Policy misperceptions, information, and the demand for redistributive tax reform: Experimental evidence from Latin American countries

Ardanaz, Martin; Hübscher, Evelyne; Keefer, Philip; Sattler, Thomas Working Paper Policy misperceptions, information, and the demand for redistributive tax reform: Experimental evidence from Latin American countries IDB Working Paper Series, No. IDB-WP-1385 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Ardanaz, Martin; Hübscher, Evelyne; Keefer, Philip; Sattler, Thomas (2022) : Policy misperceptions, information, and the demand for redistributive tax reform: Experimental evidence from Latin American countries, IDB Working Paper Series, No. IDB-WP-1385, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0004655 This Version is available at: https://hdl.handle.net/10419/289980 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-nc-nd/3.0/igo/legalcode IDB WORKING PAPER SERIES Nº IDB-WP-1385 Policy Misperceptions, Information, and the Demand for Redistributive Tax Reform: Experimental Evidence from Latin American Countries Martin Ardanaz Evelyne Hübscher Philip Keefer Thomas Sattler Inter-American Development Bank Institutions for Development Sector Fiscal Managment Division December 2022 December 2022 Policy Misperceptions, Information, and the Demand for Redistributive Tax Reform: Experimental Evidence from Latin American Countries Martin Ardanaz (Inter-American Development Bank) Evelyne Hübscher (Central European University) Philip Keefer (Inter-American Development Bank) Thomas Sattler (University of Geneva) Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 AttributionNonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/ legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this CC-IGO license. Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised version of this work may also be reproduced in any academic journal, including those indexed by the American Economic Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. Therefore, the restriction to receive income from such publication shall only extend to the publication's author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives license and these statements, the latter shall prevail. Note that the link provided above includes additional terms and conditions of the license. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Policy misperceptions, information, and the demand for redistributive tax reform: experimental evidence from Latin American countries / Martin Ardanaz, Evelyne Hübscher, Philip Keefer, Thomas Sattler. p. cm. — (IDB Working Paper Series ; 1385) Includes bibliographic references. 1. Tax auditing-Latin America. 2. Value-added tax-Latin America. 3. Fiscal policy-Latin America. 4. Taxation-Latin America. 5. Income distribution-Latin America. 6. Economic surveys-Latin America. I. Ardanaz, Martin. II. Hübscher, Evelyne, 1975III. Keefer, Philip. IV. Sattler, Thomas. V. Inter-American Development Bank. Fiscal Management Division. VI. Series. IDB-WP-1385 http://www.iadb.org Copyright © 2022 Policy Misperceptions, Information, and the Demand for Redistributive Tax Reform: Experimental Evidence from Latin American Countries∗ Martin Ardanaz (Inter-American Development Bank) Evelyne H¨ubscher (Central European University) Philip Keefer (Inter-American Development Bank) Thomas Sattler (University of Geneva) Abstract Why do individuals’ preferences for redistribution often diverge widely from their material self-interest? Using an original online survey experiment spanning eight countries and 12,000 respondents across Latin America, one of the most unequal regions in the world, we find significant evidence for an under-explored explanation: misconceptions regarding the distributional effects of current tax policy. Treated respondents who are informed that an increase in the value added tax (VAT) is regressive are significantly more likely to prefer policy reforms that make the tax more progressive. Treatment effects are driven by the large fraction of respondents who underestimate the regressivity of the VAT, even though their misperceptions are linked to fundamental views about the world. These respondents are disproportionately right-leaning and more likely to attribute success to individual effort than luck. Despite the deep-rooted nature of respondents’misperceptions, treatment effects are largest among individuals who hold these views of the world. These findings contribute both to understanding the political economy of redistribution and the potential for information interventions to shift support for fiscal adjustment policies protecting the most vulnerable. JEL Classification: D72, D90, H20, H30 Keywords: taxes, redistribution, survey experiment ∗Ardanaz and Keefer: Inter-American Development Bank, [email protected] and pk[email protected]. H¨ubscher: Central European University, huebsc[email protected]. Sattler: University of Geneva, [email protected]h. This research significantly benefited from the comments of Per Andersson, Guillermo Cruces, and Ricardo Perez Truglia. We are extremely grateful to LAPOP for its administration of the survey and, particularly, to Oscar Castorena. We are indebted to the extraordinary research assistance of Miguel Purroy. The findings and interpretations in this paper are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank or the governments it represents. 1 1 Introduction A large literature seeks to understand why individuals’ support for redistributive tax policies often diverges from their material self-interest. Previous research examines this puzzle in the United States and other advanced economies, but it is at least as profound in other regions, such as Latin America, where the challenge of inequality looms large. Two prominent lines of inquiry focus on voter ignorance regarding their place in the income distribution and the extent of inequality in society (see especially Stantcheva (2021) for the most recent advances). Slemrod (2006) though, observed that individuals also have significant misconceptions regarding the incidence of current tax policies and asks whether these misconceptions have a causal effect on preferences for progressive tax reform. Are individuals who incorrectly believe that a tax is progressive less supportive of reforms to make the tax more progressive? Is information about the distributional impact of tax policy effective at shifting citizen support for progressive reforms? We address these questions with experimental data from an original online survey spanning eight countries and 12,000 respondents across Latin America. The analysis addresses two gaps in understanding of the cognitive processes that shape voter support for economic policy reforms. One concerns the informational obstacles that voters must surmount before they are willing to support a specific policy reform. The other relates to the sources of policy misperceptions and how they can be corrected. Voter support for a specific policy reform, such as one to make the tax code more progressive, depends on whether they have sufficient information to convince them that a problem exists, but also to convince them that a specific policy reform will solve the problem and make them better off. Their support for a specific solution should therefore depend on their beliefs about its effectiveness, even if they are fully convinced of the problem and the need for a policy response of some kind.1 1For example, if they believe that tax rates on the richest are already high, they are more likely to believe that the redistributive benefits of more progressive rates are outweighed by other negative effects that they might associate with higher tax rates. Moreover, individuals, experts and non-experts alike, may struggle to draw conclusions about the effects of large and unusual changes in public policy on the income distribution. They are likely to be more certain estimating the effects of more incremental and common policy shifts. 