Fueling opposition? Yellow vests, urban elites, and fuel taxation

ABSTRACT
 The yellow vest movement in France was triggered by a fuel tax increase. The increase was designed to reduce petrol consumption and hence emissions. However, the costs of this policy were unequally distributed among citizens. The yellow vests illustrate that abstract support for public action against climate change is not enough to secure support for concrete policies. These policies run the risk of being opposed by citizens when their costs are unevenly spread across society. Through a survey experiment on Norwegian citizens, we analyse how different policy characteristics generate ‘support gaps’ within society. Particularly, we contrast two stylized profiles: ‘urban élites’ (high-educated urbanites) and ‘yellow vests’ (low-educated ruralites). We find that opposition rises with the individual cost of the policy but in a non-linear way. Environmental earmarking and invoking personal responsibility widen the gap between yellow vests, urban élites, and the rest of society. Redistribution to low-income households and invoking international commitments help to bridge such gaps. We argue that, instead of focusing on aggregate levels of support, both policymakers and academics should rather pay attention to the polarizing effects of certain policy attributes.


Introduction
The yellow vest movement in France has gained much attention from the climate and energy community (Douenne & Fabre, 2020Martin & Islar, 2021). A bottom-up, grassroots, spontaneous movement, it was mostly coordinated through social media. It has remained independent from labor unions and political parties. In both its composition and demands, it wasand still isheterogenous and in flux. In its composition, it mostly brought together lower-middle class citizens dissatisfied with political representation and the establishment. These are typically non-voters, or voters of the extreme right and extreme left (Kipfer, 2019, p. 221;Mehleb et al., 2021, p. 384). They mostly belong to rural France (de Comarmond, 2018;Vernet & de Saint Sauveur, 2019). In its demands, this movement focused on either specific issues such as the fuel tax itself, the reduced 80 km/h speed limit on secondary roads, and the resignation of President Macron, or on general issues such as purchasing power, inequalities, direct democracy, and social justice. The movement, and its Saturday demonstrations, started in November 2018 and lasted until the first Covid-19 lockdown in mid-March 2020. It has since resurfaced in weaker forms.
The yellow vests crystalized citizen opposition to climate policies. Protests were ignited by a rise in a carbon tax aimed at accelerating France's energy transition. Building on decades of discontent about inequalities and political representation, the tax increase was a triggering factor. For scholars of public policy and climate change mitigation, it provides a stark reminder that policy change does not operate in a vacuum, but interacts with questions situated beyond its immediate concerns, such as fairness, social justice, and inequalities. Indeed, France is not a climate-sceptic country. In comparative studies it often comes across as more concerned by climate change than neighboring countries such as Germany or the UK (Hornsey et al., 2018;Steentjes et al., 2017). Most of its citizens and politicians acknowledge the climate crisis and support mitigation measures. But not at all costs.
This article draws on the yellow vest movement to explore fuel taxation policies anew. To do this, it investigates how different tradeoffs affect citizen support for these policies. More specifically, it examines three factors: 1) citizens' willingness to pay, 2) their sensitivity to various revenue recycling schemes, and 3) different framings of the tax's origin. It then assesses to what extent these policy characteristics activate (or not) a yellow vest-urban élite cleavage understood as educational and urbanization divides. It does so in the context of Norway, which is both a likely case to observe such a cleavage (urban-rural tensions, educational disparities) and an unlikely case (comparatively low levels of opposition to fuel taxation, consensus-seeking style of policymaking).
We here argue that support erodes into opposition when the individual costs become too high, such as a ≥ 20% increase in fuel price. Variation regarding revenue recycling or the origin of the tax does not have an effect at the aggregate level (i.e., on the population as a whole). But it does affect opposition dynamics emanating from certain cross-sections of the population.
Respondents who have a 'yellow vest' profile (i.e., rurality and lower qualifications) are significantly less supportive of a fuel tax increase than the rest of the population. But they become indistinguishable from other citizens when the tax is accompanied by redistribution to low-income households or when it is framed in terms of international commitments. Conversely, the gap widens when the tax is framed in terms of national commitments and citizens' own personal responsibility, and when the tax is used to fund further environmental measures. Hence, some policy ingredients have polarizing effects and open 'support gaps' in society. Others don't.
Urban élites (i.e., high-educated city-dwellers), on the other hand, are unmoved by the tax's characteristics and display unconditional support for any variant of the tax. Thus, they are significantly different from the rest of the citizenry; and more so than the yellow vests. One characteristic, however, does provide an extra boost: revenue recycling towards further climate measures. For that group, the more climate-friendly the policy, the better.
This analysis highlights that some of the observational insights derived from the French case can travel to (a) an experimental setting (b) in the 'mixed' case of Norway. Indeed, though Norway exhibits clear rural-urban and educational divides (Knudsen, 2018;Mayne & Peters, 2022;Orderud & Kelman, 2011) many of the dynamics observed in France, and further afield, are fed by an ongoing crisis of representation, enduring inequalities, distributional conflicts, and a lack of social justice (Aklin & Mildenberger, 2020;Marino & Ribot, 2012). Features which are much less salient in Norway (Lister, 2009;Stein et al., 2021, p. 42). Additionally, in its climate policy, Norway tends to be characterized by 'consensus-driven policymaking' and relying on carbon taxation as a key policy instrument (Ćetković & Skjaerseth, 2019(Ćetković & Skjaerseth, , p. 1051. It was not obvious that educational and urbanization divides would structure opposition to fuel taxation in Norway too. Hence, far from being characterized by French exceptionalism, the logics underlying opposition to carbon taxation seem to travel to another case too, despite it belonging to a 'Nordic Nirvana' characterized by high wealth and welfare (Lister, 2009). This Norwegian case study highlights that urbanization and educational divides significantly structure attitudes towards climate policies, generating 'support gaps' and thereby both polarizing and hindering the necessary climate and energy transition. We do not claim that Norway faces the same levels of resistance to fuel taxation than those found in France (it does not). More simply we argue that the labels of 'yellow vests' and 'urban élites' neatly subsume urbanization and educational divides. These help us better understand the polarizing effects some climate policies may trigger.
In this article, we first provide some background information on carbon taxation and the inspiration for the cleavage analyzed in this study: the French yellow vest movement. Then, we detail the experiment's three attributes (willingness to pay, revenue recycling, origin framing) and provide some information regarding the implementation of the survey in the Norwegian context. Next, we discuss the attributes' aggregate effects and how these differ across the yellow vest-urban élite divide. Finally, we provide some overall conclusions. Within the special issue, this contribution highlights the importance of public support in the climate and energy transition. However, the contribution also outlines how consensus can be builtor at least dissensus avoidedthrough more complex policy packages. These need to deal not only with the climate and energy transition itself, but also with its societal repercussions (Boasson & Tatham, 2022).
Carbon taxation, climate change mitigation, and the yellow vests

