Cost perceptions and the support for carbon pricing

Behringer, Jan; Endres, Lukas; Korsinnek, Maike Working Paper Cost perceptions and the support for carbon pricing IMK Working Paper, No. 226 Provided in Cooperation with: Macroeconomic Policy Institute (IMK) at the Hans Boeckler Foundation Suggested Citation: Behringer, Jan; Endres, Lukas; Korsinnek, Maike (2025) : Cost perceptions and the support for carbon pricing, IMK Working Paper, No. 226, Hans-Böckler-Stiftung, Institut für Makroökonomie und Konjunkturforschung (IMK), Düsseldorf This Version is available at: https://hdl.handle.net/10419/333495 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/legalcode WORKING PAPER No. 226 • August 2025 • Hans-Böckler-Stiftung COST PERCEPTIONS AND THE SUPPORT FOR CARBON PRICING Jan Behringer 1 , Lukas Endres 2 , Maike Korsinnek 3 ABSTRACT We examine how perceptions about the costs of carbon pricing affect policy acceptance. Using a representative sample of the German population, we conduct experiments that provide randomly selected respondents with personalized information about their costs at the current carbon price or a higher future price. Participants tend to overestimate their current costs and increase their carbon price acceptance when receiving cost information. In contrast, respondents underestimate future costs and reduce their support once they learn about actual costs. This underscores the importance of personalized information in fostering current support for carbon pricing, while cautioning against potential backlash as prices rise. ————————— 1 Macroeconomic Policy Institute (IMK). Email: [email protected] 2 Macroeconomic Policy Institute (IMK). Email: [email protected] 3 Macroeconomic Policy Institute (IMK). Email: [email protected] Cost perceptions and the support for carbon pricing * Jan Behringer Lukas Endres Maike Korsinnek August 25, 2025 Abstract We examine how perceptions about the costs of carbon pricing affect policy acceptance. Using a representative sample of the German population, we conduct experiments that provide randomly selected respondents with personalized information about their costs at the current carbon price or a higher future price. Participants tend to overestimate their current costs and increase their carbon price acceptance when receiving cost information. In contrast, respondents underestimate future costs and reduce their support once they learn about actual costs. This underscores the importance of personalized information in fostering current support for carbon pricing, while cautioning against potential backlash as prices rise. JEL Classification: D12, D83, H23, Q58 Keywords: Carbon pricing, policy acceptance, perceptions, experiment * Jan Behringer, Macroeconomic Policy Institute (IMK), Georg-Glock-Straße 18, 40474 D¨usseldorf, Germany. E-Mail: [email protected]. Lukas Endres, Macroeconomic Policy Institute (IMK) and the Institute for Socio-Economics, University of Duisburg-Essen, Lotharstraße 65, 47057 Duisburg, Germany. E-Mail: [email protected]. Maike Korsinnek, Macroeconomic Policy Institute (IMK) and the Institute of Economics, University of Bamberg, Feldkirchenstr. 21, 96045 Bamberg, Germany. E-Mail: [email protected]. We thank Elena Franko for excellent research assistance. This study was registered in the American Economic Association’s registry for randomized controlled trials under ID AEARCTR-0012808. 1. Introduction Limiting global warming requires effective climate mitigation policies. However, public opposition remains a major obstacle to implementing ambitious measures. Despite being widely endorsed by economists as a cost-effective tool for reducing emissions (Climate Leadership Council, 2019), carbon pricing has encountered particularly strong resistance. In earlier instances, public opposition - driven by concerns about rising consumer prices - has limited the scope of carbon pricing schemes and even led to their reversal (Douenne and Fabre, 2022; Crowley, 2021; Anderson et al., 2023). Increasing public acceptance has thus become a key challenge. In Germany, a carbon pricing scheme for the building and transportation sectors was introduced in 2021 with fixed prices, yet public acceptance remains low. With the planned integration of the national emissions trading system into the European Union Emissions Trading System (EU ETS-II) by 2027, carbon prices are expected to rise substantially due to the transition to market-based pricing and a continuously decreasing cap on emission allowances. These anticipated price hikes will likely increase the financial burden on private households and risk intensifying concerns about personal costs. Understanding individual cost perceptions and how they shape public attitudes is therefore essential to designing effective communication strategies and policy measures that can ensure the long-term viability of carbon pricing. This paper examines how beliefs about personal costs influence the acceptance of carbon pricing. Specifically, we address two key questions: First, how accurate are individuals’ perceptions of their personal costs of carbon pricing, and do they correctly anticipate the financial implications of future price hikes for their own households? Second, does providing personalized information about the actual costs of carbon pricing affect attitudes toward the policy? To this end, we conduct tailored information provision experiments to test for the causal effect of individual cost perceptions on carbon price acceptance. Our randomized experiments are embedded in an online survey on a large sample of 4,759 respondents that is representative of the adult German population. Data collection took place shortly after 1 an unanticipated price increase that received widespread media coverage, making personal costs especially salient. Our experiments proceed as follows: We first elicit people’s acceptance of paying a carbon price. Subsequently, respondents are randomly assigned to one of two questions about their perceived additional costs of carbon pricing, based on their current energy consumption, and their uncertainty regarding this estimation, either for the current price of e45 or a projected price of e200 per ton of CO2. To generate exogenous variation in beliefs, we provide random subsamples of respondents with personalized information about their actual costs of carbon pricing, which we calculate based on previously reported household characteristics, including detailed information on their energy use. Finally, we re-elicit carbon price acceptance for all respondents. We first document a series of stylized facts about people’s attitudes toward carbon pricing and their cost perceptions: The majority of our respondents rejects the policy (54 percent) and acceptance is particularly low among those who perceive their costs to be high. At the same time, most individuals are not well informed about their current and projected future personal costs. Around 64 percent overestimate what they currently pay, while approximately 72 percent underestimate their projected future costs. On average, our respondents overestimate their current annual costs by e206.4 and underestimate future costs at the higher price by e296.4. The main finding of our paper is that personalized information about the costs of carbon pricing significantly influences acceptance of the policy. On average, individuals who overestimate their personal costs increase their carbon price acceptance when receiving cost information, while those who underestimate their costs become less supportive. The effect is stronger for those with larger initial misperceptions and greater uncertainty about their cost estimates, which is consistent with the notion that the provided information may be more valuable for ex ante less informed respondents. The aggregate implications for policy acceptance differ markedly between the current and the projected price experiment. Since most individuals overestimate their current costs, the net effect of personalized cost information on acceptance is positive. In contrast, because the majority underestimate the financial impact of the projected price, information provision leads to a decline in 2 acceptance. Importantly, we find that personalized information about current costs is particularly effective in increasing support among individuals who initially opposed carbon pricing. This highlights the potential to build broader public support by correcting current cost misperceptions. Overall, receiving personalized information about current costs increases the probability of finding carbon pricing acceptable by 4.5 percentage points. In contrast, information about projected future costs reduces support across the board, regardless of individuals’ initial stance. This results in a 9.6 percentage point increase in the likelihood of rejecting carbon pricing, suggesting that future information shocks may reinforce public opposition. We additionally identify demographic groups that drive these effects. While information about current costs increases acceptance across both government and opposition supporters, information about projected future costs disproportionately reduces support among those affiliated with the governing parties. This highlights the risk that future cost shocks may erode the political foundation of carbon pricing. Other than that, heterogeneity in treatment effects is largest based on respondents’ financial situation and exposure to carbon pricing, which aligns with systematic variations in pre-treatment beliefs. The treatment increases acceptance particularly among less affected individuals who strongly overestimate their costs at current prices, while future cost information reduces support among the more affected who substantially underestimate the financial impact of projected