of democracy yield estimates (a coefficient of −0.007 for the increase in democracy, with a standard error of 0.227 and a coefficient of 0.004 for the decrease in democracy, with a standard error of 0.739) that are not significantly different from zero at conventional confidence levels. These findings indicate that changes in a democracy have no influence, once we control for the level of democracy and other variables.

Column (2) adds the lag of life expectancy. We find evidence of persistence in life expectancy (as one would expect), with a coefficient of 0.885 (standard error = 0.000), which is both sizeable and statistically different from zero at the 99% confidence level. This indicates that the coefficient of 0.017 (standard error = 0.000) on the level of democracy now measures its short-term effect, which is still statistically different from zero at the 99% confidence level. This yields a corresponding long-term effect of 0.145.

Based on the influential contribution of Preston (1975), we have included GDP pc squared in column (3) to control for the possible nonlinear relationship that may exist between life expectancy and income. We did not find any evidence of a nonlinear effect of income on life expectancy (the coefficient (standard error) of GDP pc is 0.0232 (0.648), and the coefficient (standard error) of GDP pc squared is 0.001 (0.862)). Moreover, these estimates are not statistically different from zero at conventional significance levels. Importantly, our level of democracy effect has gone up again to the magnitude and significance levels reported in column (1).

We then look at a series of restrictions on the base sample in columns (4) through (10), using our baseline specification of column (1). Columns (4) and (5) exclude countries that were, on average, the least (falling in the first quartile) and the most (falling in the fourth quartile) democratic in the sample, while columns (6) through (10) exclude countries from various continents (Africa, Americas, Asia, Europe, and Oceania, in that order). The results in the restricted samples for the level of democracy corroborate our finding from the base sample. In the same vein, we again find no evidence that changes in democracy exert any meaningful effect on life expectancy. With a few exceptions, this is a pattern that will become synonymous with all the results that we present, that the average level of democracy is positively and statistically correlated with life expectancy; whereas, both the increase and decrease in a democracy have mostly insignificant influence on life expectancy.¹⁶

The estimated size of the association between the level of democracy and life expectancy in Panel (a) of Table 3 is not only statistically significant but are also economically considerable. As displayed, a one standard deviation increase in the level of democracy (0.35) is related to between 0.06 and 0.17 standard deviation increase in life expectancy. Using the estimated coefficient in column (1) for the world sample, this demonstrates that, for a country initially with a mean level of life expectancy of 54 years, a one standard deviation increase in the level of democracy will improve life expectancy by approximately 4.6 years. This finding is consistent with Besley and Kudamatsu (2006), who found that life expectancy is higher in democracies than autocracies by between 3.5 and 5 years.

Whilst we controlled for certain observed country-specific characteristics in Panel (a), a possible concern about the RE estimator is the potential endogeneity problem that may arise from omitted variables. We have ameliorated the impact of this in Panel (b) by including country dummies, which account for all unobserved country time-invariant features. In all the columns of Panel (b), whether we use the base sample in columns (1)–(3), or any of the sub-samples relating to the least democratic in column (4), most democratic in column (5), non-African countries in column (6), non-Americas’ countries in column (7), non-Asian countries in column (8), non-European countries in column (9), and non-Oceania countries in column (10), we find that the level, and not either of the change measures,¹⁷ of democracy remains the important measure of a political institution for explaining life expectancy.

Specifically, we see that the level of democracy is positively and statistically significant at least at the 90% confidence level (1/2 and 2/5 of the FE regressions are statistically significant at the 99% and 95% confidence levels, respectively). Moreover, the estimated coefficients (ranging from 0.023, with a standard error of 0.001, to 0.072, with a standard error of 0.017) remain economically considerable. Using the estimated coefficient in column (1), Panel (b), of Table 3, an increase in the level of democracy by 0.35 (one standard deviation) will raise life expectancy by 0.02 (over one-tenth of a standard deviation). Given these values, a country with an initial mean level of life expectancy of 54 years will now enjoy an additional 4.5 years. These FE estimates are encouraging because the inclusion of country and time dummies in the FE estimator massively reduce the numbers of potential confounders of democracy compared to when we employ the RE estimator.¹⁸

4.2 2SLS Estimates

In the analysis presented so far, we have assumed that democracy is driving life expectancy. However, it is possible to imagine a scenario where the quality of health in a country may influence its democratic outcome. The idea is that life expectancy may affect the democratic outcome; for example, a short life expectancy may diminish incentives for political participation that entails short-run outlays and long-run returns. To address this point, we first utilise regional democratisation wave as an instrument for democracy (Acemoglu et al., 2019). We briefly describe this variable next.¹⁹ Based on a number of political developments around the world in the last forty years,²⁰ Acemoglu et al. (2019) argue that democratisation and social unrest that lead to a change of regime often happens in waves across regions, as already identified in the existing literature.²¹

