Policy Research Working Paper 10167 How Does the Progressivity of Taxes and Government Transfers Impact People’s Willingness to Pay Tax? Experimental Evidence across Developing Countries Christopher Hoy Macroeconomics, Trade and Investment Global Practice September 2022 Policy Research Working Paper 10167 Abstract This paper examines how the progressivity of taxes and effects were predominantly driven by respondents in cases government transfers impacts people’s willingness to pay where the information they received was counter to their tax through a randomized survey experiment with over prior beliefs and/or consistent with their preferences. These 30,000 respondents across eight developing countries. results suggest changes in policies that increase (decrease) Respondents increased (decreased) their willingness to pay the progressivity of tax systems may also lead to increases taxes when they received accurate information that taxes (decreases) in tax compliance. in their country are progressive (not progressive). These This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at choy@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team How does the progressivity of taxes and government transfers impact people’s willingness to pay tax? Experimental evidence across developing countries* Christopher Hoy (World Bank) JEL-Classification: D31, D91, H22, H23, H24, H26 Keywords: Political Economy, Public Finance, Redistribution, Tax Compliance, Random- ized Experiment * This study was pre-registered on the American Economic Association RCT Registry (ID number AEARCTR-0008847). This work was financed as part of the Innovations for Tax Compliance (ITC) program at the World Bank. The ITC program receives funding from the Bill and Melinda Gates Foundation and the World Bank Global Tax Program. The author is incredibly grateful for comments provided on an earlier version of this paper by Steve Dav- enport, Matthew Collin, Anna Custers, Roel Dom, Will Prichard, Chiara Bronchi, Matthew Wai Poi, Pierre Bachas, Mahvish Ifrah Shaukat, Katy Bergstrom, William Dodds, Thiago Scot, Jonathan Karver, Mathieu Cloutier and seminar participants at the Australian Na- tional University and Australasian Development Economics Workshop. The views expressed in this paper are those of the author and do not necessarily reflect those of the funders. The survey would not have been possible without the excellent guidance provided by Shaelyn Laurie. The author is also very appreciative of Karin Hosking for her copy-editing services. Ethics approval for this work was processed through the Australian National University Hu- man Research Ethics Committee (Protocol 2021/792). 1 Introduction Levels of government-led redistribution from rich to poor households vary considerably across developing countries, including between those with similar levels of income and tax revenue. Figure 1 illustrates this by showing that for all developing countries where comparable data is available, the difference between the gross (i.e., pre-taxes and government transfers) and net (i.e., post-taxes and government transfers) GINI index is negligible in some countries and far more pronounced in others (CEQ, 2021). In other words, taxes and government transfers have almost no impact on income inequality in some countries and considerably reduce income inequality in others. These differences are due to whether or not tax and/or transfer systems are progressive (i.e., whether richer households pay a relatively higher share of their income in tax and whether poorer households receive a relatively higher share of their income in government transfers).1 The substantial variation in progressivity of taxes and transfers across developing countries is in stark contrast to widespread public support in most countries for government intervention to reduce income inequality (e.g., see WVS 2020; PEW 2019). This raises the question as to whether people are more (less) willing to pay tax when the tax and/or transfer system in their country is progressive (not progressive). [Figure 1] I answer this question through conducting a randomized survey experiment with over 30,000 respondents that is broadly representative of the population with internet access in eight developing countries (Colombia, Ghana, Indonesia, Jordan, Mexico, Sri Lanka, South Africa and Tanzania). This diverse set of countries makes up around 10 percent of the developing world’s population, is spread across Latin America, Africa, Asia and the Mid- dle East, and had GNI per capita ranging from USD1,080 to USD8,480 (Atlas Method) in 1 In general, tax and transfer systems are less progressive in countries that rely more heavily on indirect taxes (e.g., value added tax) compared to direct taxes (e.g., personal income tax) and/or indirect transfers (e.g., subsidies) compared to direct transfers (e.g., targeted cash transfers). 1 2020 (World Bank, 2021). Respondents in each country were randomly allocated to receive accurate information about the progressivity of taxes (“taxes treatment”), government trans- fers (“transfers treatment”) or both (“combined treatment”) in their country, or to a control group that received no information. This information was sourced from a recently released database (hereafter the “CEQ database”) that uses a standardized approach across countries to monitor progress toward Sustainable Development Goal target 10.4 about increasing the redistributive impact of fiscal policy (Lustig, Mariotti and Sánchez-Páramo, 2020). The pro- gressivity of taxes and government transfers is measured using the Commitment to Equity (CEQ) Institute methodology2 and is based on recent Household Income and Expenditure Surveys that provide detailed information about the income and consumption patterns of a nationally representative sample of households (CEQ, 2021). The impact of the information treatments on people’s willingness to pay taxes is measured using standardized questions from cross-country survey instruments (e.g., the Afrobarometer). For example, respondents were asked on a Likert scale about: whether it is important for people to pay tax, if the government always has a right to make people pay tax, and if they would still pay tax in the absence of enforcement. I illustrate the channels through which information is impact- ing people’s willingness to pay tax by examining heterogeneous treatment effects based on people’s prior beliefs and existing preferences as well as differences between the treatments and across countries. People’s willingness to pay taxes has been traditionally conceptualized as a trade-off between the punishment they face from being caught for non-compliance compared to the cost of complying (Allingham and Sandmo, 1972), however in recent years there has been growing recognition of other factors that influence people’s willingness to pay tax (Slemrod, 2019; Antinyan and Asatryan, 2019). Understanding these “quasi-voluntary” motivations for paying tax is particularly important in developing countries as there is typically weaker capacity to enforce tax legislation (Prichard et al., 2019; Dom et al., 2022). The extent 2 The CEQ approach was developed by the Commitment to Equity Institute (CEQ Institute) at Tulane University. The methodology, implementation guidelines, applications, and software of the CEQ approach can be found in Lustig (2018). 2 to which the tax and transfer system in a country reduces inequality has been proposed as one of the factors that may influence people’s willingness to pay tax (Prichard et al., 2019; Dom et al., 2022). This is because most people prefer to live in societies with lower levels of inequality than what they perceive to exist (WVS, 2020; Alesina and Giuliano, 2011) and many are supportive of the government promoting greater equality through using taxes and transfers to redistribute resources from rich to poor households (PEW 2019; Alesina and An- geletos, 2005; Hoy and Mager, 2021a). This implies that people may be more (less) willing to pay tax when they believe the tax and transfer system is progressive (not progressive). I show this formally in the paper through combining a modified version of Allingham and Sandmo’s (1972) seminal theory of what drives people’s willingness to pay tax with Alesina and Giuliano’s (2011) workhorse model that shows how beliefs and preferences about in- equality influence people’s utility. This modified conceptual framework forms the basis for the detailed pre-registered hypotheses of this study (Hoy, 2022). The overall findings of the randomized survey experiment illustrate that people’s will- ingness to pay tax is influenced by whether there is progressivity in the tax and transfer system. Respondents who received the taxes treatment in the four countries for which taxes were progressive (Colombia, Ghana, Mexico and Tanzania) were more willing to pay tax. In contrast, respondents who received the taxes treatment in the four countries for which taxes were not progressive (Indonesia, Jordan, Sri Lanka and South Africa) were less willing to pay tax. The overall effects were of a similar magnitude in each of the countries and the results are robust to a series of checks (such as comparing the results across treatments and removing respondents who took too little time or too long to complete the survey). Weaker (and often insignificant) results were attained from the transfers and combined treatments, although the point estimates were still consistent with respondents who received information that the system was progressive (not progressive) being more (less) willing to pay tax. The order of magnitude of the impact of the taxes treatment was in line with seminal cross- country randomized survey experiments (Alesina et al., 2018; Alesina et al., 2022) and if this translated into actual tax compliance behavior the effects would be non-trivial (e.g., 3 they would be of a similar size to recent work such as Balan et al., 2022). The overall treatment effects were predominantly driven by respondents in cases where the information they received was counter to their prior beliefs and/or in line with their preferences. These results are consistent with the conceptual framework that shows how prior beliefs and existing preferences about progressivity in the tax and transfer system are likely to impact people’s willingness to pay tax. Respondents who stated prior to the treatment that they prefer progressivity in the tax system and received accurate information that this was actually the case (i.e., those in Colombia, Ghana, Mexico and Tanzania) were much more willing to pay taxes. Respondents who thought the tax system was progressive but received accurate information that it was not progressive (i.e., those in Indonesia, Jordan, Sri Lanka and South Africa) were much less willing to pay taxes. There was a similar, significant effect from the combined treatment (about the net effect of taxes and transfers) among subsets of respondents in cases where the information they received was counter to their prior beliefs and/or consistent with their preferences. There were no statistically significant effects on people’s willingness to pay tax from the transfers treatment and no noteworthy trends in terms of heterogeneous treatment effects across other dimensions that were included the pre-analysis plan (e.g., by respondents’ perceived place in the national income distribution). The findings from this study shed light on how policy reforms that alter progressivity in tax and transfer systems may influence people’s willingness to pay tax. Specifically, the results suggest that efforts to improve a country’s fiscal position by increasing (decreasing) equity in the tax and transfer system may also have an additional benefit (potentially backfire ) by increasing (decreasing) people’s willingness to pay tax. This can be illustrated through the following stylized examples. Consider a tax policy reform that required richer households to pay more tax and which by doing so would make the tax and transfer system more progressive (e.g., an increase in the top marginal income tax rate). A consequence of this reform is that many taxpayers may be more likely to comply, especially if they prefer greater progressivity. Therefore, the improvement in total tax revenue collected could be greater than just the additional revenue that was intended to be gathered from richer households. 4 Another illustrative example is a tax policy reform that reduces the progressivity of the tax system, such as by increasing the tax burden dis-proportionally on poorer households (e.g., increasing the rate of value added tax on essential items). This reform could undermine many people’s willingness to pay tax and consequently not improve the fiscal position of the country as much as what was intended. In the extreme case it could be possible that any expected increase in revenue from the tax reform would be entirely offset by falls in compliance. These stylized examples show how the findings from this study are relevant for policy makers in developing countries, especially as governments face growing debt levels in the wake of the COVID-19 pandemic (World Bank, 2022a). In addition, the results show that even in the absence of a reform agenda, communicating to taxpayers about the progressive aspects of the tax system in their country would appear to be a way to boost compliance. Further, there appears to be ample scope for information campaigns to be done by policy makers to help the general population understand how taxes help fund the government transfers that benefit many households. This study makes several contributions to two broad strands of the existing literature. The first strand the study contributes to is in relation to how people’s perceptions shape their preferences regarding tax and transfer policies (Gimpelson and Treisman, 2018; Hauser and Norton, 2017). Seminal work on this topic has been conducted in recent years using large-scale, randomized survey experiments in the United States and Western Europe exam- ining a range of topics, such as inequality (Kuziemko et al., 2015), social mobility (Alesina et al., 2018) and immigration (Alesina et al., 2022). A common thread in these studies is that, in general, most people have a poor understanding of the economic circumstances in their country (e.g., about the level of inequality, see Norton and Ariely, 2011) and they have tested what happens to people’s general policy preferences when they are provided with accurate information. This study extends this literature in three ways. Firstly, I test how accurate information about existing policies (specifically the progressivity of taxes and government transfers), as opposed to existing circumstances (e.g., the level of inequality), shifts people’s preferences. In other words, I directly alter people’s beliefs about the role the 5 government currently plays in distributing resources in their country, and see how this shifts their preferences, as opposed to examining how people’s views change about what the role of the government should be once they are aware of the actual circumstances in their coun- try. Secondly, the randomized survey experiment focuses on measuring a specific intention (people’s willingness to pay tax), which is a key way people engage with the government, as opposed to general preferences. This allows for direct policy implications to emerge from this work. Thirdly, I conduct one of the first and by far the largest randomized survey experiments in this literature in developing countries (the previously largest study was in five middle-income countries by Hoy and Mager, 2021a). I incorporate best practices into the design of the randomized survey experiments from cross-country studies in high-income countries, follow a detailed, publicly available pre-analysis plan, and utilize a novel sampling methodology that allows for a more representative sample of the internet population to be collected than is typically captured in online surveys in developing countries. The second strand of the literature is in relation to a growing body of research about “quasi-voluntary” motivations for tax compliance. Examples of this work in high-income countries include how social norms (Hallsworth, 2014; De Neve et al., 2021), the provision of public goods (Giaccobasso et al., 2022) and a positive outlook on the government (Cullen et al., 2021) influence tax compliance. The extent to which “quasi-voluntary” motivations for tax compliance exist in developing countries is still unclear (Prichard et al., 2019; Dom et al., 2022). Outside of Latin America, there has been only a small number of randomized field experiments examining alternative motivations for people’s willingness to pay tax in developing countries, such as in Ethiopia (Shimeles et al., 2017), Rwanda (Mascagni and Nell, 2022), Tanzania (Collin et al., 2021) and Papua New Guinea (Hoy, McKenzie and Sinning, 2021). I contribute to this field by going well beyond existing work in three ways. Firstly, this study is the first to examine causally how progressivity (or lack thereof) in the tax and transfer system impacts people’s willingness to pay tax across countries. This be- came feasible because of the release of the CEQ database, which measures the progressivity of taxes and transfers in a standardized way across developing countries (Lustig, Mariotti 6 and Sánchez-Páramo, 2020). To the best of my knowledge, the closest example of related work is by Stantcheva (2021), who conducts randomized survey experiments in the United States that show redistributive considerations matter more to respondents than the efficiency of income and estate taxes. Secondly, the pre-registered, randomized survey experiment in this study was designed to specifically identify how a “quasi-voluntary” motivation influences people’s willingness to pay tax, which has been a challenge in prior work that largely relied on administrative data. The channels driving the treatment effects are isolated by capturing people’s prior beliefs and preferences, as well as comparing across countries and treatments, so that direct links can be made to seminal theory. Thirdly, I collect data that is repre- sentative of the internet population within each country and is comparable across a diverse set of developing countries. Consequently, the results provide rigorous insights for a much wider population and arguably have far greater external validity than previous work in these settings. This paper is structured as follows. Section 2 provides background to the study, including a conceptual framework and the hypotheses that flow from the theory as well as details about the setting of the randomized survey experiment. Section 3 describes the methodology in detail, including the sample selection, survey design and empirical analysis. Section 4 outlines the descriptive and main experimental findings. Section 5 illustrates the robustness of the main experimental results by drawing on best practices in the literature. Section 6 discusses the implications of these findings from a theoretical and policy perspective. 