price increases. Lastly, we evaluate the external validity of our findings and demonstrate their robustness regarding survey-related response biases, such as experimenter demand, survey fatigue, or distrust in the provided information. Our study contributes to a growing literature on the determinants of attitudes toward climate policy (Drews and Van den Bergh, 2016; Bergquist et al., 2022) and, more specifically, support for carbon pricing (Carattini, Carvalho and Fankhauser, 2018; MaestreAndr´es et al., 2019; Klenert et al., 2018). Among other factors, concerns about the specific design features, such as the effectiveness in reducing emissions, distributional fairness, and 3 personal costs are often cited as shaping public acceptance of carbon pricing (MaestreAndr´es et al., 2019; Carattini et al., 2017; Carattini, Carvalho and Fankhauser, 2018).1 There is mixed evidence on the (relative) importance of these factors in explaining public support for climate policy. Some studies emphasize financial self-interest, showing that support correlates with proxies for individual affectedness and exposure to costs (see, for example, Groh and Ziegler, 2018; Sommer et al., 2022). Others argue that environmental and fairness concerns matter more than personal costs (Kallbekken and Sælen, 2011; Bergquist et al., 2022). In Germany, where carbon price acceptance is relatively low, cross-country studies on hypothetical climate policies highlight the prominence of personal costs and beliefs about household-level impacts from tax-and-dividend schemes as strong correlates of opposition (Dabla-Norris et al., 2023; Dechezleprˆetre et al., 2025). These findings align with experimental evidence that consistently shows how cost-related information can causally influence support for climate policy (Schwarz et al., 2024; DablaNorris et al., 2023; Dechezleprˆetre et al., 2025; Douenne and Fabre, 2022). Most prior experimental studies on the support for carbon pricing implicitly rely on information gaps or raising the salience of costs. In contrast, we explicitly document and account for individual cost misperceptions by providing personalized cost information to manipulate beliefs.2Therefore, our approach relates closely to recent work on hypothetical carbon tax-and-dividend schemes that examines how (incorrect) individual beliefs that the own household would be a net financial loser determines policy support (Douenne and Fabre, 2022) and evidence on the role of low public awareness and underestimation of climate rebate amounts for the support of carbon pricing (Mildenberger et al., 2022). 1A related strand of the literature additionally investigates the role of revenue use and earmarking for carbon price acceptance (Baranzini and Carattini, 2017; Sælen and Kallbekken, 2011; Sommer et al., 2022; Kaestner et al., 2023; Beiser-McGrath and Bernauer, 2019). Maestre-Andr´es et al. (2019) and Klenert et al. (2018) provide reviews of the literature on the role of revenue use for policy support. 2Methodologically, our study is related to the literature that investigates the drivers of policy preferences by experimentally manipulating beliefs. For a comprehensive review of the literature utilizing information provision experiments, see Haaland et al. (2023). Our approach most closely resembles other tailored information provision experiments that provide participants with customized information based on their personal characteristics (Roth et al., 2022; Kuziemko et al., 2015; Cruces et al., 2013; Karadja et al., 2017; Hvidberg et al., 2023) or peer-group information (Card et al., 2012; Zimmermann, 2020). Thematically, we contribute to a broader literature that studies self-interested political preferences (Haaland and Roth, 2020; Kuziemko et al., 2015; Karadja et al., 2017; Stantcheva, 2021; Fanghella et al., 2023; Kaestner et al., 2025). 4 Prior work suggests that initial misperceptions may be corrected following policy implementation, potentially increasing public support over time (Konc et al., 2022).3However, in line with other studies on real-world climate policies (Mildenberger et al., 2022), we show that widespread misperceptions persist post implementation. Moreover, we find that prevalent cost overestimations at current prices cannot be extrapolated to future price developments. Instead, the substantial underestimation of costs at higher projected prices suggests that prospective information shocks may, if anything, strengthen opposition to carbon pricing. This aligns with prior evidence that higher price levels are associated with lower support (Sommer et al., 2023). Our results indicate that this negative effect may be further exacerbated by overly optimistic beliefs about future financial impacts. A key innovation of our study is this forward-looking perspective: While existing research on current or hypothetical policies consistently emphasizes that cost-related information can increase acceptance, we show that such information can also reduce support - raising concerns about the durability of climate policy under rising prices. The remainder of this paper is organized as follows. In Section 2, we describe our survey and sample as well as the design of our information provision experiments. Section 3 documents our respondents’ pre-treatment attitudes toward carbon pricing and their perceptions of personal costs. In Section 4, we present the results on the causal effect of tailored cost information on carbon price acceptance, including a heterogeneity analysis of our treatment effects, and several robustness checks. Section 5 concludes. 2. Data We use data from a representative online panel of the German adult population, collected in collaboration with GapFish, a German market research company. The survey took place between 15 January 2024 and 7 February 2024, immediately after the CO2 price in Germany was increased from e30 to e45 per ton of CO2. This price increase exceeded 3For policies beyond carbon pricing, Schuitema et al. (2010) show that acceptance of a congestion charge in Stockholm increased post-implementation as cost concerns diminished. Similarly, Carattini, Baranzini and Lalive (2018) find that acceptance of a garbage tax in Switzerland rose after initial fairness and effectiveness concerns were corrected. 5 the originally announced e40 and received widespread media attention. We use a quota sampling approach to ensure that the sample represents the adult German population in terms of age, gender, region, and household income. Table A1 in the Appendix shows that it closely resembles the German microcensus, an official, nationally representative dataset, across key characteristics.4Additional information on the microcensus is provided in Appendix B.1. Furthermore, our respondents’ reported party affiliations are closely aligned with contemporaneous national polling data. This suggests that political preferences are well represented in our sample, minimizing concerns about political bias influencing attitudes toward climate policy. In the following, we outline the structure of the survey, describe the sample restrictions, and present its main characteristics. We then detail the design of our information provision experiments and explain how we measure the key variables used in the empirical analysis. 2.1. Survey and sample The survey begins with questions on respondents’ age, gender, federal state of residence, and household income. We then collect information to estimate household-specific costs of carbon pricing, required to calculate the treatment information for our survey experiments. These variables include household size, homeowner status (owner vs. renter), living space, energy sources for space heating and hot water (i.e. electricity, gas, oil, solid fuels, or other renewables), the number of gasoline or diesel-run vehicles and annual mileage. Our survey experiments start by asking all respondents about their acceptance of carbon pricing. Respondents are then randomly assigned to one of two questions that measure their ex ante informedness about their additional costs of carbon pricing, either for the current price of e45 per ton of CO2 or a projected price of e200 per ton of CO2. Subsequently, half of all respondents are randomly selected to receive personalized information on their actual costs of carbon pricing. Finally, we re-elicit carbon price acceptance for 4The sample closely matches the microcensus in terms of age, gender, and region. However, our respondents’ household incomes are somewhat lower on average. 6 4. The causal effect of tailored cost information We now present results from the experiments, which provide respondents with personalized information about their household’s current and projected costs of carbon pricing. We first preview key preand post-treatment variables relevant to understanding acceptance revisions and then formally estimate a linear model of carbon price acceptance, leveraging the exogenous variation in cost perceptions generated by our treatments. We explore various aspects of treatment heterogeneity and close with a discussion of the robustness of our findings. 