They then conjectured that the observed regional patterns were likely reflecting the spread of the demand for democracy among countries within a region, and postulated that democracy in the cth country is shaped by democracy in the set of countries in the same region with similar histories, political cultures, practical problems, and close informational ties. This can be summarised as follows: ${I}_{c}=\left\{{c}^{^{\prime}}:{c}^{^{\prime}}\ne c, {R}_{{c}^{^{\prime}}}={R}_{c},{D}_{{c}^{^{\prime}},{t}_{0}}={D}_{c{t}_{0}}\right\}$, where ${D}_{c{t}_{0}}$ captures whether the cth country in the same region is a democracy or nondemocracy when the sample began and ${R}_{c}$ represents the geographic region of cth country. Using these sets, the authors define regional democratisation wave for each country as ${Z}_{ct}=\frac{1}{\left|{I}_{c}\right|}{\sum }_{{c}^{^{\prime}}\epsilon {I}_{c}}{D}_{{c}^{^{\prime}}t}$, which defines the jack-knifed mean of democracy in a region times the initial regime cell, which, by construction, leaves out own-country observation.

We, therefore, estimate the following regression in the first-stage:

$${D}_{ct}=\mu {Z}_{ct}+\theta {D}_{c,t-1}^{R}+\phi {D}_{c,t-1}^{F}+\Phi {{\varvec{Y}}}_{c,t-1}+\Psi {{\varvec{X}}}_{c}+{\chi }_{c}+{\omega }_{t}+{\varepsilon }_{ct}$$

(2)

where Z = regional democratisation wave, $\chi$ = country dummies, $\omega$ = time dummies, $\varepsilon$ = error term, and all the other variables are as previously defined. The exclusion restriction is that regional democratisation wave only affects life expectancy through its influence on democracy.

Table 4, columns (1)–(2) document the results from our 2SLS of the effect of democracy on life expectancy. Column (1) contains the first-stage results, which indicate that regional democratisation wave positively (coefficient = 0.553, standard error = 0.000) and statistically predict democracy at the 99% confidence level, which a priori implies a strong instrument. Moreover, we report the Kleibergen-Paap F-statistic to check for instrument quality (the value of 84.93 is much higher than 16.38, which is the most stringent Stock and Yogo weak ID critical values, such that we can reject the hypothesis that the IV size distortion is larger than 10% at the 5% level of significance).

Table 4 Alternative estimation methods

In column (2) of Table 4, we report the second-stage results. As displayed, the 2SLS estimated coefficient of 0.095 (standard error = 0.019) is statistically different from zero at the 95% confidence level, and underscores our baseline finding that the level measure of democracy raises life expectancy. The economic significance is that a one standard deviation increase in the instrumented level of democracy will yield an increase in life expectancy of 0.04 (approximately one-fifth of a standard deviation). Based on these values, a country initially with a mean level of life expectancy of 54 years can now live on average to more than 63 years.

4.3 GMM Estimates

We next provide a further robustness check on the estimation methods, as a second strategy to allay any concerns relating to our baseline RE and FE estimates. More specifically, we present findings employing the difference (Arellano & Bond, 1991) and system (Arellano & Bover, 1995; Blundell & Bond, 1998; Holtz-Eakin et al., 1988) generalized method of moments (GMM) estimators. This is of interest here because both approaches allow us to control for the possible endogeneity of all the independent variables and not only for the endogeneity of the level measure of democracy. To implement these procedures, we first difference Eq. (1) to expunge both the observed time-invariant vector of variables and unobserved country effects, obtaining:

$$\Delta {H}_{ct}=\alpha\Delta {D}_{c,t-1}+\beta\Delta {D}_{c,t-1}^{R}+\gamma\Delta {D}_{c,t-1}^{F}+\mathrm{\Theta \Delta }{{\varvec{Y}}}_{c,t-1}+\Delta {\eta }_{t}+\Delta {\epsilon }_{ct}$$

(3)

where we suppose that the variables are weakly exogenous in that they may be correlated with all shocks from the past until the present, but not with future shocks. With regards to the difference GMM estimator, Arellano and Bond (1991) derived that, under the assumption that the error terms, ${\epsilon }_{ct}$, are serially uncorrelated, Eq. (3) can be estimated based on the following moment conditions: $E\left[{\mathbf{d}}_{c,t-s}\Delta {\epsilon }_{ct}\right]=0$ for s $\ge$ 2 and t = 3, …, T, where the vector d = [D D^R D^F Y].

