Political-economic correlates of environmental policy

What are the correlates of environmental performance? In particular, does globalization lead to environmental degradation? What is the role of democracy for environmental performance and do left-wing governments really care more about the environment? Using a novel and comprehensive measure of environmental performance, we test these three hypotheses for a panel of 134 countries for the period 2007–2016. Our results are surprising—we find no evidence that democracies are cleaner, left-wing governments perform better than right-wing governments, but centrist governments clearly have the highest environmental performance, and globalization is good for the environment. The positive impact of globalization, however, is driven by social globalization—economic and political globalization do not play a role.


Introduction
What makes a country pursue sound environmental policies? Which country characteristics are correlated with a high environmental performance? Obviously, this question is of great practical relevance because environmental control is not simply a reaction to environmental problems, but the result of a political process, which differs substantially across countries. The actors, their relative strengths and ideologies, the constraints under which they operate and the institutional setup are distinctively different between nation states and therefore their environmental policies are as well.
In this paper we seek to identify the political and economic correlates of good environmental policy. In doing so, we test three theories on the determinants of environmental policy which have gained considerable currency. The pollution haven hypothesis (PHH) posits that globalization leads to increased pollution levels through the competitive erosion of environmental standards. In order to attract mobile foreign direct investment, globalized countries keep environmental standards low, which increases pollution and * It is gratefully acknowledged that the article processing charge was funded by the University of Freiburg in the funding program Open Access Publishing. harms the environment (e.g. Rauscher 1994, Ulph 1996, Umanskaya and Barbier 2008. According to the PHH, firms have incentives to shift their production to countries with low environmental regulations and export to (advanced) countries with higher standards (Copeland and Taylor 1994). However, globalization may improve access to modern, cleaner technology 3 , it may increase inbound FDI, which often adheres to the strictest environmental standards (one-fits-all technology) 4 , and it may also lead to better transmission of information and environmental awareness, which may result in stricter environmental policies. Moreover, trade and FDI pattern may be shaped by relative factor endowments and a host of other location factors, not predominantly differences in environmental standards (e.g. Helpman 1984, Markusen 1984, Millimet and List 2004. The empirical evidence is mixed. Inter alia, Brunnermeier and Levinson (2004), Levinson and Taylor (2008), Hanna (2010), Grether et al (2012) and Millimet and Roy (2016) find evidence in favor of the PHH. Levinson (2009) investigates the role of outsourcing pollution intensive industries to developing countries with less strict environmental standards. He finds only a small pollution haven effect and attributes most improvements in US manufacturing to a technology effect in the US. Brunel (2017) examines the same relationship in the context of the European Union and finds little to no evidence in favor of the PHH. Similarily, Cole and Elliott (2003), Eskeland and Harrison (2003) and Kearsley and Ridded (2010) find no effect of production being shifted towards countries with lower environmental standards 5 . For a recent literature review see Cherniwchan et al (2017). The literature so far focused on the effect of economic globalization, such as trade openness and international capital flows, on environmental policy. Our approach further considers the effect of political and social globalization, i.e. the degree to which societies are interlinked through the exchange of information, culture, and interpersonal connections.
The 'clean democracy hypothesis' posits that environmental concerns carry a larger weight in democracies as civil liberties, such as freedom of speech and of association and a free press, create larger awareness of environmental issues and allow a better organization of environmental interests. Legal and electoral accountability ensures that incumbents care about citizens' demands for better environmental quality and implement stricter standardsthus democracies should be cleaner (Payne 1995, Li andReuveny 2006). Empirical evidence is mixed with supporting evidence provided e.g. by Li and Reuveny (2006), Koubi (2009), andYou et al (2015). Povitkina (2018) finds that democracies are cleaner only if the level of corruption in democracies is low. In a limited sample of 19 developing countries, Lv (2017) shows that only democracies with higher levels of development emit less CO 2 , but not democracies as such. Bättig and Bernauer (2009) find that democracies are more willing to commit to climate change mitigation measures compared to other political regimes. However, this clear positive effect of democracy vanishes when actual air pollution levels are considered. Evidence against the clean democracy hypothesis is provided by Kammerlander and Schulze (2020a) and Gassebner et al (2011). Both studies show insignificant coefficients on their respective measures of democracy. Midlarsky (1998) even finds a negative impact of democracy on three out of six environmental measures.