4.1. Descriptive evidence Table 1 presents summary statistics for key variables across treatment and control groups in the current price (Panel A) and projected price (Panel B) experiments. Prior to the information treatment there are no statistically significant differences in mean carbon price acceptance and other pre-treatment variables. Perceived costs, actual costs, perception gaps, and the share of individuals overestimating their costs are consistent across groups, confirming successful randomization in both experiments. The lower part of each panel reports post-treatment outcomes. In the current price experiment, treated respondents exhibit significantly higher post-treatment acceptance levels than the control group, with a mean difference of 0.25 (significant at the 1 percent level). The positive effect of the information treatment is consistent with the widespread overestimation of current costs. The difference in mean acceptance revisions, which accounts for (insignificant) differences in pre-treatment acceptance, is slightly smaller but remains highly significant. Notably, the larger absolute acceptance revisions indicate significant updating behavior in both directions along the acceptance scale. The last row in Panel A shows that treated respondents are 17 percentage points more likely to revise their acceptance compared to the control group – yet a large fraction of respondents in both groups does not revise acceptance. In contrast, in the projected price experiment, where respondents predominantly underestimate their future costs, post-treatment acceptance levels of carbon pricing are 13 significantly lower among treated respondents relative to the control group, with a mean difference of 0.34 (significant at the 1 percent level). Treated respondents decrease their acceptance level by 0.29 more than those in the control group on average (significant at the 1 percent level), when accounting for pre-existing differences. The absolute acceptance revisions are larger, indicating substantial updating behavior in both directions. Finally, the share of non-revisers is 18 percentage points lower in the treatment group. Overall, these results suggest a causal effect of our treatments on carbon price acceptance. The direction of the average acceptance revision is consistent with the sign of the average perception gap. In the current price experiment, personalized information mainly corrects cost overestimation and boosts acceptance on average, whereas information about future costs reduces acceptance due to cost underestimation in the projected price experiment. We provide a more detailed analysis of the treatment effects below. 4.2. Regression evidence 4.2.1. Empirical approach To formally identify the causal effect of personalized cost information on carbon price acceptance, we estimate the following regression model using OLS: accpost i=α0+βTi+α1accprior i+δXi+ϵi(1) The dependent variable accpost icontinuously measures the post-treatment acceptance of respondent i, ranging from 1 (very unacceptable) to 5 (very acceptable). Tiis a dummy variable taking the value 1 for respondents who receive the information treatment and 0 otherwise. We control for an indicator of pre-treatment acceptance accprior i. This helps isolate treatment effects from pre-existing differences in acceptance, despite the measure not necessarily being interpersonally comparable. We also include a set of individual-level controls Xito improve the precision of our estimates and to account for minor imbalances 14 between treatment and control groups.15 ϵiis an individual-specific error term. The coefficient of interest, β, identifies the average change in carbon price acceptance among treated respondents relative to the control group. However, the model in Equation (1) incompletely characterizes how the information treatments affect policy acceptance, as the average treatment effect may mask heterogeneity regarding the direction and magnitude of cost misperceptions. Genuine belief updating suggests that the effect size is larger for respondents with less accurate priors, as the value of the signal should increase with ex ante biasedness. Thus, we expand our model by interacting the treatment dummy with respondents’ perception gap: accpost i=α0+βTi+θ1∆ϕi+γ(Ti×∆ϕi) + α1accprior i+δXi+ϵi(2) The perception gap ∆ϕiis the difference between estimated and actual costs in units of e100. Hence, positive values indicate an overestimation of costs and vice versa. βnow identifies the treatment effect for participants with no bias in cost perceptions, whereas γ captures how the treatment effect scales with the size of the perception gap. We expect the estimates of γto be positive if respondents change their policy acceptance as a result of updating their cost perceptions in the direction of the provided signal. Lastly, we explore whether participants’ responsiveness to the information treatment varies with their baseline uncertainty about the costs of carbon pricing. Prior to the information treatment, we measure uncertainty on a five-point Likert scale in the second step of the experiments, with larger values indicating higher uncertainty. Respondents with a higher cost uncertainty should place more weight on the provided signal, leading to larger revisions in policy acceptance. We test this by adding interactions with respondents’ 15Specifically, we include dummies for being male, living in East Germany, holding a high school diploma, living in a high income household, homeownership, whether the respondent’s household uses fossil fuels for heating (including hot water), owns any motor vehicles, and an indicator variable for the respondent’s political party preference. We include a quadratic polynomial in the respondent’s age, and continuously control for household size. 15 uncertainty about their costs of carbon pricing to our model: accpost i=α0+βTi+θ1∆ϕi+γ(Ti×∆ϕi) + θ2Ui+βU(Ti×Ui) +θ3(∆ϕi×Ui) + γU(Ti×∆ϕi×Ui) + α1accprior i+δXi+ϵi (3) where Uiis a continuous measure of baseline uncertainty for respondent i. Notice that βnow measures the average treatment effect for respondents with a zero perception gap and low uncertainty, whereas γis an estimate of the treatment-induced change in acceptance with respect to perception gap size for high certainty respondents. βUcaptures the change in treatment effect with increasing uncertainty at a zero perception gap. The coefficient of primary interest γUgives us the change in acceptance with respect to the perception gap and uncertainty of a treated respondent. We expect γUto be positive, if the informativeness of the provided signal is increasing with baseline uncertainty. 4.2.2. Main results The results from the main regression analyses are presented in Table 2. We begin by discussing findings from the current price experiment in Columns 1 to 3. Column 1 shows results from the baseline specification that regresses post-treatment acceptance on a treatment dummy, pre-treatment acceptance, and a set of individual-level controls (Equation 1).16 In the current price experiment, the information treatment increases respondents’ acceptance level of carbon pricing by 0.161 on average (significant at the 1 percent level). The positive estimate of βis consistent with the widespread overestimation of current costs. Supporting evidence reveals substantial heterogeneity in treatment effects based on the sign of the perception gap (Appendix Table A7). Respondents who overestimate their costs (including those with a zero perception gap) increase their acceptance level by 0.365 on average, while those who underestimate costs reduce their acceptance level by 0.228 (Column 1). These patterns highlight that the effect of information provision depends 16For completeness, we reproduce our baseline analysis from Table 2 without any additional controls in Table A6 of the Appendix, where we obtain virtually unchanged estimates. 16 critically on the direction of an individual’s bias. In Column 2 in Table 2, we examine whether the size of the treatment effect increases with the magnitude of cost misperceptions by interacting the treatment dummy with the perception gap (Equation 2). The estimated coefficient on the interaction term (γ) is positive and statistically significant at the 1 percent level, indicating that the effect of our treatment significantly increases in the size of a respondent’s bias. Specifically, for each e100 increase in the perception gap, acceptance increases by an additional 0.033, on top of a baseline treatment effect of 0.086 (β). Larger acceptance revisions among more biased respondents support the notion of meaningful updating of cost perceptions.17 Interestingly, the estimate of βsuggests that the treatment increases the acceptance of respondents with accurate prior beliefs.18 This result is consistent with behavioral evidence suggesting that individuals respond positively to affirmation. For instance, loss aversion (Tversky and Kahneman, 1991) underscores the asymmetric emotional impact of perceived gains versus losses, which may reinforce positive responses to validated expectations. Relatedly, motivated reasoning implies that confirming individuals’ beliefs or perceptions – such as their cost assessments – can elicit favorable reactions (B´enabou and Tirole, 2016). Column 3 in Table 2 extends the analysis to include respondents’ ex ante uncertainty about the perceived costs of carbon pricing (Equation 3). We find that treatment responsiveness increases with both higher uncertainty and larger misperceptions, as indicated by a positive and significant triple interaction coefficient (γU).19 The significant estimate of 17The relevance of the perception gap may depend on how it relates to a respondent’s ex-ante cost perception. That is, for the same absolute perception gap, the strength of the signal could vary across individuals with different baseline cost perceptions. To account for this, Appendix Table A8 re-estimates Equation (2) using a relative perception gap, defined as the log difference between perceived and actual costs. Results are consistent for both experiments. 18We further explore this finding by estimating heterogeneous treatment effects for respondents who overestimate, underestimate, or accurately estimate their costs of carbon pricing based on different perception accuracy thresholds. To this end, we apply two different definitions of unbiasedness. In Column 2 in Table A7 we classify respondents with a perception gap of up to 20 percent of their actual costs as unbiased. In Column 3 we allow for a deviation of e25, i.e. an absolute perception gap of up to e25, for a respondent’s perception to be defined as accurate. The estimated marginal effects for unbiased respondents are similar in magnitude to our main specification yet noisily estimated (Table A7, Columns 2 and 3). 