Paying Enough Taxes Already? Testing the Acceptability of Carbon Taxes with Survey Data

This paper analyzes the drivers of carbon taxes acceptability with survey data and semi-experimental techniques. Based on a sample of more than 300 individuals, it assesses the effect on acceptability of specific policy designs and individuals' perceptions of carbon taxes advantages and disadvantages. We find that the lack of perception of primary and ancillary benefits is one of the main barriers to the acceptability of carbon taxes. In addition, policy design matters for acceptability and in particular earmarking fiscal revenues for environmental purposes can lead to larger support. We also find an effect of labeling, comparing the wording “climate contribution” with “carbon tax”. We argue that proper policy design coupled with effective communication on the effects of carbon taxes may lead to a substantial improvement in acceptability.


Introduction
In recent decades, international negotiations have aimed at stabilizing the concentration of greenhouse gases at levels that would prevent dangerous interferences with the climate system. However, there is increasing evidence showing that current mitigation eorts are by large not enough (UNEP 2013;IPCC 2014). This result comes as no surprise.
Even though economists assessed their theoretical cost-eectiveness long time ago (cf. e.g. Baumol and Oates 1971), the implementation of powerful policy instruments to reduce greenhouse gas emissions such as carbon taxes is a rather recent phenomenon (Baranzini and Carattini 2014). A recent strand of literature has started exploring the public acceptability of carbon taxes, pointing to a series of important obstacles such as distributional impacts on low-income households and fear of competitiveness eects (cf. Baranzini et al. 2000;Zhang and Baranzini 2004). In most developed countries carbon taxes are indeed at least slightly regressive (see e.g. Roca and Serrano 2007;Brännlund and Ghalwash 2008;Sterner 2011) and the local co-benets benetting mostly the poorest households are usually given a lower weight. Despite the recent empirical evidence points to rather small competitiveness eects (Mathys and de Melo 2011), potential adverse eects on employment and competitiveness represented a real concern when the rst carbon taxes in Scandinavian countries were designed (cf. e.g. Godal and Holtsmark 2001;Bruvoll and Larsen 2004) and when similar schemes were turned down elsewhere (cf. e.g. Thalmann 2004).
More recently, the literature has devoted increasing attention towards the perceived impact of carbon taxes on the environment. While economists tend to take the benecial environmental eects as granted, the eectiveness of carbon taxes does not seem to be internalized by the general public. Based on qualitative assessments, Dresner et al. (2006) rst raised the issue of perceived environmental ineectiveness: the general public tends to miss the incentive eect of carbon taxes, thus expecting tax revenues to be earmarked for environmental purposes. When this is not the case, most people feel that carbon taxes are just a pretext to raise scal revenues. When tax revenues are earmarked for other purposes, the general public is generally disconcerted about the possibility of using the revenues of an environmental tax for something unrelated to the environment. Saelen and Kallbekken (2011) dene this problem as issue-linkage. The stylized fact of Dresner et al. (2006) is supported by the quantitative evidence of Kallbekken and Saelen (2011) and Saelen and Kallbekken (2011), both nding a negative pattern between perceived environmental ineectiveness and stated support for carbon and energy taxes, and is consistent with real voting behavior as analyzed by Thalmann (2004). Pigouvian taxes are thus perceived at the same time as coercive and ineective (Steg et al. 2006).
This may lead Pigouvian taxes to be more popular if not labelled as such: in the lab experiment of , a fee is preferred to an equivalent instrument called tax.
We improve this recent literature by using survey data and semi-experimental techniques to test the eect of several policy variables and perceptions on acceptability. First, we conrm that perceived environmental ineectiveness is one of the main barriers to the acceptability of carbon taxes. Our original approach shows that not only the expectation of main environmental eects aects acceptability, but also the perception of potential co-benets does. Second, we show that competitiveness and distributional eects may not have a role as determinants of acceptability. Third, we show that acceptability increases substantially with earmarking, in particular for environmental purposes. This is especially true among those individuals that tend to distrust the government. However, earmarking does not act as a substitute for perceived eectiveness. Even when revenues are earmarked, perceived eectiveness remains related to higher acceptability. Fourth, contrasting the labels carbon tax and climate contribution, we show that labelling can spur acceptability also in the street and not only in the lab. Climate contribution may sound as an appeal to the public good, recalling to the general public the urgency of climate change mitigation.