The third theory maintains that left-wing parties impose stricter environmental policies because they are less reluctant to impose costs on entrepreneurs and because their supporters, the working class, are more vulnerable to environmental hazards as they may be more exposed to them in the work place and can afford health care and protection against environmental hazards less. King and Borchardt (1994) and Neumayer (2003Neumayer ( , 2004 provide supporting evidence for this hypothesis. We shed new light on all three theories by using a comprehensive, multi-dimensional measure of environmental quality, the Environmental Performance Index (EPI). The EPI aggregates indicators on multiple dimensions of environmental quality, such as environmental health risks, air quality, water and sanitation, water resources, agriculture, fisheries and forests, and biodiversity, and is thus a much more comprehensive measure than the single pollution measures frequently used in the analyses of the above theories 6 .
Our results are surprising and run counter to conventional wisdom-we find no support for the clean democracy hypothesis, it is the centrist parties that have the highest EPI scores, not the left, and globalization is good for the environment. The positive impact of globalization however is driven by social globalization-economic and political globalization do not play a role.

Data and empirical approach
Our dependent variable is the EPI, a comprehensive measure of environmental quality developed jointly by the Yale Center for Environmental Law and Policy and The Center for International Earth Science Information Network at Columbia University's Earth Institute and produced in collaboration with the World Economic Forum. It is divided with equal weight into two sub-indices. The environmental health sub-index consists of indicators for environmental risk exposure (weight 1/3), air quality (1/3), and water and sanitation (1/3); the ecosystem vitality sub-index consists of indicators for water resources (25%), agriculture (10%), fisheries (5%), forests (10%), biodiversity and habitat (25%), and climate and energy (25%) 7 . The EPI uses the proximity-totarget approach where targets are defined by international treaties, standards set by international organizations such as the WHO, commonly agreed scientific thresholds or the analysis of best performers and standardizes indicators for comparability 6 Moreover, often only one or two theories are tested, which raises concerns about omitted variables. 7 These indicators consist of sub-indicators, for instance air quality is measured by household air quality (indoor solid fuel usage), the average exposure to fine particulate matter PM2.5, fine particular matters exceedance (an average of the percentage of the population exposed to PM2.5 levels at 10 µg m −3 , 15 µg m −3 , 25 µg m −3 , and 35 µg m −3 ) and average concentration of NO2; for a detailed description of the methodology see Hsu et al (2016). Data are downloadable at https://epi.yale.edu/. (Hsu et al 2016). The EPI ranges from zero (worst possible value) to 100 (best possible value) and provides a much more comprehensive picture than data on single pollution emissions or other one-dimensional measures that only describe a facet but not an overall picture of actual environmental outcomes. To circumvent problems with changes in methodology, we use the latest available backcast of the EPI2016 for the period 2007-2016, which is comparable over time 8 . Figure 1 shows the spatial distribution of the EPI averaged over 2007-2016. On average the mean EPI for all countries increased by roughly 2.5 points from 65.03 in 2007-67.43 in 2016. The two countries with the largest increase in EPI over the whole sample period are Albania (+8.81) and Bulgaria (+10.24). The largest decrease in EPI were experienced in Singapore (−3.9), Mauritius (−3), and Malaysia (−2.98).