19To put our results into perspective, a respondent overestimating costs by e300 and reporting high uncertainty (uncertainty = 4) is predicted to increase acceptance by 0.29 points in response to the treatment (0.001 + 0.003 ×3 + 0.028 ×4 + 0.014 ×3×4). 17 γUconfirms theoretical predictions about belief updating as the underlying causal mechanism. Here, we find no treatment effect on acceptance among high-certainty respondents – regardless of whether their cost perceptions are biased – reflected in insignificant estimates of both βand γ. Hence, the previously observed positive treatment effect at zero perception gap (estimate of βin Column 2) disappears once accounting for uncertainty. Instead, the positive – though imprecisely estimated – coefficient on the interaction with uncertainty (βU) suggests that the treatment effect among unbiased respondents is driven by uncertain respondents. We conduct the same analyses for the projected price experiment, presented in Columns 4 to 6 in Table 2. Column 4 provides results from our baseline specification (Equation 1), showing that personalized information about future costs of carbon pricing leads to a decrease in acceptance levels by 0.307 on average (significant at the 1 percent level). This result reflects that respondents predominantly underestimate their future costs and thereby contrasts our findings from the current price experiment. Column 4 in Appendix Table A7 confirms the previously documented asymmetry in treatment effects based on the sign of the perception gap. Among those underestimating their costs, we find an average reduction in acceptance levels of 0.471, whereas the much smaller number of overestimators increase theirs by 0.103. Column 5 in Table 2 shows that the treatment effect increases with the extent of respondents’ misperceptions, as measured by the perception gap (Equation 2). The information treatment in the projected price experiment decreases acceptance by 0.229 at a zero perception gap (β), and changes by an additional 0.026 for every e100 difference in the perception gap (γ; both significant at the 1 percent level). These results are analogous in pattern, to the amplifying effect of cost misperceptions in the current cost experiment. However, the analysis in Appendix Table A7 shows that when classifying respondents by discrete perception accuracy thresholds, the estimated treatment effect for unbiased individuals is statistically insignificant and quantitatively close to zero. This suggests that the significant negative coefficient at a zero perception gap in Column 5 in Table 2 may 18 be an artifact of the linearity assumption and should be interpreted with caution.20 Column 6 in Table 2 incorporates respondents’ ex ante uncertainty (Equation 3). We find that the effect of correcting misperceptions about future costs is driven by those with higher uncertainty. The estimated coefficient on the triple interaction term (γU) is positive and statistically significant at the 5 percent level, whereas the pairwise interaction between treatment and perception gap (γ) is not significantly different from zero.21 This result again mirrors the mechanisms in the current cost experiment. Overall, we find that respondents systematically revise their acceptance of carbon pricing in response to personalized information about either their current or future costs. Treatment responses are consistent across both experiments and align with theoretical expectations: They are larger among those with greater misperceptions and higher ex ante uncertainty about their costs. However, due to differing distributions of cost misperceptions across the two experiments, average treatment effects diverge. In the current price experiment, where most respondents overestimate their actual costs, the information treatment leads to a net increase in acceptance. By contrast, in the projected price experiment, where underestimation of future costs is more widespread, the personalized information results in a net decrease in acceptance. Our findings highlight how individual updating of policy preferences in response to personalized information depends crucially on the direction and magnitude of misperceptions, as well as individuals’ uncertainty. Thus, while personalized information-interventions can shift public support, the direction ultimately depends on the informational background of the population. 20In Columns 5 and 6 in Table A7 in the Appendix we estimate heterogeneous treatment effects for respondents who overestimate, underestimate, or accurately estimate their future costs of carbon pricing. In Column 5, we allow for a relative error of up to 20 percent in perceived costs and in Column 6 for an absolute error of up to e100 for a respondent to be classified as unbiased. Both approaches yield treatment effects for unbiased respondents that are not statistically different from zero. This contrasts with the results in Column 5 of Table 2, where the significant negative effect at a zero perception gap (β= -0.229) likely stems from the linearity assumption of the specification. Specifically, it arises from extrapolating a linear relationship across the full range of perception gaps, while the specifications in Table A7 rely on discrete classifications. Accordingly, the interaction effect in Column 5 in Table 2 should be interpreted with caution, particularly around the zero perception gap region. 21While the results in Column 6 of Table 2 should be interpreted with caution due to the discussed limitations of the linear model specification, the estimates predict that a respondent underestimating future costs by e300 and reporting high uncertainty (uncertainty = 4) decreases acceptance by 0.334 points in response to the treatment (-0.187 + 0.009 ×(-3) + (-0.009) ×4 + 0.007 ×(-3) ×4). 19 4.2.3. Heterogeneity by pre-treatment acceptance The previous analysis explores the average treatment effects and the underlying mechanisms. From a political perspective, the effectiveness of the information intervention also hinges on the ability to counteract or even reverse individuals’ stance toward carbon pricing. For example, a higher average acceptance may stem from supporters becoming more positive toward carbon pricing or, more meaningfully, from individuals initially opposed to the policy lessening their opposition or even turning supportive. Only the latter has the capacity to shift public opinion toward majority support, whereas the former would leave a dichotomous distribution of preferences unchanged. Therefore, we evaluate the intervention’s potential to build broader support for carbon pricing by examining whether the size of the treatment effect differs based on ex-ante attitude toward carbon pricing. To this end, we interact the treatment dummy with respondents’ pre-treatment acceptance (Appendix Table A9) and plot the marginal treatment effects in Figure 4.22 In the current price experiment, marginal treatment effects are positive and statistically significant among respondents who found carbon pricing very unacceptable or rather unacceptable prior to the treatment (Figure 4a). In turn, the estimated effects for survey participants with an ex-ante neutral or positive stance are not significantly different from zero.23 Hence, the positive treatment effect is driven by those with a low baseline acceptance, underscoring the political effectiveness of tailored information provision about current costs. To better understand the implications of this result for moving toward majority sup22In Table A10 of the Appendix, we address the concern that heterogeneity in treatment effects potentially results from correlations between pre-treatment acceptance and other respondent characteristics. Following Haaland and Roth (2020), we decompose the total variation in pre-treatment acceptance into a component explained by observed respondent characteristics and an unrelated residual component. This is done by regressing pre-treatment acceptance on the same set of control variables used in our main analyses. We then examine treatment effect heterogeneity separately for each component. Consistent with the main results from both experiments, we find significant heterogeneity in treatment effects with respect to the residual variation (Columns 2 and 5), suggesting that pre-treatment acceptance is independently meaningful. In contrast, we do not observe treatment heterogeneity based on the predicted component linked to respondent characteristics (Columns 3 and 6). 23We previously identified perceived costs as a strong predictor of carbon price acceptance (Appendix Table A5). Panel A in Appendix Table A12 shows that perceived current costs are greatly inflated for respondents with lower baseline acceptance levels, despite similar actual costs across groups. Consequently, the heterogeneity of our results is driven by larger perception gaps (and a larger share of overestimators) among respondents with a low pre-treatment acceptance. 