Mapping and clustering the adoption of carbon pricing policies: what polities price carbon and why?

ABSTRACT

Carbon pricing, including carbon taxes and emissions trading, has been adopted by different kinds of polities worldwide. Yet, beyond the increasing adoption over time, little is known about what polities – countries as well as sub- and supranational entities – adopt carbon pricing and why. This paper explores patterns of adoption (both implemented policies and those scheduled to be) through cluster analysis, with the purpose of investigating factors that could explain polities’ decisions to adopt carbon pricing. The study contributes empirically by studying carbon taxes and emissions trading together and by ordering the polities adopting carbon pricing into clusters. It also contributes theoretically, by exploring constellations of variables that drive the adoption of carbon pricing within individual clusters. We investigated 66 adopted policies of carbon pricing, which were divided into five clusters: early adopters, North-American subnational entities, Chinese pilot provinces, second-wave developed polities, and second-wave developing polities. The analysis indicates that the reasons for adopting carbon pricing have shifted over time. While international factors (climate commitments or influences from polities within the same region) are increasingly salient, domestic factors (including crises and income levels) were more important for the early adopters.

Key policy insights

• Carbon pricing has become a global mainstream policy instrument.

• Economic and fiscal crises provide windows of opportunity for promoting carbon pricing.

• The international climate regime can support the adoption of carbon pricing through mitigation commitments and international financial and technical assistance.

• Learning between polities from the same region is a useful tool for promoting carbon pricing.

• Carbon intensive economies tend to prefer emissions trading over carbon taxes.

KEYWORDS:

• Carbon pricing
• carbon tax
• emissions trading
• cluster analysis
• policy adoption

Introduction

Carbon pricing – carbon taxes and emissions trading – constitutes a key part of the response to climate change. It is the preferred solution for environmental economists of practically all theoretical orientations (Haites, 2018), of powerful economic actors, such as the Bretton Woods institutions, and supported by a range of actors from finance ministries to environmental non-governmental organizations (NGOs) (Rabe, 2018). Northern European countries were the first to adopt carbon pricing in the early 1990s, and especially since 2015 it has spread to all regions of the world. Yet our knowledge of which kinds of polities adopt carbon pricing and which factors influence them to do so is limited. We know that, firstly, the adoption of carbon pricing appears to have followed waves with different countries (and kinds of countries) adopting it at different times in a non-linear fashion; and that, secondly, these waves do not appear to follow the most obvious explanatory variables. Regarding these variables, there is a lack of correlation between countries being subject to international climate agreements and the adoption of carbon pricing. Furthermore, explanations based on market liberal policy paradigms, extent of fossil fuel exports or levels of income, are contradicted by the adoption of carbon pricing by corporatist and dirigiste countries such as China (2020), South Korea (2015) and the Scandinavian countries; by fossil fuel exporters such as Canada (2019) and Norway (1991); and by middle and lower income countries such as Mexico (2014) and South Africa (2019), respectively. This puzzle led us to develop our main hypothesis: different constellations of variables are at play for the decision to adopt carbon pricing in different clusters of polities.

Hence, we group the polities adopting carbon pricing into different clusters according to how similar they are to each other and how dissimilar they are to others (Wolfson, Madjd-Sadjadi, & James, 2004), based on a defined set of variables. Such analysis can tell us (1) which clusters exist; and (2) which factors distinguish the clusters and are part of the constellation of variables that drive the adoption of carbon pricing within the individual clusters. The variables that we focus on in the cluster analysis belong to four overarching categories: economic characteristics of polities; their carbon intensity; their political characteristics; and the diffusion effects across polities. The characteristics cover long-term structural variables as well as shorter-term ones such as economic and fiscal crisis or the political orientation of government.

In the first section, we discuss carbon pricing and argue why it makes sense to treat carbon taxes and emissions trading systems as both belonging to the category of carbon pricing. Second, we outline previous studies of the adoption of carbon pricing, and in the third section we clarify our use of cluster analysis as our method. Next, we describe the data used and the variables we employ. The following section is the analysis, in which we identify five clusters of polities having adopted carbon pricing instruments. In the subsequent section, we discuss the findings and their implications. In the final section, we sum up our conclusions and suggest future research.

Carbon pricing as an economic response to climate change

The pricing of greenhouse gas (GHG) emissions is the fundamental solution to climate change according to the environmental economics literature (Jacobs, 1997; Sterner & Coria, 2012; Tol, 2011). Such pricing can occur through carbon taxation or the trading of allowances to emit GHGs within an emissions trading system. Emissions trading systems are also often referred to as ‘cap-and-trade’ or ‘carbon markets’, the latter also including voluntary carbon markets and systems for purchasing carbon credits (or ‘offsets’) from entities outside the polities in question, e.g. the Clean Development Mechanism under the Kyoto Protocol (Paterson, Hoffmann, Betsill, & Bernstein, 2014). However, as offsets and voluntary carbon markets (Benwell, 2009; Segerson, 2013) do not constitute instruments adopted by polities to place a price on carbon, they are outside the scope of this study.