We have three variables of interest. Political orientation: the Database of Political Institutions, hosted by the Inter-American Development Bank, contains a variable denoting the political orientation of the largest government party (Cruz et al 2018) 9 . We create the dummy variables left and center (with right as reference category) from this database. Democracy: the polity2 score of the POLITY IV project measures the degree of democracy on a scale from +10 (fully democratic) to −10 (fully autocratic) (Center for Systemic Peace 2018). The POLITY IV series has been widely used in political science to measure the degree of democracy; it focuses on the competitiveness and openness of the elections, the nature of political participation and the constraints to the executive branch of government. Following Epstein et al (2006), we create dummy variables derived from this: Democracy (polity2 score >6), partial democracy (scores >0, but <7) and autocracy (scores ⩽0). Globalization: we use the updated KOF Globalization Index compiled by the Swiss Economic Institute at the ETH Zurich, which ranges from zero (autarky) to 100 (fully globalized), and measures multiple dimensions of globalization (Gygli et al 2019). It is composed of 43 indicators which are aggregated into three subindexes measuring economic, social and political globalization. The overall Globalization Index is the unweighted average of these subindexes. Figure 2 shows the mean value of globalization for each country 10 . Figure 3 plots histograms of the Globalization Index by the political regime and political orientation. It clearly shows that there is substantial variation of the Globalization Index in all of the categories and that the variables of interest are not too closely linked. For instance, not only democracies are highly globalized. Other control variables: we include GDP per capita and its squared term to capture possible Environmental Kuznets Curve (EKC) effects (see Grossman andKrueger 1995, Cole 2004). It is calculated as chained PPP (2011) in 1000 US$ taken from the Penn World Tables 9.1. Moreover, we include the capital-labor ratio as capital-abundant countries may specialize in capital-intensive, and possibly pollution-intensive production (Copeland and Taylor 2003), and the human capital index as higher human capital endowments may lead to cleaner production and more awareness of environmental issues (all data taken from Penn World Tables 9.1). Lastly, we control for total population (in logs) and the percentage of population living in urban areas to capture economies of scale effects in the provision of public goods and infrastructure. Our final data set covers the period 2007-2016 and contains 134 countries. Descriptive statistics are provided in table A1 in the appendix, a list of countries in the sample is found in the working paper version (Kammerlander and Schulze 2020b).
We use the following regression model where EPI i,t is the EPI of country i at time t, G i,t−1 denotes the Globalization Index, D i,t−1 is our democracy measure, PO i,t−1 the measure for the government's political orientation, X i,t−1 is a vector of control variables, µ t denotes a full set of time fixed effects, µ i a set of country fixed effects and ε i,t .a well-behaved error term. As many features of the political and economic system may move only slowly over time, we use an OLS model with time FE and one with additional country FE. Standard errors are clustered at the country level. There may be concerns that our variables of interest-measures for globalization, party orientation and democracy-may be endogenous, in particular that environmental performance may have an effect on globalization, democracy, or party orientation of the government.
It is unlikely that environmental policy or performance can affect the degree of political, social or economic globalization let alone democracy in any significant magnitude. As to whether environmental policy can affect party orientation of the government, we argue that elections are mostly won on economic issue, not environmental performance  (see e.g. Lewis-Beck and Stegmair 2000 or Kayser and Wlezien 2011). Nevertheless, since we cannot completely rule the possibility of endogeneity, we control for various explanatory variables to counteract omitted variables biases and at the same time tackle reverse causality in our data by using the first lag of all explanatory variables throughout the empirical analysis.

Baseline
Globalization is positively and significantly associated with environmental performance in all specifications (table 1). A one standard deviation increase in globalization is associated with approximately 38% of a standard deviation increase in environmental performance (models 1-4). This is a sizeable effect and contradicts the notion that globalization may lead to a race to the bottom in environmental standards 11 . We break down the Globalization Index into its three main components: economic, social and political globalization (model 5). Interestingly, social globalization, measuring interpersonal, informational and cultural aspects of globalization, drives this result-economic and political globalization are insignificant and both point estimates are substantially smaller compared to social globalization. This suggests that the transfer of information and awareness across borders and cultural proximity 11 For favorable effects of globalization in other areas see Potrafke (2015). is important for domestic environmental policy formation.
We find no support for the 'clean democracy' hypothesis. All coefficients of Democracy and Partial Democracy are insignificant in all specifications. Thus, democracies as such are not found to be cleaner than autocracies. This corroborates findings by Gassebner et al (2011) and Kammerlander and Schulze (2020a).