In any case, the existing literature has identified some potential limitations to using difference GMM. One is that the excluded cross-country vector, X, may likewise be of interest; secondly, lagged levels of persistent variables, such as the level of democracy, GDP pc, and years of schooling, will be weak instruments when models are estimated in differences; and finally, this may pronounce the measurement error problem farther (Alonso-Borrego & Arellano, 1996; Griliches & Hausman, 1986; Levine et al., 2000). The system GMM estimator, which exploits both the time series and the cross-sectional characteristics of the data, is a technique developed to resolve some of these practical difficulties.²² This method estimates Eqs. (1) and (3) as a system by employing $E\left[\Delta {\mathbf{d}}_{c,t-s}\left({{\varvec{\uplambda}}}_{c}+{\varepsilon }_{ct}\right)\right]=0$ for s $\ge$ 1 and t = 3, …, T, as additional moment conditions, with the vector ${{\varvec{\uplambda}}}_{c}={\mathbf{X}}_{c}+{\updelta }_{c}$.

We note that in both our GMM regressions, we treat all the time-varying controls as weakly exogenous and instrument for them using appropriate lags of the relevant moment conditions. In contrast, we consider all time-invariant covariates to be exogeneous. To provide a diagnostic on the weakness or invalidity of instruments and, therefore, the robustness of our results from both GMM estimators, we present two conventional specification tests. More specifically, we report: (i) the p-value for the second-order autocorrelation test to identify whether there is a serial correlation in the error term; and (ii) the p-value of Hansen’s test for joint exogeneity of the moment conditions.

Columns (3)–(4) of Table 4 document the results using both GMM estimators. In column (3), the difference GMM confirms our prior finding that the level of democracy is positively (coefficient = 0.051, standard error = 0.006) related to life expectancy and is statistically different from zero at the 99% confidence level. This estimate is economically considerable, producing slightly less than one-tenth of one standard deviation of life expectancy in response to a one standard deviation increase in the level of democracy. This implies that a country initially with a mean level of life expectancy of 54 years may now experience an additional 4.5 years. In an outcome that is reassuring, the p-values (of 0.204 and 0.413) to, respectively, check for the presence of second-order autocorrelation and test of overidentifying restrictions, suggest that one can reject the null hypothesis in both cases, thereby providing support for our identification strategy and finding.

Table 4, column (4) documents the estimates from the system GMM and is again displaying a positive effect of democracy on life expectancy, which is statistically different from zero at the 95% confidence level. The economic significance of these estimates can be illustrated as follows: a one standard deviation increase in the level of democracy (0.35) corresponds to 0.08 standard deviation increase in life expectancy. Using the estimated coefficient of 0.043 (standard error = 0.022) from the system GMM, this indicates that, for a country initially with a mean level of life expectancy of 54 years, a one standard deviation increase in the level of democracy will increase life expectancy to more than 57 years. As in the case of difference GMM, the system GMM also passes both specification tests.

4.4 Further Results

We further confirm that democracies produce better health outcomes than nondemocracies by experimenting with alternative health measures. The results are documented in Table 5, with columns (1)–(2) focusing on infant mortality, columns (3)–(4) on child mortality, and columns (5)–(6) on crude death. In this exercise, we only present estimates using our baseline methods (RE and FE estimators). The main result coming out of using other health measures is that democracy promotes good health outcomes, being negatively and significantly correlated with both rates of early life mortality and aggregate death rate.

Table 5 Alternative health indicators

To place these results concretely, we evaluate the economic implications of each estimated coefficient in Table 5. In columns (1)–(2), the coefficients of $-$ 0.195 (standard error = 0.01) and $-$ 0.158 (standard error = 0.05) on the level of democracy are both statistically different from zero at the 95% confidence level; in columns (3)–(4), the coefficients of $-$ 0.232 (standard error = 0.01) and $-$ 0.193 (standard error = 0.03) on the level of democracy are statistically different from zero at the 99% and 95% confidence levels, respectively; and in columns (5)–(6), the coefficients of $-$ 0.117 (standard error = 0.043) and $-$ 0.134 (standard error = 0.013) on the level of democracy are both statistically different from zero at the 95% confidence level.

Based on the FE estimates in columns (2), (4), and (6) of Table 5, a one standard deviation increase in the level of democracy of 0.35 will, in these instances, reduce the standard deviation of infant mortality, child mortality, and crude death by 0.05, 0.06, and 0.11, respectively. These values translate to a reduction in (i) infant mortality to around 67 per thousand live births for a country initially with a mean level of around 86 per thousand live births; (ii) child mortality to approximately 95 per thousand from approximately 132 per thousand; and (iii) crude death to about 12 per thousand from 13 per thousand. While democracy reduces the probability of all types of mortality more than nondemocracies, we observe that it is more effective in combating infant and child mortality, whereas it is less potent in dealing with crude death rate.