Whereas carbon taxes place the costs of the externality of climate change on the polluters (Jacobs, 1997; Pigou, 1932) – adhering to the ‘polluter pays’ principle (OECD, 1974) – emissions trading creates a system for allocating property rights to emit GHGs and for trading these rights (Coase, 1960; Felli, 2015). Furthermore, emissions trading regulates the quantity of emissions, while carbon taxes regulate their price (Meckling & Jenner, 2016).

We argue that it makes sense to treat carbon taxes and emissions trading as constituting one kind of policy instrument – ‘carbon pricing’ – despite the differences between them. We are interested in the political decision to adopt policies that explicitly target GHG emissions by pricing such emissions, and more specifically, the variables that influence such adoption. We assume that these decisions – no matter whether they concern carbon taxes or emissions trading – at some level are driven by attention to climate change and the notion that pricing emissions provide benefits (be they economic or environmental) – as outlined below – although other factors may be more influential. Hence there is reason to expect the adoption of carbon taxes and emission trading to be driven by similar factors, and in our analysis, we test if this is true, or if polities adopting carbon taxes and emissions trading belong to different clusters.

Besides the above-mentioned reasons for expecting the adoption of carbon taxes and emissions trading to be driven by the same factors, there are indications pointing in that direction. First, most polities with carbon taxes also have emissions trading systems in place covering specific, sometimes overlapping parts of the economy (Haites, 2018). Such use of both kinds of carbon pricing instruments by the same polities underscores that their adoption is likely to have been shaped by similar factors. They have also been increasingly promoted together as carbon pricing, rather than separately as emissions trading and carbon taxes (Carbon Pricing Leadership Coalition, 2018; Skovgaard & Canavan, n.d.).

Second, in terms of effects, carbon pricing policies give an economic signal to polluters to either reduce their emissions or to pay for the externality inherent to those emissions, unlike so-called regulatory instruments that place untradeable obligations (e.g. to reduce emissions per unit of electricity produced) on all polluters (Lauber & Schenner, 2011). Importantly, both carbon taxes and emissions trading systems (at least those with auctioning of emission allowances, which are in the majority) provide revenues for public budgets. Consequently, carbon pricing policies share key commonalities that simultaneously distinguish them from other climate policy instruments.

Third, in terms of ideational underpinnings, all carbon pricing policies are based on the notion that climate change is best mitigated by giving emitters an incentive to reduce emissions in the shape of a price signal, and that the decision on how to reduce emissions is best left to the market. These notions are underpinned by understandings of actors as economically rational, and of the response to climate change as compatible with liberal economic systems. Consequently, both emissions trading and carbon taxes have been promoted by economic actors such as academic economists, the World Bank and The Economist (Rabe, 2018).

Previous studies explaining carbon pricing

Studies of carbon pricing have historically focused on economic aspects, and in the cases in which they focus on the adoption of carbon pricing, mainly on single cases at the national, provincial or European Union level (Harrison, 2012; Heggelund, Stensdal, Duan, & Wettestad, 2019; Wettestad, 2014). Yet, a small but increasing number of publications study polities adopting emissions trading through comparative case studies (Houle, Lachapelle, & Purdon, 2015; the contributions to Biedenkopf, Müller, Slominski, & Wettestad, 2017; and Wettestad & Gulbrandsen, 2018) or large-N studies (Betsill & Hoffmann, 2011; Narassimhan, Gallagher, Koester, & Alejo, 2018). Several studies of emissions trading have adopted a critical perspective, emphasizing its neoliberal underpinnings (Lane & Newell, 2016; the contributions to Stephan & Paterson, 2012; Paterson, 2012). Studies of carbon taxes, on the other hand, tend to focus more on why such taxes were adopted and their implications (Andersen & Ekins, 2009; Haites, 2018). While a small but growing body of literature addresses the diffusion of, respectively, emissions trading (see the contributions to Biedenkopf et al., 2017; Wettestad & Gulbrandsen, 2018) and carbon taxes (Andersen, 2019; Harrison, 2010), the global adoption of carbon pricing policies as an overarching policy instrument has not been studied. Studies covering both carbon taxes and emissions trading are rare, as are comparative studies of the adoption of carbon pricing (but see Carl & Fedor, 2016; Harrison, 2012; Rabe, 2018; Rabe & Borick, 2012).