2 Previous research demonstrates that individuals often underestimate the magnitude of inequality and that this affects their support for redistributive reforms in general. For example, Stantcheva (2021) and Kuziemko et al. (2015) demonstrate with evidence from the United States that when individuals are made aware of the problem with information about inequality in society and of their own position on the income distribution, they are significantly more likely to support greater redistribution in general. However, this same information has smaller and more ambiguous effects on their support for specific progressive tax reforms. We resolve this ambiguity by providing individuals with information that is relevant to their evaluation of the effectiveness of a specific reform to make the value added tax (VAT) more redistributive, by describing the incidence of the current VAT. Individuals who receive information about the regressivity of the current VAT are far more likely to support progressive tax reform that exempts poorer deciles from paying the VAT. The magnitude of the effect is similar to that uncovered by Kuziemko et al. (2015) when looking at the shift in support for greater redistribution in general among those who are informed that inequality is a significant problem. Treatment effects are driven by individuals who have misconceptions about the incidence of the VAT that lead them to believe that the VAT is already progressive, or that make them uncertain about the incidence of the VAT. The other gap in understanding cognitive processes relates to the sources of policy misperceptions. Misconceptions can be rooted in factual knowledge gaps, but also in ideology-driven beliefs about the way the world operates. Prior research suggests that misconceptions that are rooted in ideology may be difficult to shift with simple information treatments (Luttmer and Singhal, 2011). However, not all ideological beliefs are equally deeply held and it may be that some, which turn out to be important for public policy preferences, are more susceptible to revision when new information is presented. In fact, our evidence on mechanisms indicates that policy misconceptions among respondents seem to be tied more strongly to ideologically rooted beliefs than to gaps in factual knowledge. Nevertheless, a relatively 3 simple information intervention has a significantly stronger effect on the policy preferences of these respondents. In the experiment, respondents are asked to elicit preferences over three possible reforms to the VAT: one that exempts no poor households from the tax increase, one that exempts the poorest 30 percent of households and asks the remaining households to pay more, or one that exempts the poorest 50 percent of households, raising taxes even more on the remaining households. These options are carefully constructed to hold constant the amount of tax revenue that each option collects. Treated respondents receive accurate information about how much more poorer households in the region pay as a share of their income in VAT payments compared to richer households. The data also permit us to analyze mechanisms in detail. We identify which respondents have misconceptions about the incidence of the VAT (that is, those who incorrectly believe it is progressive). We can also identify the characteristics of respondents who believe that the VAT is flat or progressive and of those who believe it is regressive. Their main distinguishing features are linked to their views about the world, reflected in their political ideology and other beliefs. For example, right-leaning respondents are more likely to hold the incorrect belief that richer households devote the same or a larger share of their income than the poor to VAT payments. The experiment yields three main results. First, learning about the regressivity of current tax policy has a large impact on support for progressive tax reform. Second, treatment effects are much stronger among those who incorrectly perceive the progressivity of current tax policy. Third, policy misperceptions are greatest among those who categorize themselves as right-leaning; treatment effects are significantly stronger among this group. These results add to the findings of a rich literature examining preferences for redistribution. Stantcheva (2021) find that informing individuals about the severity of inequality in society increases support for redistributive tax reforms in general.2Kuziemko et al. (2015) 2Fehr, Mollerstrom and Perez-Truglia (2022) report results from a two-year survey experiment in Germany that correcting respondents systematic under-estimation of their true place in the world’s income 4 examine a multi-dimensional information treatment that, among other things, informs respondents about their income relative to others.3Though the treatments have a statistically and economically large effect on preferences for redistribution in general, their impact on preferences for specific progressive tax reforms is substantively small, about one-tenth of the difference between the preferences of liberal and conservative respondents.4Our treatment, information on the incidence of the current VAT, has as large an effect on respondent preferences for a specific progressive tax reform as the effects that Stantcheva (2021) and Kuziemko et al. (2015) estimate when examining the effects of information about inequality on preferences for redistribution in general.5 Both Bartels (2005) and Slemrod (2006) observed widespread misconceptions regarding the progressivity of specific tax policies in the United States. Slemrod (2006) analyzes data from a large survey of Americans and finds that many believed that a new sales or flat tax would be more progressive than the current income tax and those who held this belief were significantly more likely to support the sales/flat tax alternatives. Slemrod (2006) closes with a question for future research to which we respond: whether these misconceptions are distribution does not affect support for policies related to global inequality. 3Their work builds on earlier contributions. For example, in Cruces, Perez Truglia and Tetaz (2013) and Fernandez-Albertos and Kuo (2018), individuals who are told that their relative income is lower than they believed demand more redistribution. Similarly, in Karadja, Mollerstrom and Seim (2017) individuals who learn they are richer relative to others demand less redistribution. In contrast to these findings, Hoy and Mager (2021) examine data from a survey experiment involving 10,000 participants in ten middle and upperincome countries and conclude that informing individuals that their position in the income distribution is lower than they thought does not increase support for redistribution. 431.1 percent of treated respondents, versus 30.21 control respondents prefer higher tax rates on the richest 1 percent; 79 percent of treated respondents, versus 74 percent of control respondents, prefer higher tax rates on millionaires. 5In contrast to our findings, Douenne and Fabre (2022) find that respondents do not change their perceptions of the progressivity of a carbon tax policy when told “this reform would increase the purchasing power of the poorest households and decrease that of the richest”. We attribute the difference to greater familiarity with the VAT, the complexity surrounding the incidence of a carbon tax, and the fact that our information treatment, though substantially lighter than others in the literature, is more detailed and entails more comprehension checks than their treatment. Numerous studies find strong effects of similar information treatments on estate taxes, which evidently do not generalize to other taxes with broader incidence. Bastani and Waldenstrom (2021) find that information about the aggregate importance of inherited wealth and its implications for the inequality opportunity in Sweden leads to a significant increase in support for estate taxation among Swedish respondents. Kuziemko et al. (2015) and Sides (2016) also find a dramatic effect of their information treatment on preferences for a higher estate tax. In Kuziemko et al. (2015), more than 50 percent of treated respondents prefer it compared to 17 percent of control respondents. We focus on a different and fiscally more important tax with much broader incidence than the estate tax. 5 When no decile is exempted, the All Pay reform option raises total VAT tax revenues by approximately 4 percent.19 For each of the other two options, Top 70% Pay and Top 50% Pay, we then calculate the additional amount by which the VAT tax rate would have to rise to ensure that total VAT tax receipts to the government still rise by 4 percent after the bottom three (“70% Pay”) or bottom five deciles (“50% Pay”) are exempted.20 To increase the salience of the policy options and reduce the cognitive burden on respondents, we tell them how much more households will pay in taxes under the new rates, not the new rates themselves. That is, respondents see the percentage increase in monthly VAT payments for an average household under