Carbon taxation
Carbon taxation is an instrument to reduce emissions. It is one instrumentamong many othersavailable in the policy toolkit to mitigate climate change (Umit & Schaffer, 2020). It is a Pigouvian tax: it seeks to internalize negative externalities. In other words, it puts a monetary value on the collateral costs of economic activities. This makes it an attractive policy instrument to disincentivize carbon emissions. However, Pigouvian carbon taxation also has some downsides. It usually adds to the tax burden of citizens (unless it is used to offset other taxes) and it can grow beyond intentions (when valueadded tax multiplies its cost or when inflation kicks in). In addition, carbon taxation can only complement other decarbonization policies. It will not be sufficient on its own. Nonetheless, it needs to be hurtful to work as a deterrent, otherwise it will not affect fuel consumption. Finally, it sometimesthough not always (see Klenert & Mattauch, 2016;Ohlendorf et al., 2021;Sterner, 2012)takes the form of a regressive tax in the sense that it can take a larger percentage of income from lower earners than from higher earners. This can raise fairness and social justice questions about the distribution of costs across society (Drews & van den Bergh, 2016, p. 862). Whilst it is a key instrument to reduce fossil fuel dependence, the crucial question is how to design a carbon tax which is sufficiently hurtful to disincentivize fuel consumption but which will, at the same time, harness wide acceptance.

The French carbon tax and the rise of discontent
The French carbon tax was put in place under the Holland presidency in 2014. The tax was intended to follow sliding rule increases over time. From 7€/ton of CO 2 (tCO 2 ) in 2014 to 100€ by 2030, with forecast increases to 600-900€ by 2050. When first introduced, the tax benefited from relatively low salience due to its small size and depressed oil market prices in 2014-6. In 2017, the Macron presidency decided to accelerate the rate of increase of the tax. The tax would almost triple in size within 5 years to reach 86 €/tCO 2 by 2022 instead of the planned 65 €/tCO 2 . This push came as the French government argued that the current taxation trajectory would fall short of achieving the 2°C global warming objective. This acceleration was unfortunate in its timing and context. Regarding timing, the hike coincided with market increases in fuel prices. This meant a double rise: in the market price of fuel and in its heavier carbon taxation. This double rise was further magnified by the multiplicative effect of the value-added tax, which is calculated on the post-carbon-tax price. Regarding context, it coincided with the implementation of unpopular reforms which were perceived as unfair or unjustified. These reforms were seen as favoring the rich, such as the abolition of the solidarity wealth tax (Impôt de solidarité sur la fortune or 'ISF') or as constraining rural citizens in unnecessary ways, such as the lowering to 80 km/h of speed limits on secondary roads.