2 Conceptual framework and Hypotheses 2.1 Conceptual framework Traditionally, people’s willingness to pay taxes has been conceptualized as a trade-off between the punishment they face from being caught for non-compliance compared to the cost of 7 complying (Allingham and Sandmo, 1972).3 This is shown formally in the utility functions below whereby yi is an individual’s household income before tax, d is the probability of being detected as non-compliant, pi is a fixed amount that represents the punishment a taxpayer will face if found to be non-compliant and ti is a fixed amount that represents a taxpayer’s tax obligation. However, in recent years this model of tax compliance has been extended to include other factors that drive compliance beyond enforcement and punishment, such as people’s desire to keep in line with social norms (Hallsworth, 2014; Slemrod, 2019; Antinyan and Asatryan, 2019). As such, the traditional model of tax compliance has been broadened to incorporate what is often referred to as “quasi-voluntary” motivations for tax compliance. Prichard et al. (2019) suggest that other than enforcement, issues to do with facilitation of tax payments and trust in the tax system impact tax compliance. They further hypothesize that trust in the tax system is built on four related concepts of equity, reciprocity, accountability and fairness. Formally, these “quasi-voluntary” motivations for tax compliance can be expressed as the utility gain an individual receives from paying tax ai . As such, for a single point in time an individual’s utility from complying with taxes (Uci ) and from not complying (Uni ) can be expressed as: Uci = yi − ti + ai (1) and Uni = yi − dpi , (2) According to this model, taxpayers comply if Uci >Uni , which requires that: ti < dpi + ai (3) 3 In some respects, this work involved applying Becker’s (1968) seminal work on crime and punishment to tax compliance. 8 I extend this basic model by decomposing quasi-voluntary motivations for paying tax (shown as ai in the model above) to specifically identify how “equity” can play a role in driving people’s willingness to pay tax (Prichard et al., 2019). By doing so I separate this reason from other quasi-voluntary motivations (shown as bi in the revised model below). Equity, more precisely articulated as vertical equity by Prichard et al. (2019), is considered to be a driver of tax compliance because many people would prefer lower levels of inequality in their country and consequently are supportive of the role taxes and transfers can play in redistributing resources from rich to poor (WVS, 2020). This is formally integrated into the model by drawing on the “workhorse” utility function by Alesina and Giuliano (2011) that shows how differences between actual and preferred levels of inequality (Q − Q∗ i) impact people’s utility (the weight an individual places on deviations from their ideal level of inequality is captured in the term γi ). The revised model of people’s utility from paying tax can be expressed as follows: Uci = yi − ti + bi − γi (Q − Q∗ i) 2 (4) I dis-aggregate this utility function further by continuing to draw on Alesina and Giu- liano’s (2011) seminal work as they argue that people’s utility is largely (if not exclusively) influenced by differences between actual and preferred levels of inequality that are due to ∗ factors outside an individual’s control (Ql − Ql i ), as opposed to overall levels of inequality (Q − Q∗ i ). I identify that one of the key determinants of inequality outside an individual’s control is the degree of progressivity in the tax and transfer system in their country. I reflect ∗ this in the model with the term (Qt − Qt t i ), whereby Q is the level of progressivity in the tax ∗ and transfer system, Qt i is people’s preferred levels of progressivity in the tax and transfer system, and γit reflects the weighting people place on this (all other differences in inequality are captured in the terms denoted with o). As such, holding everything else constant, people who prefer the existing level of progressivity in the tax and transfer system will be more 9 willing to comply with taxes than those who do not. Consequently, the revised model of people’s utility from paying tax can be expressed as follows: ∗ 2 o∗ 2 Uci = yi − ti + bi − γit (Qt − Qt o o i ) − γi (Q − Qi ) (5) The final substantive modification I make is to incorporate the fact that it is people’s beliefs about how taxes and transfers are distributed, as opposed to what is actually the case, that will influence their willingness to pay tax. Previous research has shown that people tend to have a poor understanding of both the level of inequality in their country and their position in the national income distribution (see, for example, Hoy and Mager, 2021a) and there is evidence from the United States to suggest these misperceptions also extend to tax policies (Stantcheva, 2021). Consequently, I rewrite the utility function to factor in that people’s willingness to pay tax will be influenced by the extent to which they believe the tax and transfer system is progressive (Qt bi ): t∗ 2 o∗ 2 Uci = yi − ti + bi − γit (Qt o o bi − Qi ) − γi (Qbi − Qi ) (6) This utility function provides guidance as to how people’s willingness to pay tax (Uc − i) will be influenced by accurate information (I ) about the progressivity of taxes and/or transfers in their country (Qt ). In other words, it is possible to make predictions about how people’s utility from paying taxes varies when they have accurate information (i.e., Uci |I ). The two main dimensions in which heterogeneity would be expected are in terms of people’s prior beliefs and existing preferences about the progressivity of tax and transfer policies t∗ (captured formally as (Qt − Qt t bi ) and (Q − Qi ) respectively). These dimensions form the basis of the primary hypotheses that are discussed in the following subsection. 10 2.2 Hypotheses Three groups of primary hypotheses emerge from the conceptual framework. Group A of Hypotheses are based on a key implication from the theory and existing empirical literature suggesting that progressivity (a lack of progressivity) in the tax and transfer system will, on average, make people more (less) willing to pay tax. Group B of Hypotheses summarizes how people’s willingness to pay tax is likely to vary by their prior beliefs about the progressivity of the tax and transfer system. Group C of Hypotheses outlines how people’s willingness to pay tax is expected to vary by their preferences for progressivity in the tax and transfer system. All of these hypotheses were pre-registered on the American Economic Association RCT Registry prior to field work commencing (Hoy, 2022).4 Group A – People’s willingness to pay tax varies by the degree of progressiv- ity in the tax and transfer system Hypothesis A1: Informing people that the distribution of taxes is progressive (not progres- sive), will increase (decrease) their willingness to pay tax. Hypothesis A2: Informing people that the distribution of transfers is progressive (not pro- gressive), will increase (decrease) their willingness to pay tax. Hypothesis A3: Informing people that the net impact of taxes and transfers is progressive (not progressive), will increase (decrease) their willingness to pay tax. Group B – People’s willingness to pay tax varies by their prior beliefs about the progressivity of the tax and transfer system Hypothesis B1: Informing people that the distribution of taxes is progressive (not progres- 4 A fourth set of primary hypotheses were pre-registered that examined how the impact of people’s will- ingness to pay tax from the treatments varies based on their perceived position in the national distribution. However during the implementation of the survey it became clear that there were inadequate numbers of respondents in the poorest and richest quintiles to be adequately powered to detect statistically significant effects. The results for this dimension are presented in Section 5 of the paper, but given the challenges involved in getting respondents from across the national income distribution it is not the primary focus of the paper. 11 sive) when they thought it was not progressive (progressive), will increase (decrease) their willingness to pay tax. Hypothesis B2: Informing people that the distribution of transfers is progressive (not pro- gressive) when they thought it was not progressive (progressive), will increase (decrease) their willingness to pay tax. Hypothesis B3: Informing people that the net impact of taxes and transfers is progressive (not progressive) when they thought taxes and/or transfers were not progressive (progres- sive), will increase (decrease) their willingness to pay tax. Group C – People’s willingness to pay tax varies by their preferences for the progressivity of the tax and transfer system Hypothesis C1: Informing people that the distribution of taxes is progressive (not progres- sive) when they prefer it to be progressive (not progressive), will increase (decrease) their willingness to pay tax. Hypothesis C2: Informing people that the distribution of transfers is progressive (not pro- gressive) when they prefer it to be progressive (not progressive), will increase (decrease) their willingness to pay tax. Hypothesis C3: Informing people that the net impact of taxes and transfers is progressive (not progressive) when they prefer taxes and/or transfers to be progressive (not progressive), will increase (decrease) their willingness to pay tax. These hypotheses do not focus on differences between how the treatments may impact willingness to pay tax, but ex-ante it is conceivable differences would exist. As noted in the pre-analysis plan, survey respondents may be more likely to respond to the taxes treatment than to the government transfers treatment for a number of reasons. Firstly, on average, the share of household income collected in taxes is much higher than what is provided in transfers, which means people may be more concerned about how taxes are distributed compared to transfers. Secondly, there is reason to believe that “loss aversion” could exist 12 where people’s utility is more likely to be influenced by “losing” from paying tax than by “gaining” from receiving a transfer. Thirdly, people’s awareness of when they pay tax may be higher than their awareness about when they receive a transfer. For example, people are likely to be more conscious of paying income tax compared to receiving a subsidy for their fuel consumption, and consequently this could make them more responsive to information about who pays taxes as opposed to who receives transfers. 2.3 Setting of the study 2.3.1 Selection of countries The eight countries (Colombia, Ghana, Indonesia, Jordan, Mexico, Sri Lanka, South Africa and Tanzania) focused on in this study were selected for the following reasons. Firstly, there is very limited, standardized, cross-country data available about the progressivity of taxes and government transfers in developing countries. By far the largest effort that has been made to collect and disseminate this information has been through the Commitment to Equity Institute at Tulane University, which is headed by Nora Lustig (CEQ, 2021). Estimates have been produced of the difference between the gross and net GINI index in over 55 developing countries through this work program in partnership with the World Bank (see Figure 1). These estimates are based on standardized household income and expenditure surveys and in 2020 a cross-country database that provided dis-aggregated information in a standardized way for many countries was publicly released through a joint initiative between universities, civil society and international organizations (Lustig, Mariotti and Sánchez- Páramo, 2020). However due to a range of factors, including governments’ reluctance to make certain information publicly available, information about the progressivity of direct and indirect taxes as well as direct and indirect government transfers (including subsidies) is restricted to a far smaller subset of these countries. This subset of countries was the starting point for selecting which countries to include in this study. Secondly, the time and costs involved in collecting data online in low- and middle-income 13 countries are considerably lower when there is a high internet population in absolute terms. As such, countries with high populations and/or high internet penetration rates were focused on as part of this study. For example, some countries in this study like Tanzania have relatively high total populations (60 million), but low internet penetration rates (20 percent), whereas other countries like Jordan have a relatively low total population (10 million) but relatively high internet penetration rates (67 percent) (World Bank, 2021). Thirdly, due to funding reasons it was necessary to collect a diverse set of countries in each of the major regions with low- and middle-income countries (i.e., Latin America, West Africa, East Africa, the Middle East, South Asia and East Asia) as well as across various income levels. This restricted the choice set considerably in some regions; for example, Indonesia was the only country in East Asia with publicly available data about the distribution of direct and indirect taxes, as well as direct and indirect government transfers (including subsidies) (CEQ, 2021). Finally, efforts were made to ensure that the information about the distribution of taxes and government transfers in the database was still likely to provide a reasonable estimate of what would exist in 2022. There were some countries, such as the Islamic Republic of Iran, where there have been significant changes to the tax and transfer system since the survey included in the database took place and as a result it would not be a realistic approximation of how taxes and government transfers were likely to be distributed in 2022. 2.3.2 Progressivity of the tax and transfer systems in the countries in this study The extent of progressivity in the tax and transfer system in the eight countries (Colombia, Ghana, Indonesia, Jordan, Mexico, Sri Lanka, South Africa and Tanzania) in this study varied considerably. This is illustrated in Figure 2, which summarizes variations in the degree of progressivity in the tax and transfer systems according to the CEQ database (CEQ, 2021). For presentational purposes this figure shows the distribution of taxes and/or transfers across quintiles (whereas the CEQ database focuses on deciles) and combines both direct and indirect taxes and government transfers. In general, tax and transfer systems 14 are less progressive in countries that rely more heavily on indirect taxes (e.g., value added tax) compared to direct taxes (e.g., personal income tax) and/or indirect transfers (e.g., subsidies) compared to direct transfers (e.g., targeted cash transfers).5 Importantly, the CEQ database is based on actual levels of tax paid and transfers received (i.e., this already factors in existing levels of tax compliance by households). The tax system is progressive in four of the eight countries (Colombia, Mexico, Ghana and Tanzania) and the transfer system is progressive in six of the eight countries (Colom- bia, Mexico, Indonesia, Jordan, Sri Lanka and South Africa). The net impact of taxes and transfers is “weakly” progressive in all countries, however in Ghana and Tanzania the net impact is negative across all quintiles (i.e., all households pay more in tax than they receive in transfers). As such, for the purposes of this study, the net impact of taxes and transfers is only considered to be progressive in the six countries (Colombia, Mexico, Indonesia, Jordan, Sri Lanka, South Africa) where poorer households receive more in transfers than they pay in taxes. In summary, there are three groups of countries, one where both taxes and transfers are progressive (Colombia and Mexico), another where taxes are progressive and transfers are not (Ghana and Tanzania) and a final group where taxes are not progressive and transfers are (Indonesia, Jordan, Sri Lanka and South Africa). Collectively, these eight countries span the set of developing countries with comparable data available about the difference between the gross and net GINI index, ranging from around half a percentage point in Sri Lanka to almost ten percentage points in South Africa (see Figure A1). [Figure 2] 5 It is important to note that in each of these countries almost all households indirectly pay tax (e.g., through paying value added tax) and/or indirectly receive government transfers (e.g., through receiving fuel subsidies) (CEQ, 2021). 15 3 Methodology 3.1 Sample Selection and Sample Size The randomized survey experiment collected a broadly representative sample of the popu- lation with internet access in each country during the first three months of 2022 using an internationally respected online survey firm, RIWI (see Appendix A for details about the survey methodology). This is a similar approach to what was used by Alesina et al. (2018) and Alesina et al. (2022) in their seminal cross-country randomized survey experiments in the United States and Western Europe. A key difference is that in high-income countries internet access is near universal, whereas in the developing countries in this study the in- ternet penetration rate varies from 20 to 67 percent of the total population (World Bank, 2021). This resulted in a sample of respondents where younger people and men were over represented compared to the total population (see Table A1 in Appendix B). To address concerns about how representative the sample is of the total population, throughout the body of the paper I weight the descriptive and experimental results by the age and gender of the total population. As a robustness check, I present the sample average treatment effects (i.e., the unweighted findings) in Appendix B (see Tables A2–A4 in Appendix B). In general, the effects are almost identical. I also examine whether particular types of internet users were more likely to participate in and complete the survey. In general, respondents using a smartphone were less likely to begin the survey and conditional on starting the survey they were slightly less likely to complete all the questions (see Table A5 in Appendix B). This is to be expected as the visual components of the survey are easier to see on a larger screen. To address concerns about this an additional robustness check was conducted using "device type weights". This involved adjusting the results to ensure that in each country the share of respondents using a smartphone at the end of the survey was the same as the share of respondents using a smartphone that were exposed to the survey (i.e. they saw the invitation to participate in the survey). I show that using these "device type weights" had 16 no meaningful impact on the results (see Tables A6–A8 in Appendix B). The sample size in each of the eight countries is at least 3,600 respondents who com- pleted the survey. In total, over 30,000 respondents participated in the randomized survey experiment. The sample size of at least 900 respondents in each treatment and the control group in each country is in line with best practice in the existing literature. This is a similar sample size to other cross-country randomized survey experiments in this field by Alesina et al. (2018) and Alesina et al. (2022). Other studies in this field that have focused on a single country have tended to have smaller sample sizes (e.g., Cruces et al., 2013; Karadja et al., 2017) and have still detected significant heterogeneous treatment effects based on prior beliefs. Furthermore, a Journal of Economic Literature article summarizing best practices in online randomized survey experiments providing information interventions suggests that having in the order of 700 to 800 respondents per treatment/control group should provide adequate power to detect an effect (Haaland et al., forthcoming). 