each policy option. The increased VAT payments are simply the product of the tax rate established for each policy option and total household consumption in all non-exempt deciles. Our vignettes inform respondents of these increases: 5% in the case of the Top 70% Pay option, and 6% in the Top 50% Pay option. All respondents see the three policy alternatives and are then asked to evaluate three pairwise comparisons: Option 1 against Option 2; Option 1 against Option 3; Option 2 against Option 3. The order in which respondents see these vignettes is randomized. In each comparison, respondents indicate on a 5-point scale if they are more likely to vote for the government if the government implemented Option xvs. Option y. A value of 1 indicates most support for Option x, 5 indicates most support for Option y, and 3 indicates that the respondent is indifferent between Option xand Option y. After respondents completed these comparisons, they are guided to a new screen that asks them to choose their most preferred option among the following four alternatives: Option 1, Option 2, Option 3, or an additional Option 4, which suggests that the government does not adjust the VAT to address the fiscal crisis. mention reduced rates. Our policy options incorporate only increases in the standard VAT rate, not the reduced rates. 19While in practice the exact number varies from country to country depending on their prior VAT tax rate and base, the cross-country variation is trivial, lending credence to the use of a common figure across the experiment. 20As is standard in this type of analysis, we assume that household consumption is inelastic with respect to the changes in the VAT tax rate and that households bear the full burden of the tax (see Lustig (2018) and IDB (2022)). 12 3.2 Information treatment The hypothesis motivating the study is that individuals’ reluctance to embrace progressive tax reforms can be traced to their uncertainty about the progressivity of the existing tax system. Therefore, before choosing between the policy options, the survey includes an information treatment that tells respondents randomly assigned to the treatment group the distributive impact of the VAT in a typical Latin American country. To build the treatment, we again use the tax incidence analysis described above providing information on the fraction of monthly income devoted to VAT payments by different income groups. Thus, our treatment involves telling respondents about the incidence of the VAT across the income distribution and, specifically, that lower income households devote a higher share of their income to VAT payments than higher income households, as shown by Figure 1. We highlight the fact that the poorest households pay up to 23% of their income on the VAT and the richest households pay only 11% of their income on VAT payments. To increase salience and comprehension, the information is presented both verbally and graphically.21 Figure 1: Impact of VAT on income deciles; averages across selected countries in Latin America 0 .05 .1 .15 .2 .25 Share of income paid on VAT 1 2 3 4 5 6 7 8 9 10 Income decile 21Appendix Figure A1 shows a screenshot from the survey with the actual graph shown to the treatment group. In order to generate a common treatment across countries, the figure presents average values across the countries in our sample, thus representing the distributive impact of VAT in a typical or representative country in Latin America. 13 Treatment length or content could affect attrition in the treatment group relative to the control group, potentially biasing responses. However, attrition was nearly the same in treatment and control groups, on average 6.2% of respondents in the control group and 6.6% in the treatment group. The difference is entirely insignificant, whether or not we control for country fixed effects and respondent characteristics (see Table 1).22 Table 1: Survey attrition rate by treatment status Outcome: D(Dropped out from the survey) (1) (2) (3) Treated 0.004 0.004 0.001 (0.006) (0.006) (0.003) Observations 12,985 12,985 12,371 Country Fixed Effects No Yes Yes Individual controls No No Yes Mean Dep Var (Control) .0622 .0622 .0171 Notes: This table presents the estimates of the effect of being treated on survey attrition. The outcome is a dummy taking the value of 1 if the individual dropped out from the survey. Individual controls include sex, age, household size, the education level, the income level, employment and formality status, whether the participant is retired, and a subsidies reception dummy. Standard errors are clustered at the country level. 3.3 Control variables The survey collected a wide range of household and respondent characteristics that might influence their support for more progressive tax reform. These included basic data about education, age, gender, household size, and employment status. In addition, respondents provided information that allow us to place their actual and perceived location in the income distribution, as well as their attitudes on key issues. These are all balanced across treatment and control groups. We include them to identify empirical regularities in the data that link 22In their work, using more intensive treatments, Kuziemko et al. (2015) and Stantcheva (2021) experienced overall attrition rates of 15% and 19%-20%, respectively. In Kuziemko et al. (2015), treated individuals were 11.3 percentage points less likely to finish the survey and in Stantcheva (2021) the treated were between two and six percentage points less likely. In our experiment, the treated are 0.4 percentage points less likely. 14 this research to prior work, yield surprising new regularities, or, most importantly, help to estimate heterogeneous treatment effects that are useful to explore mechanisms. Actual and perceived position in the income distribution. Substantial theoretical and empirical attention has been given to household income as a determinant of redistributive preferences. If respondents are only motivated by their material self-interest, respondents in the top half of the income distribution should prefer the All Pay option over the other two; those in the fourth or fifth deciles should prefer the Top 50% Pay option over the other two; and those in the first, second or third deciles should prefer either of the redistributive options over the first. Two questions capture households’ actual and perceived location in the income distribution. First, prior to entering the VAT portion of the survey, respondents were asked to imagine a staircase with ten steps, with the poorest located on the first step and the richest on the tenth step. Their self-location on the staircase constitutes their perceived location in the income distribution. Second, we derived their actual location by asking them, at the end of the survey, for their household income. Specifically, we computed the thresholds for each income decile in the survey countries using Latin American household survey data from Soci´ometro-IDB and SEDLAC. Respondents were asked to place themselves in one of the 10 income categories.23 The 10-step scales allowed us to group respondents into the three income groups that are relevant for the fiscal policy questions: the lower 30% group including respondents who classify themselves into the first three income percentiles; the middle 40-50% group including respondents who classify themselves into the fourth and fifth income percentiles; and the top 60% group including respondents who classify themselves into sixth through the tenth percentiles. Tax incidence misperceptions. We expect treatment effects to be strongest among those who have incorrect perceptions of the incidence of tax policy. Therefore, before introducing 23The distribution of respondents for these two variables is in Appendix B, Figure B1. 