The yellow vest movement
The yellow vest movement was a spontaneous movement initiated through social media. It took the form of Saturday demonstrations and road/roundabout blockades, lasting from November 2018 till March 2020 (first corona lockdown). It quickly broadened into a movement for economic justice, with rising fuel prices an emblem of high living costs (Salies, 2019). It was also combined with a sentiment that a disproportionate burden of the government's tax reforms was falling on the lower-middle classes and on inhabitants of rural and peri-urban areas (who are dependent on fossil cars and fuel heating). This section of society, dominated by rural lower-middle class citizens (Kipfer, 2019, p. 212), also corresponds to those who 'have been hit hard by the great recession and international competition' (Vona, 2019, p. 524). In other words, the losers of globalization.
In terms of its immediate demands regarding fuel prices, the yellow vest movement was successful. The French government had to backdown, withdraw the carbon price increase, and freeze any further increases for the 2019-2022 period. In terms of its wider demands regarding purchasing power, inequalities, direct democracy, and social justice, the movement's success is far from clear. In this sense, the yellow vest movement illustrates a number of dilemmas common in energy and climate policies. The most crucial being, perhaps, how different Sustainable Development Goals (SDGs) can be combined, such as achieving the energy and climate transition on the one hand (SDG7, SDG13) and reducing inequalities on the other (SDG10) (Martin & Islar, 2021;Salies, 2019).
Fostering support or triggering opposition: what policy mixes for whom? Policy success is affected by a variety of variables. Carbon taxes are no exception in this regard. An important element regarding carbon pricing is public attitudes (Levi et al., 2020;Umit & Schaffer, 2020). These attitudes are, in turn, affected by numerous factors (Baranzini & Carattini, 2017;Carattini et al., 2018;. They range from specific policy factors (e.g., perceived effectiveness of the instrument), to contextual factors (e.g., trust in government) and social-psychological factors (e.g., climate change beliefs and risk perception). Though crucial, many of these are difficult to alter. For instance, it is difficult to change levels of trust in government or climate beliefs. They all matter, but policymakers will have a tough time changing them through legislation, at least in the short run. In contrast, we focus on three factors we consider both relevant and more malleable: a) the individual cost of the tax, b) the different types of revenue recycling schemes that accompany the tax, and c) the framing of the tax in terms of its origin. We focus on these, since they can be realistically altered in different policy packages (see Klenert et al., 2018;Kotchen et al., 2013;Schaffer & Umit, 2022) and have come across as relevant in the yellow vest movement. 1 Learning from the French experience, we expect that these factors play out differently across societal groups (see Driscoll, 2021;Kipfer, 2019;Martin & Islar, 2021;Mehleb et al., 2021). Indeed, a conclusion from the yellow vest movement is that one cannot expect homogenous responses to different policy characteristics. Societies are heterogenous. And we expect that certain policy attributes can activate certain cleavages, whilst others will leave them dormant. We therefore distinguish two groups within the wider population. On the one hand, a 'yellow vest' profile, understood as citizens with low educational attainments and rural residence (as defined by Corre, 2018;de Comarmond, 2018;Vernet & de Saint Sauveur, 2019). And on the other, an 'urban élite' profile, understood as urbanites with high educational attainments (see Driscoll, 2021, pp. 12-13;Martin & Islar, 2021, p. 603). We do not wish to reify either category. Nor do we consider them as uniform and fixed. Both yellow vests and urban élites are much more diverse than these simplifying dichotomies. But we do believe that education and rurality are key components in the yellow vesturban élite divide (see Corre, 2018;de Comarmond, 2018;Driscoll, 2021, pp. 12-13;Martin & Islar, 2021, p. 603;Vernet & de Saint Sauveur, 2019). And we expect that certain policy characteristics will activate such a cleavage whilst others won't. Awakening such a cleavage can lead to resistance to change and potentially policy failure. We therefore express our hypotheses both in general terms and regarding a possible yellow vest-urban élite divide.

Individual willingness to pay
One explanation of support for climate policies is citizens' 'willingness to pay'. This factor plays a role across different policy instruments, including carbon taxation (Kotchen et al., 2013). It is directly related to the (perceived or real) individual costs of the policy. Whilst some climate policies have individual costs which are difficult to assess, others have more obvious costs. The clarity of costs to an individual plays a crucial part in affecting support (Bechtel & Scheve, 2013). As summarized by Drews and van den Bergh (2016, p. 861) many studies fail to specify the policy's personal costs when investigating public opinion support for climate mitigation. This, in turn, results in biases towards favorable responses when personal costs are unclear or uncertain. Conversely, clearly specifying personal (monetary) costs to individuals may well give a better indication of support for a concrete policy.
We hypothesize that higher individual costs will decrease support (Bechtel & Scheve, 2013;Umit & Schaffer, 2020). Following the French experience, we expect that the monetary costs of the fuel tax will affect the yellow vests' support more dramatically than that of the rest of the population, and certainly far more than urban élites who are likely to exhibit a greater (capacity and hence) willingness to pay (Driscoll, 2021, p. 12).

H1. Individual willingness to pay
H1a. Support will decline as individual costs rise Compared to the rest of the population: H1b. Yellow vest support will be more sensitive to individual costs H1c. Urban élite support will be less sensitive to individual costs