20 port, we additionally estimate treatment-induced changes in the probability of selecting each acceptance level using an ordered probit model. The marginal effects in Panel A in Appendix Table A11 indicate that the likelihood to consider carbon pricing very unacceptable drops by 5.1 percentage points for treated respondents. We find a corresponding increase in the probabilities to find carbon pricing acceptable (4.5 percentage points) or express neutrality (0.6 percentage points). Thus, providing tailored information on current costs can meaningfully broaden public support for carbon pricing. An analogous analysis of the projected price experiment yields a less positive outlook. Here, we find large and statistically significant reductions in acceptance across almost all levels of pre-treatment acceptance, except for those least favorable toward carbon pricing at baseline (Figure 4b).24 Marginal effects from the ordered probit model in Panel B of Table A11 show that the probability of finding carbon pricing very or somewhat unacceptable increases by 9.6 percentage points in response to the treatment, while the likelihoods of neutrality and policy acceptance decline by 1.7 and 7.9 percentage points, respectively. Hence, our results suggest that information shocks regarding the future costs of carbon pricing may reinforce opposition to the policy. In sum, personalized information on current costs can build broader support by shifting views among those most critical of carbon pricing, yet information about future costs tends to reduce support across the board. Moreover, the adverse effects from informing about future costs exceed the increase in carbon price acceptance from personalized information on current costs. 4.2.4. Other heterogeneity Identifying the populations that drive average treatment effects can inform more targeted and cost-effective communication strategies (Allcott, 2011). To better understand which groups are most responsive, we interact the treatment indicator with standard sociodemographic characteristics (dummies for age, education, gender, household size, region), 24This finding is consistent with limited differences in mean perceived costs, actual costs, and perception gaps by pre-treatment acceptance levels (Appendix Table A12, Panel B). Apart from those with the lowest baseline acceptance, who hold more accurate beliefs, all other survey participants tend to underestimate their future costs to a similar degree. 21 political affiliation (ruling party vs. opposition), and indicators of a respondent’s financial situation and exposure to carbon pricing (income, homeownership, households’ carbon price exposure in the building and transport sectors).25,26 When examining heterogeneity along standard sociodemographic dimensions, we find limited systematic variation. Most interaction effects are imprecisely estimated and close to zero. In the current price experiment (Panel A, Table 3), men show a significantly smaller increase in acceptance (Column 1), while respondents living in East Germany appear slightly more responsive (Column 3). However, the latter effect is not statistically significant. In the projected price experiment (Panel B), older individuals and supporters of the governing coalition become significantly more opposed following treatment (Columns 2 and 6). Overall, these patterns align with differences in pre-treatment beliefs, such as mean perception gaps or the share of respondents overestimating costs (Appendix Table A13 and Table A14). The differential response by political affiliation is particularly notable. While supporters of both government and opposition parties respond positively to information about current costs, government supporters exhibit a stronger negative reaction when exposed to projected future costs. This divergence underscores the political risk that future cost shocks may disproportionately reduce support among those most likely to back carbon pricing politically, thereby weakening its support base. Furthermore, we consistently observe pronounced heterogeneity based on financial characteristics and those directly related to carbon pricing exposure. Columns 7 to 10 in Panel A of Table 3 reveal that the positive treatment effect in the current price experiment is concentrated among lower-income individuals, renters and those not reliant on fossil fuels for heating and transportation. Appendix Table A13 shows that these respondents face significantly lower actual costs than their counterparts, whereas cost perceptions are more aligned across groups. This implies a significantly higher prevalence of cost over25We include homeowner status as a measure of financial exposure, as renters face substantially lower costs of carbon pricing due to regulations that impose a share of the costs on their landlords. 26The ruling coalition comprises the Social Democratic Party (SPD), the Green Party (B¨undnis 90/Die Gr¨unen), and the liberal Free Democratic Party (FDP), which introduced the carbon price in 2021 and was in power when we conducted the survey. 22 taxes in a post-Paris world: are millions of nays inevitable?’, Environmental and Resource Economics 68, 97–128. Carattini, S., Carvalho, M. and Fankhauser, S. (2018), ‘Overcoming public resistance to carbon taxes’, Wiley Interdisciplinary Reviews: Climate Change 9(5), e531. Card, D., Mas, A., Moretti, E. and Saez, E. (2012), ‘Inequality at work: The effect of peer salaries on job satisfaction’, American Economic Review 102(6), 2981–3003. Climate Leadership Council (2019), ‘Economists’ Statement on Carbon Dividends’. Crowley, K. (2021), ‘Fighting the future: The politics of climate policy failure in Australia (2015–2020)’, Wiley Interdisciplinary Reviews: Climate Change 12(5), e725. Cruces, G., Perez-Truglia, R. and Tetaz, M. (2013), ‘Biased perceptions of income distribution and preferences for redistribution: Evidence from a survey experiment’, Journal of Public Economics 98, 100–112. Dabla-Norris, M. E., Helbling, M. T., Khalid, S., Khan, H., Magistretti, G., Sollaci, A. and Srinivasan, M. K. (2023), Public perceptions of climate mitigation policies: Evidence from cross-country surveys, Staff Discussion Note 2023/002, International Monetary Fund, Washington, DC. Dechezleprˆetre, A., Fabre, A., Kruse, T., Planterose, B., Sanchez Chico, A. and Stantcheva, S. (2025), ‘Fighting climate change: International attitudes toward climate policies’, American Economic Review 115(4), 1258–1300. Douenne, T. and Fabre, A. (2022), ‘Yellow vests, pessimistic beliefs, and carbon tax aversion’, American Economic Journal: Economic Policy 14(1), 81–110. Drews, S. and Van den Bergh, J. C. (2016), ‘What explains public support for climate policies? A review of empirical and experimental studies’, Climate Policy 16(7), 855– 876. 29 Fanghella, V., Faure, C., Guetlein, M.-C. and Schleich, J. (2023), ‘What’s in it for me? Self-interest and preferences for distribution of costs and benefits of energy efficiency policies’, Ecological Economics 204, 107659. Gentzkow, M. and Shapiro, J. M. (2006), ‘Media bias and reputation’, Journal of Political Economy 114(2), 280–316. Groh, E. D. and Ziegler, A. (2018), ‘On self-interested preferences for burden sharing rules: An econometric analysis for the costs of energy policy measures’, Energy Economics 74, 417–426. Haaland, I. and Roth, C. (2020), ‘Labor market concerns and support for immigration’, Journal of Public Economics 191, 104256. Haaland, I., Roth, C. and Wohlfart, J. (2023), ‘Designing information provision experiments’, Journal of Economic Literature 61(1), 3–40. Hvidberg, K. B., Kreiner, C. T. and Stantcheva, S. (2023), ‘Social positions and fairness views on inequality’, Review of Economic Studies 90(6), 3083–3118. Jacksohn, A., Gr¨osche, P., Rehdanz, K. and Schr¨oder, C. (2019), ‘Drivers of renewable technology adoption in the household sector’, Energy Economics 81, 216–226. Kaestner, K., Pahle, M., Schwarz, A., Sommer, S. and St¨unzi, A. (2023), Experts’ conjectures, people’s statements and true preferences: The case of carbon price support, USAEE Working Paper 23-591, USAEE. Kaestner, K., Sommer, S., Berneiser, J., Henger, R. and Oberst, C. (2025), ‘Cost sharing mechanisms for carbon pricing: What drives support in the housing sector?’, Energy Economics 142, 108134. Kalkuhl, M., Kellner, M., Bergmann, T. and R¨utten, K. (2023), CO2-Bepreisung zur Erreichung der Klimaneutralit¨at im Verkehrsund Geb¨audesektor: Investitionsanreize und Verteilungswirkungen, Technical report, Mercator Research Institute on Global Commons and Climate Change, Berlin. 30 Kallbekken, S. and Sælen, H. (2011), ‘Public acceptance for environmental taxes: Selfinterest, environmental and distributional concerns’, Energy Policy 39(5), 2966–2973. Karadja, M., Mollerstrom, J. and Seim, D. (2017), ‘Richer (and holier) than thou? The effect of relative income improvements on demand for redistribution’, Review of Economics and Statistics 99(2), 201–212. Klenert, D., Mattauch, L., Combet, E., Edenhofer, O., Hepburn, C., Rafaty, R. and Stern, N. (2018), ‘Making carbon pricing work for citizens’, Nature Climate Change 8(8), 669–677. Konc, T., Drews, S., Savin, I. and Van Den Bergh, J. C. (2022), ‘Co-dynamics of climate policy stringency and public support’, Global Environmental Change 74, 102528. Kuziemko, I., Norton, M. I., Saez, E. and Stantcheva, S. (2015), ‘How elastic are preferences for redistribution? Evidence from randomized survey experiments’, American Economic Review 105(4), 1478–1508. Maestre-Andr´es, S., Drews, S. and Van Den Bergh, J. (2019), ‘Perceived fairness and public acceptability of carbon pricing: a review of the literature’, Climate Policy 19(9), 1186–1204. Mildenberger, M., Lachapelle, E., Harrison, K. and Stadelmann-Steffen, I. (2022), ‘Limited impacts of carbon tax rebate programmes on public support for carbon pricing’, Nature Climate Change 12(2), 141–147. Nerini, F. F., Keppo, I. and Strachan, N. (2017), ‘Myopic decision making in energy system decarbonisation pathways. A UK case study’, Energy Strategy Reviews 17, 19– 26. Roth, C., Settele, S. and Wohlfart, J. (2022), ‘Risk exposure and acquisition of macroeconomic information’, American Economic Review: Insights 4(1), 34–53. Roth, C. and Wohlfart, J. (2020), ‘How do expectations about the macroeconomy affect 