15 our experiment, we asked respondents whether they believe rich households spend a higher, the same, or lower fraction of their income on VAT compared to poor households.24 Only 35 percent of our sample is aware that poor households tend to pay a higher fraction of their income in VAT than the rich.25 Attitudinal controls. The survey collected additional information that is particularly useful for understanding mechanisms. We asked respondents where they located themselves ideologically, on a 10 point scale from left to right; right-leaning respondents were significantly more likely to underestimate the regressivity of the VAT. Individuals’ support for redistribution can also depend on whether they believe that success in life depends on one’s own efforts. Those who believe this is the case turn out to have the misconception that the VAT is progressive.26 Beliefs about the potential for upward mobility in society might also affect support for progressive tax reform, and perceptions about the progressivity of the VAT. Respondents therefore indicated which of four statements they most agreed with, from “almost all children from poor households have the same opportunities as children from rich households” to “almost no child from a poor household has the same opportunities as children from rich households.” Again, these beliefs are strongly associated with misconceptions about the incidence of the VAT. 24The exact question wording is as follows: “Over the course of a year, all households will have dedicated a certain percentage of their income to paying VAT for the goods and services they purchased. What do you think is the percentage of income paid by poor households and rich households on VAT? Do you think that rich households spend a higher percentage of their income paying VAT, or a lower percentage, compared to poor households?” Respondents choose one of five possible answers: 1) Rich households spend a much higher percentage of their income in VAT payments; 2) Rich households spend a higher percentage; 3) Rich and poor households spend about the same percentage; 4) Poor households spend a higher percentage; and 5) Poor households spend a much higher percentage of their income in VAT payments. 25Krupnikov et al. (2006) argue that survey data on respondents’ factual knowledge, in their case knowledge of the incidence of the estate tax, likely underestimate knowledge. When they offer one dollar to respondents who correctly answer the question about estate tax incidence, the number of correct responses increase by more than 30 percent. Although we do not reward respondents for correct answers on the incidence of the VAT, the possible inaccuracy of responses does not bias our experimental results. Random assignment of respondents into treatment and control groups ensures balance in the number of respondents who accurately and inaccurately respond to the question regarding VAT tax incidence. 26The exact question wording is as follows: “With which of statements A or B are you more in agreement? A. People’s incomes are the product of their individual efforts; or B. People’s incomes are the product of factors outside of their control?” 16 Respondents had an opportunity to indicate which two potential problems in their country, out of a list of 14, most concerned them. The list was randomly reordered for each respondent. From these choices, we constructed a dummy variable to indicate those respondents who were most concerned about inequality and poverty. They were more in favor of progressive tax reform, and more likely to say that the VAT was regressive. Finally, the survey also asked various questions related to trust in others and in government. These did not have systematic effects on either preferences for tax reform nor misconceptions regarding the incidence of the VAT. 4 Empirical Strategy We examine whether the information treatment has a significant effect on VAT reform preferences by estimating an empirical specification with the following form: yic =αc+θ1Treatedi+θ2Xi+εic (1) The variable yic captures the VAT reform options preferred by respondent iin country c. There are different versions of the variable, capturing respondent preferences across binary comparisons of the three options All Pay,70% Pay, or 50% Pay.T reatediis an indicator variable that equals 1 if the respondent ireceived the information treatment, and 0 otherwise. The coefficient of interest, θ1, captures the average differential change between those who received the information treatment and those who did not. We include a complete set of country fixed effects αcto control for any source of crosscountry heterogeneity. The term Xiin equation (1) represents control variables. The group of basic socioeconomic controls consists of the respondent’s actual position in the income distribution, education, age, gender, employment status, whether the worker is informal or retired, whether the respondent receives any government subsidy, and household size. Some specifications also control for respondents’ attitudes: their perceived location in the income 17 distribution, whether they consider inequality and poverty as main problems in their country, trust in the current government, beliefs about the determinants of economic success (luck vs. effort), beliefs about the life opportunities of poor children, previous knowledge about who decides tax policy, and political alignment (left vs. right ideological dimensions). Finally, εic is the clustered error term that allows correlation within countries. To investigate the heterogeneity of the information treatment based on respondent’s characteristics, we use an augmented version of the main specification, equation (1). We estimate the following equation: yic =αc+θ1Treatedi+θ2(T reatedi×Zi) + θ3Zi+µic, (2) The coefficient of interest in this equation is θ2, which captures the differential effect on tax policy preferences of those who received the information treatment that also share characteristic Zi. 5 Results 5.1 Graphical evidence Before examining respondent support for redistribution we confirm in Figure C1 in Appendix C that the treatment and control groups are balanced with respect to all observable variables. This is unsurprising given the random assignment of respondents to treatment and control groups. The analysis provides reassurance that the two groups are likely to be balanced as well with respect to unobservable characteristics. Simple comparisons of respondent support for the various reform options reveals a significant preference for the redistributive over the non-redistributive reform options. This preference is stronger among treated respondents. Figure 2 describes these differences. Subfigure 2(a) shows that more respondents favor the options that include a compensation 18 component, Options 2 (70% Pay) and 3 (50% Pay), than Option 1 (All Pay). More specifically, respondents first made a pairwise comparison between Option 1, All Pay and Option 2, 70% Pay, on the 5-point scale described in the previous section, where 1 or 2 expressed support for the first option, 4 or 5 for the second, and 3 reflects indifference. The columns Support for O1/O2are the shares of respondents who prefer All Pay or 70% Pay, respectively. The shares for the other two pairwise comparisons are computed similarly. The fraction of respondents who prefer the more redistributive options over the non-redistributive option is between 10 and 15 percentage points. When respondents choose between the two redistributive options, they slightly favor Option 2, which exempts the bottom 30%, over Option 3, which exempts the bottom 50% of the income distribution. Between 20% and 25% of the respondent are indifferent between the various options. Respondents then indicated their preferred policy from among any of the three options plus the added option of no VAT reform at all, despite the fiscal stringencies that the government confronts. Subfigure 2(b) reports the responses to this question. Option 1, All Pay, receives less support than the other two options. The most redistributive option, Option 3, 50% Pay, receives more overall support than all other options, including 70% Pay, which differs from the pairwise comparisons in subfigure 2(a). We also find that only a small share of respondents, about 7%, favors no fiscal adjustment.27 27We do not attach a strong interpretation to the fact that more respondents prefer an uncompensated tax hike to doing nothing. It is possible that the framing of the vignette, emphasizing that serious fiscal problems threaten economic stability, employment, and family incomes, could account for weak support for no action. Experimenter demand, though, is another plausible explanation, since all of the focus of the section is on VAT policy changes. 