We do find support for the notion that left-wing governments obtain higher EPI scores than rightwing governments (reference category). Left-wing government have a 2 points higher EPI scores on average, about 13% of a standard deviation. Yet, centrist governments have an even higher EPI score-plus 3.3 points (22% of a standard deviation). This is a novel and interesting result which had so far been buried as political orientation had been measured only in the left-right dichotomy, which is clearly too restrictive.
Human capital-abundant economies have higher EPI scores, either because of a cleaner production structure or because politically relevant environmental awareness is higher. More capital-intensive countries have higher EPI scores; countries with larger populations and higher GDP per capita have lower EPI scores, but we find no support for an EKC. The negative marginal effect of higher GDP per capita diminishes in absolute terms, but the turning point is practically outside the support of the GDP per capita (model 4) 12 . Higher urbanization is correlated with higher EPI performance.

Globalization and pollution embodied in trade
The focus on domestic indicators, as used by the EPI could be misleading if countries, for instance as a result of environmental policies, shifted their production of pollution-intensive goods abroad and imported these goods instead. Kanemoto et al (2014) find that, although official targets have been met, many developed countries have in fact increased their emissions when imported emissions are taken into account. Kolcava et al (2019) investigate whether trade liberalization leads to an outsourcing of pollution from developed to developing countries. They find that trade agreements lead to an increased export of ecological footprint in low-income countries, but surprisingly, not to an increase in imported ecological footprint in wealthy countries. Brandi et al (2020) show that environmental provisions may help to reduce pollution intensive and increase greener exports of developing countries. Aichele and Felbermayr (2012) find that the ratification of the Kyoto protocol on the one hand decreased domestic emissions, but on the other hand also increased 12 It is beyond the 99 percentile. imported emissions via trade leading to a zero net effect at best. There is a sizable literature that conceptualizes and estimates 'imported' emissions. For a literature review on CO 2 embodied emissions in trade, see for example Sato (2014). More globalized countries thus could be better able to become greener in terms of domestic emissions by offloading pollutionintensive production to other countries. Clearly, this possibility could not explain the positive coefficient of social globalization, which captures interpersonal, informational and cultural globalization, but it could bias the coefficient for economic globalization upwards.
To shed some light on this issue, we complement our analysis of the determinants of environmental performance by an analysis of the determinants of trade-embedded pollution using the same explanatory variables and the same time span 13 . To measure trade-embedded emissions, we rely on the Our World in Data CO 2 and Greenhouse Gas Emissions database compiled by Ritchie and Roser 13 The trade embedded emission data are not available for all countries used in the baseline regression. We use the largest common sample of 110 (instead of 134) countries.
(2017) 14 . They measure trade embedded CO 2 emissions as the difference between exported and imported emissions as percentage of domestic production emissions. Imported emissions are defined as emissions that are emitted elsewhere in the production of goods that are subsequently imported to the home country. Unfortunately, these data exist only for CO 2 emissions and not for other pollutants that also affect the EPI, such as PM 2.5 . Table 2 mirrors table 1, but uses CO 2 emissions embedded in trade as the dependent variable instead of the EPI. In contrast to the results on the EPI, we find no significant coefficient on globalization in (1) to (3). When splitting globalization into its three main components in (4), we find a significant increase in trade embedded emissions for economic globalization and a significant decrease in trade embedded emissions for social globalization. A one standard deviation increase in economic globalization 14 Ritchie and Roser (2017) compile a large dataset on emissions from many different sources. The sources for the trade embedded data are Peters et al (2011)  (which translates to 16.55 points in the original 0-100 scale) is associated with an increase in trade embedded emissions of about 15.4 percentage points. In contrast, a one standard deviation increase in social globalization (18.29 points) is associated with a decrease of about 42.57 percentage points. Thus, comparing the magnitude of both associations, the estimated impact of social globalization on trade embedded emissions is 2.75 larger than for economic globalization.