3.2 Survey Design The survey consisted of two sections and the treatments were provided in between the two sections of the survey. The first collected people’s demographic characteristics as well as prior beliefs and existing preferences (13 questions were asked prior to the treatment). The second included questions about people’s willingness to pay tax (five questions were asked following the treatment). The survey was designed to be quite focused and brief, which enabled the median respondent to complete the entire survey in less than 11 minutes. To maximize the likelihood respondents would provide honest answers, at the start of the survey they were informed that the answers they provide would be restricted to a team of independent, non- partisan researchers and that they would remain entirely anonymous. Ensuring respondent anonymity was a crucial part of this study, given the sensitivity of the topic, and this meant that no identifying information was collected. As a result, it was not possible to conduct a follow up survey to the original study as this would have required respondents to 17 provide details about how to be contacted and consequently their identity. Given that similar randomized survey experiments in the past have consistently shown that follow up surveys detect a persistence of treatment effect, there is no reason to believe that the same would not occur in this case. The survey instrument in English is provided in full in Appendix C and the exact treatments in each country are provided in Appendix D. This survey instrument was translated into the following languages: Spanish (for use in Mexico and Colombia), Arabic (for use in Jordan), Bahasa (for use in Indonesia), Swahili (for use in Tanzania) and Sinhala and Tamil (for use in Sri Lanka). 3.2.1 Questions measuring people’s prior beliefs and preferences Prior to the treatments, respondents were asked to provide information about their beliefs about their household’s position in the income distribution in their country, their beliefs and preferences for the distribution of taxes and government transfers in their country, preferences for the level of inequality in their country, and whether they viewed their households as being net contributors to or beneficiaries from the tax and transfer system. These questions were either sourced from existing studies in a series of developing countries (Q4 and Q5) (e.g., see Hoy and Mager, 2021a) or were specifically developed for this study (Q6–Q11). The questions developed for this study were based on the structure of standardized questions in the literature (e.g., they use a Likert scale), informed by expert feedback and modified based on the piloting process to ensure these new questions were adequately comprehended by respondents (see Appendix A for details of the piloting process). 3.2.2 Questions measuring people’s willingness to pay tax People’s willingness to pay taxes was measured using standardized questions from cross- country surveys (e.g., Afrobarometer) as well as drawing on the experience of previous survey instruments focusing on “quasi-voluntary” motivations for why people pay tax in developing countries (e.g., those referred to in Prichard et al., 2019). There is no single “ideal” question on this complex topic and as a result five questions focusing on slightly different ways of 18 measuring people’s willingness to pay tax were used. This ensured that if a treatment effect was detected across most, or even all of these questions, there would be good reason to be very confident the findings of the randomized survey experiment were not due to measurement error. More complicated questions (including list experiments that “implicitly” capture willingness to pay tax) were considered but ultimately discarded as the piloting process illustrated that keeping questions simple was by far the best way to maximize data quality and minimize attrition (see Appendix A for details). The first question to measure willingness to pay tax (Q14) directly asks respondents whether they would pay tax if they knew that they would not be caught for non-compliance. A potential shortcoming of a direct measure of “quasi-voluntary” tax compliance is that people may be unwilling to provide honest answers and as a result the share of respondents claiming they will pay tax in the control group (i.e., in the absence of additional information provided through the treatment) will almost certainly be higher than what will actually be the case. Therefore, this question is likely to suffer from “ceiling effects” (Po, 1998), which means it provides a lower bound estimate of the impact of the treatments on people’s willingness to pay tax. The remaining four “indirect” questions about people’s willingness to pay tax capture slightly different aspects of what is sometimes referred to in the literature as “tax morale”. All of these questions have been sourced from existing studies in low- and middle-income countries on this topic. The second question (Q15) measures the degree to which respondents believe people not paying is understandable. It was used in the Afrobarometer (2012; 2013; 2015) as well as by Ali, Fjeldstad and Sjursen (2014). The third question (Q16) measures whether people believe paying taxes is important and was used by Khwaja et al. (2020). The fourth question (Q17) measures people’s unconditional beliefs about the extent to which the government has the right to make people pay taxes and it has been included in many rounds of the Afrobarometer (2002; 2003; 2004; 2008; 2012; 2013; 2014; 2015; 2017). The fifth question measures the degree to which people believe that paying tax should be conditional on what the government spends tax on and is a slightly modified version of what was used 19 by Prichard, Jibao, and Orgeira (forthcoming). 3.2.3 Treatments The treatments were designed to provide people with accurate information about the pro- gressivity of taxes and/or government transfers in their country. Specifically, the treatments provided an indication of whether taxes and/or transfers were progressive in their country but did not provide information about the level of taxes and transfers as a percentage of household income. This is because it would not be possible to clearly isolate the channels through which the treatments were impacting people’s willingness to pay tax if information was provided about both the progressivity and level of taxes and transfers. For example, if both aspects were included it is possible some respondents may react to the degree of progressivity, while others may react to the level of taxes and transfers, and it would not be possible to differentiate between these. Efforts were also made to ensure that respon- dents were likely to trust the content of the treatments by following a similar approach to seminal work by Alesina et al. (2018). For example, respondents were informed that the information they were provided with recently became publicly available online through a collaboration between universities, civil society and international organizations (see Lustig, Mariotti and Sánchez-Páramo, 2020). In addition, given the extensive analysis in prior work to illustrate that experimental demand effects are unlikely to be present in these types of randomized survey experiments (e.g., see Kuziemko et al., 2015 on a related topic and de Quidt, Haushofer and Roth, 2018 more broadly), it is extremely unlikely to be an issue in this study. Survey respondents in each country were randomly allocated either to one of three treatment groups or to a control group that received no information (i.e., the multiple treatment arms were exclusive of one another). The first treatment (hereafter the “taxes treatment”) provided information from the CEQ database about the distribution of taxes (both direct and indirect taxes, such as income tax and value added tax) in their coun- try. The second treatment (hereafter the “transfers treatment”) provided information from the CEQ database about the distribution of government transfers (both direct and indirect 20 transfers, such as cash payments and energy subsidies). The third treatment (hereafter the “combined treatment”) provided information from the CEQ database about the net effect of the distribution of taxes and government transfers. 3.3 Empirical analysis I conducted a randomized survey experiment to test the impact of accurate information about the distribution of taxes, government transfers or both on people’s willingness to pay tax. Specifically, Hypotheses A1, B1, C1, are tested by the taxes treatment. Hypotheses A2, B2, C2, are tested by the transfers treatment. Hypotheses A3, B3, C3, are tested by the combined treatment. Randomization allows for the impact of the treatments to be deter- mined by comparing differences in mean outcomes between the control group and treatment groups. The randomization process was stratified by the age and sex of respondents. The balance tables for each country based on all answers provided prior to the treatment are in Appendix B (see Tables A9–A11), including measures of both individual and joint signifi- cance (i.e., both t-statistics for every variable and an f-statistic across all variables within a given country). The survey experiment has only five outcomes, which means that the risk of multiple hypothesis testing being an issue is very low. However, to address potential concerns all five outcomes are aggregated into an index. This is the identical approach to what was used in related randomized survey experiments by Alesina et al. (2018) and Karadja, Moller- strom and Seim (2017). Specifically, I create a “Willingness to Pay Tax” Index, which is an unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. The answers to each outcome question and the “Willingness to Pay Tax” Index are presented in the tables of results. The main results of the survey experiment are based on pooled Ordinary Least Squares (OLS) regressions with country fixed effects across all countries for which the treatment is in the same direction. For example, respondents across the four countries for which taxes were 21 progressive (Colombia, Ghana, Mexico and Tanzania) are pooled together and respondents across the four countries for which taxes were not progressive (Indonesia, Jordan, Sri Lanka and South Africa) are pooled together. This approach is in line with what was undertaken by Alesina et al. (2018) and Alesina et al. (2022). Specifically, it involves using an OLS regression in the form of a linear probability model by creating a dummy variable for each outcome of interest (see Section 3.2.2 and Appendix A for details) and a dummy variable for each treatment group that takes on the value one if the respondent belongs to the specific treatment group and the value zero if the respondent belongs to the control group. This can be expressed formally as follows: Yij = β0 + β1 T1 + β2 T2 + β3 T3 + Xi γ + θj + ϵij (7) where i denotes individuals, j denotes countries, β1 captures the average difference be- tween respondents in the taxes treatment group (T1 ) and the control group in regard to the outcome of interest (Y ), β2 captures the average difference between respondents in the transfers treatment group (T2 ) and the control group in regard to the outcome of interest (Y ), and β3 captures the average difference between respondents in the combined treatment group (T3 ) and the control group in regard to the outcome of interest (Y ). Further, Xi is a vector of variables that controls for potential imbalances in background characteristics (age, sex, location, education level, employment status and perceived place of household in the national income distribution) between treatment and control groups, θj captures country level fixed effects, ϵij is the model error term and β0 is the intercept. As per the pre-registered hypotheses, heterogeneous effects of the treatments are explored by conducting the regression analysis outlined in Equation (7) on subsets of respondents based on their responses provided prior to the treatments. Specifically, Group B of hypothe- ses is tested by conducting the regression analysis outlined in Equation (7) on respondents who believed the tax system was progressive and then separately reproducing this analysis 22 on respondents who did not believe the tax system was progressive. Group C of Hypotheses is tested by conducting the regression analysis outlined in Equation (7) on respondents who prefer the tax system to be progressive and then separately reproducing this analysis on respondents who do not prefer the tax system to be progressive. 4 Findings 4.1 Descriptive statistics 4.1.1 Willingness to pay tax across countries People’s willingness to pay tax varied between the different questions asked and across coun- tries. Figure 3 shows that, depending on the specific question and the country, between 19 and 89 percent of respondents stated that they were willing to pay tax. In five of the eight countries in this study (Colombia, Indonesia, Mexico, Sri Lanka and South Africa) there was broadly similar willingness to pay tax as there was only a 6 to 16 percentage point difference across countries for a given question. On average, people’s willingness to pay tax was more than 10 percentage points higher in Ghana and Tanzania than in the other countries, while the opposite was the case in Jordan. The findings for each question are broadly consistent with the general patterns in the surveys that the questions were sourced from. For example, across multiple rounds of the Afrobarometer (2012, 2013, 2015) between 45 and 63 percent of survey respondents in Tanzania stated that not paying tax was wrong and punishable, while 47 percent of people in Tanzania in this survey agreed with this statement. It is important to keep in mind that these descriptive survey findings about willingness to pay tax across countries cannot be directly compared to actual taxpayer behavior as this information is not available in a standardized way (PWC, 2022; USAID, 2019; World Bank, 2022b). [Figure 3] 23 The characteristics associated with being willing to pay tax varied considerably across the questions that were asked. Multivariate regression analysis shows that the most common pattern across questions is that those aged between 18 and 34 years old were less likely to state they would be willing to pay tax compared to those aged 35 years and older (see Table A12 in Appendix B). In addition, respondents who perceived themselves to be in the middle of the income distribution were more likely to state that they were willing to pay tax and, interestingly, people who thought they were in the richest quintile were often the least likely to state they would be willing to pay tax (although differences were typically not statistically significant). No other background characteristics were consistently associated with answers to the various questions that were used to measure willingness to pay tax across countries. 4.1.2 Beliefs and preferences regarding progressivity in the tax and transfer system across countries On average, almost two-thirds of respondents across the eight countries in this study stated that they prefer richer households to pay a higher share of their income in tax than poorer households, but less than half stated that they believed this was currently the case. In addi- tion, another 15 to 30 percent of respondents stated that they neither agreed nor disagreed with these statements. Figure 4 shows for each country the share of respondents who stated they currently believe the tax system is progressive and those who would prefer this to be the case. Across countries, around one-third of respondents stated a difference between what they believe to be the case and what they would prefer to exist. Respondents in Jordan were the least likely to believe that the tax system was progressive (30 percent) and those in South Africa were the most likely (62 percent). Across these eight countries people’s beliefs and preferences about whether the tax system was progressive were largely unrelated to what is actually the case. These descriptive trends suggest people have limited understanding of how progressive the tax system is in their country, but regardless of people’s beliefs, a sizable majority of people would prefer to have a progressive tax system. 24 [Figure 4] Interestingly, this pattern of preferring progressive taxes and transfers did not vary dra- matically across the income distribution. On average, richer respondents were less supportive of progressive taxation by five to ten percentage points, but still in all countries, in every quintile more respondents agreed than disagreed that richer households should pay a higher share of their income in tax than poorer households (see Figure A2 in Appendix B).6 The consistency of this preference among rich and poor across countries is particularly striking given that it is only actually the case for four of the eight countries in this study (Colombia, Ghana, Mexico and Tanzania). Support for progressive transfers was even more consistent across the income distribution in each country (see Figure A3 in Appendix B). Collectively, these results imply that there is less hostility toward progressive taxes and transfers among people who perceive themselves to be rich than what many may believe to be the case. On the other hand, there is far from universal support for progressive taxes and transfers among people who perceive themselves to be poor. It is important to note that the income distri- bution is constructed based on where people perceive their household to be as opposed to being determined based on reported household income and matching this to national income distribution. Consistent with the finding in Hoy and Mager (2021a), this resulted in most respondents perceiving themselves to be in the middle quintiles and only a very small share stating that they were in the poorest or richest quintile in each country. 4.1.3 Relationship between willingness to pay tax and beliefs about the pro- gressivity in the tax and transfer system across countries The relationship between people’s willingness to pay tax and beliefs about progressivity in the tax and transfer system varies somewhat based on the type of question asked. Figure 5 shows that across all eight countries people who believe the tax system is progressive are more willing to pay tax based on the four indirect questions. This descriptive finding is en- 6 The exception is the richest quintile in Mexico. 