Altogether, we argue that our study will contribute to the existing literature by (1) testing to what degree decisions to adopt emissions trading and carbon taxes are influenced by the same variables, (2) studying the entire universe of cases of carbon pricing rather than selected cases, and (3) exploring a full range of factors potentially influencing the adoption of policies.

Method

Cluster analysis is designed to create groups not evident in advance and identify common characteristics within each group. The objective is to organize incidences of the adoption of carbon pricing policies – the unit of analysis in this study – into groups based on a larger set of relevant variables pertaining to the adopting polity. The identification of commonalities and trends is a useful method for early, explorative research. In this case, cluster analysis allows us to identify different sets of variables that correlate with clusters of cases. Hence, it does not assume that the same variables correlate equally with all cases, but facilitates future studies of different constellations of variables (or causal pathways) to the adoption of carbon pricing policies. Moreover, by reducing the complexity of the data constituted by the particularities of individual cases, we are able to identify general patterns correlated with the adoption process.

Cluster analysis techniques group observations that resemble each other, while differing from others (Everitt, Landau, Leese, & Stahl, 2011). This enabled us to divide data so that each adoption of a policy belongs to a single cluster, with the sum of all clusters containing all policies in the dataset. We used the SPSS Two-Step Cluster Component that incorporates the use of categorical variables (such as policy choice) without treating them as continuous, which would have reduced the validity of the analysis. The Two-Step Clustering method resembles traditional hierarchical approaches of similarity measures, but includes a pre-step based on a sequential approach (SPSS, 2001).

Data

Data was collected with the objective of determining if a polity had adopted or considered a carbon pricing policy. The basic unit of analysis is the individual decision to adopt a carbon pricing instrument within a given polity. The dataset is mostly composed of individual countries’ policies, although subnational policies were also included for Brazil, Canada, China, Japan, Spain and the United States, since at least one of the subnational units had implemented a carbon pricing policy. With the exception of the Chinese subnational policies, all subnational carbon pricing policies were decided on the subnational level.

Status of carbon pricing policies

For the main variable, we first mapped national, subnational or supranational units that had currently operating or scheduled carbon pricing policies, from 1990 to 2018. All policies placing charges based on the carbon content of a given fuel were examined, regardless of their price level or the sectors included. Although this broad definition led to a dataset consisting of diverse policies, we were interested in identifying policy proposals and decisions that specifically aim to place a price on emitting carbon, regardless of its scope. We found 66 adopted policies, of which five are scheduled to be implemented, and 61 are implemented.

We used the World Bank’s Carbon Pricing Dashboard (The World Bank, 2018a) to identify whether polities have adopted carbon pricing instruments, complemented with information from the Intended Nationally Determined Contributions (INDCs) presented by countries in the context of the Paris Agreement, as well as national legislation.

Polity variables

Additional to the main variable of the status of carbon pricing policies, information on four sets of variables connected to the polity was gathered to act as grouping factors for the cluster analysis. These variables were selected based on theoretically-informed assumptions of factors influencing the adoption of carbon pricing policies (drawing on literature on topics including carbon pricing, environmental politics in general, policy diffusion and others, as discussed in the context of individual categories below). All data was gathered for the year of adoption; at the level of the adopting polity when available, and as an average of the countries or states that make up the EU and the Regional Greenhouse Gas Initiative (RGGI, in the Northeast of the US) in the cases of these two emissions trading systems. For sources of data, see Appendix.

The first set of variables covers economic characteristics of polities. These are the GDP per capita and total GDP of national and subnational units; account balance; industry’s share of GDP; and ability to produce and export manufactured goods competitively, measured using the United Nations Industrial Development Organization composite index (UNIDO, 2017). The assumption is that a higher level of income could be one of the factors inducing polities to adopt carbon pricing, but that an open economy facing high levels of global competition could have the opposite effect. We also included fiscal and economic crises. The underlying assumption is that a crisis can act as an external shock, affecting the likelihood of adoption (Nohrstedt & Weible, 2010). We expect that fiscal crises could in some clusters make the adoption of carbon pricing more likely, while economic crises could have the opposite effect. The reason is that fiscal crises, defined as a ‘fiscal period of heightened budgetary distress resulting in the sovereign taking exceptional measures’ (Gerling, Medas, Poghosyan, Farah-Yacoub, & Xu, 2017), lead to a demand for policies that increase revenue (or cut expenditure), including carbon pricing. Economic crises, on the other hand, defined as an ‘event in which the annual growth rate of real GDP per capita becomes negative’ (Hausmann, Rodríguez, & Wagner, 2008), increases the sensitivity of economic sectors to higher taxes. The occurrence of fiscal crises in up to three years prior to the adoption of the carbon pricing initiative was considered. For economic crises, we considered negative growth up to three years prior to the adoption, identified by yearly growth data.