Revenue recycling
Carbon taxes generate new revenue. This raises the question of the use of this additional revenue. The reinvestment of carbon tax monies has been called 'revenue recycling'. A variety of revenue recycling schemes exist, from no particular revenue recycling (i.e., allocation to general public expenditures), to funding further environmental measures, to redistribution to affected parties. There is a consensus in the literature that lack of revenue recycling dampens support for carbon taxation (Beiser-McGrath & Bernauer, 2019). Klenert et al. (2018, p. 671) have summarized that when carbon revenue goes towards the general government budget, public acceptability tends to be lower. However, if the carbon revenue is clearly earmarked for a specific purpose, such as green investments or support of affected groups, then acceptability tends to be higher. We therefore expect that clearly communicated revenue recycling will increase support.
There is growing evidence that two particular types of revenue recycling schemes boost support. These are earmarking to green or environmental projects and redistribution to affected groups. Green earmarking has been shown to increase support for fuel taxation (Baranzini & Carattini, 2017;. Such environmental earmarking has also been argued to make the tax appear more environmentally effective. This increases the tax's legitimacy as it becomes more than yet another revenue generating instrument (Mehleb et al., 2021, p. 383).
At the same time, there is a recognition among climate scholars that a central issue in the energy transition is about distributive justice and inequalities (Aklin & Mildenberger, 2020). A strand of scholarship has recognized that climate policies receive more support if the distribution of costs is progressive and that any additional revenue is redistributed to the more vulnerable members of society (Drews & van den Bergh, 2016, p. 862). Klenert et al. (2018) argue that the best type of revenue recycling is in fact context dependent. They outline that 'when distributional concerns are the greatest obstacle to higher carbon prices, transfers directed to the poor outperform other recycling mechanisms. (…) Earmarking the revenue for green spending might be the option of choice if the main obstacle is that citizens are unconvinced of the environmental benefits of higher carbon prices' (2018: 675).
We take up this line of reasoning but scale it down to individuals. If individuals are more concerned by distributional and fairness issues, redistributive measures will boost support. If individuals are more concerned about accelerating climate change mitigation, green earmarking will boost support. We therefore expect economic redistribution to increase yellow vest support, whilst further financing of environmental measures would increase support among urban élites (Driscoll, 2021;Mehleb et al., 2021).

H2. Revenue recycling
H2a. Support will increase with explicit revenue recycling Compared to the rest of the population: H2b. Yellow vest support will increase with economic redistribution to affected parties H2c. Urban élite support will increase with the financing of further environmental measures

Origin of the policy
Policies may come across as varyingly legitimate and efficient depending on their source of origin. This is especially true for climate policies. One strand of the literature perceives climate policy as requiring collective action at the global level (Keohane & Victor, 2011). This would limit free-riding and help coordinate both domestic and international mitigation efforts (Keohane & Victor, 2016). In this sense, emission-reducing policies such as a carbon tax may well come across as more efficient and credible when emanating from international agreements within established frameworks such as that of the United Nations Framework Convention on Climate Change (UNFCCC). Indeed, past research has indicated greater support for climate mitigation policies when a large number of countries participates and when monitoring is carried out by an independent commission or the United Nations itself (as opposed to one's own government) (Bechtel & Scheve, 2013, p. 13766).
On the other hand, the failures of global coordination to slowdown climate change have also led to disillusionment as to the effectiveness of international cooperation. This has encouraged some to put greater hope in national initiatives rather than lowest-common-denominator international attempts (Rayner, 2010). Simultaneously, 'green nationalism' has emerged as a vehicle to harness support among more reticent citizens (Conversi, 2020). Indeed, nationalism is mostly associated with climate skepticism and opposition to heavier taxes on fossil fuels (Kulin et al., 2021). Framing a fuel tax in terms of a national obligation may boost support in ways that other framings may not. Such 'green nationalism' has been observed in a number of settings, from minority nationalism in Scotland and Catalonia (Conversi & Friis Hau, 2021) to incantations of national pride in the Nordic countries (Ridanpää, 2021).
Finally, some research has highlighted the potential of invoking citizens' moral responsibility in fighting the climate crisis. There are two sides to this argument. First, personal moral responsibility can act as a driver for behavioral change. As individual-level change is observed, it then diffuses through emulation and reciprocity processes, triggering an increase in collective capability (Kyllönen, 2018). Second, individual appetite for climate mitigation will elicit responses from politicians and business leaders. They will be responsive to the resulting electoral and market pressures, i.e., to citizens acting as voters and consumers. This 'ecological citizenship' (Wolf et al., 2009) has led some scholars to argue that 'behavior based on moral or social responsibility may hold out the best prospects for sustainable action' (Weber, 2010, p. 332).
Within the context of a fuel tax, where elasticity of demand is weak and individual agency faces greater constraints, we expect that invoking international commitments is the most likely framing to boost support. Regarding the yellow vest-urban élite divide, we expect that yellow vests will be less receptive to personal responsibility and national commitments (Driscoll, 2021, p. 7). This is because yellow vests already consider that they are personally shouldering a disproportionate amount of the burden in the mitigation effort, especially compared to economic élites or richer members of society. And that this inegalitarian regime is being maintained with the connivance of the political class. To their eyes, responsibility lies rather with polluting economic élites (not the common people) and with national political élites (who entrench the establishment) (Kipfer, 2019;Mehleb et al., 2021). We expect the yellow vests to better respond to a broad international commitment, as embodied by the (less domestically politicized) UNFCCC COP21 Paris Agreement.
One should, however, clarify this expectation. Although we expect that yellow vests will 'better' respond to international commitments than to national commitments or personal responsibility, we do not expect them to respond particularly 'well' to such international commitments. Indeed, recent literature has highlighted the triangular link between those who feel 'left behind' by globalization (such as the yellow vests), rightwing populism, and climate skepticism; the latter also expressing itself through anti-cosmopolitan, anti-liberal, and anti-élite narratives (Lockwood, 2018, p. 719, 722, 726) which may well take aim at international agreements such as the UN's Paris Agreement. Similarly, a link has been observed between the growth of non-majoritarian institutions (including international organizations such as the UN) and an authoritarian populist backlash (Zürn, 2021). Although we do not expect international organizations such as the UN to win popularity contests among less educated ruralites, we do expect that highlighting personal responsibility or national commitments will fare even worse.
Conversely, we expect that urban éliteswho have shorter commutes and greater access to alternatives, such as collective transport or electric carswill be more sensitive to invocations of personal responsibility. Such personal responsibility will echo these élites' normative desires to act as 'ecological citizens' within a cosmopolitan world they experience and identify with (Valencia Sáiz, 2005). As educated urbanites, they also have stronger voting habits and greater economic leverage, hence a more powerful voter-consumer card to play as an individual (Wolf et al., 2009).