While economic globalization does lead to the net export of CO 2 emissions (affected by the import of more CO 2 intensive goods), this effect is not large enough to affect overall environmental performance as measured by the EPI or it may be offset by other globalization-related factors (table 1, model 4). More striking however is the effect of social globalization: not only does it lead to improved environmental performance, it also reduces strongly the export of CO 2 emissions through international trade. Again, the effect of social globalization is much stronger than the effect of economic globalization.
Controls are identical to those in table 1, corresponding column.

Robustness checks
We confirm the robustness of our results, using our model 4 of table 1 as baseline. First, we use alternative cut-offs for democracy (polity2 scores 6-10) and autocracy (−6 to −10) with the group in between named anocracy, as suggested by the authors of PolityIV (Center for Systemic Peace 2018). Second, we use the raw polity2 scores (instead of the categorical variables).
Third, we include the Political Corruption Index taken from the V-Dem database (McMann et al 2016) as corruption may compromise a government's ability to effectively implement environmental regulations (Povitkina 2018). None of these changes affects our three central results-democracy is uncorrelated to environmental performance, left governments have a better environmental performance but centrist governments even more so, and (social) globalization increases environmental performance (table 3) 15 .
Lastly, we run regressions with country fixed effects. Using only the within country variation, coefficients keep their signs, but mostly lose their significance. Given the short time span of a decade (dictated by data availability), finding insignificant results is unsurprising. Only social globalization 15 We have also used six measures from the World Governance Indicators (voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law, control of corruption) instead of democracy, a dummy for oil producing countries, and an indicator for presidential systems as opposed to parliamentary ones. We also use the party of the country's CEO as an alternative measure for political orientation. The findings from the baseline are confirmed in all cases (see the working paper, Schulze and Kammerlander 2020b). remains significantly positive. This again underscores the importance of social globalization for environmental performance.

Discussion
We have analyzed the political and economic correlates of environmental performance using a novel measure of environmental performance, which incorporates a wide array of important variables concerning environmental outcomes and policies rather than single emission measures. Our three major findings are important and challenge conventional wisdom, two of which are completely novel.
The dominating view on the globalizationenvironment nexus has been that globalization is bad for the environment. It has been said to constrain governments to set low environmental standards as they would otherwise lose their competitive edge on the product markets and mobile capital would flow to other countries. This argument focuses on mobility of goods and capital as relevant transmission mechanism of globalization. We do not find any significant negative association of globalization with environmental performance. On the contrary, we find a robust positive correlation of globalization with the environment through an entirely different transmission channel: it is the exchange of information, of culture and of people travelling in and out, i.e. social globalization; that is associated with high environmental performance of a country.
Our finding suggests that the mobility of ideas rather than the mobility of goods and investment matter for environmental performance. Concepts of environmental protection adopted from the global world feed effectively into the domestic political process if a large part of the society is exposed to them. This transmission mechanism has not been studied in the globalization-environment debate, but it is obviously important and needs further investigation.
Second, we find that democracies' environmental performance is not better compared to nondemocracies, which may seem counterintuitive. The clean democracy hypothesis posits that democracies should be cleaner as environmental awareness can be more easily created and conveyed, environmental interests can be better organized and accountability of democratically elected decision-makers ensures that environmental interests feed into the political process. Our finding is, however, in line with a small emerging literature that finds evidence contradicting the clean democracy hypothesis (Gassebner et al 2011, Kammerlander andSchulze 2020a). Either democracies provide room for counterbalancing interests that effectively offset environmental interests (Midlarsky 1998) or non-democracies have similar interests in a clean environment or may be able to enforce regulations more effectively-democratic constituencies per se may not have a larger interest  Our third central result shows that political orientation matters, yet in an unexpected way: left-wing governments pursue stricter environmental protection than right-wing governments, but centrist governments outshine both. This is again a novel finding as political orientation has only been tested in the framework of the left-right dichotomy, which our research proves to be too restrictive.
We do not find no evidence for an EKC. So, neither growth nor democracy as such are a panacea for the huge environmental problems we face today; yet the creation of environmental awareness through ideational spill-overs in the course of social globalization may be part of the solution.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// epi.yale.edu/.