We use the Canton of Geneva, Switzerland, as eld and interview more than 300 individuals between December 2012 and January 2013. The Swiss context may be particularly salient to simulate voting behavior, since Swiss people are used to express their opinions in poll and ballots, but we consider all of our ndings of general interest. The lack of popularity of carbon taxes has indeed limited its implementation in virtually all political contexts in developed countries.
2 Survey design and data description

Hypotheses
We formulate a series of main hypotheses to be tested with the econometric model.
We expect positive perceived impacts of carbon taxes to be positively associated to carbon tax acceptability. Positive impacts consist of carbon emissions abatements (i.e. environmental eectiveness) and improvements in local outcomes such as air quality, health and road externalities. As shown by Dresner et al. (2006) and Bristow et al. (2010), the acceptability of a given carbon tax design can be inuenced by how agents perceive the instrument as eective (see also Brouwer et al. 2008 on a carbon travel tax).
We are not aware of previous studies examining the link between perceived co-benets and acceptability, but Longo et al. (2012) suggest for instance that the willingness-to-pay for climate change mitigation is about 50%-70% higher when co-benets are considered by respondents.
Concerning drawbacks, distributional concerns aect acceptability in Thalmann (2004), in which the probability of a yes-vote for a green proposal is substantially lower when the proposal implies a clear increase in inequalities. Bristow et al. (2010) nd a marked preference for a carbon credit up to 4 tons of CO 2 per capita to reduce distributional effects. Inequality aversion is also present in Brannlund andPersson (2012).
Acceptability is supposed to increase with earmarking and we expect earmarking for environmental purposes to have the largest impact (cf. e.g. Dresner et al. 2006;Steg et al. 2006;Kallbekken and Aasen 2010;Saelen and Kallbekken 2011). When earmarking is not specied, we expect people distrusting the government to be less likely to approve carbon taxes. Since most people fail to understand the dierence between Pigouvian and Ramsey taxes and thus perceive environmental taxes as a mean to raise more revenues , they may be willing to give up some of their income only if the use of revenues is clearly made explicit. Because the tax is perceived as environmentally ineective, the only way to impact the environment is by earmarking the scal revenues for the environment.
Based on the literature, we forecast dierent acceptability depending on the labeling of the tax. According to Steg et al. (2006), environmental taxes are perceived by most of the general public as penalties, i.e. coercive measures imposing a change in behavior.
Higher acceptability with dierent labeling is found not only in , but also in the online choice experiment of Brannlund and Persson (2012), in which a policy called tax is opposed to another policy simply framed as other. To make the contrast even more manifest, in our survey we opt for the labeling climate contribution as opposed to tax. We expect climate contribution to sound as an appeal to the public good, which may crowd in motivation as predicted by persuasive advertising models as in Becker and Murphy (1993) and Nyborg et al. (2006)

Economic context
From a climate policy perspective, the context of Switzerland is of particular interest.
Switzerland lobbies beside the European Union in the post-Kyoto negotiations, urging for ambitious agreements. In addition, in the aftermath of the Fukushima accident, Switzerland decided to start to phase-out nuclear energy. Since currently 40% of electricity is from nuclear sources, Switzerland has very low carbon emissions from electricity in international comparison and thus little room for maneuver to replace nuclear energy without increasing carbon emissions. This implies the need for substantial reductions in consumption (Baranzini et al. 2013).