H3. Origin of the policy
H3a. Support will increase if the policy originates from international commitments Compared to the rest of the population: H3b. Yellow vest support will increase if the policy originates from international commitments H3c. Urban élite support will increase if the policy originates from personal responsibility

The survey experiment
We test our expectations using a survey experiment in the context of Norway. We provide information on the Norwegian case, its relevance to understand opposition to fuel taxation, and details on the setup and analysis of the multifactorial survey experiment.

Norway
Norway is a mixed case in the sense that it is both an unlikely and a likely case of opposition to fuel taxation. One systemic determinant of support for carbon taxes is political trust (Klenert et al., 2018;Umit & Schaffer, 2020). Trust levels in Norwayand in the Nordics more generallyare among the highest in the world, and strongly associated with high levels of wealth, social security, income equality, civil liberties, and gender equality (Lister, 2009;Stein et al., 2021, p. 42). In this sense, many of the ingredients at the root cause of the yellow vest movement (Driscoll, 2021;Kipfer, 2019;Martin & Islar, 2021;Mehleb et al., 2021), are simply much less pronounced in the Norwegian case. Additionally, the style of climate policymaking in Norway is widely presented as being 'consensus-seeking' and 'partnership-based' (Ćetković & Skjaerseth, 2019, p. 1044) with a long tradition of carbon pricing as a key policy instrument (Ćetković & Skjaerseth, 2019(Ćetković & Skjaerseth, , p. 1051. Certainly, fuel taxes are not popular in Norway , p. 2972. However, they are significantly less unpopular than in other countries. Available comparative research shows Norwegians are among those most in favor of climate-driven fuel tax increases (Umit & Schaffer, 2020, p. 4, see Figure 1) and certainly much more than the French, German, or British (Steentjes et al., 2017, p. 32).
On the other hand, Norway is a likely case for exhibiting a yellow vesturban élite cleavage, due to ongoing rural-urban tensions and educational disparities, which impact questions of representation and policymaking (Knudsen, 2018;Mayne & Peters, 2022;Orderud & Kelman, 2011). Additionally, Norway is a global outlier when it comes to non-fossil cars. In 2020, it was ranked number one in the world with about 18.1% of its cars in use being electric; well ahead of runners-up Iceland (5.5%), Sweden (3.7%), and the Netherlands (3.2%). 2 This factor is double edge. It may potentially decrease the general salience of a fuel tax since it applies to fewer people and there are viable alternatives to fossil cars. But it may also increase the divisiveness of such a tax: higher educated urbanites are more likely to be owners of electric vehicles whilst fossil car drivers are predominant in rural areas and among the less educated (Kester et al., 2020;Sovacool et al., 2018).

Multifactorial survey experiment set-up
To test our hypotheses, we constructed a multifactorial survey experiment. Technically, this design is best described as a single-profile multifactorial vignette experiment, which is probably the most widely used factorial survey design in the social sciences (Hainmueller et al., 2015(Hainmueller et al., , p. 2396. It took the form of a web-based survey. We informed respondents that we would like to know what they think about fuel taxes. We indicated that an increase in the existing fuel tax (for petrol and diesel) can lead to people driving less with fossil cars. This would in turn reduce CO 2 emissions and help reduce climate change. We subjected respondents to three different attributes. We asked respondents to imagine a tax increase which would make fuel more expensive by a certain percentage (levels = 5%, 20%, 35%). This corresponds to Attribute 1 to evaluate different thresholds of willingness to pay (hypothesis 1). We indicated that the income generated through this increase will be used for different purposes (levels = general expenditures, green projects, economic redistribution). This corresponds to Attribute 2 to evaluate different revenue recycling schemes (hypothesis 2). We specified that this tax increase is a consequence of different commitments (levels = international commitments, national commitments, personal responsibility as citizens). This corresponds to Attribute 3 to evaluate different framing effects of the origin of the policy (hypothesis 3). We rounded off the experiment by asking respondents to what extent they would, based on the presented information, oppose or support such a fuel tax increase if it were to be proposed. Respondents then situated themselves on a 7-point scale ranging from 'oppose very strongly' to 'support very strongly' (0-6). This is the study's dependent variable.
Each level (or scenario) within each attribute was exogenously and randomly ascribed to participants (see supplementary material OSM-2.A and OSM-2.B). We provide the text of the experiment in the supplementary material (English translation and Norwegian original, see OSM-1). There, we also provide a more detailed discussion of the three different levels for each attribute, how they relate to the French and Norwegian cases, and how the thresholds and alternatives were chosen (see OSM-2).