The second set of variables covers carbon intensity. The assumption was that a high degree of the carbon intensity of an economy constitutes a factor discouraging the adoption of carbon pricing policies, due to opposition from powerful interest groups or the public. Six variables were included at the national level: CO₂ emissions per capita; CO₂ intensity, measured in CO₂ emissions per GDP unit; share of CO₂ emissions coming from manufacturing industries and construction; total energy consumption, and rents coming from oil as a share of GDP, as a proxy for the importance of non-renewable energies for the economy.

The third set of variables covers the political characteristics of polities. A key theoretical assumption is that domestic variables supporting environmental policies could be a factor acting in favour of the adoption of carbon pricing policies. This was measured through the existence of an environment/climate change ministry; the polity’s green party representation in the cabinet (also capturing that parliamentary systems are more likely to have such representation); and an increase in the green party’s share of the vote prior to the discussion or adoption. Other variables captured political characteristics of polities. Following Andersen (2019), a corporatism index was included using the data of Jahn (2016) for the available countries and sub-national units. High levels of corporatism would indicate many veto points for interest groups, making it more difficult to adopt carbon pricing policies. A composite index for rule of law covering the capacity of polities to govern was also included, with the assumption that capacity to govern would be more important to the adoption of a trading scheme than of a tax (the former requiring more administration than the latter). In terms of political short-term factors surrounding carbon pricing adoption, besides a green party in cabinet and its share of votes, the left-right orientation of the polity’s legislative body was used, based on the Comparative Political Data Set (Armingeon et al., 2017) and case-by-case examinations.

Finally, a fourth set of variables attempts to capture the diffusion of policies. Policy diffusion concerns factors influencing interrelated decisions in several polities to adopt a given policy, unlike the previously discussed factors, which concern the influence of similar but unrelated factors on such decisions (Elkins & Simmons, 2005). Hence, whereas the previous variables focused on factors specific to each polity (e.g. its level of income), the diffusion variables are influential across several polities (e.g. entry into force of the Kyoto Protocol). This influence can take place either in terms of polities influencing each other or an international factor influencing several polities (Marsh & Sharman, 2009). Diffusion variables relevant to the adoption of carbon pricing studied here include factors that emanate from the international climate regime, specifically the existence of a binding climate change commitment or pledge at the year of adoption or consideration, and the status as recipient of climate finance or under a policy programme for climate (e.g. the Partnership for Market Readiness). The diffusion variables also include the factor of a polity learning by observing another polity’s experience with carbon pricing and the consequences of this policy. This could make the adoption of carbon pricing in other polities more likely. We operationalize learning by including a regional variable, since such learning is expected to be more likely among geographically proximate polities (Elkins & Simmons, 2005).

Analysis

The cluster analysis was undertaken to identify characteristics between polities that have adopted carbon pricing policy instruments. The data includes 66 policies coming from 62 polities covering all regions of the world.

Five clusters are identified; one consisting of early adopters and four from the last decade, including North-American subnational entities, Chinese pilot provinces, developed polities, and developing polities respectively.

The first cluster consists of early adopters and shows the lowest intra-class similarity, i.e. the highest degree of internal divergence. Mostly, this group comprises polities implementing carbon pricing from 1990 to 2006. Members of this group are typically – but not exclusively – European. Importantly, they have all adopted carbon taxes, either applying to the whole economy or restricted to vehicles. Economically, at adoption, they had a balanced account of payments, but show larger dissimilitude in income levels than any other group. Among the European states adopting economy-wide carbon taxes, most of them are relatively small and, according to Andersen (2019), are characterized by neo-corporatist styles of policy-making. Most of them experienced fiscal or economic crises prior to adoption: 71% and 64% respectively, a factor that has been identified as driving the adoption of the first carbon taxes in Finland (1990) and Sweden (1991) (see Appendix). There is great variation in the carbon intensity variables. Politically, they are more left leaning than other groups, but not greener.

The second cluster covers Canadian provinces and US states that implemented emissions trading rather than taxes from 2007 onwards, showing a high intra-class similarity. Members of this group have relatively high GDP and have been less exposed to financial and economic crisis than other groups. Comparatively, they score higher on carbon intensity variables: they have a higher industry share of GDP than most groups, higher CO₂ emissions per capita, and higher CO₂ per GDP unit. The political desire to address climate change in the absence of federal level carbon pricing in both the US and Canada (which did not adopt federal carbon pricing until 2018) appears to be an important factor, in spite of the polities not being subject to international climate commitments (Rabe, 2018; Wettestad & Gulbrandsen, 2018). Examples of this cluster are the Californian emissions trading scheme and the British Columbia carbon tax.