25 tirely consistent with the conceptual framework in Section 2. However, Figure 5 also shows that the opposite is the case in many countries when a direct measure is used. Regression analysis confirms these patterns exist, however the positive relationship between believing the tax system is progressive and indirect measures of willingness to pay tax is stronger after controlling for background characteristics (see Table A13 in Appendix B). [Figure 5] 4.2 Main experimental results 4.2.1 Overall effects of the treatments (testing Group A of Hypotheses) The overall impact of the tax treatment illustrates that people’s willingness to pay tax is influenced by whether or not the tax system is progressive (this is consistent with Hypothesis A1). Table 1 shows that respondents who received the taxes treatment in the four countries for which taxes were progressive (Colombia, Ghana, Mexico and Tanzania) were more willing to pay tax. For example, respondents in the taxes treatment group in these countries were 2.3 percentage points more likely to state that they thought paying taxes was important, compared to respondents in the control group. In contrast, respondents who received the taxes treatment in the four countries for which taxes were not progressive (Indonesia, Jordan, Sri Lanka and South Africa) were less willing to pay tax. For example, respondents in the taxes treatment group in these countries were 2.2 percentage points less likely to state that they thought not paying taxes should be punishable, compared to respondents in the control group. Weaker (and often insignificant) results were attained from the transfers and com- bined treatments, although the point estimates were still consistent with respondents who received information that the system was progressive (not progressive) being more (less) willing to pay tax. 26 [Table 1] The results of the taxes treatment across each of the outcome variables are summarized in Figure 6 and show that, in general, there was a consistent result in the order of two to three percentage points. The exception is the direct measure of willingness to pay tax where there were no impacts from the tax treatment in the countries where the tax system was progressive. This is consistent with the descriptive findings that show the direct measure of willingness to pay tax was not positively associated with believing the tax system was progressive in most countries. [Figure 6] 4.2.2 Heterogeneous effects of the treatments based on prior beliefs (testing Group B of Hypotheses) The impact of the taxes and combined treatments varied based on people’s prior beliefs in several instances, in line with Group B of hypotheses. Table 2 shows that the overall neg- ative effect of the taxes treatment in countries where the tax system was not progressive is entirely driven by respondents who held a prior belief that the tax system was progressive. For example, respondents in the taxes treatment group who held a prior belief that the tax system was progressive, but were in countries where this was not the case, were 6.3 percent- age points less likely to state that the government has the right to tax. As such, the taxes treatment appears to be correcting people’s prior beliefs and this impacts their willingness to pay tax in the expected direction (i.e., consistent with the conceptual framework discussed in Section 2). Similarly, there is a positive effect on respondents from the combined treatment who were in countries where the tax and transfer system was progressive, but they did not think this was the case. For example, this subset of respondents in the combined treatment group were 6.0 percentage points more likely to state that government has the right to tax. 27 There were no significant heterogeneous treatment effects in regard to people’s prior beliefs from the transfers treatment. [Table 2] 4.2.3 Heterogeneous effects of the treatments based on existing preferences (testing Group C of Hypotheses) The impact of the taxes and combined treatments also varied based on people’s existing preferences in line with Group C of Hypotheses. Table 3 shows that the overall positive effect of the taxes treatment in countries where the tax system was progressive is entirely driven by respondents who held an existing preference for the tax system to be progressive. For example, respondents in the taxes treatment group who had an existing preference for the tax system to be progressive and were in countries where this was the case, were 2.7 percentage points more likely to state that they thought paying taxes was important. As such, the treatment appears to be increasing people’s willingness to pay tax when it is high- lighting that the actual circumstances in their country are consistent with their preferences. In addition, there is a clear negative effect across multiple outcomes from the treatment among respondents in the taxes treatment group who had an existing preference for the tax system to be progressive, but were in countries where this was not the case. There is also a positive effect from the combined treatment in countries where the tax and transfer system was progressive among respondents who prefer this to be the case. For example, this subset of respondents in the combined treatment group were 2.9 percentage points more likely to state that people should not refuse to pay tax until they receive more in govern- ment transfers. There were no significant heterogeneous treatment effects regarding people’s preferences from the transfers treatment. [Table 3] 28 The heterogeneous treatment effects from the taxes and combined treatments are sum- marized in Figure 7 below. There is a clear pattern whereby the treatment effects were pre- dominantly driven by respondents in cases where the information they received was counter to their prior beliefs and/or consistent with their preferences. [Figure 7] 5 Robustness of the experimental results 5.1 Country level results The main findings of the tax treatment did not vary greatly across countries, which means the pooled regression results discussed in the prior section are not driven by a small number of countries. Figure A4 shows how the point estimates of the overall treatment effects were somewhat similar across countries based on the willingness to pay tax index (the exception is in regards to Ghana). This result clearly shows that in a diverse set of countries a desire for progressivity of taxes is clearly linked to people’s willingness to pay tax (see Table A14 in Appendix B). The heterogeneous treatment effects in each country were also largely con- sistent with the pooled estimates discussed above. 5.2 Differences between the treatments The main results of this study appear to be driven by the content of the treatments, as opposed to simply receiving a treatment. The experiment was designed in a way that al- lowed for comparisons to be made across treatments to rule out concerns that the overall effects were purely due to receiving any information about taxes and transfers. This was 29 possible in six of the eight countries (Ghana, Indonesia, Jordan, Sri Lanka, South Africa and Tanzania) where the direction of the taxes and transfers treatments were opposing one another (e.g., in Ghana the taxes treatment was highlighting that the system was progressive whereas the transfers treatment was stating the opposite) (see Table A15 in Appendix B). Figure A5 summarizes the results showing that across these six countries, there was a clear difference between respondents’ willingness to pay tax, depending on whether the treatment they received indicated that the tax and transfer system was progressive or not progressive (whereby receiving the taxes treatment is coded as 1 and receiving one of the other treat- ments is coded as 0). 5.3 Impact of the treatments across the income distribution The point estimates of the impact of the tax treatment were somewhat consistent across the poorest four quintiles of the income distribution, but there was some evidence of an op- posing effect for the richest quintile (although differences are rarely statistically significant) (see Table A16 in Appendix B). Figure A6 summarizes this by showing that in countries where taxes were progressive, the taxes treatment effect on the willingness to pay (WTP) tax index for the poorest three quintiles was similar and positive (around five percentage points); it was twice as large for the second richest quintile (around ten percentage points) and negative for the richest quintile (around 12 percentage points). In countries where taxes were not progressive, the taxes treatment effect on the WTP tax index for the poorest three quintiles was somewhat similar and negative (around five percentage points) and close to zero for the richest two quintiles. These patterns are consistent with respondents understanding the implications of how a progressive (or not progressive) tax system would likely impact their household income based on their perceived position in the national income distribution. However, as noted in the descriptive results, most respondents perceive themselves to be in the middle quintiles. Only 5 percent of respondents perceived their households to be in the 30 richest quintile, which greatly reduces the precision of the estimates for the richest quintiles, resulting in wide standard errors. As such, it is important to interpret these findings across the income distribution with caution. 5.4 Representativeness of the survey The main results of the randomized survey experiment hold with and without weights applied to adjust the data to match the general population and with and without weights applied to adjust the data to match the characteristics of the internet population that were invited to participate in the survey. Firstly, as described in the methodology (Section 3), the results presented throughout the body of the paper have weights for age and sex to adjust the data to match the general population. In Appendix B, the results are also presented without these weights and the findings are very similar (see Tables A2-A4). Secondly, the characteristics of the population that were invited to participate in the survey were compared to those that completed the survey to examine whether differences existed. The main dimension that was identified is whether people were participating in the survey via a smartphone. Those that were tended to be less likely to participate in the survey in the first place and less likely to complete the survey conditional on starting (see Table A5 in Appendix B). To examine whether this was driving the results I re-weighted the data to match the original composi- tion of respondents (smartphone vs other device types) that were invited to participate in the survey and this did not have a noteworthy impact on the results (see Tables A6-A8 in Appendix B). 5.5 Consideration of alternative dimensions of heterogeneity The dimensions of heterogeneity examined in the previous section (prior beliefs and prefer- ences for progressivity) appear to be the main ways in which the treatments effects varied, 31 especially when compared to alternative dimensions of heterogeneity included in the pre- analysis plan. There were four characteristics that were included as “secondary” hypotheses in the pre-analysis plan and the treatments did not vary greatly on these dimensions. Specif- ically, prior to the treatments respondents were asked about whether they paid a large share of their household income in tax, were net contributors to the tax and transfer system, whether they preferred lower levels of inequality, and if they were working. These were noted as potential dimensions for which heterogeneous treatment effects could exist in the pre-analysis plan, however the direction of heterogeneity were not necessarily clear. The main results on these dimensions are presented in Tables A17–A20 in Appendix B. In gen- eral, there is limited variation between respondents on these dimensions of heterogeneity. It is worth noting that the point estimates regarding the treatment effect from being informed that the tax system was not progressive were larger for respondents who were working and for whom a large share of their household income was paid in tax. A particularly large negative effect on these types of individuals is potentially most concerning for policy makers as they are more likely to be contributing substantially to government revenue. 5.6 Robustness checks The main results of the randomized survey experiment did not vary considerably when con- ducting a series of robustness checks. These checks involved removing respondents who took too long or short a period of time to complete the survey as well as conducting the analysis using alternative econometric specifications (see Tables A21–A22 in Appendix B). In addition, I show that the results are unlikely to be due to differential attrition between the treatment and control groups by using Lee (2009) bounds analysis (see Table A23 in Appendix B). 32 6 Discussion 6.1 Summary of the experimental results and relationship to the pre-registered hypotheses This study has illustrated that progressivity in the tax system influences whether people are willing to pay tax. Respondents who were informed that the tax system was progressive (not progressive) were more (less) likely to pay tax. These results were predominantly driven by respondents in cases where the information they received was counter to their prior beliefs and/or consistent with their preferences. There were weaker (and often insignificant) overall effects from treatments that provided information only about the progressivity of transfers or the net effect of taxes and transfers (although the point estimates were in the same direction). Specifically, the impacts of the taxes treatment were consistent with the related hypotheses (A1, B1 and C1 in Section 2), whereas there was only limited statistically significant evidence related to the combined treatment (A3, B3 and C3 in Section 2) and no significant results in line with the hypotheses related to the transfers treatment (A2, B2 and C2 in Section 2). The differences between the size of the effects of the taxes and other treatments were somewhat anticipated as noted in Section 2 and in the pre-analysis plan.7 The most straight- forward explanation for these differences is that the questions were about the tax system and consequently people were more responsive to information about taxes than transfers. In other words, respondents’ elasticity of willingness to pay tax is higher for information about taxes. Another, potentially compatible, explanation for these results is that people may not necessarily link the taxes they pay with the transfers people receive. It may not be clear to respondents that the structure and generosity of the transfer system has anything to do with paying tax. For example, they may be in favor of the transfer system being progressive 7 Three reasons were noted in the pre-analysis plan. Firstly, on average, the share of household income collected in taxes is much higher than what is provided in transfers, which means people may be more concerned about how taxes are distributed compared to transfers. Secondly, there is reason to believe that “loss aversion” could exist where people’s utility is more likely to be influenced by “losing” from paying tax than by “gaining” from receiving a transfer. Thirdly, people’s awareness of when they pay tax may be higher than their awareness about when they receive a transfer. 33 and want this to continue, but they do not respond to this situation by being more willing to pay tax for a range of reasons, such as not believing the tax they will pay will help pay for transfers. There were some differences between the impact of the treatments on the questions that were used to measure people’s willingness to pay tax, but these differences should be interpreted with caution. Differences across various measures of people’s willingness to pay tax have been observed in the existing literature (Prichard, forthcoming) and this was noted as likely to occur in the pre-analysis plan (it was for this reason that a diverse set of outcomes were included to reduce the risk that measurement error would impact the results). While the tax treatment was more likely to positively impact indirect measures (i.e., all questions other than the direct measure) there were no statistically significant differences between the treatment effects on each of the outcomes, although the lack of effect from the tax treatment on the direct measure of willingness to pay tax is also consistent with the descriptive findings that show a somewhat unexpected negative relationship with beliefs about progressivity. 6.2 Theoretical implications from this study This study has generated rigorous evidence that the progressivity of the tax system shapes people’s willingness to pay tax across countries. As discussed throughout this paper, prior to this study there was limited empirical evidence about how progressivity of taxes and govern- ment transfers shapes people’s willingness to pay tax, particularly in developing countries. The results provide clear evidence supporting a conceptual framework that combines semi- nal theoretical models of tax compliance (Allingham and Sandmo, 1972) and preferences for redistribution (Alesina and Giuliano, 2011) to illustrate the channels through which equity in the tax and transfer system is likely to influence people’s willingness to pay tax. The most immediate theoretical implication from these findings is that research on tax compli- ance needs to engage further with how the progressivity of taxes impacts people’s utility. To put this formally using the utility function in Section 2, the weighting (γit ) people place 34 on the difference between their perceived and preferred levels of progressivity in the tax and t∗ transfer system (Qt bi − Qi ) is non-trivial. As such, the role of equity in the tax system should be considered alongside more commonly cited “quasi-voluntary” motivations for why people pay (or do not pay) tax, such as to keep up with social norms (Hallsworth, 2014), to contribute to the provision of public goods (Giaccobasso et al., 2022), and because they have a positive outlook on the government (Cullen et al., 2021). While there has been some related work along these lines in the United States (Stantcheva, 2021), this study builds on these foundations to illustrate how progressivity in the tax and transfer system in general impacts people’s willingness to pay tax, as well as by showing how generalizable these trends are across a diverse set of developing countries. The order of magnitude of the impact of the taxes treatment on willingness to pay tax was in line with seminal cross-country randomized survey experiments (Alesina et al., 2018; Alesina et al., 2022) and if this translated into actual tax compliance behavior the effects would be non-trivial (e.g., they would be of a similar size to recent work such as Balan et al., 2022). Given the novelty of this study it is challenging to precisely compare the order of magnitude of the treatment effects to related work, however there is a further limitation regarding the nature of the treatment. Ultimately, the information provided to respondents in the randomized survey experiment is largely binary (taxes and/or transfers are either progressive or not) and as a result this means it is not possible to estimate how the order of magnitude of progressivity in the tax and transfer system matters (technically the figures in the treatments provide this information, but it is unlikely to have been fully comprehended by some respondents). The similarity in the impact of the taxes treatment within the two groups of countries (with taxes being either progressive or not) would suggest that the order of magnitude of progressivity was not necessarily a particularly important consideration for respondents. Rather, what influenced respondents was purely whether or not the tax system was progressive (i.e., it was a binary consideration). 