5 Conclusion

This paper has considered the causal effect of democracy on the health of nations. We have represented democracy using the level and change measures based on data from Marshall et al. (2018), Cheibub et al. (2010), and Bormann and Golder (2013). Our core health measure is life expectancy from WDI (2016). Overall, we find that healthier countries are those with more consolidated democratic values. For instance, after accounting for the various country and time features, a one standard deviation increase of 0.35 in the level of democracy is associated with a 0.11 standard deviation increase in life expectancy in the baseline analysis. This is equivalent to an increase in life expectancy of around 5 years for a country which initially had a mean level of life expectancy of 54 years. As shown, these results are robust to a variety of extended econometric investigations. In most of these regressions, however, we find little or no evidence of a meaningful effect for the change measures of democracy. Whenever there is a substantial impact, the evidence seems to reveal that the political status quo is preferred to a transition.

We have also utilised infant mortality, child mortality, and crude death from WDI as alternative measures of a country’s health status. Our results continue to hold that democracy is pro-health. More specifically, we obtain that a one standard deviation increase in the level of democracy of 0.35 will, in these instances, reduce the standard deviation of infant mortality, child mortality, and crude death by 0.05, 0.06, and 0.11, respectively. These values lead to a decline in (i) infant mortality to around 67 per thousand live births for a country which initially had a mean level of around 86 per thousand live births; (ii) child mortality to approximately 95 per thousand from approximately 132 per thousand; and (iii) crude death to about 12 per thousand from 13 per thousand.

So, how do we interpret our results in light of the existing literature? Importantly, our paper has gone beyond confirming that the level of democracy has robustly significant and positive effects on the health of a nation to establish that these effects are causal, thereby reinforcing and extending the findings of Safaei (2006) that democracy is beneficial to a country’s health conditions. While Safaei presented evidence based on cross-sectional data, this paper has overcome several limitations with such cross-country regression analyses by using data panels. Our findings are also consistent with the results of Wang et al. (2019), who, as we did, used panel data and methods in their empirical investigation of the role of democracy on population health, focussing on infant mortality. A key advancement of our paper is that we have adopted a more thorough identification strategy in implementing the 2SLS estimates.

These results, however, do not represent a consensus in the political economy literature. For instance, some authors (see, for example, Gauri & Khaleghian, 2002; Shandra et al., 2004; Besley & Kudamatsu, 2006; Ross, 2006; Gerring et al., 2012) have either shown that there is no robust relationship between democracy and health, or that, if such relationships exist, it is historical and not contemporaneous. In the much-cited work of Ross (2006), he finds that the level measure of democracy is sometimes statistically significant for explaining child mortality but that this effect is not economically large, whereas, he finds that the history of democracy in a country never reached statistical significance. Contrary to Ross (2006), Besley and Kudamatsu (2006) and Gerring et al. (2012) found evidence supporting the view that democratic history is the more important measure of democracy.

Using an alternative approach espoused by Minier (1998), our change measures of democracy are employed to capture the history of democracy in a country. As already discussed above, it is the level, rather than the change, measure of democracy that has a robustly significant and causal effect on health. Given our results vis-a-vis the ones in the existing literature, we are led to conclude that the material role of democracy, as a system of political organisation, and the deep institutional values it represents in promoting human welfare must be taken more seriously. Whether it is the level/change (or the contemporaneous/historical) measure of democracy, the evidence suggests that some dimension of democracy is vital for improving the health of a nation as found also by Mejia (2022).

We conclude with two parting remarks. First, future empirical research endeavours in this area should focus more on the mechanisms via which democracy may be playing the crucial roles of increasing life expectancy and decreasing all types of mortality rates. Although we have established a direct effect of democracy on health outcomes, it is presumed that it is what democracies do differently, when compared to autocracies, that matters for the health of their citizens. A promising avenue to take this investigation further would be for researchers to do more case studies of countries that have successfully and unsuccessfully transitioned to democracy in recent years. Second, this paper, and such studies, hold pertinent health policy implications for both national governments and international agencies. In particular, our results are suggestive that the United Nations, as it aims to progress the 2030 agenda, can peddle political reform as well as economic adjustment programmes as preconditions for its agencies to offer necessary development assistance that is focussed on ensuring good health and promoting well-being at all ages.

Appendix

See Table

Table 6 List of countries and sample averages of main variables

6.