Chinese provinces constitute their own cluster, with the highest intra-class similarity measure. They differ from the other subnational adoptions in being driven by central government decisions rather than by subnational ones (Lo & Chang, 2014; Qi, Wang, & Zhang, 2014). They have all implemented pilot ETS since 2012, which include similar sectors (industry and power as a minimum). The one different case in this cluster is the nationwide Chinese national ETS, scheduled to be operational by 2020, which incorporates the existing sub-national ones.

The fourth cluster captures what we call ‘developed second-wavers’, and includes national and subnational units of developed countries with an average GDP per capita of 44,000 USD, the second highest after the US-Canadian cluster. These have been adopted since 2005, but with most of the cluster adopting carbon pricing around 2014. Carbon taxes and emissions trading have been adopted to equal degrees. Although a relevant proportion of this group experienced economic crises prior to adoption (47%), fiscal crises are rare. Typically, their economies are less carbon-intensive than the other groups. Politically, 42% of them had green parties in government at the time of adoption, and 72% show an increase in the green party vote, while scoring very high on rule of law. These environmentally-oriented political factors fit with existing explanations of these polities mainly being driven by political concerns over climate change (Wettestad & Gulbrandsen, 2018). Examples of this cluster are the European Union and Japanese provincial emissions trading systems.

Finally, although the time of adoption is similar to the previous clusters, polities in the fifth cluster usually are middle and upper-middle income countries, with an average GDP per capita of 14,000 USD. This group has a low intra-class similarity, due to the differences in economic and political variables. However, they do share a strong preference for carbon taxes (81%). Comparatively, they show lower industrial competitiveness, and more negative balance of payments. Fiscal and economic crises, although far from being the norm, are more usual than in previous groups: 46% of polities experienced fiscal crises and 36% economic ones prior to adoption. There is considerable variation on most carbon intensity variables. Politically, governments adopting these policies comprise the entire spectrum, and show no increased voting for green parties. Members of this group usually have low indexes of corporatism and score lower on rule of law compared to other polities adopting carbon pricing. However, 63% of them received international support and have signed international agreements on climate change action, in line with the explanations that highlight the role of international commitments for the adoption of carbon pricing (Ortega Díaz & Gutiérrez, 2018). Members of this group include Kazakhstan’s ETS adopted in 2013, Mexico’s carbon tax adopted in 2014, and South Africa’s in 2019.

Table 1 shows the structure found for this subset, with a silhouette average of 0.5, which shows a solid cluster structure for the presented cases.¹

Table 1. Cluster structure for implemented and scheduled policies (number of policies per polity included).

Early adopters		US & Canada		Chinese provinces		Developed second wavers		Developing second wavers
Albania	1	Alberta	2	Beijing	1	Australia	1	Argentina	1
Cyprus	1	British Columbia	2	Chongqing	1	Austria	1	Chile	1
Denmark	1	California	1	China	1	Catalonia	1	Colombia	1
Estonia	1	Canada	1	Fujian	1	Croatia	1	India	1
Finland	1	Colorado	1	Guangdong	1	EU/EEA	1	Kazakhstan	1
Greece	1	Massachusetts	1	Hubei	1	France	1	Mexico	1
Latvia	1	Ontario	1	Shanghai	1	Germany	1	Portugal	1
Luxembourg	1	Quebec	1	Shenzhen	1	Iceland	1	Singapore	1
Norway	1	RGGI	2	Tianjin	1	Ireland	1	South Africa	1
Poland	1	Washington	1			Japan	1	Sri Lanka	1
Slovenia	1					Korea	1	Ukraine	1
Sweden	1					Liechtenstein	1
Zambia	1					New Zealand	1
Zimbabwe	1					Saitama	1
						Spain	1
						Switzerland	2
						Tokyo	1
						United Kingdom	1
Total: 14		Total: 13		Total: 9		Total: 19		Total: 11

Note: Italics are used for scheduled policies. In the case of RGGI, although a carbon tax has been in place since 2009, an ETS is in scheduled for 2018. Details in the appendix.

Discussion

The cluster analysis underscores that it makes sense to describe the polities adopting carbon pricing as belonging to different clusters characterized by different variables, indicating that the decisions have been driven by different constellations of variables. First, regarding the economic variables, experiencing economic and/or fiscal crisis was more common among the early adopters than within any other cluster. For this particular group, although diverging in levels of income, regional location and economic composition, economic or fiscal crisis was experienced by most of them. This arguably indicates a change of focus over time, from carbon pricing policies as a revenue-gathering solution to other considerations recently becoming more important. This transformation may coincide with a change in the framing of carbon pricing from defining it as a mainly economic policy instrument to a mainly environmental one. Furthermore, the analysis does not support the expectation that fiscal and economic crises differ in their impacts on adoption.