35 6.3 Implications for policy makers A key implication for policy makers from this study is that changes to the degree of equity in the tax system will impact people’s willingness to pay tax. As discussed in the introduc- tion, reforms to taxes that intend to improve a country’s fiscal position are likely to change the degree of progressivity in the tax system and it is necessary for policy makers to bet- ter understand people’s responses to such reforms. Tax reforms that improve progressivity could have an additional benefit by increasing people’s willingness to pay tax. The opposite could also be the case, whereby tax reforms that reduce progressivity could in turn decrease people’s willingness to pay tax. In the most extreme case, it is possible that tax reforms that reduce progressivity, which were intended to improve the fiscal position of a country, could undermine tax compliance to a point whereby the net impact on revenue is negative. Ultimately, the exact order of magnitude of these “second round” effects that have an ad- ditional benefit (backfire effect ) from increasing (decreasing) progressivity in the tax system will likely vary over time and across countries. However, the results of this study do suggest that policy makers should take these “second round” effects of tax reform quite seriously. Policy makers can also learn from this research about the benefits from communicating effectively with the general population about the purposes of tax reforms, especially when they are implemented in tandem with changes to the government transfer system. Clearly the results show that most people have a preference for progressive taxes and this can be utilized by policy makers to justify changes to the tax system. Alongside other reasons for tax reform (e.g., improving a country’s fiscal position), communicating the role of taxes in promoting greater equality (when this is actually the case) appears to be an important tool in policy makers’ arsenal, particularly in democratic regimes. Even in the absence of a reform agenda, communicating to taxpayers about the progressive aspects of the tax system in their country would appear to be a way to boost compliance. Further, there appears to be ample scope for information campaigns to be done by policy makers to help the general population understand how taxes help fund the government transfers that benefit so many 36 households. A potential reason why this approach may have been under exploited by policy makers is they are most interested in richer individuals paying tax as ultimately that will collect the most revenue and they are concerned that these taxpayers may be the least receptive to messages about progressivity. Our study presents mixed results on this point. While those that perceived their households to be in the richest quintile were the least supportive of progressive taxes, those in the second richest quintile were the most supportive (although differences were not statistically significant). Further, there was evidence to suggest that people who were working and paid a large share of their household income in tax were particularly supportive of progressive taxes (although once again differences were not statis- tically significant). Collectively, this would seem far from clear evidence to suggest that this concern regarding upsetting richer taxpayers warrants discarding communication campaigns about progressive reforms to taxation. 6.4 Directions for future research A key area for future research that the findings from this study would suggest is worth pursuing is testing how equity in tax and transfer systems influences taxpayer behavior using administrative data (ideally across countries). Randomized survey experiments, including seminal studies by Kuziemko et al. (2015), Alesina et al. (2018), Alesina et al. (2022) and Stantcheva (2021) rely on the use on self-reported outcomes. While this is incredibly useful, a natural next step is to try to link these outcomes to administrative data. This may be more straightforward in a high-income country setting where data about taxpayer behavior is publicly available (e.g., in Scandinavian countries). Another area worthy of greater attention is exploring the extent to which other aspects of “quasi-voluntary” motivations for paying tax exist. For example, a randomized survey experiment examining issues to do with fairness in the tax system would make a large contribution to the literature and may matter more to taxpayers than equity in the tax system. 37 7 Bibliography Afrobarometer, 2012, 2013, 2015. Available online at https://afrobarometer.org/surveys- and-methods/questionnaires Alesina, A., and G. Angeletos, 2005. “Fairness and Redistribution.” American Economic Review, 95(4), 960–980. Alesina, A., and P. Giuliano, 2011. “Preferences for Redistribution.” In Handbook of Social Economics, edited by J. Benhabib, A. Bisin, and M. Jackson, 93–131. Amsterdam: North-Holland. Alesina, A., S. Stantcheva, and E. Teso, 2018. “Intergenerational Mobility and Prefer- ences for Redistribution.” American Economic Review, 108(2), 521–54. Alesina, A., A. Miano, and S. Stantcheva, 2022. “Immigration and Redistribution”. The Review of Economic Studies, rdac011, https://doi.org/10.1093/restud/rdac011 Ali, M., O. Fjeldstad, and I. H. Sjursen, 2014. “To Pay or Not to Pay? Citizens’ Attitudes Toward Taxation in Kenya, Tanzania, Uganda, and South Africa.” World Development, 64, 828–842. Allingham, M., and A. Sandmo, 1972. “Income Tax Evasion: A Theoretical Analysis.” Journal of Public Economics, 1(3–4), 323–338. Antinyan, A., and Z. Asatryan, 2019. “Nudging for Tax Compliance: A Meta-Analysis.” ZEW Discussion Paper No. 19-055. Balan, P., A. Bergeron, G. Tourek, and J. Weigel. 2022. “Local Elites as State Capacity: How City Chiefs Use Local Information to Increase Tax Compliance in the Democratic Republic of the Congo.” American Economic Review, 112(3): 762–97. Becker, G., 1968. “Crime and Punishment: An Economic Approach.” Journal of Political Economy, 76(2), 169–217. Collin, M., V. Di Maro, D. K. Evans, and F. Manang, 2021. “Property Tax Compliance in Tanzania: Can Nudges Help?” Technical Report, Brookings Global Working Paper No.161. Commitment to Equity (CEQ) Institute, 2021. Commitment to Equity Handbook: 38 Estimating the Impact of Fiscal Policy on Inequality and Poverty. Available online at https://commitmentoequity.org/ Cruces, G., R. Perez-Truglia, and M. Tetaz, 2013. “Biased Perceptions of Income Distri- bution and Preferences for Redistribution: Evidence from a Survey Experiment.” Journal of Public Economics, 98, 100–112. Cullen, J. B., N. Turner, and E. Washington, 2021. “Political Alignment, Attitudes to- ward Government, and Tax Evasion.” American Economic Journal: Economic Policy, 13(3), 135–66. De Neve, J., C. Imbert, J. Spinnewijn, T. Tsankova, and M. Luts, 2021. “How to Improve Tax Compliance? Evidence from Population-Wide Experiments in Belgium.” Journal of Political Economy, 129(5), 1425–1463. de Quidt, J., J. Haushofer, and C. Roth, 2018. “Measuring and Bounding Experimenter Demand.” American Economic Review, 108(11), 3266–3302. Dom, R., A. Custers, S. Davenport, and W. Prichard, 2022. Innovations in Tax Compli- ance: Building Trust, Navigating Politics, and Tailoring Reform. Washington, DC: World Bank. Gimpelson, V., and D. Treisman, 2018. “Misperceiving Inequality.” Economics and Poli- tics, 30(1), 27–54. Giaccobasso, M., B. C. Nathan, R. Perez-Truglia, and A. Zentner, 2022. “Where Do My Tax Dollars Go? Tax Morale Effects of Perceived Government Spending.” NBER Working Paper No. 29789, February 2022. Hauser, O., and M. Norton, 2017. “(Mis)perceptions of Inequality.” Current Opinion in Psychology, 18, 21–25. Haaland, I., C. Roth, and J. Wohlfart, Forthcoming. “Designing Information Provision Experiments.” Journal of Economic Literature. Hallsworth, M., 2014. “The use of Field Experiments to Increase Tax Compliance.” Oxford Review of Economic Policy, 30(4), 658–679. Hoy, C., and F. Mager, 2021a. “Why Are Relatively Poor People Not More Supportive 39 of Redistribution? Evidence from a Randomized Survey Experiment across Ten Countries.” American Economic Journal: Economic Policy, 13(4), 299–328. Hoy, C., and F. Mager, 2021b. “American exceptionalism? Differences in the elasticity of preferences for redistribution between the United States and Western Europe.” Journal of Economic Behavior and Organization, 192, 518–540. Hoy, C., L. McKenzie, and M. Sinning, 2021. “Improving Tax Compliance without In- creasing Revenue: Evidence from Population-Wide Randomized Controlled Trials in Papua New Guinea.” Policy Research Working Paper No. 9539. World Bank. Hoy, C., 2022. “How does the progressivity of taxes and government transfers impact people’s willingness to pay tax?” AEA RCT Registry. January 21. https://doi.org/10.1257/rct.8847-1.0 IRIS, 2021. IRIS Network. Available online at http://www.irisnetwork.org/ Karadja, M., J. Mollerstrom, and D. Seim, 2017. “Richer (and Holier) than Thou? The Effect of Relative Income Improvements on Demand for Redistribution.” Review of Economics and Statistics, 99(2), 201–12. Khwaja, A., O. Haq, A. Khan, B. Olken, and M. Shaukat, 2020. “Rebuilding the Social Compact: Urban services delivery and property taxes in Pakistan.” International Initiative for Impact Evaluation. Available online at: https://www.3ieimpact.org/sites/default/files/2020-04/IE117-DPW1.1005-Property-taxes -Pakistan.pdf Kuziemko, I., M. Norton, E. Saez, and S. Stantcheva, 2015. “How Elastic Are Preferences for Redistribution: Evidence from Randomized Survey Experiments.” American Economic Review, 105(4), 1478–1508. Lee, D., 2009. “Training, Wages, and Sample Selection: Estimating Sharp Bounds on Treatment Effects.” Review of Economic Studies 76(3), 1071–1102. Lustig, N. (Ed.), 2018. “Commitment to Equity Handbook. Estimating The Impact of Fiscal Policy on Inequality and Poverty”. Brookings Institution Press. Lustig, N., C. Mariotti, and C. Sánchez-Páramo, 2020. “The redistributive impact of fiscal 40 policy indicator: A new global standard for assessing government effectiveness in tackling inequality within the SDG framework.” World Bank Blogs, June 11, 2020, Available on- line at https://blogs.worldbank.org/opendata/redistributive-impact-fiscal-policy-indicator- new-global-standard-assessing-government Mascagni, G., and C. Nell, 2022. “Tax Compliance in Rwanda: Evidence from a Message Field Experiment.” Economic Development and Cultural Change, 70(2), 587–623. Norton, M., and D. Ariely, 2011. “Building a Better America – One Wealth Quintile at a Time.” Perspectives on Psychological Science, 6(9), 9–12. OECD, 2022. “Income Distribution Database.” Available online at https://stats.oecd.org/Index.aspx?DataSetCode=IDD Pew Research Center. 2019. “Spring 2019 Survey Data.” Pew Research Center: Global Attitudes and Trends. https://www.pewresearch.org/global/dataset/spring-2019-survey-data/ (accessed July 17, 2022). PWC, 2022. “Paying Taxes.” Available online at https://www.pwc.com/gx/en/services/ tax/publications/paying-taxes-2020/explorer-tool.html Po, A. L. W., (1998). Dictionary of Evidence-based Medicine. Radcliffe Publishing. Prichard, W., S. Jibao, and N. Orgeira, Forthcoming. “Sub-National Tax Reform and Tax Bargaining: A Quasi-Randomized Evaluation from Sierra Leone.” ICTD Working Paper. International Centre for Tax and Development. Prichard, W., Forthcoming. “Unpacking ‘Tax Morale’: Distinguishing Between Condi- tional and Unconditional Views of Tax Compliance.” ICTD Working Paper. International Centre for Tax and Development. Prichard, W., A. Custers, R. Dom, S. Davenport, and M. Roscitt, 2019. “Innovations in Tax Compliance: Conceptual Framework.” Policy Research Working Paper No. 9032. World Bank. Shimeles, A., D. Gurara, and F. Woldeyes, 2017. “Taxman’s Dilemma: Coercion or Per- suasion? Evidence from a Randomized Field Experiment in Ethiopia.” American Economic Association: Papers and Proceedings, 107(5), 420–424. 41 Slemrod, J., 2019. “Tax Compliance and Enforcement.” Journal of Economic Literature, 57(4), 904–954. Stantcheva, S., 2021. “Understanding Tax Policy: How do People Reason?” The Quar- terly Journal of Economics, 136(4), 2309–2369. USAID, 2019. "Collecting Taxes.” Available online at https://www.usaid.gov/documents /1865/collecting-taxes-2017-2018-program-document-pdf World Bank, 2021. “World Development Indicators.” Washington, DC: World Bank. Available online at https://databank.worldbank.org/source/world-development-indicators World Bank, 2022a. “World Development Report 2022: Finance for an Equitable Recov- ery.” Washington, DC: World Bank. World Bank, 2022b. “Enterprise Surveys.” Washington, DC: World Bank. Available online at https://www.enterprisesurveys.org/en/custom-query World Values Survey (WVS), 2020. “Wave 7.” Available online at: https://www.worldvaluessurvey.org/WVSDocumentationWV7.jsp 42 Tables and Figures Table 1: Overall effects of the treatments Direct Punishable Important Right to Tax Do not Refuse INDEX b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Progressive) 0.008 0.022 0.023** 0.016 0.013 0.036** (0.01) (0.02) (0.01) (0.01) (0.01) (0.01) p-value 0.594 0.272 0.032 0.140 0.255 0.030 Observations 7605 7605 7605 7605 7605 7605 Taxes (Not Progressive) -0.022 -0.022* -0.012 -0.027 -0.028 -0.048** (0.01) (0.01) (0.02) (0.01) (0.02) (0.01) p-value 0.213 0.061 0.517 0.131 0.328 0.011 Observations 7435 7435 7435 7435 7435 7435 Transfers (Progressive) -0.006 0.009 0.016 -0.002 0.004 0.009 (0.01) (0.01) (0.01) (0.02) (0.01) (0.02) p-value 0.569 0.461 0.259 0.914 0.611 0.653 Observations 11318 11318 11318 11318 11318 11318 Transfers (Not Progressive) -0.014 0.007 0.002 -0.002 -0.008 0.000 (0.01) (0.01) (0.01) (0.03) (0.01) (0.01) p-value 0.358 0.652 0.853 0.966 0.423 0.976 Observations 3810 3810 3810 3810 3810 3810 Combined (Progressive) 0.000 0.008 0.021** 0.020 0.012* 0.025* (0.01) (0.01) (0.01) (0.01) (0.00) (0.01) p-value 0.997 0.479 0.049 0.101 0.052 0.098 Observations 11066 11066 11066 11066 11066 11066 Combined (Not Progressive) -0.011 -0.024 0.012 0.009 -0.021 -0.010 (0.02) (0.01) (0.01) (0.03) (0.02) (0.02) p-value 0.714 0.217 0.354 0.806 0.405 0.732 Observations 3769 3769 3769 3769 3769 3769 Note: This table is based on Equation 7 in Section 3 of the paper. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the overall impact of each of the treatments relative to the control group, where countries are pooled based on whether the tax and/or transfer system is progressive. Table 2: Heterogeneous effects of the treatments based on prior beliefs about progressivity Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents believed richer households pay more tax than poorer households Taxes (Progressive) 0.018 0.038** 0.010 0.016 0.037 0.053 (0.01) (0.01) (0.01) (0.03) (0.02) (0.03) Taxes (Not Progressive) -0.017 -0.047** -0.038 -0.063** -0.026 -0.080** (0.03) (0.01) (0.02) (0.01) (0.04) (0.02) Transfers (Progressive) 0.032 0.015 0.004 -0.019 0.018 0.021 (0.03) (0.01) (0.02) (0.03) (0.02) (0.03) Transfers (Not Progressive) -0.005 0.034 -0.018 -0.010 0.004 -0.019 (0.00) (0.02) (0.00) (0.00) (0.03) (0.03) Combined (Progressive) 0.009 -0.004 -0.005 -0.005 0.013 0.003 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.026 -0.031 0.015* 0.020 -0.044 -0.021 (0.01) (0.02) (0.00) (0.02) (0.02) (0.01) Panel B - Respondents did not believe richer households pay more tax than poorer households Taxes (Progressive) 0.002 0.010 0.032 0.015 -0.006 0.023 (0.01) (0.02) (0.01) (0.02) (0.02) (0.02) Taxes (Not Progressive) -0.024 0.001 0.012 0.005 -0.029 -0.015 (0.02) (0.01) (0.02) (0.02) (0.01) (0.02) Transfers (Progressive) -0.033* 0.005 0.024* 0.010 -0.007 0.000 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Transfers (Not Progressive) 0.004 -0.007 0.013 0.001 -0.014 0.006 (0.00) (0.01) (0.01) (0.04) (0.02) (0.02) Combined (Progressive) -0.016 0.022 0.059*** 0.060** 0.008 0.057** (0.02) (0.01) (0.01) (0.02) (0.01) (0.02) Combined (Not Progressive) 0.008 -0.012 0.005 -0.012 0.009 0.001 (0.04) (0.00) (0.01) (0.04) (0.01) (0.03) Note: This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on their prior beliefs about progressivity. Beliefs about progressivity are based on Q8, which asks respondents whether they believe that richer households pay a higher share of their income in tax than poorer households. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects based on respondents prior beliefs about whether taxes were progressive, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table 3: Heterogeneous effects of the treatments based on existing preferences about progressivity Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents preferred richer households to pay more tax than poorer households Taxes (Progressive) 0.018 0.031 0.027* 0.016 0.025 0.051*** (0.01) (0.02) (0.01) (0.01) (0.01) (0.01) Taxes (Not Progressive) -0.022 -0.040*** -0.030 -0.045* -0.018 -0.066*** (0.02) (0.00) (0.02) (0.02) (0.03) (0.01) Transfers (Progressive) 0.000 0.022 -0.001 -0.005 0.006 0.010 (0.01) (0.02) (0.02) (0.03) (0.01) (0.02) Transfers (Not Progressive) -0.005 -0.000 -0.007 -0.015 -0.005 -0.016 (0.01) (0.01) (0.01) (0.01) (0.01) (0.00) Combined (Progressive) 0.014 0.007 0.005 0.015 0.029** 0.030*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.006 -0.028 0.017* 0.008 -0.028 -0.009 (0.01) (0.01) (0.00) (0.02) (0.01) (0.02) Panel B - Respondents did not prefer richer households to pay more tax than poorer households Taxes (Progressive) -0.011 0.004 0.012 0.014 -0.006 0.006 (0.02) (0.02) (0.02) (0.01) (0.01) (0.02) Taxes (Not Progressive) -0.024 0.003 0.011 -0.000 -0.045 -0.023 (0.03) (0.01) (0.02) (0.02) (0.02) (0.03) Transfers (Progressive) -0.015 -0.009 0.039** -0.000 0.002 0.007 (0.02) (0.02) (0.01) (0.02) (0.01) (0.02) Transfers (Not Progressive) 0.011 0.023 0.014 0.020 -0.008 0.031 (0.02) (0.01) (0.03) (0.06) (0.01) (0.03) Combined (Progressive) -0.053* 0.004 0.056** 0.025 -0.049 -0.007 (0.02) (0.02) (0.02) (0.03) (0.03) (0.04) Combined (Not Progressive) -0.027 -0.004 -0.011 0.003 0.016 -0.009 (0.06) (0.01) (0.02) (0.07) (0.03) (0.06) Note: This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on their prior preferences for progressivity. Preferences about progressivity are based on Q9, which asks respondents whether they think that richer households should pay a higher share of their income in tax than poorer households. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects based on respondents existing preferences about whether taxes should be progressive, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Figure 1: Difference between the Gross and Net GINI index in developing countries Note: This figure is based on 45 low- and middle-income countries for which comparable data exists (CEQ, 2021). The level of tax revenue and income classification for each country is based on the year in which the survey took place that was used to measure the difference between the Gross and Net GINI index. For presentational purposes, South Africa is excluded as the difference between the Gross and Net GINI index is substantially larger than for any other country (around 10 percentage points). This figure shows the difference between the gross (i.e., pre-taxes and government transfers) and net (i.e., post-taxes and government transfers) GINI index is negligible in some developing countries and far more substantial in others. This variation exists even between countries with similar levels of income and tax revenue as a share of GDP. Source: CEQ, 2021 Figure 2: Taxes and government transfers (both direct and indirect) as a fraction of household income Note: Taxes are displayed as negative because they reduce household income. In South Africa, taxes and transfers as a fraction of household income is greater than 1 for the poorest quintile. This is possible because household consumption is higher than household income for the these households. This figure shows the average fraction of household income for each quintile in each country that is directly or indirectly paid in taxes or received in government transfers as well as the net impact of taxes and transfers on household income. Source: CEQ, 2021 Figure 3: Willingness to pay tax across countries according to different indicators Note: CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Pay w/o enforcement: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse to Pay: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). This figure shows the share of respondents stating they are willing to pay tax in each country according to different questions. Figure 4: Beliefs and preferences regarding the progressivity of taxes across countries Note: CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Beliefs about progressivity are based on Q8, which asks respondents whether they believe that richer households pay a higher share of their income in tax than poorer households. Preferences about progressivity are based on Q9, which asks respondents whether they think that richer households should pay a higher share of their income in tax than poorer households. This figure shows the share of respondents stating they had a prior belief and/or an existing preference that the tax system is progressive in their country. Figure 5: Relationship between willingness to pay tax and beliefs about the progressivity of taxes across countries Note: CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Believe progressive: Based on Q8, which asks respondents whether they believe that richer households pay a higher share of their income in tax than poorer households (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Prefer progressive: Based on Q9, which asks respondents whether they think that richer households should pay a higher share of their income in tax than poorer households (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). This figure shows how willingness to pay tax varies based on prior beliefs about whether the tax system is progressive. Figure 6: Overall impact of the tax treatment on each of the measures of willingness to pay tax Note: Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). WTP tax INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. 90 percent confidence intervals are displayed in this figure. This figure shows the overall impact of the tax treatment on each of the measures of willingness to pay tax, where countries are pooled based on whether taxes are actually progressive. Figure 7: Summary of the heterogeneous treatment effects from the taxes and combined treatments Note: Only the results for the combined treatment are shown for countries in which the tax and transfer system is progressive. There are no effects from the combined treatment in countries for which the tax and transfer system is not progressive (see Tables 2 and 3). 90 percent confidence intervals are displayed in this figure. Perceive currently progressive: Based on Q8, which asks respondents whether they believe that richer households pay a higher share of their income in tax than poorer households (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Prefer progressive: Based on Q9, which asks respondents whether they think that richer households should pay a higher share of their income in tax than poorer households (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). WTP tax INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This figure summarizes the heterogeneous treatment effects from the tax and combined treatments based on respondents prior beliefs and existing preferences for progressivity of taxes. List of Appendices Appendix A - Details about the survey methodology Appendix B - Additional tables and figures Appendix C - Survey instrument (English version) Appendix D - Country specific treatments (English version) Appendix A - Details about the survey methodology Approach to data collection Ideally, face-to-face surveys collecting a representative sample of the general pop- ulation using a sampling frame, such as a recent census, would have been conducted in each of the countries in this study. Not only are the costs involved in doing this prohibitive, but there are also issues with conducting face-to-face surveys during a pandemic. While phone surveys present a popular alternative, this is not an appro- priate format for a survey along these lines. The treatments are designed to be visual in nature and it is not possible to communicate these messages fully via a phone call. This left an online survey as the most promising option for data collection, even though there are challenges with representativeness that need to be recognized and can be overcome to some extent. A major challenge with conducting an online randomized survey experiment in low- and middle-income countries is collecting a representative sample of the total population. Unlike high-income countries where internet access is near universal, the share of the total population with internet access in the countries in this study varies from 20 to 67 percent. Furthermore, there is limited existing online survey “infras- tructure”, such as what exists in many high-income countries where market research firms run online opinion polls daily from a large pool of pre-registered respondents who regularly complete surveys. This is far less common in low- and middle-income countries and there are reasons to be concerned about just how narrow a subset of the population would participate in an engagement like this. Similar concerns exist regarding the use of online labor platforms, such as MTurk, in a low- and middle- income country context. Alternative approaches to online data collection in low- and middle-income coun- tries can crudely be categorized as providing “opt-in” or “opt-out” options. An ex- ample of the former would be to use social media advertisements to invite people to participate in an online survey. While this “opt-in” approach may be attractive as it is easy to implement, I identified at least two shortcomings that I felt meant this approach was not ideal for this study. Firstly, there is a clear concern regarding selection bias as people who would “opt in” to a survey based on a social media adver- tisement potentially have some unobservable characteristics that make them distinct from the rest of the population. It is challenging to estimate the extent to which these unobservable characteristics exist without gaining access to administrative data from social media providers. Secondly, as I was asking about a sensitive topic (tax compliance) it is possible that participants would not provide honest answers as they could easily be identified through the platform that they were opting into the survey on (e.g., Facebook). As such, on balance I felt that an “opt-in” approach along these lines would not be ideal for this study. Despite these concerns regarding “opt-in” online data collection via social media, I still attempted to pilot the randomized survey experiment via Facebook and In- stagram in the countries with the two smallest “internet” populations in this study (Tanzania and Jordan). These countries were chosen as reaching a large enough sample size to have statistical power to detect effects from the treatments would be the most challenging in these settings. The survey was non-incentivized (to minimize concerns about experimenter demand effects and to ensure respondents did not need to provide identifiable information) and to comply with research ethics protocols the social media advertisements stated that respondents would be asked questions about taxes. Partly because of these constraints it was not possible to solicit even half of the total respondents required for the survey via this sampling method in Tanzania and Jordan, despite the social media advertisements reaching millions of unique so- cial media users over a period of two months. These challenges that were faced when trying to pilot an “opt-in” approach to the survey provided further rationale behind using an alternative approach for this study. Data was collected for the online randomized survey experiment in this paper us- ing an “opt-out” approach offered by the survey firm, RIWI. They capture a sample of respondents that is broadly representative of the internet population in each coun- try by using Random Domain Intercept Technology. This involves sampling internet users who incidentally access expired or inactive domains (i.e., which often result in a “404 error”). As domain names regularly change and they often do not automat- ically redirect internet users it is commonplace for the internet using population to incidentally access inactive domains. Research suggests the likelihood of accessing an inactive domain is approximately proportional to having access to the internet (IRIS, 2021). RIWI exploits this by redirecting users from inactive domains to a website inviting them to take part in a survey. At this point people can decide whether to continue to participate in the survey or “opt out”. RIWI tracks information about the device used and operating system used by people who are redirected towards to the survey platform, even if they do not answer a single question. In addition, the first question people are asked is about their age and sex. As a result, I observe how “opt-out” rates from a representative sample of the internet population vary based on the characteristics of respondents (for example, I am able to measure whether people using smartphones disproportionally opt-out of the survey). A shortcoming of this “opt-out” approach is that high rates of attrition occur early in the survey. However, given that I track how attrition varies by the characteristics of respondents and the survey experiment is at the back end of the survey, this does not undermine the integrity of the study. Pilot data The proposed survey instrument went through an extensive review process within the World Bank prior to being piloted in December 2021. The internal review pro- cess identified ways in which the survey instrument could reflect best practice in the literature (e.g., avoiding ceiling effects on the outcome variables by phrasing ques- tions to ensure greater variation of responses across a Likert scale). Reviewers also emphasized that during the piloting process it will be crucial to examine whether respondents adequately comprehend the treatments and the questions. As such the primary focus of the piloting that took place was to ensure the responses that were gathered indicated the respondents understood the survey instrument. In addition, piloting provided an opportunity to verify the assumptions made about the size of the treatment effects in the statistical power calculations and to identify ways in which the experiment could be designed in a manner to minimize attrition. These three issues are discussed one by one below following a description of the piloting process. Implementation of the piloting process The survey instrument and experiment were piloted with 1,061 respondents (who completed the survey) that made up a representative sample of the internet popu- lation in India in December 2021. India was selected as an appropriate location to pilot the survey as this is where the survey firm typically conducts pilots (due to the diverse, but very large, population where English is commonly used on the internet); it has a similar level of development to many of the countries in the full study and as I was not including India in the full study, I did not need to be concerned about contaminating the pool of respondents. There were two phases to the pilot. The first phase involved using visual stimuli for some of the questions (somewhat similar to what Hoy and Mager (2021b) used in high-income countries) capturing people’s prior beliefs and preferences about the distribution of taxes and transfers as well as levels of inequality in their country. In this version of the survey instrument that had been approved through the internal review process at the World Bank, respondents were required to select the distri- bution of taxes and transfers that exists in their country based on actual examples. Specifically, the options provided for respondents to select from were based on the actual progressivity of taxes in Tanzania in 2011, Colombia in 2014 and Jordan in 2017. In addition, respondents were randomly allocated to receive questions from a pool of seven potential questions about their willingness to pay tax. This process helped to inform which five questions should be included in the full study. In total, 511 respondents completed this phase of the pilot. The main change in the second phase of the pilot was replacing the questions from the first phase that involved visual stimuli with basic questions that aimed to capture people’s prior beliefs and preferences about the distribution of taxes and transfers as well as levels of inequality in their country on a Likert scale. This ap- proach brought the format of these questions into line with the rest of the survey. A shortcoming of this approach was that it was no longer possible to identify whether people’s beliefs and preferences matched examples of the actual level of progressivity of taxes in some low- and middle-income countries. In the second phase of the pi- lot, respondents continued to be randomly allocated to receive a subset of questions about their willingness to pay tax. In total, 550 respondents completed this phase of the pilot. Lessons learned through the piloting process There were three key lessons that emerged from the two phases of the pilot that informed the final survey instrument. Firstly, there was a clear need to keep the survey instrument as simple as possible. Answers to the questions that included vi- sual stimuli in the first phase of the pilot suggested respondents did not adequately comprehend the options they were presented with. Responses were very evenly dis- tributed across the options in each of the four questions about people’s beliefs and preferences in regard to the distribution of taxes and transfers in their country. To test whether this was primarily due to measurement error, in the second phase of the pilot respondents were randomly allocated to either the question format from phase one or basic questions about their views on how taxes and transfers are dis- tributed using a Likert scale. The results were substantially different between these approaches with the basic question format returning results far more consistent with previous literature. Specifically, the results showed most people tend to prefer richer households to pay more taxes than poorer households and poorer households to re- ceive more government transfers than richer households (i.e., most people tend to prefer progressivity in the tax and transfer system). As such I decided that the final survey instrument should rely on these basic questions to capture people’s prior be- liefs and preferences, even though this means that the options provided are not based on actual progressivity of taxes and transfers in countries. I believe that capturing higher quality, reliable responses is of greater importance. Secondly, the results of the piloting process provided me with confidence that the sample size in the final study would be adequate. The point estimates of the treatment effects were promising as they indicated variation between respondents in the treatment and control groups of an order of magnitude that I would be powered to detect at standard levels (i.e., an alpha of 0.05 and beta of 0.2) when the full sample of respondents is reached (i.e., 3,600 as opposed to 1,061). The direction of the treatment effects was also often in line with the primary hypotheses of this study. Thirdly, the piloting process highlighted ways to minimize attrition during the survey experiment and the most straightforward way was by removing list experi- ments from the study. Specifically, there was low attrition for the outcome variable questions included in the final survey instrument, whereas around one quarter of respondents dropped out during the two list experiments that were included in the pilot. Removing the list experiments from the randomized survey experiment was not a major issue for our study as there is debate in the literature about the value of this approach in general and I would have potentially faced considerable issues with inadequate statistical power. In the second phase of the pilot I also randomized al- ternative data quality check questions between respondents immediately prior to the survey experiment and found that our original question from the first phase of the pilot outperformed an alternative question that was used by Alesina et al. (2018). As such, I felt confident that including a question that asks respondents to drop out prior to the treatment if they are unwilling to complete the survey experiment would serve as an effective way to minimize attrition post treatment. I was also reassured by the lack of differential attrition observed throughout the piloting process. Coding of variables Q0 – Age - age1834= 1 if respondent aged 18–34 years (respondents under 18 auto- matically discarded), 0 if respondent aged 35 years or older Q0 – Sex - male = 1 if respondent male, 0 otherwise Q1 – Education - edusecorless = 1 if respondent selected primary or secondary edu- cation, 0 otherwise Q2 – Location - largecity = 1 if respondent selected large city or suburb, 0 otherwise Q3 – Employment type - working= 1 if respondent selected employee or self-employed/small business owner, 0 otherwise Q4 – Prefer lower