Implementation and methods
The survey experiment was implemented through the Norwegian Citizen Panel (NCP) in its wave 18 (2020). This panel is a general population web panel established for academic purposes. The participants have been recruited via random sampling from the official national population registry. 1,267 respondents completed the survey experiment. In the analysis, we use the weights provided by the NCP to enhance the sample's representativeness and the experiment's external validity. Details regarding the technical aspects of the survey, the data collection wave, representativity, and weighing are discussed in the wave 18 methodology report of the NCP (available open acces: Skjervheim et al., 2020).
Yellow vest profiles are defined as citizens with low educational attainments and rural residence (following Corre, 2018;de Comarmond, 2018;Vernet and de Saint Sauveur, 2019). In practical terms, this means citizens who have only completed secondary education or less and who, in addition, live in a rural area understood as a sparsely inhabited area, a village, or a small/medium-sized town. This represents about 41.6% of respondents. In contrast, the urban élite profile is defined as metropolitan-dwellers with high educational attainments (following Martin & Islar, 2021, p. 603). This means citizens who have higher education and who, in addition, live in urban areas understood as a city, a city suburb/outskirt, or a small/ medium-sized town. 3 This represents about 27.1% of respondents. We provide further details in the supplementary material (OSM-3).
We analyse the data by estimating a series of OLS models 4 and plotting the predicted values. We plot these predicted values with 95% and 84% confidence intervals. The latter is recommended by Julious (2004) for the comparison of point estimates within experimental settings where non-overlapping 84% confidence intervals indicate that differences between estimates are significant at the 0.05 level (see also Cumming (2009, p. 206) and Greenland et al. (2016, p. 344)). In Figure 2 we report predicted values for the experiment's different attributes and levels (see numerical values in OSM-5). In Figure 3, we plot predicted values for three-way interaction terms between the different combinations of attributes and levels resulting in 27 unique policy mixes (see . Finally, in Figure 4 we report predicted values from two-way interaction effects between the various attributes and levels, on the one hand, and the yellow vest-urban élite divide, on the other (see OSM-7).

Aggregate effects and policy cocktails
Norwegians are mildly against increasing fuel taxation to mitigate climate change. As shown in Figure 1, about 46.9% of respondents would oppose such a tax (from very strongly to somewhat). 40% would support it (same scale). 13.1% would be indifferent to it (neither/nor). The overall mood is one of slight opposition. This is because support is mostly tepid ('somewhat support', coded 4), whilst opposition is more evenly spread across the three categories ('somewhat', 'strongly', 'very strongly', coded 2-0) resulting in a sample mean of 2.78. This relative negativity towards climate taxes is common when individual costs are specified in understandable terms, such as an explicit price increase (Drews & van den Bergh, 2016;.
When looking at aggregate patterns, reported in Figure 2, individual costs come across as the most important factor. Respondents are mildly in favor of Figure 2. Support for increasing fuel taxation: average predicted effects.
Notes: Predicted values of support with 95% and 84% percent confidence intervals. 84% confidence intervals are recommended by Julious (2004) for the comparison of point estimates within experimental settings (see also Cumming (2009, p. 206) and Greenland et al. (2016, p. 344)). 7-point scale where 0 is 'oppose very strongly' and 6 is 'support very strongly'. N = 1,267. Vertical dashed line = grand mean. Vertical solid line = indifference. Numerical values reported in the supplementary material (OSM-5) together with a full x-axis version of the figure (OSM-8).
a carbon tax rendering fuel 5% more expensive (b = 3.23). However, if the tax implies an increase of 20% or 35%, then support morphs into opposition. Levels of opposition are similar in both scenarios (b = 2.56 and 2.51). This suggests that the relationship between individual costs and willingness to pay is not strictly linear but probably has a breakpoint somewhere between 5% and 20%.
And indeed, this corresponds to the French experience. The French carbon tax initially remained under the radar as oil was cheap in 2014-5. As the price of fuel jumped, due to higher market prices and the scaling-up of the carbon tax, so did discontent. According to Salies (2019, p. 1), the '25% increase in car fuel prices' is key to understanding the uprise of the yellow vests. It is no happenstance that the experiment's 5% increase generates mild support, whilst a 20% or 35% increase generates opposition. Individuals are relatively unmoved by a fuel tax aimed at mitigating climate changeas long as it is not too costly.
The other attributes have weak effects. Lack of earmarking generates opposition which is softened by recycling towards further environmental measures or redistribution to lower-income households. However, these effects are not significantly different from one another in the aggregate analysis. Results are even weaker regarding the source of the tax. Whether it is a consequence of international commitments, national commitments, or of citizens' personal responsibility is of little consequence when considered at the aggregate level.
An advantage of survey experiments is to observe how attributes and their varying levels interact among themselves. The resulting 27 unique policy mixes are reported in Figure 3. These cocktails generate various levels of support and opposition. The most popular policy consists of the 5%-environmental recycling-personal responsibility mix (b = 3.85, CI(84) = 3.45-4.25). The least popular consists of the 20%-general expenditure-personal responsibility mix (b = 1.73, CI(84) = 1.34-2.11). Hence, we would advise policymakers to only invoke personal responsibility together with lower price increases. Indeed, the top-5 scenarios include three instances of personal responsibility framing, but always in conjunction with the 5% increase. Coupled with 20 or 35% increases, personal responsibility risks backfiring: the bottom-5 scenarios have three instances of personal responsibility, each time with 20% or 35% increases.
Unsurprisingly, the 6 top combinations all include the lowest price increase. The top-10, however, also includes instances of the 35% and 20% price increase. This top-10 ranges from 3.85 to 2.96, representing mild forms of support verging on indifference, and it generates significantly more support than the bottom-4. The bottom-4 ranges from 1.97 to 1.73, hence between moderate and strong-ish opposition. Surprisingly, the bottom-4 includes only one 35% scenarios.
Overall, willingness to pay certainly comes across as a crucial factor. The top-6 are exclusively 5% scenarios whilst none of the bottom-5 are.
However, willingness to pay is not everything. The most popular 35% configuration (ranked 7th) is significantly different from the bottom-5 which also includes two 35% scenarios. We draw from this that, if policymakers want to implement a 35% price hike, they would be well advised to: combine it with redistributive measures helping low-income households and frame it as a consequence of international commitments (b = 3.18, CI (84) = 2.71-3.66). This will generate significantly more support than other 35% scenarios, such as funding further environmental measures and drawing on personal responsibility, or funding general public expenditures and responding to national commitments (b = 2.18 and 1.97; CI(84) = 1.71-2.65 and 1.47-2.48). Hence, revenue recycling and origin framing significantly affect levels of support for a 35% price hike. These results echo other findings on how pairing different policy attributes can affect support, including how international commitments increase the acceptance for greater emission Notes: Predicted values of support for 27 unique policy cocktails. 95% and 84% percent confidence intervals. 84% confidence intervals are recommended by Julious (2004) for the comparison of point estimates within experimental settings (see also Cumming (2009, p. 206) and Greenland et al. (2016, p. 344 cuts, even when these come at a greater cost to individual citizens (Tingley & Tomz, 2020, p. 1143, 1147.