When it comes to the carbon intensity of the polities, there is considerable variation. Energy intensity of the economy, CO₂ per capita or per GDP unit, or CO₂ from industrial sources, show large variation both within the sample and compared with countries that have not considered carbon pricing mechanisms. No evidence was found for any of these indicators to be related negatively with the likelihood of adopting carbon pricing, even controlling for GDP levels. However, with few exceptions, it seems that carbon intensive economies opt for emissions trading rather than carbon taxation, in line with literature finding that fossil fuel companies have tended to push for emissions trading rather than carbon taxes (Meckling, 2011a, 2011b).

Political characteristics play a minor role in defining clusters. Government composition covers the full political spectrum within all clusters, with both left- and right-wing governments, presidential and parliamentary democracies as well as authoritarian regimes adopting carbon pricing. This underscores that carbon pricing today is a global mainstream policy instrument. Nonetheless, among early adopters, left-wing governments seem to have played a leading role, as have green parties in government, and an increase in voting for green parties, among second wave developed polities (mostly parliamentarian systems). This suggests that domestic environmental factors have been more important in these polities, which further supports the change to carbon pricing being framed as an environmental policy.

Regarding diffusion mechanisms, adoption of carbon pricing policies appears to follow geographical patterns. Geographically proximate polities will often adopt carbon pricing closely following each other. A clear example of this is the Latin American wave of carbon pricing policies occurring since 2014 when Mexico adopted a carbon tax, which was then followed by Colombia and Chile in 2017, Argentina in 2018, and now being considered in Panama, Peru and Costa Rica. A similar situation is found with the early adopters in Europe, starting with Northern Europe and spreading to other parts of Europe. This indicates that, especially among the early adopters, learning between polities in relation to the use of carbon taxes as a tool to address financial (and economic) crises was in play. For later patterns of diffusion, it is more difficult to isolate learning between polities from international influences such as adoption of the Paris Agreement or international support. Diffusion from the international level seems to play an increasing role in the context of the first commitment period of the Kyoto Protocol and the 2009 and 2015 Copenhagen and Paris Climate Conferences, as well as international support for carbon pricing, inter alia, by the World Bank. Overall, this indicates that diffusion from the international level has played a role, especially among the two latter clusters.

Finally, in terms of instrument choice, few variables distinguished the polities adopting carbon taxes from those adopting emissions trading, and the choice of instrument was not as important a characteristic of any of the clusters compared to timing or level of income. With the exception of the Chinese cluster (which constitutes an outlier as it was driven by one set of national decisions rather than polity-specific ones), all clusters include both carbon taxes and emissions trading. Although taxes were more common among early adopters and developing second wavers, and emission trading more common among US and Canadian states and provinces, our study does not identify a clear overarching pattern regarding carbon taxes versus emissions trading (except for carbon intensive economies preferring emissions trading). Thus, our decision to include both policies under the carbon pricing heading is supported.

Conclusions

Our analysis mapped the development of carbon pricing from the first wave of adoptions in Northern Europe in the 1990s to its current status as a mainstream global climate policy instrument. Carbon pricing is now operational, or scheduled, in polities covering developed and developing, and more and less carbon intensive countries, as well as supra- and subnational entities from all regions. The analysis demonstrated the usefulness of employing cluster analysis to study the adoption of environmental policy instruments. The clusters were respectively early adopters, North American polities, Chinese provinces, developed second wavers and developing second wavers. The analysis has identified the characteristics of each one of these clusters, including the differences between them.

Crucially, our results show that there have been different pathways to the adoption of carbon pricing, where different constellations of variables were in operation. Thus, the differences among the clusters underscore that the factors that caused the adoption of carbon pricing in the 1990s in Europe were not identical to those currently driving the adoption of carbon pricing in Latin America, Asia or American states, although there are some similarities. Among the more recent clusters, international environmental factors are among these variables, whereas at earlier stages, economic and fiscal crises were salient. The latter crises may have acted as external shocks enabling adoption of new policies. Regarding international environmental commitments, there seems to be no difference between binding commitments (the Kyoto Protocol) and bottom-up pledges (the Paris Agreement). Whereas diffusion from the international level through treaty commitments seems to constitute one constellation of variables (or pathway) leading to the adoption of carbon pricing (in the second wave developing polities cluster), another pathway seems to consist of diffusion acting together with domestic political factors. These political factors are more constant between the clusters. Yet, increased voting for green parties and green parties in government are common characteristics among second wave developed polities. Importantly, the analysis also showed that carbon taxes and emissions trading have been adopted due to similar sets of factors: except for the Chinese cluster, all clusters included both carbon taxes and emissions trading. Furthermore, whether carbon pricing was implemented for environmental or economic concerns does not seem to influence the choice of policy instrument.