19 Figure 2: Support for fiscal adjustment options, in % (a) Pairwise comparisons 0 5 10 15 20 25 30 35 40 45 Support for All Pay Indifferent Support for 70% Pay All vs. 70% Pay 0 5 10 15 20 25 30 35 40 45 Support for 70% Pay Indifferent Support for 50% Pay 70% vs. 50% Pay 0 5 10 15 20 25 30 35 40 45 Support for All Pay Indifferent Support for 50% Pay All vs. 50% Pay Untreated Treated (b) Direct comparison of all options 0 5 10 15 20 25 30 35 All Pay 70% Pay 50% Pay No adjustment Untreated Treated All options Figure 2 also summarizes how the information treatment, which manipulates respondents’ knowledge about the regressive impact of the VAT, affects their policy preferences. It compares the average policy preference for the treated (red bar) with those for the nontreated (gray bar). For all three pairwise comparisons, information about the regressive impact of the VAT increases support for a policy that compensates citizens in lower income brackets. The increase in support is most pronounced for Option 3, 50% Pay which proposes 20 an increase in VAT of 6% for citizens who belong to the upper half (top 50%) of the income distribution while exempting the bottom half. This result is consistent across both outcome variables in panels (a) and (b) of Figure 2. The share of respondents that are indifferent remains almost identical between treated and untreated respondents. 5.2 Regression results: average treatment effects Table 2 further examines the treatment effect using a series of OLS regression models with country fixed effects.28 We use different outcome variables for the analysis of the pairwise comparisons in columns (1) to (6): the original, 5-scale categorical variable where higher values indicate greater support for the option mentioned first in the top row29 and a dummy version that takes the value 1 if the respondent supports the option mentioned first in the top row. In columns (7) to (10), the outcome variables are dummy variables that take the value 1 if the respondent chose the option listed on top of the column and 0 otherwise. Finally, Panel A simply regresses the respective outcome variable on the treatment dummy; Panel B does the same but includes a series of socioeconomic control variables; and Panel C includes variables capturing a respondent’s subjective perceptions, beliefs and knowledge in addition to the socioeconomic controls.30 This is a version of the table just above, but with stars indicating statistical significance The results confirm the graphical analysis of Figure 2. The information treatment has a consistent and statistically significant impact on support for the different adjustment options. The negative signs on the coefficients in the Treated row of 2 indicate that respondents who were informed about the regressive impact of the VAT are less likely to choose Option 1, All Pay, over Option 2, 70% Pay or over Option 3, 50% Pay (columns (1)-(2) and (5)- (6)). In column 6, information reduces support for the least redistributive option by 3.4 28The summary statistics of all relevant variables are in Appendix Table B1. 29For example, for the comparison of ‘Option1to2’, higher values indicate greater support for Option 1. 30The results for the control variables are in the Appendix, Table C1. 21 Table 3: Heterogenous effect - Misperception of VAT incidence (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Preferred Choice Cat D Cat D Cat D All Pay 70% Pay 50% Pay No Adj Treated * PerceptVAT -0.098 -0.028 -0.097** -0.043 -0.111** -0.040** -0.028 0.008 0.022 -0.002 (0.061) (0.020) (0.038) (0.023) (0.042) (0.017) (0.017) (0.019) (0.021) (0.012) Treated 0.014 0.001 0.013 0.010 -0.040 -0.009 -0.021 0.016 0.011 -0.007 (0.048) (0.015) (0.023) (0.013) (0.041) (0.015) (0.019) (0.016) (0.015) (0.009) PerceptVAT 0.185** 0.053** 0.194*** 0.055** 0.220*** 0.066*** 0.065*** 0.008 -0.068** -0.005 (0.066) (0.018) (0.047) (0.022) (0.044) (0.009) (0.013) (0.014) (0.021) (0.010) Constant 2.811*** 0.291*** 2.966*** 0.362*** 2.807*** 0.301*** 0.255*** 0.292*** 0.370*** 0.083*** (0.044) (0.012) (0.030) (0.012) (0.031) (0.007) (0.008) (0.008) (0.015) (0.006) Observations 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 R-squared 0.014 0.013 0.008 0.009 0.019 0.016 0.018 0.005 0.011 0.005 Mean Dep. Var. 3.094 0.317 2.934 0.389 3.107 0.326 0.277 0.308 0.338 0.076 Mean Dep. Var. (control) 2.931 0.326 3.091 0.398 2.949 0.344 0.297 0.297 0.326 0.080 Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Notes: This table presents the results of heterogeneous treatment effects. “All vs. 70% Pay” measures respondents’ preferences for option 1 over option 2. Cat means that it is the categorical measure, which takes values from 1 to 5, where 5 is a greater preference for option 1. Dis an indicator variable that takes values of 1 if option 1 was preferred, and 0 otherwise. Same logic for the dependent variables in columns 3 to 6. All Pay in column 7 is an indicator variable that takes values of 1 if option 1 was chosen, and 0 otherwise. The same logic for columns 8, 9 and 10. All Pay being the least redistributive option and 50% Pay the most redistributive. No Adj means no action. Treated is an indicator variable that takes values of 1 if it received the information treatment, and 0 otherwise. PerceptVAT is an indicator variable equal to 1 if respondent believes that rich households spend a higher or similar percentage of their income on VAT payments relative to poor households, and 0 otherwise. Clustered standard errors at the country level are reported in parentheses. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1% level. significant. For those who (incorrectly) believe that rich people pay a greater share of their income on VAT, the point estimate is considerably larger and the 95% confidence interval does not span the zero line. In short, the treatment effect is driven by misinformed people who change their attitude towards compensated fiscal adjustment when they learn that their beliefs about the incidence of the VAT across income groups are wrong. 28 Figure 3: Treatment Effects by Perception about VAT Incidence -.2 -.15 -.1 -.05 0 .05 Treatment effect Correct Incorrect Perception VAT (a) All vs. 50% Pay -.06 -.04 -.02 0 .02 Treatment effect Correct Incorrect Perception VAT (b) D(All vs. 50% Pay) Notes: This figure presents the impact of the information treatment on support for the more redistributive option 3, 50% Pay, conditional on the respondent’s perception of the impact of the VAT. The results are based on models (5) and (6) in Table 3. “All vs. 50% Pay” is a categorical variable that measures preferences for option 1 over option 3. D(All vs. 50% Pay) is an indicator variable that takes values of 1 if the respondent prefers option 1 more, and 0 if he/she prefers option 3 more. Perception VAT is an indicator variable that takes the value 1 if the respondent (incorrectly) believes that richer households spend the same or a higher percentage of their income on VAT relative to poorer households, and 0 otherwise. 6.2 Drivers of misperceptions In the next step, we examine the determinants of tax incidence misperceptions. Table 4 examines to what extent these misperceptions correlate with cognitive and ideological factors, including education, political alignment (left vs. right), concerns about inequality, and beliefs about the sources of poverty and individual success. Beliefs and ideology, rather than cognitive factors (education), appear to drive misperceptions. The positive coefficient on political alignment in Table 4 shows that people on the center/right tend to be more likely to believe the rich pay more than the poor. In contrast, people who think inequality and poverty are a significant concern are more likely to believe that the poor pay more than the rich in VAT, as the negative coefficient on concern for inequality in Table 4 shows. We would therefore expect that the information treatment should mostly affect people on the center and right of the ideological spectrum since the perceptions of VAT incidence of those on the left are more consistent with its actual distributive impact. Similarly, the treatment should affect respondents who believe that inequality is the main problem less than those 29 who believe other issues to be more salient. This is because the perceptions of the latter about the VAT impact is less accurate than the perceptions of the former. Table 4: Determinants of VAT misperceptions (1) (2) (3) (4) (5) Perception of VAT impact Educated 0.005 -0.009 -0.007 -0.015 -0.015 (0.016) (0.016) (0.015) (0.015) (0.015) PoliticalAlign 0.027*** 0.027*** 0.026*** 0.026*** 0.022*** (0.004) (0.004) (0.004) (0.004) (0.003) Bottom30Actual -0.053*** -0.036** -0.043** (0.014) (0.014) (0.013) B40and50Actual -0.005 0.002 -0.001 (0.016) (0.015) (0.015) Bottom30Perceived -0.095*** -0.083*** -0.070*** (0.009) (0.007) (0.008) B40and50Perceived -0.022 -0.017 -0.012 (0.012) (0.011) (0.010) KnowledgeTaxes -0.001 (0.010) ConcernIneqPov -0.044*** (0.010) TrustGov 0.021 (0.018) BeliefsLuck -0.028*** (0.007) PoorChildOpportunity -0.092*** (0.017) Constant 0.499*** 0.526*** 0.542*** 0.553*** 0.654*** (0.019) (0.021) (0.023) (0.025) (0.028) Observations 12,152 12,152 12,152 12,152 12,152 R-squared 0.039 0.041 0.043 0.044 0.054 Mean Dep. Var. 0.580 0.580 0.580 0.580 0.580 Country FE Yes Yes Yes Yes Yes Notes: This table presents the determinants of respondents’ misperceptions about VAT. BottomXXActual and BottomXXPerceived identify respondents whose reported income puts them in the bottom XXth percentile and whose perceived income puts them in the bottom XXth percentile, respectively. Clustered standard errors at the country level are reported in parentheses. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1% level. This is what Figures 4 and 5 show.34 Figure 4 illustrates the impact of the information treatment for respondents who place themselves on different locations on the left-right political dimension. The treatment does not affect respondents on the left: the marginal effect for these respondents is zero, which means that treated and untreated respondents from the 34The figures are based on the results in Appendix C, Tables C4 and C5. 30 left, on average, do not differ when they compare Options 1 and 3. In contrast, the information treatment has a strong effect on respondents on the right; they are correspondingly less likely to select the least redistributive Option 1, All Pay, over the most redistributive Option 3, 50% Pay, when they learn about the regressive impact of the VAT. Figure 4: Treatment effects by political alignment -.3 -.2 -.1 0 .1 Treatment effect 0 1 2 3 4 5 6 7 8 9 10 Political Alignment 0 10 20 30 40 % of observations (a) All vs. 50% Pay -.1 -.05 0 .05 Treatment effect 0 1 2 3 4 5 6 7 8 9 10 Political Alignment 0 10 20 30 40 % of observations (b) D(All vs. 50% Pay) Notes: This figure presents the impact of the information treatment on support for the more redistributive option 3, 50% Pay, conditional on the respondent’s political alignment. The results are based on models (5) and (6) in Table C4. “All vs. 50% Pay” is a categorical variable that measures preferences for option 1 over option 3. D(All vs. 50% Pay) is an indicator variable that takes values of 1 if the respondent prefers option 1 more, and 0 if he/she prefers option 3 more. Political alignment is the respondent’s position on the left-right political dimension. Figure 5 compares treatment effects for respondents with strong and weak concern for inequality and poverty. The figure shows that the treatment effect is not statistically significant for respondents who have a strong concern for inequality. In contrast, it is large for those with a small concern, indicating that respondents who are not much concerned with inequality are less likely to choose Option 1 over Option 3 when they are informed about the regressive impact of VAT. 7 Conclusions and Policy Implications Prior research has found a strong relationship between respondents’ knowledge of their location in the income distribution and support for more redistributive policies in general, but much weaker effects on support for specific measures to redistribute. Our results suggest that 31 Figure 5: Treatment effects by concern about inequality and poverty as problems -.2 -.1 0 .1 Treatment effect Weak Strong Concern Inequality and Poverty (a) All vs. 50% Pay -.06 -.04 -.02 0 .02 Treatment effect Weak Strong Concern Inequality and Poverty (b) D(All vs. 50% Pay) Notes: This figure presents the impact of the information treatment on support for the more redistributive option 3, 50% Pay, conditional on the respondent’s concern for inequality and poverty. The results are based on models (5) and (6) in table 5. “All vs. 50% Pay” is a categorical variable that measures preferences for option 1 over option 3. D(All vs. 50% Pay) is an indicator variable that takes values of 1 if the respondent prefers option 1 more, and 0 if he/she prefers option 3 more. Concern for inequality is an indicator variable that takes the value 1 if the respondent answers that s/he is strongly concerned about inequality and poverty, and 0 otherwise. misperceptions about the distributional incidence of different policy measures could account for this. In theory, individuals might reasonably have a difficult time inferring significant changes in the distribution of income and their position within it from any specific change in the tax code. However, when they are informed about the general incidence of a salient tax across all households in the income distribution, their support for more redistributive tax policies increases. The effects are large. In addition, they are driven in part by a group of respondents who might reasonably be considered as hard-to-reach: respondents whose ideologies and view of the world lead them to assume that the VAT is not regressive and that redistribution is an inappropriate goal for public policy. In fact, treatment effects are stronger among this group. These results have policy implications both regarding how to inform citizens about complex fiscal policy reforms with easy-to-interpret facts, but also about how to design fiscal adjustment packages. More progressive policy responses are more popular, but only when individuals are informed about how progressive they are. 32 References Alesina, Alberto, Armando Miano and Stefanie Stantcheva. 2022. “Immigration and Redistribution.” The Review of Economic Studies pp. 1–39. Alesina, Alberto, Gabriele Ciminelli, Davide Furceri and Giorgio Saponaro. 2021. “Austerity and Elections.” IMF Working Paper No. 2021/121. Alesina, Alberto, Stefanie Stantcheva and Edoardo Teso. 2018. “¨ Intergenerational mobility and preferences for redistribution.” American Economic Review 2(2):521–554. Ardanaz, Martin, Mark Hallerberg and Carlos Scartascini. 2020. “Fiscal consolidations and Electoral Outcomes in Emerging Economies: Does the Policy Mix Matter? Macro and Micro Level Evidence from Latin America.” European Journal of Political Economy 64(September):1–28. Bachas, Pierre, Lucie Gadenne and Anders Jensen. 2021. Informality, Consumption Taxes and Redistribution. NBER Working Papers 27429 National Bureau of Economic Research. Barreix, Alberto, Martin Bes, Oscar Fonseca, Maria Belen Fontenez, Dalmiro Moran, Emilio Pineda and Jeronimo Roca. 2022. “Revisiting Personalized VAT: A Tool for Fiscal Consolidation with Equity.” IDB Discussion Paper 939. Bartels, Larry. 2005. “Homer Gets a Tax Cut: Inequality and Public Policy in the American Mind.” Perspectives on Politics 3(1):15–31. Bastani, Spencer and Daniel Waldenstrom. 2021. “Perceptions of Inherited Wealth and the Support for Inheritance Taxation.” Economica 88:532–556. Boudreau, Cheryl and Scott MacKenzie. 2018. “Wanting What is Fair: How Party Cues and Information about Income Inequality Affect Public Support for Taxes.” Journal of Politics 80(2):367–381. 33 Cruces, Guillermo, Ricardo Perez Truglia and Martin Tetaz. 2013. “Biased perceptions of income distribution and preferences for redistribution: evidence from a survey experiment.” Journal of Public Economics 98(February):100–112. David, Antonio and Daniel Leigh. 2018. “A New Action-Based Dataset of Fiscal Consolidation in Latin America and the Caribbean.” IMF Working Paper 2018/094. de Bresser, Jochem and Marike Knoef. 2022. “Eliciting preferences for income redistribution: A new survey item.” Journal of Public Economics 214. Douenne, Thomas and Adrien Fabre. 2022. “Yellow Vests, Pessimistic Beliefs, and Carbon Tax Aversion.” American Economic Journal: Economic Policy 14(1):81 – 110. Fehr, Dietmar, Johanna Mollerstrom and Ricardo Perez-Truglia. 2022. “Your Place in the World: Relative Income and Global Inequality.” American Economic Journal: Economic Policy 14(4):232 – 268. Fernandez-Albertos, Jos´e and Alexander Kuo. 2018. “Income Perception, Information, and Progressive Taxation: Evidence from a Survey Experiment.” Political Science Research and Methods 6(1):83–110. Gasparini, Leonardo. 1998. “Incidencia distributiva del sistema impositivo argentino.” In La Reforma Tributaria en Argentina. FIEL. Hoy, Christopher. 2022. “How Does the Progressivity of Taxes and Government Transfers Impact People’s Willingness to Pay Tax?: Experimental Evidence across Developing Countries.” World Bank Policy Research Working Papers (10167). Hoy, Christopher and Franziska Mager. 2021. “Why Are Relatively Poor People Not More Supportive of Redistribution? Evidence from a Randomized Survey Experiment across Ten Countries.” American Economic Journal: Economic Policy 13(4):299 – 328. 