Given the public rejection of three energy-tax proposals in 2000, Switzerland adopted a climate strategy based on voluntary agreements and only in 2008 introduced a carbon tax limited to heating fuels (cf. Thalmann 2004; Baranzini et al. 2004). After that Switzerland missed in 2012 its objective of CO 2 emissions reduction, the carbon tax rate was revised upward, but an extension of the tax base to all fuels may be desirable to reach

Qualitative survey
To dene the questions of the quantitative survey, we rst administer through semistructured interviews a qualitative survey to a small and unrepresentative sample of about 40 adults living in the Canton of Geneva. We report the main ndings. First, private actions (e.g. improvements in energy eciency) are preferred to public intervention to curb energy consumption and emissions. Second, when interviewers introduce explicitly the role of the public sector, the general opinion is to limit its intervention to communication and education (i.e. suasion). Market instruments are mentioned only by few, and subsidies (e.g. for public transportation) are by large preferred to taxes, as in Cherry et al. (2012). Third, interviewers face some resistance when they propose the implementation of a generalized carbon tax, related to distrust in the government and a presumed ineectiveness in changing behavior. It follows that when they ask how tax revenues should be used, most respondents suggest to keep them in the environmental domain. Fourth, social cushioning for low-income households is regarded as important, but it seems that the way of nancing it should make abstraction of environmental tax revenues. It also asks whether respondents generally trust their government (93%) and if they are aware of the existing CO 2 tax on heating fuels (only 40%). The lack of awareness concerning current taxation may be explained by the limited salience of both taxation and lump-sum refunds to households through reduced healthcare bills. Respondents are also asked what role they would attribute to the public sector to spur energy conservation. Only a tiny fraction of individuals (2%) contends that there is no need for energy conservation at all. Similarly, only 3% believe that energy consumption does not need to be regulated. That is, the large majority expects the government to intervene to curb energy consumption. However, as for the qualitative survey, market instruments do not represent the favored tool. Preferences go rather to informational campaigns raising awareness and to policies funding public transportation. However, already at this stage 35% of respondents believe that the Swiss government should intervene with a broader carbon tax to lower energy consumption.

Policy variables
In what follows we present the main policy variables included in the survey (cf. Table   B.3 for the full descriptive statistics).
Carbon taxes' environmental eectiveness In the survey we introduce a hypothetical carbon tax (or climate contribution) with a tax rate of 120 CHF per ton of CO 2 , implying a price increase of gasoline of about 15% and of heating fuel of about 30%.
The majority of the sample thinks that the tax would lead to a reduction of their level of energy consumption, but a non-negligible proportion of respondents (37%) expects no change in behavior. A small minority (7%) even expects larger consumption. These either represent protest answers or suggest that worries of a possible motivational crowding-out may be justied. That is, economic instruments may turn out to have counterproductive eects on intrinsically-motivated agents (Deci and Ryan 1985), if individuals that already provide large eorts for a given public good in absence of any economic incentive feel frustrated for being taxed despite their eorts (no behavior is good enough not to be penalized, Goeschl and Perino 2012) or feel less responsible toward the provision of the public good as they think that since I pay, I can consume and thus pollute (Bazin et al. 2004).
Next, the questionnaire enlarges the focus and asks whether people expect the tax to be eective, i.e. if it would lead to a decrease in the energy consumption and greenhouse gas emissions of Switzerland. A short majority (52%) expects the tax not to be eective.
Co-benets Respondents are asked to spontaneously mention a list of ancillary benets of carbon taxes, if any, without having access to the list of potential answers in the questionnaire, to avoid to inuence their opinions. About half of the sample (56%) expects better air quality as an ancillary benet from carbon taxes. The proportion is lower for congestion issues (27%), health improvements (42%) and road accidents (18%). People thus exhibit a relatively low awareness of co-benets, which may need to be targeted through improved communication, along with primary benets.
Disadvantages In line with the literature, regressive eects seem to represent a real issue for the people in this sample. However, we nd that only a minority (25%) is concerned about distributional eects on rural households, which are also expected to be particularly aected, given the limited possibilities of substitution between private and public transportation in the countryside. This may be specic to the context of Geneva, whose countryside hosts many high-income households attracted by calmness, green spaces and possibly interesting scal conditions. Instead, only relatively small proportions of respondents are concerned about employment and competitiveness eects.
In particular, the number of people concerned about employment issues (11%) is only slightly larger than those concerned about their own job (5%, correlation of 0.46). We note that at the time of the survey the level of unemployment in the Canton of Geneva (in Switzerland) was about 5.5% (3%).