On the basis of our analysis, future research should focus on pinpointing the influence of individual variables within the constellations causing the adoption of carbon pricing within the different clusters. Whereas cluster analysis can identify patterns of correlation between factors and the adoption of carbon pricing, other methods are required to assess the causal influence of these individual factors on the adoption of carbon pricing within each cluster. Process-tracing within single case studies can assess precisely which casual role each factor played, whereas comparative analysis and regression analysis can assess broader patterns of causal influences within clusters or for the entire case universe.

We argue that it would be particularly interesting to study the role of economic and environmental factors, as well as domestic and international ones, within the individual clusters. More specifically, our analysis points to the need to assess the interaction between these factors, including the influence of fiscal and economic crisis and which role they play vis-a-vis other factors (e.g. acting as external shocks); the interaction of international and domestic environmental factors (e.g. international commitments bringing carbon pricing onto the political agenda); and the degree to which the adoption of carbon pricing instruments is increasingly motivated by environmental, rather than economics, concerns.

Disclosure statement

No potential conflict of interest was reported by the authors.

Acknowledgments

We would like to thank Roger Hildingsson, Matthew Paterson, the participants in the annual meeting of the Working Group on Environmental Politics of the Swedish Political Science Association (3rd–5 October 2018) and the three anonymous reviewers for their very useful comments at various stages of writing this paper. This work was supported by the Swedish Research Council Formas under Grant number 2017-01638.

ORCID

Jakob Skovgaard http://orcid.org/0000-0002-8814-2534

Additional information

Funding

This work was supported by the Swedish Research Council Formas under [grant number 2017-01638].

Notes

1. Silhouette coefficients range from 1 to −1. A value above 0.2 usually indicates the existence of a cluster structure (Everitt et al. 2011).

Appendix

Table A1. Policies included in cluster analysis.

Country	Jurisdiction	Status	Policy	Country	Jurisdiction	Status	Policy
Albania	Albania	Implemented	Carbon tax	Ireland	Ireland	Implemented	Carbon tax
Argentina	Argentina	Implemented	Carbon tax	Japan	Japan	Implemented	Carbon tax
Australia	Australia	Implemented	ETS	Japan	Saitama	Implemented	ETS
Austria	Austria	Implemented	Carbon tax	Japan	Tokyo	Implemented	ETS
Canada	Alberta	Implemented	Carbon tax	Kazakhstan	Kazakhstan	Implemented	ETS
Canada	Alberta	Implemented	ETS	Korea, Republic of	Korea, Republic of	Implemented	ETS
Canada	British Columbia	Implemented	Carbon tax	Latvia	Latvia	Implemented	Carbon tax
Canada	British Columbia	Implemented	ETS	Liechtenstein	Liechtenstein	Implemented	Carbon tax
Canada	Canada	Scheduled	Carbon tax	Luxembourg	Luxembourg	Implemented	Carbon tax
Canada	Ontario	Implemented	ETS	Mexico	Mexico	Implemented	Carbon tax
Canada	Quebec	Implemented	ETS	New Zealand	New Zealand	Implemented	ETS
Chile	Chile	Implemented	Carbon tax	Norway	Norway	Implemented	Carbon tax
China	Beijing	Implemented	ETS	Poland	Poland	Implemented	Carbon tax
China	Chongqing	Implemented	ETS	Portugal	Portugal	Implemented	Carbon tax
China	China	Scheduled	ETS	Singapore	Singapore	Scheduled	Carbon tax
China	Fujian	Implemented	ETS	Slovenia	Slovenia	Implemented	Carbon tax
China	Guangdong	Implemented	ETS	South Africa	South Africa	Scheduled	Carbon tax
China	Hubei	Implemented	ETS	Spain	Spain	Implemented	Carbon tax
China	Shanghai	Implemented	ETS	Spain	Catalonia	Implemented	Carbon tax
China	Shenzhen	Implemented	ETS	Sri Lanka	Sri Lanka	Implemented	ETS
China	Tianjin	Implemented	ETS	Sweden	Sweden	Implemented	Carbon tax
Colombia	Colombia	Implemented	Carbon tax	Switzerland	Switzerland	Implemented	Carbon tax
Croatia	Croatia	Implemented	Carbon tax	Switzerland	Switzerland	Implemented	ETS
Cyprus	Cyprus	Implemented	Carbon tax	Ukraine	Ukraine	Implemented	Carbon tax
Denmark	Denmark	Implemented	Carbon tax	United Kingdom	United Kingdom	Implemented	Carbon tax
Estonia	Estonia	Implemented	Carbon tax	United States	California	Implemented	ETS
EU	EU/EEA	Implemented	ETS	United States	RGGI	Scheduled	ETS
Finland	Finland	Implemented	Carbon tax	United States	RGGI	Implemented	ETS
France	France	Implemented	Carbon tax	United States	Colorado	Implemented	Carbon tax
Germany	Germany	Implemented	Carbon tax	United States	Massachusetts	Implemented	ETS
Greece	Greece	Implemented	Carbon tax	United States	Washington	Implemented	ETS
Iceland	Iceland	Implemented	Carbon tax	Zambia	Zambia	Implemented	Carbon tax
India	India	Implemented	Carbon tax	Zimbabwe	Zimbabwe	Implemented	Carbon tax