31 personal expectations and behavior?’, Review of Economics and Statistics 102(4), 731– 748. Sælen, H. and Kallbekken, S. (2011), ‘A choice experiment on fuel taxation and earmarking in Norway’, Ecological Economics 70(11), 2181–2190. Schuitema, G., Steg, L. and Forward, S. (2010), ‘Explaining differences in acceptability before and acceptance after the implementation of a congestion charge in Stockholm’, Transportation Research Part A: Policy and Practice 44(2), 99–109. Schwarz, A., St¨unzi, A., Kaestner, K., Pahle, M. and Sommer, S. (2024), Tailored information and the public support for carbon pricing, Working paper. Schyns, B. and Paul, T. (2002), ‘Deutsche Self-Monitoring Skala’, Available at https: //zis.gesis.org/DoiId/zis55 (2025/06/11). Snyder, M. (1974), ‘Self-monitoring of expressive behavior.’, Journal of Personality and Social Psychology 30(4), 526–537. Sommer, S., Konc, T. and Drews, S. (2023), How resilient is public support for carbon pricing? Longitudinal evidence from Germany, Ruhr Economic Papers 1017, RWI - Leibniz-Institut f¨ur Wirtschaftsforschung. Sommer, S., Mattauch, L. and Pahle, M. (2022), ‘Supporting carbon taxes: The role of fairness’, Ecological Economics 195, 107359. Stantcheva, S. (2021), ‘Understanding tax policy: How do people reason?’, The Quarterly Journal of Economics 136(4), 2309–2369. Tversky, A. and Kahneman, D. (1991), ‘Loss aversion in riskless choice: A referencedependent model’, The Quarterly Journal of Economics 106(4), 1039–1061. Wasi, N. and Carson, R. T. (2013), ‘The influence of rebate programs on the demand for water heaters: The case of New South Wales’, Energy Economics 40, 645–656. Zimmermann, F. (2020), ‘The dynamics of motivated beliefs’, American Economic Review 110(2), 337–363. 32 Intro questions: gender, age, federal state, household income Questions on households’ heating and driving habits Q: Prior acceptance of carbon pricing Q: Costs of carbon pricing at price of AC45/t CO2 + uncertainty Q: Costs of carbon pricing at price of AC200/t CO2 + uncertainty Treatment: Actual costs No treatment Treatment: Actual costs No treatment Q: Posterior acceptance of carbon pricing Other questions: socio-demographics, political party preferences, survey feedback p≈0.5 p ≈0.5 p≈0.5 p ≈0.5 p ≈0.5 p ≈0.5 Notes: The figure provides an overview of the structure of the survey experiments. Figure 1: Structure of the survey experiments 0.0 0.1 0.2 0.3 0.4 Fraction Very acceptable Rather acceptable Neither nor Rather unacceptable Very unacceptable Notes: The figure shows the pre-treatment acceptance of carbon pricing for all respondents. Figure 2: Pre-treatment acceptance of carbon pricing 33 0.00 0.05 0.10 0.15 0.20 0.25 Fraction −4000 −3000 −2000 −1000 0 1000 2000 3000 4000 5000 Perception gap (a) Current price 0.00 0.05 0.10 0.15 0.20 0.25 Fraction −4000 −3000 −2000 −1000 0 1000 2000 3000 4000 5000 Perception gap (b) Projected price Notes: The figure shows the distributions of the respondents’ cost perception gaps (in bins of e100) for the current price of e45 per ton of CO2 (a) and the projected price of e200 per ton of CO2 (b). The perception gap is defined as the difference between perceived and actual costs. Positive (negative) values indicate overestimation (underestimation) of costs. Figure 3: Distribution of cost perception gaps −1.0 −0.5 0.0 0.5 Marginal effect 1 2 3 4 5 Pre−treatment acceptance (a) Current price −1.0 −0.5 0.0 0.5 Marginal effect 1 2 3 4 5 Pre−treatment acceptance (b) Projected price Notes: The figure shows marginal treatment effects on respondents’ post-treatment acceptance of carbon pricing by respondents’ pre-treatment acceptance (1=very unacceptable to 5=very acceptable). The figure displays the marginal treatment effects with 95 percent confidence intervals. The corresponding estimates are presented in Appendix Table A9. Figure 4: Heterogeneity by pre-treatment acceptance 34 Table 1: Average acceptance and cost perceptions by treatment Panel A: 45 e/t CO2 Control (1) Treated (2) Diff. (1)-(2) Pre-treatment acceptance 2.40 2.51 -0.11 Perceived costs 392.29 413.53 -21.24 Actual costs 199.99 193.00 6.99 Perception gap 192.30 220.53 -28.23 Overestimation 0.64 0.65 -0.01 Post-treatment acceptance 2.37 2.62 -0.25*** Acceptance revision -0.03 0.11 -0.14*** Abs. acceptance revision 0.22 0.47 -0.25*** Non-reviser 0.81 0.64 0.17*** Observations 814 816 1630 Panel B: 200 e/t CO2 Control (1) Treated (2) Diff. (1)-(2) Pre-treatment acceptance 2.53 2.48 0.05 Perceived costs 559.36 576.28 -16.91 Actual costs 864.11 864.65 -0.55 Perception gap -304.74 -288.38 -16.37 Overestimation 0.28 0.29 -0.01 Post-treatment acceptance 2.43 2.08 0.34*** Acceptance revision -0.10 -0.39 0.29*** Abs. acceptance revision 0.26 0.54 -0.29*** Non-reviser 0.80 0.61 0.18*** Observations 795 829 1624 Notes: The table presents summary statistics for key variables across treatment and control groups in the current price (Panel A) and projected price (Panel B) experiments. Columns 1 and 2 show means for the control and treatment groups, and Column 3 shows the difference in means between the two groups. Overestimation includes a small number of respondents with a zero perception gap. * p<0.1, ** p<0.05, *** p<0.01 35 Table 2: Treatment effects on carbon price acceptance 45 e/t CO2 200 e/t CO2 (1) (2) (3) (4) (5) (6) T 0.161*** 0.086** 0.001 -0.307*** -0.229*** -0.187** (0.034) (0.037) (0.099) (0.035) (0.036) (0.090) Perception gap -0.004 0.009 -0.006*** -0.001 (0.004) (0.007) (0.002) (0.005) T×Perception gap 0.033*** 0.003 0.026*** 0.009 (0.008) (0.016) (0.004) (0.009) Uncertainty -0.019 -0.026 (0.018) (0.019) T×Unc. 0.028 -0.009 (0.032) (0.030) Perception gap ×Unc. -0.006** -0.002 (0.003) (0.002) T×Perception gap ×Unc. 0.014** 0.007** (0.006) (0.003) Controls Yes Yes Yes Yes Yes Yes R-squared 0.741 0.747 0.749 0.712 0.724 0.725 Observations 1630 1630 1620 1624 1624 1609 Notes: The table presents estimation results from OLS regressions. Results for the current price experiment are shown in Columns 1 to 3 and results for the projected price experiment are shown in Columns 4 to 6. The dependent variable is the post-treatment acceptance of carbon pricing, measured on a five-point Likert scale (1=very unacceptable to 5=very acceptable). T is a dummy variable indicating that a respondent received personalized information about the costs of carbon pricing. The perception gap is defined as the difference between perceived and actual costs (divided by 100). Uncertainty regarding perceived costs is measured on a five-point Likert scale (0=very uncertain to 4=very certain). All regressions include the pre-treatment acceptance of carbon pricing and the set of controls described in Appendix Table A5. Robust standard errors are in parentheses. * p<0.1, ** p<0.05, *** p<0.01 36 Table 3: Heterogeneity by respondent characteristics (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Panel A: High Ruling High Fossil Motor 45 e/t CO2 Male Aged 45+ East school Single party income Homeowner heating vehicles T 0.262*** 0.124** 0.138*** 0.178*** 0.151*** 0.129*** 0.233*** 0.254*** 0.412*** 0.438*** (0.052) (0.057) (0.036) (0.049) (0.040) (0.045) (0.042) (0.047) (0.079) (0.087) T×C -0.191*** 0.057 0.154 -0.034 0.037 0.033 -0.250*** -0.221*** -0.318*** -0.341*** (0.068) (0.068) (0.108) (0.067) (0.067) (0.078) (0.071) (0.067) (0.087) (0.094) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes R-squared 0.742 0.741 0.741 0.741 0.741 0.765 0.742 0.742 0.743 0.743 Observations 1630 1630 1630 1630 1630 1334 1630 1630 1630 1630 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Panel B: High Ruling High Fossil Motor 200 e/t CO2 Male Aged 45+ East school Single party income Homeowner heating vehicles T -0.261*** -0.229*** -0.312*** -0.294*** -0.325*** -0.220*** -0.265*** -0.201*** -0.115 -0.083 (0.053) (0.057) (0.039) (0.049) (0.041) (0.042) (0.042) (0.046) (0.077) (0.092) T×C -0.086 -0.121* 0.035 -0.026 0.056 -0.212** -0.142* -0.269*** -0.241*** -0.274*** (0.069) (0.067) (0.085) (0.069) (0.067) (0.085) (0.075) (0.069) (0.087) (0.099) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes R-squared 0.713 0.713 0.712 0.712 0.712 0.729 0.713 0.715 0.714 0.714 Observations 1624 1624 1624 1624 1624 1344 1624 1624 1624 1624 Notes: The table presents estimation results from OLS regressions. Results for the current price experiment are shown in Panel A and results for the projected price experiment are shown in Panel B. The dependent variable is the post-treatment acceptance of carbon pricing, measured on a five-point Likert scale (1=very unacceptable to 5=very acceptable). T is a dummy variable indicating that a respondent received personalized information on the costs of carbon pricing. C is a dummy variable representing the following individual-level characteristics: Male (1=yes, 0=no), Aged 45+ (1=aged 45 or over, 0=otherwise), East (1=living in East Germany, 0=living in West Germany), High school (1=yes, 0=no), Single (1=Single household, 0=otherwise), Ruling party (1=supporters of SPD, B90/Die Gr¨unen or FDP, 0=supporters of CDU/CSU, Linke, AfD, other parties), High income (1=household income of e4,000 or more, 0=less than e4,000), Homeowner (1=yes, 0=no), Fossil heating (1=yes, 0=no), Motor vehicles (1=yes, 0=no). All regressions include the pre-treatment acceptance of carbon pricing and the set of controls described in Appendix Table A5. Robust standard errors are in parentheses. * p<0.1, ** p<0.05, *** p<0.01 37 Online appendix: Cost perceptions and the support for carbon pricing Jan Behringer, Lukas Endres, Maike Korsinnek Summary of the appendix In Section A we show summary statistics, balance tests, and additional estimation results. Table A1 compares summary statistics of the initial sample with the German microcensus for key demographic variables. Table A2 contrasts