Finally, one of the main perceived drawbacks is represented by the private cost of climate change mitigation, which is strictly positive for all citizens absent any earmarking.
The most generalized fear for respondents in this sample (67%) is to be constrained to reduce the overall level of consumption due to the higher energy prices. Interestingly, expecting lower purchasing power does not necessarily imply a loss of comfort. It appears that people in the sample feel that they could live comfortably even with less purchasing power, although they may not like it. Further data inspections show that expecting losses of purchasing power is negatively correlated with the highest income category and expecting less comfort is positively associated with the lowest income category. Acceptability conditional on recycling We retest the level of acceptability conditional on earmarking and revenue recycling. In detail, the survey asks whether the respondent would accept a CT/CC if revenues were to be recycled according to her preferred recycling option. Yes-votes reach now 64%, i.e. about 15% more than without earmarking. This result is consistent with the literature. This level of support may however be misleading since obtained by assuming that the preferred recycling options of each individual can be implemented simultaneously. We also stress the relative importance of the residual 36% of respondents for which earmarking revenues is not sucient to have them accepting the CT/CC, at least at the tax rate proposed by the questionnaire.
Econometric analyses are performed in section 3.
Tax rate So far, the questionnaire refers to a CT/CC with a tax rate of 120 CHF/tCO 2 , causing with full pass-through an increase of about 15% (30%) in the price of gasoline (heating fuels). The questionnaire thus asks to the respondents what would be their highest acceptable CT/CC tax rate, expressed in terms of energy price increases. The aim is to measure the intensity of acceptability. The distribution of answers is bounded by the minimum and maximum possible answers given in the questionnaire (0% and 30%) and centered in the 5%-10% interval (using interval means, the average is 7% and the median 7.5%). Hence, albeit 64% of the sample supports the tax in the previous question, when asked about dening themselves the tax rate, respondents tend to indicate more moderate energy price increases than what proposed by the survey. The two results are not necessarily in contradiction. Voters are indeed supposed to select the option that is closer to their preferences. In this respect, the Norwegian choice experiment of Saelen and Kallbekken (2011) shows that respondents would prefer to decrease the current level of environmental taxes, absent earmarking for environmental purposes. In a similar spirit, Godal and Holtsmark (2001) suggest to always start with a low tax rate and increase it regularly once the policy is in place.
3 Econometric analysis, results and discussion 3.1 Carbon tax acceptability In this section we analyze the determinants of carbon tax acceptability when earmarking is not specied. Since the outcome variable is binary, we apply a probit estimation strategy. We use as controls both socioeconomic characteristics and the policy perceptions Estimations results are reported in Table 1 Some other socioeconomic variables are clearly not signicant. We do not nd for instance any statistically signicant eect for age (both as a continuous variable or using specic groups such as e.g. youth, retired people), gender and political positioning.
Column (1) shows that the number of cars held by respondents is negatively and signicantly linked with the probability of accepting a carbon tax, whereas green membership and education have a positive impact. That is, as in Thalmann (2004), it is not only car ownership, but the number of vehicles that is related to political behavior.
Ecologists are of course expected to be relatively more in favor of climate policy tightening, assuming no motivational crowding-out. The eect of education is also as predicted.
Since education is a long-run investment, educated people may possess a lower discount rate than the average citizen, according to Bornstein and Lanz (2008 In column (2) we introduce variables on environmental attitudes. This allows us to test the hypotheses H3 and H5. We nd that being a priori in favor of a carbon tax has a very large eect on acceptability. This conrms H5 and provides evidence for the internal validity in the questionnaire. As expected, trust in the government engenders higher acceptability, providing rst evidence in favor of H3.