A bridge too far? Yellow vests and urban elites
Successful policies do not need to benefit from the support of a majority. Indifference is sufficient. But they do need to avoid alienating a sizeable subset of the population, especially if this subset self-identifies as a group and expresses its discontent. We draw on the two profiles of the yellow vests and urban élites to analyse these dynamics. We seek to understand what affects support and opposition for fuel taxation in these two stylized camps, but also which policy characteristics reduce or widen the gap between these groups and the rest of society.
Yellow vestssupport gaps can be closed Yellow vest profiles are crucially different from the rest of the population when it comes to individual costs. Even a small price increase (5%) generates some opposition. Greater increases magnify this opposition. The rest of the population is much less averse to heavier taxation: a small increase is met with support, larger increases are met with indifference. However, as highlighted in Figure 4 (left panel), whilst any tax increase opens a gap, this gap is smaller when the tax is low. The yellow vests are simply much more sensitive to price increases than the rest of the population. Their opposition significantly grows when fuel costs rise by 20% or 35%. In other words, higher Notes: Predicted values of support for yellow vests (left panel, grey), urban élites (right panel, grey), and the rest (black). 95% and 84% percent confidence intervals. 84% confidence intervals are recommended by Julious (2004) for the comparison of point estimates within experimental settings (see also Cumming (2009, p. 206) and Greenland et al. (2016, p. 344)). Oppose-support 7-point scale ( levels of taxation generate a significant support gap between the yellow vests and the rest. Willingness to pay for climate policies quickly vanishes in this segment of the population characterized by remoteness, a limited economy, and few perspectives of higher earnings through continued education (Corre, 2018;de Comarmond, 2018;Vernet and de Saint Sauveur, 2019).
Regarding revenue recycling, the yellow vests are again significantly different from the rest of the population. None of the revenue recycling schemes in themselves do much to convince them of the acceptability of a fuel tax. Meanwhile, the rest of the population is either indifferent (general public spending and redistribution) or supportive (green earmarking). Two policy options are nonetheless remarkable. First, the gap between yellow vests and the rest is widest when it comes to green earmarking. This is the policy characteristic which generates the largest support gap in terms of the distance between confidence intervals (Δ in CIs(84) = 0.874). Second, one recycling option, however, does close this gap: redistribution to lowincome households. This policy is met by indifference by the rest of the population but it does bridge the gap with the yellow vests, as this is their least disliked scenario. Although it comes across as a lowest-common denominator compromise, it does generate a convergence of citizen preferences and avoids activating polarization between the yellow vests and the rest. This result is very much in line with conclusions put forward by Klenert et al. (2018) who highlight the importance of compensation to those especially hurt by higher carbon prices, 'such as rural or low-income households' (p. 675).
Interestingly, the yellow vests are not significantly different from the rest of the population when the tax is framed as emanating from international commitments. These yellow vests may suffer from globalization in terms of job prospects or perceived threats to their cultural environment (Lockwood, 2018;Salies, 2019, p. 2). However, invoking the UN's Paris Agreement as a legitimizing frame does not generate a divide between them and the rest. This is very different from the two other framings, which do open a substantial support gap.
Referring to national legal frames and to personal responsibility generate the second and third largest support gap between the yellow vests and the rest (Δ in CIs(84) = 0.775 and 0.728, respectively). These results are very much in line with observations derived from the French case. The aversion to national frames reflects the yellow vests' dissatisfaction with domestic democratic processes and a feeling that national institutions do not represent their interests well enough (Kipfer, 2019). They are disillusioned by the political class, its perceived collusion with business élites and the upper-class establishment, which together maintain a status quo which does not reflect their policy preferences (Driscoll, 2021;Mehleb et al., 2021). The yellow vests are also different from their peers when it comes to personal responsibility. Justifying a tax increase through 'personal responsibility' simply does not cut it. It triggers significant resentment towards the policy. Again, this is very much in line with existing empirical insights. This is a segment of the population which feels that a disproportionate burden of the government's tax reforms is falling on the working lower-middle classes, especially in rural and peripheral areas (Mehleb et al., 2021). They feel they are already contributing enough, maybe more than their fair share in light of their financial resources and of big business emissions (Driscoll, 2021, pp. 13-14).
In sum, the largest support gaps between the yellow vests and the rest emerge when green earmarking, national commitments, and personal responsibility are invoked. These gaps are bridged (and the cleavage deactivated) when there is economic redistribution to lower-income households and when international commitments are invoked. The support gap is also noticeably smaller in the case of a low tax increase.