inequality - lowerineq = 1 if respondent selects strongly agree or agree, 0 otherwise Q5 – Perceived position in national income distribution - pB40 = 1 if respondent selected poorest or second poorest quintile, 0 otherwise Q6 – Household paid large share of income in tax - largetax = 1 if respondent selects strongly agree or agree, 0 otherwise Q7 – Household paid more in tax than received in transfers - netcont = 1 if respon- dent selects strongly agree or agree, 0 otherwise Q8 – Perceived taxes as currently progressive - curprogtax = 1 if respondent selects strongly agree or agree, 0 otherwise Q9 – Prefer taxes to be progressive - progtax = 1 if respondent selects strongly agree or agree, 0 otherwise Q10 – Perceived transfers as currently progressive - curprogtrans = 1 if respondent selects strongly agree or agree, 0 otherwise Q11 – Prefer transfers to be progressive - progtrans = 1 if respondent selects strongly agree or agree, 0 otherwise Q12 – Data quality check - willcomplete = 1 if respondent selected yes, 0 otherwise TREATMENT PROVIDED Q14 – Will not pay without enforcement - willpaytax = 0 if respondent selects strongly agree or agree, 1 otherwise Q15 – Not paying tax is wrong and punishable - wrongpunish = 1 if respondent selected wrong and punishable, 0 otherwise Q16 – Paying taxes is important - importanttopay = 1 if respondent selects strongly agree or agree, 0 otherwise Q17 – Government has right to pay tax - righttotax = 1 if respondent selects strongly agree or agree, 0 otherwise Q18 – Do not Refuse - donotrefusepaytax = 1 if respondent selects strongly disagree or disagree, 0 otherwise Appendix B - Additional Tables Table A1: Age and sex of survey respondents and the general adult population Male (%) 18-34 years (%) Male (%) 18-34 years (%) survey survey population population Colombia 59.7 56.2 49.1 43.8 Ghana 78.7 78.1 50.7 55.6 Indonesia 67.2 73.3 50.4 43.4 Jordan 57.2 70.2 50.6 53.3 Mexico 62.5 53.5 48.9 45.1 South Africa 58.7 60.9 49.3 49.1 Sri Lanka 76.9 62.3 48.0 36.6 Tanzania 70.8 79.4 50.0 60.0 This table shows the age and sex of survey respondents compared to the general adult population in each country. Source: World Bank, 2021 Table A2: Overall effects of the treatments (without weights) Direct Punishable Important Right to Tax Do not Refuse INDEX b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Progressive) 0.005 0.014 0.015* 0.018* 0.010 0.028** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) p-value 0.420 0.332 0.083 0.092 0.260 0.020 Observations 7605 7605 7605 7605 7605 7605 Taxes (Not Progressive) -0.025 -0.013 -0.007 -0.019 -0.024 -0.038* (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) p-value 0.159 0.387 0.454 0.180 0.126 0.088 Observations 7435 7435 7435 7435 7435 7435 Transfers (Progressive) -0.005 0.015 0.012 -0.008 0.008 0.009 (0.01) (0.01) (0.01) (0.01) (0.00) (0.02) p-value 0.683 0.144 0.314 0.493 0.121 0.610 Observations 11318 11318 11318 11318 11318 11318 Transfers (Not Progressive) -0.014 0.002 0.002 -0.005 -0.007 -0.008 (0.01) (0.01) (0.00) (0.02) (0.01) (0.00) p-value 0.358 0.899 0.567 0.848 0.449 0.137 Observations 3810 3810 3810 3810 3810 3810 Combined (Progressive) -0.001 0.011 0.014 0.010 0.011* 0.019 (0.01) (0.01) (0.01) (0.01) (0.00) (0.01) p-value 0.924 0.401 0.169 0.294 0.068 0.196 Observations 11066 11066 11066 11066 11066 11066 Combined (Not progressive) -0.019 -0.025 0.011 -0.001 -0.020 -0.018 (0.01) (0.01) (0.00) (0.02) (0.02) (0.02) p-value 0.297 0.168 0.132 0.977 0.454 0.444 Observations 3769 3769 3769 3769 3769 3769 Note: This table is directly comparable to Table 1 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 oth- erwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the overall impact of each of the treatments (without weights) relative to the control group, where countries are pooled based on whether the tax and/or transfer system is progressive. Table A3: Heterogeneous effects of the treatments based on prior beliefs about progressivity (without weights) Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents believed richer households pay more tax than poorer households Taxes (Progressive) 0.012** 0.011** 0.005 0.011 0.029* 0.031 (0.00) (0.00) (0.01) (0.03) (0.01) (0.02) Taxes (Not Progressive) -0.039 -0.044* -0.026* -0.039** -0.029 -0.074** (0.02) (0.02) (0.01) (0.01) (0.02) (0.01) Transfers (Progressive) 0.019 0.017 -0.001 -0.015 0.019* 0.016 (0.02) (0.01) (0.02) (0.02) (0.01) (0.02) Transfers (Not Progressive) -0.030 0.023 -0.004 -0.022 -0.008 -0.019 (0.01) (0.01) (0.01) (0.01) (0.04) (0.03) Combined (Progressive) -0.001 -0.004 -0.003 -0.007 0.007 -0.004 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.020 -0.027 0.014 0.011 -0.031 -0.017 (0.01) (0.01) (0.01) (0.01) (0.03) (0.02) Panel B - Respondents did not believe richer households pay more tax than poorer households Taxes (Progressive) 0.000 0.015 0.023 0.023 -0.006 0.024 (0.01) (0.02) (0.01) (0.02) (0.02) (0.02) Taxes (Not Progressive) -0.011 0.015 0.013 -0.001 -0.019 -0.002 (0.02) (0.02) (0.01) (0.01) (0.01) (0.02) Transfers (Progressive) -0.020 0.015* 0.022* -0.004 0.000 0.006 (0.02) (0.01) (0.01) (0.01) (0.01) (0.01) Transfers (Not Progressive) -0.004 -0.009 0.007 0.003 -0.006 -0.001 (0.01) (0.02) (0.00) (0.03) (0.03) (0.01) Combined (Progressive) -0.002 0.030 0.040*** 0.036** 0.015 0.050** (0.02) (0.02) (0.00) (0.01) (0.01) (0.02) Combined (Not Progressive) -0.016 -0.023 0.002 -0.024 -0.005 -0.027 (0.03) (0.01) (0.02) (0.02) (0.00) (0.01) Note: This table is directly comparable to Table 2 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on their prior beliefs about progressivity. Beliefs about progressivity are based on Q8, which asks respondents whether they believe that richer households pay a higher share of their income in tax than poorer households. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects (without weights) based on respondents prior beliefs about whether taxes were progressive, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A4: Heterogeneous effects of the treatments based on existing preferences about progressivity (without weights) Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents preferred richer households to pay more tax than poorer households Taxes (Progressive) 0.015* 0.021 0.019** 0.019 0.024 0.044*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.00) Taxes (Not Progressive) -0.030** -0.033* -0.017 -0.025** -0.017 -0.052** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Transfers (Progressive) 0.003 0.020 -0.004 -0.012 0.008 0.006 (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) Transfers (Not Progressive) -0.025 0.000 0.001 -0.008 -0.015 -0.019* (0.01) (0.00) (0.00) (0.01) (0.00) (0.00) Combined (Progressive) 0.015 0.004 0.007 0.009 0.025*** 0.025** (0.01) (0.01) (0.01) (0.01) (0.00) (0.01) Combined (Not Progressive) -0.017* -0.028 0.017 0.007 -0.028 -0.014 (0.00) (0.00) (0.01) (0.02) (0.02) (0.02) Panel B - Respondents did not prefer richer households to pay more tax than poorer households Taxes (Progressive) -0.011 0.003 0.006 0.015** -0.012* -0.001 (0.01) (0.02) (0.02) (0.00) (0.00) (0.02) Taxes (Not Progressive) -0.019 0.015 0.001 -0.014 -0.035 -0.022 (0.03) (0.02) (0.02) (0.03) (0.02) (0.04) Transfers (Progressive) -0.014 0.011 0.036** -0.003 0.010* 0.017 (0.02) (0.02) (0.01) (0.02) (0.00) (0.02) Transfers (Not Progressive) 0.002 0.006 0.004 -0.001 0.007 0.009 (0.02) (0.04) (0.01) (0.04) (0.02) (0.00) Combined (Progressive) -0.051* 0.027 0.030 0.011 -0.032 -0.007 (0.02) (0.03) (0.02) (0.02) (0.02) (0.04) Combined (Not Progressive) -0.020 -0.011 -0.017 -0.041 0.021 -0.034 (0.04) (0.01) (0.03) (0.03) (0.02) (0.01) Note: This table is directly comparable to Table 3 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on their prior preferences for progressivity. Preferences about progressivity are based on Q9, which asks respondents whether they think that richer households should pay a higher share of their income in tax than poorer households. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects (without weights) based on respondents existing preferences about whether taxes should be progressive, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A5: Share of participants using a smartphone at various stages of the survey Exposed to survey (%) Began experiment (%) Completed survey (%) Colombia 62.2 54.0 53.8 Ghana 50.4 66.7 66.6 Indonesia 72.4 78.0 77.2 Jordan 79.9 73.0 72.5 Mexico 62.5 52.6 52.5 South Africa 64.7 63.2 63.0 Sri Lanka 75.9 70.7 70.2 Tanzania 83.6 81.2 80.7 This table shows the share of participants using a smartphone that were exposed to the survey, begin the survey experiment and completed the survey. Table A6: Overall effects of the treatments (with device type weights) Direct Punishable Important Right to Tax Refuse to Pay INDEX b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Progressive) 0.003 0.014 0.014 0.017 0.010 0.027** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) p-value 0.637 0.370 0.102 0.108 0.286 0.020 Observations 7605 7605 7605 7605 7605 7605 Taxes (Not Progressive) -0.026 -0.015 -0.007 -0.019 -0.025 -0.039* (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) p-value 0.147 0.352 0.383 0.179 0.123 0.079 Observations 7435 7435 7435 7435 7435 7435 Transfers (Progressive) -0.007 0.015 0.011 -0.009 0.007 0.007 (0.01) (0.01) (0.01) (0.01) (0.00) (0.02) p-value 0.583 0.162 0.362 0.487 0.134 0.683 Observations 11318 11318 11318 11318 11318 11318 Transfers (Not Progressive) -0.019 -0.002 0.002 -0.008 -0.007 -0.013 (0.01) (0.01) (0.00) (0.03) (0.01) (0.01) p-value 0.190 0.876 0.674 0.797 0.442 0.267 Observations 3810 3810 3810 3810 3810 3810 Combined (Progressive) -0.004 0.008 0.012 0.010 0.011* 0.016 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) p-value 0.625 0.507 0.210 0.312 0.080 0.254 Observations 11066 11066 11066 11066 11066 11066 Combined (Not progressive) -0.021 -0.027 0.012 -0.002 -0.021 -0.021 (0.01) (0.01) (0.00) (0.02) (0.02) (0.02) p-value 0.327 0.137 0.182 0.932 0.465 0.453 Observations 3769 3769 3769 3769 3769 3769 Note: This table is directly comparable to Table 1 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 oth- erwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the overall impact of each of the treatments (with device type weights) relative to the control group, where countries are pooled based on whether the tax and/or transfer system is progressive. Table A7: Heterogeneous effects of the treatments based on prior beliefs about progressivity (with device type weights) Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents believed richer households pay more tax than poorer households Taxes (Progressive) 0.012* 0.014** 0.005 0.012 0.029** 0.033 (0.00) (0.00) (0.01) (0.03) (0.01) (0.02) Taxes (Not Progressive) -0.040* -0.043* -0.027* -0.041** -0.029 -0.076** (0.02) (0.02) (0.01) (0.01) (0.02) (0.01) Transfers (Progressive) 0.020 0.017 -0.001 -0.015 0.020* 0.017 (0.02) (0.01) (0.02) (0.02) (0.01) (0.02) Transfers (Not Progressive) -0.037 0.017 -0.006 -0.024 -0.011 -0.028 (0.01) (0.01) (0.02) (0.01) (0.04) (0.03) Combined (Progressive) -0.004 -0.004 -0.003 -0.007 0.009 -0.004 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.022 -0.029 0.014 0.011 -0.033 -0.018 (0.01) (0.01) (0.01) (0.01) (0.04) (0.02) Panel B - Respondents did not believe richer households pay more tax than poorer households Taxes (Progressive) -0.003 0.012 0.021 0.022 -0.006 0.021 (0.01) (0.02) (0.01) (0.02) (0.02) (0.02) Taxes (Not Progressive) -0.011 0.012 0.012 0.001 -0.019 -0.003 (0.02) (0.02) (0.01) (0.01) (0.01) (0.02) Transfers (Progressive) 0.023 0.014 0.019 -0.005 -0.001 0.002 (0.02) (0.01) (0.01) (0.01) (0.01) (0.01) Transfers (Not Progressive) -0.008 -0.012 0.007 -0.001 -0.005 -0.005 (0.00) (0.02) (0.01) (0.03) (0.03) (0.01) Combined (Progressive) -0.007 0.024 0.036*** 0.037** 0.014 0.044* (0.02) (0.02) (0.01) (0.01) (0.01) (0.02) Combined (Not Progressive) -0.018 -0.025 0.003 -0.029 -0.004 -0.031 (0.03) (0.01) (0.02) (0.03) (0.00) (0.01) Note: This table is directly comparable to Table 2 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on their prior beliefs about progressivity. Beliefs about progressivity are based on Q8, which asks respondents whether they believe that richer households pay a higher share of their income in tax than poorer households. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects (with device type weights) based on respondents prior beliefs about whether taxes were progressive, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A8 - Heterogeneous effects of the treatments based on existing preferences about progressivity (with device type weights) Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents preferred richer households to pay more tax than poorer households Taxes (Progressive) 0.012 0.020 0.021** 0.019 0.025 0.045*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.00) Taxes (Not Progressive) -0.030** -0.034* -0.018 -0.026** -0.017 -0.053** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Transfers (Progressive) 0.004 0.019 -0.005 -0.011 0.008 0.006 (0.01) (0.01) (0.01) (0.02) (0.01) (0.02) Transfers (Not Progressive) -0.033*** -0.003 0.002 -0.009 -0.020** -0.025 (0.00) (0.01) (0.00) (0.02) (0.00) (0.01) Combined (Progressive) 0.012 0.003 0.006 0.009 0.024*** 0.023** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.019 -0.029* 0.020 0.009 -0.029 -0.013 (0.00) (0.00) (0.00) (0.02) (0.02) (0.02) Panel B - Respondents did not prefer richer households to pay more tax than poorer households Taxes (Progressive) -0.012* 0.004 0.002 0.014 -0.014* -0.005 (0.01) (0.02) (0.02) (0.01) (0.01) (0.02) Taxes (Not Progressive) -0.021 0.012 0.001 -0.014 -0.036 -0.024 (0.02) (0.02) (0.02) (0.03) (0.02) (0.04) Transfers (Progressive) -0.017 0.010 0.034** -0.006 0.011* 0.014 (0.02) (0.02) (0.01) (0.02) (0.00) (0.02) Transfers (Not Progressive) 0.002 0.002 0.002 -0.005 0.012 0.007 (0.02) (0.04) (0.01) (0.04) (0.03) (0.00) Combined (Progressive) -0.058** 0.022 0.028 0.010 -0.030 -0.012 (0.02) (0.03) (0.02) (0.02) (0.02) (0.04) Combined (Not Progressive) -0.023 -0.015 -0.024 -0.054 0.019 -0.048 (0.04) (0.01) (0.03) (0.04) (0.02) (0.02) Note: This table is directly comparable to Table 3 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on their prior preferences for progressivity. Preferences about progressivity are based on Q9, which asks respondents whether they think that richer households should pay a higher share of their income in tax than poorer households. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects (with device type weights) based on respondents existing preferences about whether taxes should be progressive, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A9: Balance table for the taxes treatment group relative to the control group CO GH ID JO LK MX TZ ZA b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Male 0.000 -0.073*** 0.032 0.009 -0.054* -0.013 0.017 0.034 (0.02) (0.03) (0.03) (0.02) (0.03) (0.02) (0.03) (0.02) p-value 0.983 0.007 0.198 0.714 0.058 0.596 0.510 0.151 18-34 years -0.014 0.001 -0.015 -0.008 0.013 0.037 0.011 -0.019 (0.02) (0.03) (0.03) (0.03) (0.03) (0.02) (0.03) (0.02) p-value 0.548 0.976 0.572 0.742 0.594 0.118 0.713 0.437 Sec edu or less -0.007 -0.026 0.021 0.003 -0.025 -0.020 -0.079*** 0.010 (0.02) (0.02) (0.03) (0.02) (0.02) (0.03) (0.02) (0.02) p-value 0.767 0.267 0.444 0.884 0.300 0.460 0.001 0.676 Large city 0.046* 0.043* 0.001 0.011 -0.031 0.018 0.010 -0.015 (0.03) (0.02) (0.02) (0.02) (0.03) (0.03) (0.02) (0.03) p-value 0.072 0.071 0.961 0.669 0.230 0.489 0.660 0.554 Working 0.008 -0.031 -0.009 -0.023 -0.014 -0.005 0.036 -0.041 (0.02) (0.02) (0.02) (0.02) (0.03) (0.02) (0.02) (0.03) p-value 0.741 0.200 0.715 0.350 0.569 0.845 0.123 0.102 Believe B40 -0.005 0.009 0.052* 0.004 0.030 -0.007 0.005 -0.027 (0.02) (0.03) (0.03) (0.02) (0.03) (0.03) (0.03) (0.02) p-value 0.846 0.729 0.062 0.876 0.231 0.806 0.846 0.262 Observations 1923 1878 1864 1887 1799 1874 1930 1885 F-statistic 0.774 2.282 1.315 0.193 1.419 0.673 2.608 0.924 Note: This table presents the results of an OLS regression whereby the dependent variable is a dummy variable based on whether a respondent received the taxes treatment and the independent variables are characteristics of respondents. * p < 0.1, ** p < 0.05, *** p < 0.01. CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Male: Based on Q0, which asks respondents whether they are male or female (variable takes value of 1 if they select "Male" and 0 otherwise). 18-34 years: Based on Q0, which also asks respondents their age (variable takes value of 1 if they select between 18-34 years and 0 if they select 35 or older, noting respondents under the age of 18 years were automatically excluded). Sec edu or less: Based on Q1, which asks whether respondents their level of education (variable takes value of 1 if they select "Primary or less" or "Secondary" and 0 otherwise). Large city: Based on Q2, which asks respondents about where they live (variable takes value of 1 if they select "Large city" and 0 otherwise). Working: Based on Q3, which asks whether respondents their current employment status (variable takes value of 1 if they select "Employee" or "Self employed" and 0 otherwise). Believe B40: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they select "Poorest group" or "Second poorest group" and 0 otherwise). This table presents the differences in means for demographic characteristics between the taxes treatment group and the control group in each country. There are few meaningful differences. It is important to note that these differences in demographic characteristics are controlled for in the main regression analysis (see Equation (7) in Section 3 of body of the paper). Table A10: Balance table for the transfers treatment group relative to the control group CO GH ID JO LK MX TZ ZA b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Male 0.015 -0.032 -0.015 0.020 -0.050* 0.005 0.077*** -0.016 (0.02) (0.03) (0.02) (0.02) (0.03) (0.02) (0.03) (0.02) p-value 0.522 0.258 0.535 0.405 0.075 0.831 0.002 0.504 18-34 years -0.021 -0.001 0.013 -0.017 0.032 0.006 0.002 -0.010 (0.02) (0.03) (0.03) (0.03) (0.02) (0.02) (0.03) (0.02) p-value 0.382 0.972 0.629 0.509 0.198 0.805 0.930 0.687 Sec edu or less -0.029 0.004 -0.012 0.047** -0.038 -0.049* -0.066*** 0.036 (0.02) (0.02) (0.03) (0.02) (0.02) (0.03) (0.02) (0.02) p-value 0.220 0.881 0.654 0.047 0.117 0.070 0.004 0.133 Large city -0.006 0.015 0.025 0.018 -0.041 0.005 0.036 -0.033 (0.03) (0.02) (0.02) (0.02) (0.03) (0.03) (0.02) (0.03) p-value 0.816 0.540 0.286 0.454 0.105 0.834 0.120 0.213 Working 0.009 -0.001 0.047* 0.017 -0.006 -0.021 -0.002 0.019 (0.02) (0.02) (0.02) (0.02) (0.03) (0.03) (0.02) (0.02) p-value 0.707 0.970 0.051 0.497 0.820 0.399 0.934 0.442 Believe B40 -0.028 -0.009 0.034 -0.007 0.018 0.019 0.013 0.019 (0.02) (0.03) (0.03) (0.02) (0.03) (0.03) (0.03) (0.02) p-value 0.244 0.735 0.229 0.769 0.465 0.500 0.632 0.427 Observations 1905 1837 1901 1917 1849 1873 1973 1873 F-statistic 1.029 0.331 1.119 1.129 1.756 0.689 3.566 0.951 Note: This table presents the results of an OLS regression whereby the dependent variable is a dummy variable based on whether a respondent received the transfers treatment and the independent variables are characteristics of respondents. * p < 0.1, ** p < 0.05, *** p < 0.01. CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Male: Based on Q0, which asks respondents whether they are male or female (variable takes value of 1 if they select "Male" and 0 otherwise). 