We also nd that being highly concerned about the climate and paying no attention to energy consumption have a signicant and economically meaningful impact on acceptability, consistently with Kallbekken and Saelen (2011). Since the eect of education disappears once introduced the role of information and concern, we may suggest that its eect is rather driven by information asymmetry than a dierence in discounting. Expectation of cooperation in energy conservation's eorts from fellow-citizens is associated with a positive eect on acceptability. We relate this nding with the recent works in Column (3) allows us to further test our main hypotheses. The eects related with environmental eectiveness and perceived co-benets are striking and partly conrm H1. If the tax is expected to be eective in reducing emissions, acceptability rises by about 30%. The impact of co-benets on acceptability has a similar magnitude, since this variable ranges from 0 to 8. Hence, our ndings strongly support the literature on the perceived eectiveness of carbon taxes and provide a quantitative estimate of the magnitude of its linkage with acceptability, which is shown to depend also on perceived co-benets 5 .
Neglecting co-benets would clearly imply an overestimation of the net policy costs, since most studies providing monetary estimates of co-benets suggest that they are relatively conspicuous compared to mitigation costs, also in the case of Switzerland and other developed countries (cf. e.g. OECD 2014). In fact, co-benets are in the order of several tens of dollars per ton of CO 2 and may well exceed abatement costs (Baranzini and Carattini 2014). According to Pittel and Rübbelke (2008), co-benets may be sufciently large to justify cooperation in international negotiations and lead to binding international agreements, of course provided that their existence (and magnitude) is recognized. That is, co-benets may be a game changer in the political economy of climate change mitigation, if fully internalized in people's beliefs.
Regarding policy drawbacks, all included variables have the expected negative sign, but only loss of purchasing power is signicant. Being one of the 67% of the sample 5 The coecient for crowding-out of intrinsic motivation or protest answers does not attain statistical signicance. This may suggest that what we face is indeed some motivational crowding-out rather than simply protest, although with no impact on acceptability. arming that carbon taxes are an issue for purchasing power is linked with about 15% lower probability of accepting the instrument. The concern of losing purchasing power makes sense in particular given that at this stage the use of tax revenues is not specied.
However, this concern may vanish over time after that the carbon taxes are implemented, since the real eect of loss in consumption on well-being may be lower than actually perceived, as income relative to others may be rather unchanged (Gowdy 2008, see also Howarth 2006).
Regressive impacts do not seem to matter in this context. Distributional eects are an issue for an important number of individuals as indicated in the descriptive statistics, but fail to signicantly impact acceptability. This result dier from those of most of the literature but provides quantitative evidence for one of the nding from the focus group of Kallbekken and Aasen (2010), which stress that respondents seem to be concerned by distributional eects, but not enough to pretend environmental taxes not to be regressive.
Competitiveness and employment eects are clearly non-signicant. We see four possible explanations for their non-signicance. First, individuals may not be concerned about competitiveness issues, consumers having dierent interests than rms, and about unemployment, since they may have very small empathy for potential jobless and perhaps limited fear of how rising levels of unemployment could aect their own situation.
However, this may be in contradiction with the evidence based on voting behavior of Thalmann (2004), in which concern about employment issues contributed to the rejection by the Swiss population of three energy tax proposals in spite of an unemployment rate below 2%. Second, individuals may not expect competitiveness eects to be suciently large to become a real problem. This may make sense in the light of the modeling exercise of Sceia et al. (2012), which nd very limited terms-of-trade eects for Switzerland when simulating the impact of unilateral moves towards more stringent climate policy. Third, respondents may expect Swiss climate policy to be part of a concerted move undertaken with other countries, e.g. under the umbrella of a renewed Kyoto-like agreement. In such scenario, terms-of-trade eects as modeled by Sceia et al. (2012) become positive. Fourth, the very low concern for employment and competitiveness eects may also be due to the low prole of corporate interest groups at the time of the survey.
Indeed, once approaching important votes, the latter tend to employ massive lobbying eorts to have their vested interests internalized by the public, leading the industry ight argument to gain a very important weight in the political discourse (Spash and Lo 2012).
This interpretation would call for green lobbying to oppose the industry ight argument and preserve the current outcome (see Dietz et al. 2012).
Altogether, this evidence may question the rationale for the large exemptions and privileges given to some industries by the carbon schemes of some Scandinavian countries or Australia, which have eventually watered down the environmental impact of the tax itself (see Lin and Li 2011;Baranzini and Carattini 2014). Partially rejecting H1, Table   1 suggests that the popularity of carbon taxes may not necessarily imply a trade-o between environmental, distributional and competitiveness eects.