34 H¨ubscher, Evelyne and Thomas Sattler. 2017. “Fiscal consolidation under electoral risk.” European Journal of Political Research 56(1):151–168. H¨ubscher, Evelyne, Thomas Sattler and Markus Wagner. 2021. “Voter Responses to Fiscal Austerity.” British Journal of Political Science 51(4):1751–1760. IDB. 2022. “Analysis of the Incidence of Consumption Taxes in Selected Countries of Latin America and the Caribbean.” Inter-American Development Bank, Unpublished Report. Karadja, Mounir, Johana Mollerstrom and David Seim. 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. Krupnikov, Yanna, Adam Levine, Arthur Lupia and Markus Prior. 2006. “Public Ignorance and Estate Tax Repeal: The Effect of Partisan Differences and Survey Incentives.” National Tax Journal 59(3):425–437. Kuziemko, Ilyana, Michael Norton, Emmanuel Saez and Stefanie Stantcheva. 2015. “How elastic are preferences for redistribution? Evidence from randomized survey experiments.” American Economic Review 105(4):1478–1508. Lustig, Nora. 2018. Commitment to Equity (CEQ) Handbook: Estimating the impact of fiscal policy on inequality and poverty. Brookings Institution Press. Lustig, Nora, Carola Pessino and John Scott. 2014. “The Impact of Taxes and Social Spending on Inequality and Poverty in Argentina, Bolivia, Brazil, Mexico, Peru, and Uruguay: Introduction to the Special Issue.” Public Finance Review 42(2):287–303. Luttmer, Erzo and Monica Singhal. 2011. “Culture, Context, and the Taste for Redistribution.” American Economic Journal: Economic Policy 3(1):157–179. Metcalf, Gilbert. 1994. Lifecycle vs. Annual Perspectives on the Incidence of a Value Added Tax. NBER Working Papers 4619 National Bureau of Economic Research. 35 Rastelleti, Alejandro. 2021. “IVA personalizado: Experiencia de 5 pa´ıses y su importancia estrat´egica para la pol´ıtica y la admonistraci´on tributaria.” Recaudando Bienestar Blog, IDB. Sausgruber, Rupert and Jean-Robert Tyran. 2011. “Are We Taxing Ourselves: How Deliberation and Experience Shape Voting on Taxes.” Journal of Public Economics 95(1-2):164– 176. Sides, John. 2016. “Stories or Science? Facts, Frames, and Policy Attitudes.” American Politics Research 44(3):387–414. Slemrod, Joel. 2006. “The Role of Misconceptions in Support for Regressive Tax Reform.” National Tax Journal 1(LIX):57–75. Stantcheva, Stefanie. 2021. “Understanding tax policy: How do people reason?” The Quarterly Journal of Economics 136(4):2309–2369. Weisstanner, David and Klaus Armingeon. 2022. “Redistributive preferences: Why actual income is ultimately more important than perceived income.” Journal of European Social Policy 32(2):151–168. 36 Screen 1 A menudo, los países de América Latina se encuentran con problemas fiscales serios, amenazando la estabilidad económica, el empleo, y los ingresos familiares. Una opción a la cual los gobiernos recurren para salir del callejón fiscal es subir los impuestos al consumo - el Impuesto al Valor Agregado (IVA). El IVA afecta más a los pobres que a los ricos. Como muestra el gráfico abajo, un individuo perteneciente al 10% de hogares más pobres destina alrededor de 23% de sus ingresos mensuales en concepto de pagos de IVA. En cambio, un individuo perteneciente al 10% de los hogares más ricos paga sólo 11% de sus ingresos en concepto de IVA. Pagos de IVA como porcentaje del ingreso en América Latina Screen 2 Queremos saber su opinión de tres opciones que tienen los gobiernos para recaudar más a través de del IVA. Dos de ellas buscan proteger a los hogares más pobres del impacto del aumento, al recaudar más de los demás hogares. Opción 1: Aumentar el IVA de manera que el gobierno recaude lo suficiente para evitar la crisis fiscal. Cada persona pagaría 4% más que ahora en concepto de IVA. Por ejemplo, si una persona actualmente paga mensualmente 3.000 pesos, después del aumento del IVA pagaría 3.120 pesos. A Survey Screenshots Figure A1: Information treatment and VAT adjustment options Figure C7: Treatment effects by country (with and without controls) Treatment estimate -.4 -.2 0 .2 -.4 -.2 0 .2 -.4 -.2 0 .2 All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Total Argentina Brazil Chile Colombia C. Rica Guatemala Mexico Peru (a) Without controls Treatment estimate -.4 -.2 0 .2 -.4 -.2 0 .2 -.4 -.2 0 .2 All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Total Argentina Brazil Chile Colombia C. Rica Guatemala Mexico Peru (b) With controls Notes: Treatment effects, equivalent to the results in Table 2, Panels A (without controls) and C (including socioeconomic controls and knowledge, beliefs, and perceptions), when all countries are pooled, and by country. Total treatment refers to the pooled estimation. Outcome variables are the categorical measures described in Table 2. The standard errors are clustered at the country level for the pooled sample, and are robust for the country specifications. Point estimates with 95% confidence intervals. Figure C8: Country Exclusion Argentina Brazil Chile Colombia Costa Rica Guatemala Mexico Peru -.15 -.1 -.05 0 -.15 -.1 -.05 0 -.15 -.1 -.05 0 All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Excluded country Treatment effect Notes: Treatment effects, equivalent to the results in Table 2, Panel A, when countries are excluded one-by-one. Outcome variables are the categorical measures described in Table 2. Point estimates with 95% confidence intervals 44 Table C1: Main effect - showing controls (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Preferred Choice Cat D Cat D Cat D All Pay 70% Pay 50% Pay No Adj Treated -0.051*** -0.017*** -0.051** -0.016 -0.114*** -0.035*** -0.040** 0.022 0.027** -0.008 (0.013) (0.005) (0.020) (0.012) (0.021) (0.008) (0.014) (0.014) (0.009) (0.005) Bottom30Actual 0.238*** 0.049*** 0.128** -0.001 0.228** 0.035 0.036** -0.023** -0.043** 0.030*** (0.046) (0.012) (0.040) (0.016) (0.067) (0.023) (0.014) (0.009) (0.015) (0.008) B40and50Actual 0.199*** 0.034** 0.094** -0.001 0.183*** 0.025 0.020* -0.011 -0.023* 0.014 (0.039) (0.010) (0.030) (0.010) (0.037) (0.014) (0.009) (0.011) (0.011) (0.008) Educated -0.088*** -0.020* -0.063 -0.014 -0.082* -0.018 -0.031*** -0.002 0.032* 0.000 (0.022) (0.010) (0.036) (0.010) (0.043) (0.012) (0.008) (0.015) (0.016) (0.005) Age 0.000 -0.000 -0.001 -0.001** -0.001 -0.001 0.000 -0.000 -0.001 0.000 (0.001) (0.000) (0.001) (0.000) (0.001) (0.001) (0.001) (0.000) (0.001) (0.000) Female -0.024 -0.017 -0.024 -0.015 -0.013 -0.018 -0.004 0.027*** -0.022*** -0.002 (0.039) (0.014) (0.015) (0.008) (0.031) (0.010) (0.007) (0.005) (0.006) (0.006) Unemployed -0.050 -0.014 0.003 0.005 -0.017 -0.002 -0.009 0.004 -0.005 0.010** (0.033) (0.009) (0.019) (0.006) (0.052) (0.013) (0.009) (0.005) (0.006) (0.003) InformalWorker -0.084** -0.021** -0.048 -0.014 -0.038 -0.009 -0.014 0.022** -0.005 -0.003 (0.033) (0.007) (0.027) (0.010) (0.035) (0.013) (0.011) (0.008) (0.013) (0.008) Retired -0.035 0.005 -0.058 -0.014 -0.008 -0.013 -0.013 0.008 0.030 -0.025 (0.090) (0.023) (0.043) (0.010) (0.098) (0.026) (0.034) (0.031) (0.021) (0.015) GovernmentSub 0.001 0.015 0.053 0.031** -0.029 -0.001 -0.007 0.028 0.001 -0.022* (0.043) (0.012) (0.038) (0.010) (0.041) (0.012) (0.009) (0.019) (0.014) (0.012) HouseholdSize 0.011* 0.001 0.002 -0.002 0.016* 0.005 0.002 -0.001 -0.002 0.001 (0.005) (0.002) (0.004) (0.002) (0.008) (0.003) (0.002) (0.002) (0.002) (0.002) Bottom30Perceived -0.033 -0.016 -0.074 -0.021 -0.096** -0.025* 0.005 -0.028* 0.018* 0.005 (0.037) (0.012) (0.044) (0.014) (0.032) (0.012) (0.013) (0.014) (0.009) (0.007) B40and50Perceived 0.004 0.000 -0.040 -0.017 -0.067** -0.015 0.003 -0.008 0.005 0.000 (0.017) (0.009) (0.032) (0.010) (0.025) (0.008) (0.007) (0.011) (0.012) (0.003) KnowledgeTaxes -0.019 -0.003 -0.039 -0.008 -0.042* -0.003 0.007 -0.007 0.008 -0.008 (0.022) (0.008) (0.023) (0.009) (0.019) (0.006) (0.008) (0.010) (0.009) (0.006) ConcernIneqPov -0.175*** -0.033** -0.066* -0.004 -0.145*** -0.030*** -0.050*** 0.021** 0.044* -0.014** (0.040) (0.012) (0.031) (0.008) (0.036) (0.008) (0.011) (0.009) (0.019) (0.004) TrustGov 0.019 0.017 0.071 0.043** 0.053 0.027 0.022 0.013 -0.026 -0.009 (0.052) (0.014) (0.048) (0.013) (0.064) (0.017) (0.013) (0.009) (0.017) (0.008) BeliefsLuck 0.005 0.011 -0.028 0.003 -0.024 0.001 -0.038** 0.017* 0.009 0.012* (0.031) (0.006) (0.038) (0.012) (0.035) (0.008) (0.015) (0.007) (0.016) (0.005) PoorChildOpportunity -0.207*** -0.062*** -0.161*** -0.037*** -0.277*** -0.080*** -0.038*** -0.006 0.054*** -0.009 (0.027) (0.012) (0.018) (0.006) (0.020) (0.009) (0.011) (0.009) (0.011) (0.009) PoliticalAlignment 0.063*** 0.018*** 0.045*** 0.011*** 0.064*** 0.017*** 0.018*** -0.005** -0.016*** 0.003 (0.013) (0.003) (0.008) (0.001) (0.014) (0.003) (0.003) (0.002) (0.003) (0.002) Observations 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 R-squared 0.045 0.030 0.023 0.015 0.051 0.034 0.034 0.008 0.026 0.012 Mean Dep. Var. 3.094 0.317 2.934 0.389 3.107 0.326 0.277 0.308 0.338 0.076 Mean Dep. Var. (control) 2.931 0.326 3.091 0.398 2.949 0.344 0.297 0.297 0.326 0.080 Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Notes: This table presents the results of the main treatment effects. “All vs. 70% Pay” measures respondents’ preferences for option 1 over option 2. Cat means that it is the categorical measure, which takes values from 1 to 5, where 5 is a greater preference for option 1. Dis an indicator variable that takes values of 1 if option 1 was preferred, and 0 otherwise. Same logic for the dependent variables in columns 3 to 6. All Pay in column 7 is an indicator variable that takes values of 1 if option 1 was chosen, and 0 otherwise. The same logic for columns 8, 9 and 10. All Pay being the least redistributive option and 50% Pay the most redistributive. No Adj means no action. Treated is an indicator variable that takes values of 1 if it received the treatment of experiment 1, and 0 otherwise. Clustered standard errors at the country level are reported in parentheses. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1% level. 45 Table C2: Attention checks by country (1) (2) (3) (4) (5) (6) Country ATT. Check I ATT. Check II At least one correct Both correct Wrong Correct Wrong Correct Wrong Correct Wrong Correct Argentina 44.68 55.32 48.12 51.88 29.54 70.46 63.25 36.75 Brazil 45.77 54.23 49.11 50.89 29.84 70.16 65.05 34.95 Chile 47.91 52.09 45.18 54.82 29.34 70.66 63.75 36.25 Colombia 46.15 53.85 50.72 49.28 30.64 69.36 66.23 33.77 Costa Rica 48.80 51.20 57.31 42.69 34.44 65.56 71.68 28.32 Guatemala 45.44 54.56 57.03 42.97 33.38 66.62 69.09 30.91 Mexico 48.92 51.08 53.57 46.43 33.99 66.01 68.50 31.50 Peru 46.04 53.96 49.74 50.26 31.64 68.36 64.13 35.87 All 46.72 53.28 51.32 48.68 31.59 68.41 66.45 33.55 Notes: This table presents the percentages of respondents by country who chose each of the options in the attention control question. ATT. Check I corresponds to the question: Under which reform option do the poorest households pay more?. Respondents were expected to choose the 4% increase in VAT payments applied to all households. ATT. Check II corresponds to the question Under which reform option do the richest households pay more?. Respondents were expected to choose the 6% increase in VAT payments that exclude the bottom 50%. Option 1 is “4% increase”, Option 2 is “5% increase”, and Option 3 is “6% increase”. Table C3: Main effect - subsample attention checks (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Preferred Choice Cat D Cat D Cat D All Pay 70% Pay 50% Pay No Adj Treated -0.090*** -0.032*** -0.058* -0.019 -0.136*** -0.041*** -0.045*** 0.021 0.028** -0.004 (0.013) (0.006) (0.029) (0.014) (0.025) (0.011) (0.012) (0.015) (0.010) (0.005) Observations 8,313 8,313 8,313 8,313 8,313 8,313 8,313 8,313 8,313 8,313 R-squared 0.029 0.022 0.011 0.010 0.029 0.020 0.025 0.009 0.015 0.011 Mean Dep. Var. 3.094 0.317 2.934 0.389 3.107 0.326 0.277 0.308 0.338 0.076 Mean Dep. Var. (control) 2.931 0.326 3.091 0.398 2.949 0.344 0.297 0.297 0.326 0.080 Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Notes: This table presents the results of the main treatment effects in a subsample of those who responded well to care checks. “All vs. 70% Pay” measures respondents’ preferences for option 1 over option 2. Cat means that it is the categorical measure, which takes values from 1 to 5, where 5 is a greater preference for option 1. Dis an indicator variable that takes values of 1 if option 1 was preferred, and 0 otherwise. Same logic for the dependent variables in columns 3 to 6. All Pay in column 7 is an indicator variable that takes values of 1 if option 1 was chosen, and 0 otherwise. The same logic for columns 8, 9, and 10. All Pay being the least redistributive option and 50% Pay the most redistributive. No Adj means no action. Treated is an indicator variable that takes values of 1 if it received the treatment of experiment 1, and 0 otherwise. Clustered standard errors at the country level are reported in parentheses. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1% level. 46 Table C4: Heterogeneous effect - political alignment (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Preferred Choice Cat D Cat D Cat D All Pay 70% Pay 50% Pay No Adj TreatedXPoliticalAlignment -0.011 -0.006* 0.001 -0.000 -0.021*** -0.009** -0.003 0.004 0.001 -0.002 (0.006) (0.003) (0.018) (0.005) (0.006) (0.003) (0.003) (0.002) (0.004) (0.002) Treated 0.007 0.013 -0.058 -0.015 -0.002 0.013 -0.023 -0.000 0.022 0.001 (0.037) (0.017) (0.080) (0.020) (0.040) (0.022) (0.029) (0.022) (0.026) (0.015) PoliticalAlignment 0.081*** 0.024*** 0.054*** 0.013*** 0.090*** 0.025*** 0.022*** -0.007** -0.020*** 0.004 (0.016) (0.003) (0.013) (0.003) (0.017) (0.004) (0.004) (0.002) (0.005) (0.003) Observations 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 R-squared 0.037 0.026 0.017 0.012 0.040 0.027 0.029 0.007 0.020 0.011 Mean Dep. Var. 3.094 0.317 2.934 0.389 3.107 0.326 0.277 0.308 0.338 0.076 Mean Dep. Var. (control) 2.931 0.326 3.091 0.398 2.949 0.344 0.297 0.297 0.326 0.080 Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Notes: This table presents the results of heterogeneous treatment effects. “All vs. 70% Pay” measures respondents’ preferences for option 1 over option 2. Cat means that it is the categorical measure, which takes values from 1 to 5, where 5 is a greater preference for option 1. Dis an indicator variable that takes values of 1 if option 1 was preferred, and 0 otherwise. Same logic for the dependent variables in columns 3 to 6. All Pay in column 7 is an indicator variable that takes values of 1 if option 1 was chosen, and 0 otherwise. The same logic for columns 8, 9 and 10. All Pay being the least redistributive option and 50% Pay the most redistributive. No Adj means no action. Treated is an indicator variable that takes values of 1 if it received the treatment of experiment 1, and 0 otherwise. Clustered standard errors at the country level are reported in parentheses. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1% level. Table C5: Heterogenous effect - concern about inequality and poverty (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) All vs. 70% Pay 70% vs. 50% Pay All vs. 50% Pay Preferred Choice Cat D Cat D Cat D All Pay 70% Pay 50% Pay No Adj TreatedXConcernIneqPov 0.099** 0.016 0.054 -0.010 0.119* 0.036* 0.027 -0.009 -0.039* 0.021** (0.040) (0.017) (0.082) (0.027) (0.063) (0.018) (0.015) (0.015) (0.020) (0.008) Treated -0.073*** -0.021** -0.063** -0.014 -0.142*** -0.043*** -0.046** 0.024 0.036*** -0.014** (0.010) (0.006) (0.025) (0.013) (0.018) (0.007) (0.015) (0.015) (0.005) (0.005) ConcernIneqPov -0.301*** -0.062*** -0.151* -0.012 -0.290*** -0.070*** -0.084*** 0.029* 0.083** -0.028*** (0.056) (0.012) (0.078) (0.021) (0.067) (0.013) (0.020) (0.012) (0.029) (0.006) Observations 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 12,152 R-squared 0.028 0.018 0.011 0.009 0.029 0.020 0.023 0.007 0.016 0.010 Mean Dep. Var. 3.094 0.317 2.934 0.389 3.107 0.326 0.277 0.308 0.338 0.076 Mean Dep. Var. (control) 2.931 0.326 3.091 0.398 2.949 0.344 0.297 0.297 0.326 0.080 Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Notes: This table presents the results of heterogeneous treatment effects. “All vs. 70% Pay” measures respondents’ preferences for option 1 over option 2. Cat means that it is the categorical measure, which takes values from 1 to 5, where 5 is a greater preference for option 1. Dis an indicator variable that takes values of 1 if option 1 was preferred, and 0 otherwise. Same logic for the dependent variables in columns 3 to 6. All Pay in column 7 is an indicator variable that takes values of 1 if option 1 was chosen, and 0 otherwise. The same logic for columns 8, 9, and 10. All Pay being the least redistributive option and 50% Pay the most redistributive. No Adj means no action. Treated is an indicator variable that takes values of 1 if it received the treatment of experiment 1, and 0 otherwise. Clustered standard errors at the country level are reported in parentheses. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1% level. 47