Urban elitesconsistent outliers
Urban élites have a remarkably distinct and stable set of preferences. As summarized in Figure 4 (right-panel), they are systematically different from the rest of the population. In that sense, urban élites are more different from the rest of the population than the yellow vests are. Their estimates never significantly overlap. In this study, it is the urban élites who are the outliers, not the yellow vests.
These élites are relatively insensitive to financial costs. The 5%, 20%, and 35% price hikes leave them unmoved. They are always supportive. This may be a consequence of a higher willingness to pay for climate mitigation, of greater financial capacity tout court, or of this segment of the population being relatively unaffected by the tax due to access to alternatives such as collective transport and non-fossil cars (Kester et al., 2020;Sovacool et al., 2018). When it comes to revenue recycling, they are somewhat sensitive to green earmarking. This significantly boosts their support (compared to general public expenditures). This is also where the support gap is widest compared to the rest of the population (Δ in CIs(84) = 0.897). Interestingly, and as was the case for the yellow vests, the preference gap with the rest of the population is smallest in the context of redistribution to low-income households and largest in the context of green earmarking. Finally, urban élites are as supportive of a fuel tax increase when framed as originating from international commitments, national commitments, or citizens' personal responsibility. One nonetheless remarks that the gap between urban élites and the rest of the population is widest in the scenario invoking personal responsibility (Δ in CIs(84) = 0.823).
Overall, urban élites are clearly in favor of fuel taxation. Similarly to the yellow vests, the gap between urban élites and the rest of the population widens with higher price increases, green earmarking, and when personal responsibility is invoked. It decreases in the case of a 5% hike, redistribution to low-income households, and national commitments.

Conclusion
The results from this study are clear. If policymakers wish to avoid activating cleavages along the yellow vest-urban élite divide, then they should avoid high fuel taxes, green earmarking, and invoking citizens' personal responsibility. If gaps are to be bridged and cleavages left dormant, then a lower tax hike, economic redistribution, and invoking international commitments should be prioritized. Obviously, too low carbon taxes will be ineffective. To act as a deterrent, carbon taxes need to escalate, meaning higher prices for consumers. We argue that support for a 35% increase can be harnessed if it is combined with economic redistributive measures and international commitments (see 7th most popular cocktail in Figure 3). It is no coincidence that redistribution and international commitments are also the attributes that bridge the gap between the yellow vests and the rest. In other words, yellow vest activation is not a fatality. Support gaps can be bridged, and opposition to high climate taxes can be mitigated by different policy mixes.
More generally, we argue that climate policies run the risk of triggering social unrest if they do not cater to the preferences of some of the polity's more neglected citizens. These are the peripheral, less qualified folk, who often tend to be the losers of globalization compared to high-educated white-collar city-dwellers (Salies, 2019, p. 2;Wanvik & Haarstad, 2021). In advanced democracies, two old-fashioned cleavages seem to have been reactivated by the climate and globalization challenges. On the one hand, the rural-urban cleavage and, on the other, the education cleavage. These two cleavages are important to understand many political outcomes such as voting patterns, attitudes towards immigration, or towards European integration. They also seem increasingly important to understand reactions towards climate policies.
In this sense, the present study very much aligns with Aklin and Mildenberger (2020). These authors claim that a key challenge in climate policy is not overcoming collective action dilemmas, but rather those generated by distributive conflicts. They outline that, as climate policies create winners and losers, distributive conflicts will arise among them. We argue that the cleavages these conflicts reawaken are crucial to understand the acceptance of climate policies (see also Aasen & Saelen, 2022;Wanvik & Haarstad, 2021). If left unaddressed, gaps will widen within society, leading not only to policy failure but also to social unrest. Climate policies which neglect questions of rising inequalities and social justice will face acceptance challenges and may well activate cleavages eroding the very democratic sustainability of our polities.