Labeling is not signicant at this stage. Regarding the relative performance of the three specications of Table 1, we see that the goodness-of-t increases as more variables are added, conrming that policy perception does matter for acceptability, beside individual characteristics.

Impact of earmarking on carbon tax acceptability
In this section, we focus on the individuals that do not accept the policy proposal in subsection 3.1 and examine whether their choice changes conditional on the implementation of their preferred option of revenue recycling. The variable to be explained takes value 1 for those changing opinion in favor of the CT/CC and 0 otherwise. We then rely again on a probit model, conditional on choosing 0 in the rst acceptability question.
Consistently with the descriptive evidence, we select tax rebates to households and rms as the reference case.
The estimation reported in Table 2 includes variables for revenue recycling and labeling, now signicant. The literature gives no priors on the potential impact of socioeconomic characteristics on acceptability conditional on the preferred use of tax revenues.
We nd that in general none is statistically signicant, except for a positive eect again of education (p-value of 0.098). Policy variables seem instead conrming our set of hypotheses. Trust in the government is associated with a negative sign and a fairly large coecient. This makes sense in the light of the positive coecient in Table 1. That is, trust in the government is positively associated with being in favor of the CT/CC regardless of how tax revenues are used, whereas respondents distrusting the government are relatively more likely to reject the rst proposal and potentially change opinion in Table 2, once revenues are earmarked. Hence, we can conrm hypothesis H3.
Compared to the rest of respondents, those suggesting that the government should address the issue of energy consumption with taxation do not have a higher probability to change their opinion with earmarking. This is consistent with the result of Table 1 and hypothesis H4. In contrast, those asking the government to better communicate the need and possibilities for energy conservation are, everything else equal, more likely to support the tax both unconditionally and conditionally on earmarking. This may point again to the complementarity between taxation and communication.
The eect of perceived eectiveness is in line with our hypothesis. Expecting the tax to work is linked to a positive eect on acceptability, also among those that rejected the rst CT/CC proposal. As suggested by the qualitative analysis of Kallbekken and Aasen (2010), being aware of how the incentive eect works does not necessarily imply no demand for earmarking. The marginal eect in Table 2 is still pretty large and implies that in this sub-sample the likelihood of voting yes once the use of revenue is dened is about 25% larger for those believing the CT/CC to work than for those that do not. On top of that, there seems to be again a positive eect of co-benets.
We observe that the probability of reconsidering the CT/CC is larger for those selecting recycling for environmental purposes than for those opting for tax rebates (the dummy of reference), everything else equal. The coecient for recycling through social cushioning (as dened by the respondent) is not statistically signicant. Therefore, it seems that earmarking for environmental purposes really matters for acceptability. Since we control for e.g. trust in the government and perceived eectiveness, we relate this demand for environmental recycling with the issue-linkage, i.e. the need for the public to see a straightforward and logical nexus between the tax and the use of revenues

Conclusion
Carbon taxes are an eective instrument for curbing greenhouse gas emissions, yet are seldom implemented (Baranzini and Carattini 2014). This paper uses survey data to assess drivers and barriers to public acceptability of carbon taxes. Albeit the political discourse generally focuses on the negative impacts on competitiveness and distributional eects (cf. e.g. Spash and Lo 2012), the data analyzed here indicate that individuals are more concerned by the environmental eectiveness of the tax. Indeed, we show that perceived environmental eectiveness and expectation of local co-benets are the main drivers of acceptability. Competitiveness eects are almost completely neglected, whereas distributional issues (in particular regarding poor households) seem to represent a real concern for the general public, but with little impact on acceptability.
According to our ndings, communicating both primary and ancillary benets of carbon taxes seems to be essential for improving acceptability. Along with earmarking, this could be very useful to reduce the opposition related to mistrust in the government and Ramsey-type tax aversion (see . In this respect, we nd that in terms of acceptability the best way of recycling the tax revenues is to give the priority to environmental spending. In the same vein, we also provide evidence that using a dierent label, viz. climate contribution rather than carbon tax, can be benecial in terms of acceptability.