the initial sample of our survey with the final sample on which we estimate our main results. Table A3 and Table A4 provide evidence of covariate balance in the treatment and control groups of our experiments. Table A5 examines determinants of pre-treatment acceptance. Table A6 replicates our main regression results without additional controls. Table A7 shows differential treatment effects for respondents who overestimate, underestimate, or accurately estimate their costs of carbon pricing. Table A8 shows how treatment effects vary with relative perception gaps. Table A9 examines heterogeneous treatment effects by pre-treatment acceptance. Table A10 shows that the estimated heterogeneous effects by pre-treatment acceptance are robust to using a residual component of pre-treatment acceptance. Table A11 provides ordered probit estimates of our main treatment effects. Table A12 shows cost perceptions by pre-treatment acceptance. Table A13 and Table A14 show cost perceptions by respondent characteristics. Table A15 evaluates the external validity of the main results by reweighting our sample to represent the general population and demonstrates the robustness of our main results regarding survey related response biases such as experimenter demand effects, distrust in the provided information, and survey fatigue. Section B.1 provides background information on the 2022 microcensus. In Section B.2 we describe in detail the method for the calculation of respondents’ CO2 costs. In Section B.3 we show English translations of our main survey questions. Section B.4 provides screenshots of the main experiment questions from the online survey. 38 Table A9: Heterogeneity by pre-treatment acceptance (1) (2) 45 e/t CO2 200 e/t CO2 T 0.253*** 0.042 (0.054) (0.037) T×Acceptance=2 0.015 -0.511*** (0.102) (0.078) T×Acceptance=3 -0.178* -0.371*** (0.099) (0.099) T×Acceptance=4 -0.227** -0.694*** (0.091) (0.105) T×Acceptance=5 -0.252** -0.323*** (0.100) (0.125) Controls Yes Yes R-squared 0.742 0.722 Observations 1630 1624 Notes: The table presents estimation results from OLS regressions. Results for the current price experiment are shown in Column 1 and results for the projected price experiment are shown in Column 2. The dependent variable is the posttreatment acceptance of carbon pricing, measured on a fivepoint Likert scale (1=very unacceptable to 5=very acceptable). T is a dummy variable indicating that a respondent received personalized information about the costs of carbon pricing. All regressions include the set of controls described in Appendix Table A5. Robust standard errors are in parentheses. * p<0.1, ** p<0.05, *** p<0.01 45 Table A10: Heterogeneity by residual pre-treatment acceptance 45 e/t CO2 200 e/t CO2 (1) (2) (3) (4) (5) (6) T 0.346*** 0.160*** 0.255 0.066 -0.299*** -0.209 (0.067) (0.034) (0.199) (0.058) (0.035) (0.190) Acceptance 0.878*** 0.865*** (0.017) (0.018) T×Acceptance -0.076*** -0.146*** (0.023) (0.026) Res. acceptance 0.899*** 0.867*** (0.020) (0.020) T×Res. acceptance -0.115*** -0.153*** (0.032) (0.032) Pred. acceptance 1.188*** 0.960*** (0.206) (0.236) T×Pred. acceptance -0.007 -0.040 (0.080) (0.077) Controls Yes Yes Yes Yes Yes Yes R-squared 0.742 0.743 0.250 0.711 0.710 0.246 Observations 1630 1630 1630 1624 1624 1624 Notes: The table presents estimation results from OLS regressions. Results for the current price experiment are shown in Columns 1 to 3 and results for the projected price experiment are shown in Columns 4 to 6. In Columns 1 and 4, we include the pre-treatment acceptance of carbon pricing. For the other regressions, we decompose the total variation in pre-treatment acceptance of carbon pricing into a component predicted by the set of control variables we use throughout the paper, and a residual component that is not explained by these variables. In Columns 2 and 5, we include the residual component of pre-treatment acceptance. In Columns 3 and 6, we include the predicted component of pre-treatment acceptance. The dependent variable is the post-treatment acceptance of carbon pricing, measured on a five-point Likert scale (1=very unacceptable to 5=very acceptable). T is a dummy variable indicating that a respondent received personalized information about the costs of carbon pricing. All regressions include the set of controls described in Appendix Table A5. Robust standard errors are in parentheses. * p<0.1, ** p<0.05, *** p<0.01 46 Table A11: Ordered probit estimations of treatment effects (1) (2) (3) (4) (5) Panel A: 45 e/t CO2 Acc.=1 Acc.=2 Acc.=3 Acc.=4 Acc.=5 T -0.051*** 0.000 0.006*** 0.028*** 0.017*** (0.011) (0.002) (0.002) (0.006) (0.004) Controls Yes Yes Yes Yes Yes Pseudo R-squared 0.415 0.415 0.415 0.415 0.415 Observations 1630 1630 1630 1630 1630 (1) (2) (3) (4) (5) Panel B: 200 e/t CO2 Acc.=1 Acc.=2 Acc.=3 Acc.=4 Acc.=5 T 0.090*** 0.006 -0.017*** -0.055*** -0.024*** (0.011) (0.004) (0.003) (0.007) (0.004) Controls Yes Yes Yes Yes Yes Pseudo R-squared 0.410 0.410 0.410 0.410 0.410 Observations 1624 1624 1624 1624 1624 Notes: The table presents average marginal treatment effects from ordered probit regressions. Results for the current price experiment are shown in Panel A and results for the projected price experiment are shown in Panel B. The dependent variable is the post-treatment acceptance of carbon pricing, measured on a five-point Likert scale (1=very unacceptable to 5=very acceptable). T is a dummy variable indicating that a respondent received personalized information about the costs of carbon pricing. All regressions include the pre-treatment acceptance of carbon pricing and the set of controls described in Appendix Table A5. Robust standard errors are in parentheses. * p<0.1, ** p<0.05, *** p<0.01 Table A12: Cost perceptions by pre-treatment acceptance (1) (2) (3) (4) (5) Panel A: 45 e/t CO2 Acc.=1 Acc.=2 Acc.=3 Acc.=4 Acc.=5 Perceived costs 519.441 402.866 335.878 324.316 250.654 Actual costs 216.931 197.086 175.599 191.394 169.149 Perception gap 302.510 205.780 160.279 132.921 81.505 Overestimation 0.713 0.653 0.631 0.572 0.579 Observations 540 343 320 320 107 (1) (2) (3) (4) (5) Panel B: 200 e/t CO2 Acc.=1 Acc.=2 Acc.=3 Acc.=4 Acc.=5 Perceived costs 786.310 514.375 475.833 479.536 263.364 Actual costs 949.527 881.252 791.325 834.672 717.996 Perception gap -163.217 -366.877 -315.492 -355.137 -454.632 Overestimation 0.349 0.266 0.274 0.243 0.200 Observations 510 349 317 338 110 Notes: The table presents summary statistics for key variables across different levels of pretreatment acceptance in the current price (Panel A) and projected price (Panel B) experiments. 47 Table A13: Cost perceptions by respondent characteristics (current price experiment) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) No high High Female Male Aged <45 Aged 45+ West East school school No single Single Perceived costs 426.306 382.292* 436.879 384.215* 390.387 473.443** 396.723 409.229 450.504 279.670*** Actual costs 177.805 212.975*** 195.151 197.229 197.977 188.127 182.214 211.014*** 230.443 108.545*** Perception gap 248.501 169.317*** 241.728 186.986** 192.410 285.316*** 214.508 198.215 220.062 171.125* Overestimation 0.688 0.612*** 0.660 0.641 0.631 0.744*** 0.674 0.621** 0.620 0.720*** Observations 764 866 579 1051 1384 246 822 808 1176 454 (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) Non-ruling Ruling Low High No homeHomeNo fossil Fossil No motor Motor party party income income owner owner heating heating vehicles vehicles Perceived costs 460.318 288.309*** 384.450 448.376** 369.389 448.836*** 360.917 413.952 214.275 445.822*** Actual costs 213.217 176.362*** 166.348 270.663*** 130.297 287.122*** 101.489 221.437*** 49.449 229.929*** Perception gap 247.101 111.946*** 218.102 177.713 239.092 161.714*** 259.429 192.515** 164.826 215.893 Overestimation 0.663 0.589*** 0.687 0.552*** 0.743 0.517*** 0.820 0.603*** 0.808 0.611*** Observations 884 450 1159 471 942 688 339 1291 302 1328 Notes: The table presents summary statistics for key variables across respondent characteristics in the current price experiment. We conducted t-tests for equality of means within each demographic subgroup (e.g., female vs. male). * p<0.1, ** p<0.05, *** p<0.01 48 Table A14: Cost perceptions by respondent characteristics (projected price experiment) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) No high High Female Male Aged <45 Aged 45+ West East school school No single Single Perceived costs 574.255 562.479 699.578 493.911*** 549.400 671.177** 544.967 590.914 640.437 403.732*** Actual costs 815.510 907.487*** 918.554 833.889** 862.943 872.400 774.185 954.146*** 1031.309 485.873*** Perception gap -241.255 -345.009** -218.976 -339.977** -313.542 -201.222* -229.218 -363.232*** -390.873 -82.141*** Overestimation 0.323 0.250*** 0.333 0.257*** 0.275 0.335* 0.295 0.274 0.254 0.354*** Observations 761 863 585 1039 1376 248 810 814 1127 497 (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) Non-ruling Ruling Low High No homeHomeNo fossil Fossil No motor Motor party party income income owner owner heating heating vehicles vehicles Perceived costs 668.588 406.331*** 485.687 765.920*** 503.322 666.163*** 526.930 578.351 304.220 627.765*** Actual costs 947.273 798.459*** 690.612 1282.249*** 580.431 1295.383*** 471.656 963.402*** 236.197 1006.726*** Perception gap -278.685 -392.127** -204.925 -516.329*** -77.110 -629.220*** 55.273 -385.052*** 68.023 -378.961*** Overestimation 0.296 0.229*** 0.303 0.241** 0.346 0.191*** 0.520 0.225*** 0.503 0.235*** Observations 864 480 1147 477 979 645 327 1297 300 1324 Notes: The table presents summary statistics for key variables across respondent characteristics in the projected price experiment. We conducted t-tests for equality of means within each demographic subgroup (e.g., female vs. male). * p<0.1, ** p<0.05, *** p<0.01 