Table A2. Variables, categories and distribution of categorical data used in carbon policies cluster analysis.

Variable	Level of aggregation	Categories	N	Percentage	Source
Pricing mechanism	Polity	Carbon tax	39	59.10%	Individually assessed
		ETS	27	40.90%
Type of jurisdiction	Polity	National	42	63.64%	V-Dem (Coppedge et al., 2018)
		Subnational	23	34.85%
		Supranational	1	1.52%
Economic crisis	Polity	No economic crisis	46	69.69%	The World Bank (2018b)
		Economic crisis	20	30.30%
Fiscal crisis	Polity	No fiscal crisis	50	75.75%	Gerling et al. (2017)
		Fiscal crisis	16	24.24%
Ministry of environment/climate change	Polity	No	21	31.80%	Individually assessed
		Yes	45	68.20%
Increased green party vote	Polity	Regular vote for green party	45	68.20%	Comparative Political Data Set (Armingeon et al., 2017)
		Increased vote for Green Party	21	31.80%
Green party in government	Polity	No green party in government	56	84.80%	Comparative Political Data Set (Armingeon et al., 2017). Individual assessments
		Green party in government	10	15.20%
Government composition	Polity	Right	12	18.20%	Comparative Political Data Set (Armingeon et al., 2017)
		Center-right	16	24.20%
		Center	2	3.00%
		Center-left	11	16.70%
		Left	19	28.80%
		Wide coalition	4	6.10%
		Not applicable	2	3.00%
Corporatism index scale	Polity	Low	39	60.00%	Jahn (2016) and individual assessments. Countries not included in the index and subnational units were analyzed case-by-case, using Jahn’s definition.
		Middle	15	23.10%
		High	11	16.90%
Climate change commitment	Polity	No	28	42.40%	United Nations Climate Change (2018)
		Yes	38	57.60%
Under international climate programme	National before 2015, polity after	No	49	74.20%	Individual assessment
		Yes	17	25.80%
Region	Polity	East Asia & Pacific	16	24.20%	The World Bank (2018e)
		Europe & Central Asia	26	39.40%
		Latin America & Caribbean	4	6.10%
		Middle East & North Africa	1	1.50%
		North America	14	21.20%
		South Asia	2	3.00%
		Sub-Saharan Africa	3	4.50%

Table A3. Descriptive statistics of continuous data used in the cluster analysis of carbon policies.

Variable	N	Minimum	Maximum	Mean	Std. Deviation	Source
CO2 per capita (metric tons per capita)	66	0.19	19.93	8.82	5.19	The World Bank (2018c)
CO2 per GDP unit (kg per 2010 US$ of GDP)	66	0.01	1.69	0.51	0.46	The World Bank (2018d)
Industry share of GDP (value added)	66	11.41	39.81	28.26	7.14	The World Bank (2018f)
Competitiveness Industrial Performance Index	66	0.01	0.56	0.23	0.15	UNIDO (2017)
Account balance at adoption (US$, billions)	66	−711.04	304.16	−14.29	176.55	International Monetary Fund (2018)
GDP at adoption (current US$, millions)	66	5081.43	11064666.28	2228211.20	3547758.17	International Monetary Fund (2018)
GDP per capita at adoption (current US$)	66	857.36	73111.78	29057.89	21612.52	The World Bank (2017)
Oil share of GDP	66	0.00	5.63	0.31	0.83	The World Bank (2016)
Total energy consumption (kWatt/hour, millions)	66	386.30	4831403.30	1151246.90	1727884.56	UNData (2018)
CO2 from manufacturing (% of total fuel combustion)	66	0.00	34.69	16.98	8.09	The World Bank (2018g)
Rule of Law Index	66	0.19	1.00	0.81	0.26	Coppedge et al. (2018)