49 Table A15: Robustness (1) (2) (3) (4) (5) (6) Panel A: SelfFeedback Feedback 45 e/t CO2 Baseline Reweighting monitoring Plausibility interest length T 0.161*** 0.123*** 0.157*** 0.158*** 0.159*** 0.176*** (0.034) (0.037) (0.036) (0.037) (0.035) (0.043) Controls Yes Yes Yes Yes Yes Yes R-squared 0.741 0.741 0.756 0.745 0.742 0.743 Observations 1630 1630 1453 1519 1596 1062 (1) (2) (3) (4) (5) (6) Panel B: SelfFeedback Feedback 200 e/t CO2 Baseline Reweighting monitoring Plausibility interest length T -0.307*** -0.304*** -0.298*** -0.311*** -0.307*** -0.311*** (0.035) (0.038) (0.036) (0.036) (0.035) (0.044) Controls Yes Yes Yes Yes Yes Yes R-squared 0.712 0.712 0.720 0.720 0.712 0.702 Observations 1624 1624 1451 1497 1576 1058 Notes: The table presents estimation results from OLS regressions. Results for the current price experiment are shown in Panel A and results for the projected price experiment are shown in Panel B. Columns 1 and 2 show the results for the unweighted and reweighted sample and Columns 3 to 6 show the results for various subsamples. In Column 3, we exclude respondents with high levels of self-monitoring. For this purpose, we construct an index based on four items from the German version of the self-monitoring scale by Schyns and Paul (2002) and exclude the top decile. In Column 4, we exclude respondents that find the provided information unplausible. In Column 5, we exclude respondents that find the survey (rather) uninteresting. In Column 6, we exclude respondents that find the survey (rather) too long. The dependent variable is the post-treatment acceptance of carbon pricing, measured on a five-point Likert scale (1=very unacceptable to 5=very acceptable). T is a dummy variable indicating that a respondent received personalized information about the costs of carbon pricing. All regressions include the pre-treatment acceptance of carbon pricing and the set of controls described in Appendix Table A5. Robust standard errors are in parentheses. * p<0.1, ** p<0.05, *** p<0.01 50 B. Data appendix B.1. Additional information on the 2022 microcensus The microcensus is Germany’s largest annual general population survey, conducted by the official statistical authorities. It employs a stratified cluster sampling design in which all members of households in randomly selected districts are legally required to participate. With approximately 810,000 respondents, the survey covers roughly 1 percent of Germany’s total population. For our analysis we use the most recent available dataset from 2022. Due to data protection reasons, the Scientific Use File consists of a representative 70 percent subsample, totalling 683,588 individual observations. We restrict the sample to respondents aged 18 to 75, residing at their main residence to match our own survey. This leaves us with 488,363 individual observations in our final dataset. A detailed documentation of the 2022 microcensus is provided by the Federal Statistical Office (2023). B.2. Cost calculation To illustrate the cost calculation method, consider the following example: A family of four rents a 120 m2apartment with an oil heating system for both space heating and hot water supply. The family owns two cars - one with a gasoline engine and one with a diesel engine - and has driven 5,000 kilometers in the past twelve months. To determine total additional annual costs (H6), we first calculate several auxiliary variables (H1-H5). We start by estimating the family’s emissions from space heating and hot water supply (H1). To approximate the family’s energy use for space heating, we multiply the size of their dwelling by the average energy consumption per square meter, as provided by the Environmental-Economic Accounts of the Federal Statistical Office (2022). We obtain the family’s energy use for hot water consumption by multiplying the household size by the average hot water consumption per person also sourced from the Environmental-Economic Accounts. Both values are then multiplied by an energy-source-specific emission coefficient for oil to translate consumption into CO2 emissions in tons. The family’s total emissions 51 from heating and hot water use are therefore given by: H1 = 120 m2×(133 ×0.000266)t CO2 m2+ (4 ×1,280 ×0.000266)t CO2= 5.60728 t CO2 German law (CO2 Cost Allocation Act) dictates that the costs of carbon pricing for emissions from heat consumption in rental housing have to be shared between landlords and tenants. The share of carbon costs borne by the tenant is determined based on the dwelling’s energy efficiency, measured in terms of CO2 emissions per square meter of living space. To adjust tenants’ costs accordingly, we first determine the energy efficiency of each tenant’s dwelling (H2) by dividing total emissions from space heating and hot water supply by the dwelling’s size: H2 = 5.60728 t CO2 120m2= 0.046727 t CO2 m2 As stated by the CO2 Cost Allocation Act, the tenant’s share of CO2 costs (H3) is determined according to the following scale: H3 =                                                          1,if H2 <0.012 0.9,if 0.012 ≤H2 <0.017 0.8,if 0.017 ≤H2 <0.022 0.7,if 0.022 ≤H2 <0.027 0.6,if 0.027 ≤H2 <0.032 0.5,if 0.032 ≤H2 <0.037 0.4,if 0.037 ≤H2 <0.042 0.3,if 0.042 ≤H2 <0.047 0.2,if 0.047 ≤H2 <0.052 0.05,if H2 ≥0.052 Because of the low energy efficiency of our exemplary family’s housing, they only bear the costs for 30 percent of total emissions from space heating and hot water supply. Therefore, we calculate the CO2 emissions that the household has to effectively pay for 52 (H4) by multiplying household emissions by 0.3 (H3): H4 = 0.3×5.60728 t CO2= 1.682184 t CO2 Next, we calculate the family’s carbon emission from transportation (H5). We begin by estimating the household’s gasoline and diesel consumption in liters by apportioning total mileage across gasoline and diesel vehicles based on the household’s fleet and applying average diesel and gasoline consumption per kilometer from the “Transport in Figures 2022/2023” report by the Federal Ministry for Digital and Transport (2022). We multiply these averages by fuel-specific CO2 emissions per liter, which are 0.00265 for diesel and 0.00237 for gasoline. For the family owning one diesel and one gasoline car, the formula for total transport emissions for a travelled distance of 5000 kilometers is: H5 = 5000 km ×(0.07 L km ×0.00265 t CO2 L×1 2+ 0.077 L km ×0.00237 t CO2 L×1 2) = 0.919975 t CO2 Finally, we compute total additional current costs of carbon pricing (H6) by aggregating emissions from household heating and transportation and multiplying total emissions by the current carbon price of e45/t CO2:28 H6 = (0.919975 t CO2+ 1.682184 t CO2)× e45 t CO2 =e117.097155 Thus, the family has current additional annual costs of carbon pricing of e117 (rounded to the nearest euro). B.3. Main experiment: Survey questions Q1. Age How old are you? Age in years 28Alternatively, we calculate projected future costs by multiplying total emissions by the projected carbon price of e200/t CO2. Our example family would have future costs of e520. 53 Q2. Gender Please enter your gender: [ ] Male [ ] Female [ ] Diverse Q3. State In which federal state do you live? [ ] Baden-W¨urttemberg [ ] Bavaria [ ] Berlin [ ] Brandenburg [ ] Bremen [ ] Hamburg [ ] Hesse [ ] Mecklenburg-Western Pomerania [ ] Lower Saxony [ ] North Rhine-Westphalia [ ] Rhineland-Palatinate [ ] Saarland [ ] Saxony [ ] Saxony-Anhalt [ ] Schleswig-Holstein [ ] Thuringia [ ] I do not live in Germany Q4. Household income What is your household’s current total monthly net income? This refers to the total amount from wages, salaries, income from self-employment, retirement pensions or civil service pensions, each after deducting taxes and social security contributions. Please also include income from public assistance, rental or lease income, housing benefits, child benefits, and any other sources of income. A household is defined as people who live together and share finances, that is, they cover daily living expenses together and do not account for their purchases separately. If you don’t know the exact amount, please provide an estimate. [ ] Less than e500 [ ] e500 to less than e1,000 [ ] e1,000 to less than e1,500 [ ] e1,500 to less than e2,000 [ ] e2,000 to less than e2,500 [ ] e2,500 to less than e3,000 [ ] e3,000 to less than e3,500 [ ] e3,500 to less than e4,000 [ ] e4,000 to less than e4,500 54 61 62 B.4.2. Projected price experiment 63 64 References Appendix Federal Ministry for Digital and Transport (2022), ‘Verkehr in Zahlen 2022/2023’. Federal Statistical Office (2022), ‘Umwelt¨okonomische Gesamtrechnungen - Private Haushalte und Umwelt - Berichtszeitraum 2000-2020’. Federal Statistical Office (2023), ‘Qualit¨atsbericht Mikrozensus 2022’. 65 Imprint Publisher Macroeconomic Policy Institute (IMK) of Hans-Böckler-Foundation, Georg-Glock-Str. 18, 40474 Düsseldorf, Germany, phone +49 211 7778-312, email [email protected] IMK Working Paper is an irregular online publication series available at: https://www.imk-boeckler.de/de/imk-working-paper-15378.htm The views expressed in this paper do not necessarily reflect those of the IMK or the Hans-Böckler-Foundation. ISSN 1861-2199 This publication is licensed under the Creative commons license: Attribution 4.0 International (CC BY). Provided that the author's name is acknowledged, this license permits the editing, reproduction and distribution of the material in any format or medium for any purpose, including commercial use. The complete license text can be found here: https://creativecommons.org/licenses/by/4.0/legalcode The terms of the Creative Commons License apply to original material only. The re-use of material from other sources (marked with source) such as graphs, tables, photos and texts may require further permission from the copyright holder.