18-34 years: Based on Q0, which also asks respondents their age (variable takes value of 1 if they select between 18-34 years and 0 if they select 35 or older, noting respondents under the age of 18 years were automatically excluded). Sec edu or less: Based on Q1, which asks whether respondents their level of education (variable takes value of 1 if they select "Primary or less" or "Secondary" and 0 otherwise). Large city: Based on Q2, which asks respondents about where they live (variable takes value of 1 if they select "Large city" and 0 otherwise). Working: Based on Q3, which asks whether respondents their current employment status (variable takes value of 1 if they select "Employee" or "Self employed" and 0 otherwise). Believe B40: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they select "Poorest group" or "Second poorest group" and 0 otherwise). This table presents the differences in means for demographic characteristics between the transfers treatment group and the control group in each country. There are few meaningful differences. It is important to note that these differences in demographic characteristics are controlled for in the main regression analysis (see Equation (7) in Section 3 of body of the paper). Table A11: Balance table for the combined treatment group relative to the control group CO GH ID JO LK MX TZ ZA b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Male -0.014 -0.035 0.035 0.021 -0.070** -0.004 0.034 0.010 (0.02) (0.03) (0.03) (0.02) (0.03) (0.02) (0.03) (0.02) p-value 0.563 0.203 0.157 0.396 0.012 0.880 0.184 0.680 18-34 years -0.026 -0.014 -0.004 -0.016 -0.000 0.004 -0.019 -0.022 (0.02) (0.03) (0.03) (0.03) (0.02) (0.02) (0.03) (0.02) p-value 0.275 0.633 0.881 0.540 0.995 0.872 0.510 0.376 Sec edu or less 0.016 -0.037 -0.020 -0.009 0.010 -0.011 -0.056** 0.024 (0.02) (0.02) (0.03) (0.02) (0.02) (0.03) (0.02) (0.02) p-value 0.510 0.125 0.473 0.719 0.668 0.696 0.018 0.321 Large city 0.002 0.039 -0.019 -0.005 -0.055** 0.017 -0.037 -0.009 (0.03) (0.02) (0.02) (0.02) (0.03) (0.03) (0.02) (0.03) p-value 0.923 0.101 0.416 0.825 0.029 0.514 0.114 0.728 Working 0.033 -0.051** -0.016 -0.034 0.019 0.019 0.004 -0.023 (0.02) (0.02) (0.02) (0.03) (0.02) (0.03) (0.02) (0.03) p-value 0.184 0.034 0.502 0.180 0.451 0.449 0.862 0.357 Believe B40 -0.035 -0.002 0.049* -0.013 0.017 0.014 0.010 -0.031 (0.02) (0.03) (0.03) (0.02) (0.03) (0.03) (0.03) (0.02) p-value 0.156 0.947 0.078 0.589 0.493 0.619 0.708 0.210 Observations 1849 1900 1878 1865 1850 1794 1869 1830 F-statistic 1.108 1.927 1.207 0.442 1.939 0.237 1.656 0.604 Note: This table presents the results of an OLS regression whereby the dependent variable is a dummy variable based on whether a respondent received the combined treatment and the independent variables are characteristics of respondents. * p < 0.1, ** p < 0.05, *** p < 0.01. CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Male: Based on Q0, which asks respondents whether they are male or female (variable takes value of 1 if they select "Male" and 0 otherwise). 18-34 years: Based on Q0, which also asks respondents their age (variable takes value of 1 if they select between 18-34 years and 0 if they select 35 or older, noting respondents under the age of 18 years were automatically excluded) . Sec edu or less: Based on Q1, which asks whether respondents their level of education (variable takes value of 1 if they select "Primary or less" or "Secondary" and 0 otherwise). Large city: Based on Q2, which asks respondents about where they live (variable takes value of 1 if they select "Large city" and 0 otherwise). Working: Based on Q3, which asks whether respondents their current employment status (variable takes value of 1 if they select "Employee" or "Self employed" and 0 otherwise). Believe B40: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they select "Poorest group" or "Second poorest group" and 0 otherwise). This table presents the differences in means for demographic characteristics between the combined treatment group and the control group in each country. There are few meaningful differences. It is important to note that these differences in demographic characteristics are controlled for in the main regression analysis (see Equation (7) in Section 3 of body of the paper). Table A12: Characteristics associated with willingness to pay tax Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Smartphone 0.018 0.029 0.059** 0.034** 0.006 0.063** (0.02) (0.02) (0.02) (0.01) (0.02) (0.02) Male -0.017 0.033 0.077*** 0.045** 0.021 0.069* (0.02) (0.02) (0.02) (0.01) (0.03) (0.03) 18-34 years -0.080*** -0.071*** -0.037 -0.063** -0.088*** -0.140*** (0.02) (0.02) (0.03) (0.02) (0.01) (0.04) Sec edu or less -0.051** -0.041 -0.021 -0.030* -0.039* -0.076*** (0.02) (0.03) (0.01) (0.01) (0.02) (0.02) Large city 0.027** -0.010 0.021 0.035 -0.008 0.029 (0.01) (0.01) (0.02) (0.02) (0.01) (0.02) Working -0.019 0.024 -0.008 0.017 0.009 0.010 (0.02) (0.02) (0.01) (0.01) (0.02) (0.03) Poorest quintile 0.048 0.057 -0.079 0.042 -0.006 0.029 (0.06) (0.06) (0.05) (0.04) (0.05) (0.08) Second poorest quintile 0.151** 0.036 -0.118* 0.080 -0.017 0.060 (0.06) (0.05) (0.06) (0.06) (0.05) (0.09) Middle quintile 0.135* 0.075* -0.086 0.116** -0.001 0.106 (0.06) (0.04) (0.05) (0.05) (0.04) (0.08) Second richest quintile 0.050 0.099 -0.093 0.053 0.043 0.064 (0.07) (0.05) (0.06) (0.05) (0.06) (0.09) Observations 7933 7933 7933 7933 7933 7933 Note: This table presents the results of OLS regressions whereby the dependent variable is based on various measures of respondents (in the control group) willingness to pay tax and the independent variables are characteristics of respondents. Country fixed effects are used. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. Smartphone: Based on data provided by survey firm (variable takes value of 1 if they accessed the survey via smartphone and 0 otherwise). Male: Based on Q0, which asks respondents whether they are male or female (variable takes value of 1 if they select "Male" and 0 otherwise). 18-34 years: Based on Q0, which also asks respondents their age (variable takes value of 1 if they select between 18-34 years and 0 if they select 35 or older, noting respondents under the age of 18 years were automatically excluded) Sec edu or less: Based on Q1, which asks whether respondents their level of education (variable takes value of 1 if they select "Primary or less" or "Secondary" and 0 otherwise). Large city: Based on Q2, which asks respondents about where they live (variable takes value of 1 if they select "Large city" and 0 otherwise). Working: Based on Q3, which asks whether respondents their current employment status (variable takes value of 1 if they select "Employee" or "Self employed" and 0 otherwise). Poorest quintile: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they selected the "Poorest group" or and 0 if they selected the “Richest group”). Second poorest quintile: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they selected the "Second poorest group" or and 0 if they selected the “Richest group”). Middle quintile: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they selected the "Middle group" or and 0 if they selected the “Richest group”). Second richest quintile: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they selected the "Second richest group" or and 0 if they selected the “Richest group”). This table presents the characteristics that are associated with respondents in the control group stating they were willing to pay tax across countries. Table A13: Relationship between willingness to pay tax and beliefs about the progressivity of taxes Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Believe progressive -0.076** -0.019 0.130*** 0.202*** -0.025 0.096** (0.03) (0.01) (0.02) (0.03) (0.03) (0.04) Male -0.017 0.030 0.040** 0.069*** 0.020 0.061* (0.02) (0.02) (0.01) (0.02) (0.02) (0.03) 18-34 years -0.077** -0.070*** -0.065** -0.040 -0.088*** -0.141*** (0.02) (0.02) (0.02) (0.03) (0.02) (0.04) Sec edu or less -0.050* -0.035 -0.035** -0.021 -0.037* -0.074** (0.02) (0.03) (0.01) (0.01) (0.02) (0.02) Large city 0.022 -0.011 0.033 0.022 -0.007 0.026 (0.01) (0.01) (0.02) (0.02) (0.01) (0.02) Working -0.017 0.024 0.012 -0.015 0.010 0.006 (0.02) (0.02) (0.01) (0.01) (0.02) (0.03) Believe B40 -0.011 -0.028 -0.039*** -0.015 -0.015 -0.047* (0.02) (0.02) (0.01) (0.02) (0.01) (0.02) Observations 7933 7933 7933 7933 7933 7933 Note: This table presents the results of OLS regressions whereby the dependent variable is based on various measures of respondents (in the control group) willingness to pay tax and the independent variables are characteristics of respondents and a dummy variable based on their prior beliefs about the progressivity of taxes. Country fixed effects are used. * p < 0.1, ** p < 0.05, *** p < 0.01. Believe progressive: Based on Q8, which asks respondents whether they believe that richer households pay a higher share of their income in tax than poorer households (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Male: Based on Q0, which asks respondents whether they are male or female (variable takes value of 1 if they select "Male" and 0 otherwise). 18-34 years: Based on Q0, which also asks respondents their age (variable takes value of 1 if they select between 18-34 years and 0 if they select 35 or older, noting respondents under the age of 18 years were automatically excluded). Sec edu or less: Based on Q1, which asks whether respondents their level of education (variable takes value of 1 if they select "Primary or less" or "Secondary" and 0 otherwise). Large city: Based on Q2, which asks respondents about where they live (variable takes value of 1 if they select "Large city" and 0 otherwise). Working: Based on Q3, which asks whether respondents their current employment status (variable takes value of 1 if they select "Employee" or "Self employed" and 0 otherwise). Believe B40: Based on Q5, which asks respondents about their households place in the national income distribution (variable takes value of 1 if they select "Poorest group" or "Second poorest group" and 0 otherwise). Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the relationship between being willing to pay tax and believing the tax system is progressive after controlling for demographic characteristics. Table A14: Overall impact of each treatment on the WTP tax index by country CO GH ID JO LK MX TZ ZA b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes treatment 0.058* 0.013 -0.020 -0.042 -0.045 0.029 0.040 -0.071** (0.03) (0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03) p-value 0.055 0.651 0.646 0.115 0.181 0.347 0.179 0.013 Observations 1923 1878 1864 1887 1799 1874 1930 1885 Transfers treatment 0.069** -0.011 0.010 -0.033 -0.045 0.040 0.013 -0.008 (0.03) (0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03) p-value 0.022 0.718 0.799 0.222 0.194 0.215 0.676 0.783 Observations 1905 1837 1901 1917 1849 1873 1973 1873 Combined treatment 0.052* -0.032 0.064 0.004 0.012 0.039 0.011 -0.017 (0.03) (0.03) (0.04) (0.03) (0.04) (0.03) (0.03) (0.03) p-value 0.080 0.233 0.117 0.880 0.730 0.236 0.726 0.574 Observations 1849 1900 1878 1865 1850 1794 1869 1830 Note: This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for each country (i.e. there are no country fixed effects). * p < 0.1, ** p < 0.05, *** p < 0.01. CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. WTP tax INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the overall impact of the treatments in each of the countries. Table A15: Differences in the impact of the treatments in countries where either taxes or transfers are not progressive Direct Punishable Important Right to Tax Refuse to Pay INDEX b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Not Progressive) × Transfers (Progressive) -0.014 -0.018 -0.018 -0.023 -0.029 -0.043** (0.01) (0.01) (0.01) (0.01) (0.02) (0.01) p-value 0.286 0.203 0.270 0.182 0.259 0.015 Observations 10689 10689 10689 10689 10689 10689 Taxes (Progressive) × Transfers (Not progressive) 0.013 0.010 0.005 0.020 0.027 0.033** (0.01) (0.00) (0.00) (0.01) (0.02) (0.00) p-value 0.264 0.126 0.409 0.269 0.361 0.043 Observations 5150 5150 5150 5150 5150 5150 Note:. This table is based on Equation 7 in Section 3 of the paper, except the treatment dummy is coded such that it takes on the value of 1 if the respondent received the taxes treatment and 0 if they respondent received either the transfers or combined treatment. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the impact of tax treatment compared to the transfers and combined treatments in the six countries for which these treatments were in opposing directions (Ghana, Indonesia, Jordan, Sri Lanka, South Africa and Tanzania). Table A16 - Impact of the treatments on WTP tax index across the income distribution Q1 Q2 Q3 Q4 Q5 b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Progressive) 0.066* 0.047 0.033** 0.102 -0.115 (0.03) (0.03) (0.01) (0.10) (0.13) p-value 0.081 0.158 0.022 0.368 0.453 Observations 668 1454 4924 354 205 Taxes (Not Progressive) -0.026 -0.042 -0.065** 0.034 0.005 (0.06) (0.03) (0.01) (0.12) (0.06) p-value 0.698 0.280 0.021 0.792 0.945 Observations 984 1672 4349 247 183 Transfers (Progressive) -0.047 0.030 0.005 0.041 0.054 (0.04) (0.03) (0.02) (0.11) (0.09) p-value 0.318 0.321 0.768 0.717 0.577 Observations 1360 2482 6850 378 248 Transfers (Not Progressive) 0.039 0.021 0.013 -0.169 -0.035 (0.05) (0.09) (0.01) (0.17) (0.06) p-value 0.559 0.852 0.460 0.502 0.642 Observations 294 661 2507 216 132 Combined (Progressive) 0.018 0.022 0.033 -0.038 -0.032 (0.04) (0.03) (0.02) (0.14) (0.12) p-value 0.638 0.522 0.113 0.797 0.799 Observations 1318 2413 6716 373 246 Combined (Not progressive) -0.023 0.029 -0.008 -0.077 -0.030 (0.02) (0.06) (0.04) (0.15) (0.22) p-value 0.510 0.717 0.894 0.690 0.913 Observations 299 650 2489 207 124 Note:. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for each quintile. * p < 0.1, ** p < 0.05, *** p < 0.01. WTP tax index: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. Q1: Poorest quintile, based on answer to Q5. Q2: Second poorest quintile, based on answer to Q5. Q3: Middle quintile, based on answer to Q5. Q4: Second richest quintile, based on answer to Q5. Q5: Richest quintile, based on answer to Q5. This table shows the overall impact of the tax treatment on the willingness to pay tax index for each quintile, where countries are pooled based on whether taxes are actually progressive. Table A17 – Heterogeneous effects of the treatments based on whether households paid a large share of their income in tax Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents stated their household paid a large share of their income in tax Taxes (Progressive) 0.006 0.038** 0.024 0.015 0.017 0.042* (0.01) (0.01) (0.01) (0.01) (0.02) (0.02) Taxes (Not Progressive) -0.025 -0.022* -0.020 -0.045** -0.017 -0.054** (0.03) (0.01) (0.02) (0.01) (0.02) (0.01) Transfers (Progressive) 0.006 0.013 0.007 -0.011 0.024** 0.016 (0.01) (0.01) (0.01) (0.02) (0.01) (0.02) Transfers (Not Progressive) -0.021*** -0.008 -0.012** -0.032* -0.031 -0.048 (0.00) (0.03) (0.00) (0.00) (0.01) (0.02) Combined (Progressive) -0.001 0.005 0.002 0.003 0.023** 0.014 (0.01) (0.02) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) - 0.047 0.003 0.008 -0.002 -0.031 -0.026 (0.03) (0.01) (0.01) (0.01) (0.03) (0.03) Panel B - Respondents stated their household did not pay a large share of their income in tax Taxes (Progressive) 0.014 -0.009 0.015 0.017 0.010 0.027 (0.02) (0.05) (0.02) (0.03) (0.02) (0.04) Taxes (Not Progressive) -0.016 -0.022 -0.004 -0.002 -0.039 -0.036 (0.04) (0.01) (0.01) (0.03) (0.03) (0.03) Transfers (Progressive) -0.024 0.006 0.023 0.008 -0.021 -0.002 (0.02) (0.02) (0.02) (0.01) (0.01) (0.02) Transfers (Not Progressive) 0.041 0.037 0.033 0.067 0.041* 0.105 (0.01) (0.01) (0.03) (0.10) (0.01) (0.06) Combined (Progressive) 0.003 0.015 0.044** 0.044** 0.002 0.045 (0.01) (0.01) (0.02) (0.02) (0.02) (0.02) Combined (Not Progressive) 0.075* -0.089 0.016* 0.027 0.007 0.023 (0.01) (0.06) (0.00) (0.07) (0.03) (0.01) Note: This table is directly comparable to Tables 2 and 3 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on their prior beliefs about the share of their household income that is paid in tax. Beliefs about the share of household income that is paid in tax are based on Q6. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects based on respondents beliefs about the share of their household income that is paid in tax, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A18 – Heterogeneous effects of the treatments based on whether respondents claimed their household was a net contributor to the tax and transfer system Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents stated their household paid more in taxes than they received in transfers Taxes (Progressive) 0.001 0.031** 0.013 0.007 0.023 0.033** (0.02) (0.01) (0.01) (0.01) (0.02) (0.01) Taxes (Not Progressive) -0.022* -0.031* -0.022 -0.043* -0.016 -0.056** (0.01) (0.01) (0.02) (0.02) (0.02) (0.02) Transfers (Progressive) 0.005 0.009 0.006 -0.008 0.012 0.011 (0.01) (0.01) (0.02) (0.02) (0.01) (0.02) Transfers (Not Progressive) -0.014 0.006 -0.001 -0.024* -0.005 -0.017** (0.01) (0.01) (0.00) (0.00) (0.01) (0.00) Combined (Progressive) -0.009 0.002 -0.009 -0.001 0.015 0.000 (0.01) (0.02) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.045 -0.018* 0.013 -0.001 -0.034 -0.030 (0.03) (0.00) (0.02) (0.02) (0.02) (0.02) Panel B - Respondents stated their household did not pay more in taxes than they received in transfers Taxes (Progressive) 0.020 0.007 0.040* 0.034 -0.003 0.044 (0.01) (0.04) (0.02) (0.02) (0.01) (0.02) Taxes (Not Progressive) 0.021 -0.012 -0.004 -0.009 -0.044 -0.040 (0.03) (0.01) (0.03) (0.03) (0.04) (0.04) Transfers (Progressive) -0.023 0.009 0.025 0.004 -0.009 0.002 (0.03) (0.02) (0.01) (0.02) (0.02) (0.03) Transfers (Not Progressive) 0.022 0.009 0.004 0.033 -0.002 0.033 (0.02) (0.03) (0.02) (0.07) (0.01) (0.02) Combined (Progressive) 0.016 0.017 0.058** 0.045* 0.009 0.059 (0.02) (0.02) (0.02) (0.02) (0.02) (0.03) Combined (Not Progressive) 0.043 -0.035 0.007 0.027 0.010 0.027 (0.02) (0.02) (0.02) (0.05) (0.02) (0.01) Note: This table is directly comparable to Tables 2 and 3 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately based on whether respondents claimed their household was a net contributor to the tax and transfer system. Respondents’ views about whether their household was a net contributor to the tax and transfer system is based on Q7. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects based on whether respondents claimed their household was a net contributor to the tax and transfer system, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A19 – Heterogeneous effects of the treatments based on whether respondents’ preferred lower levels of inequality Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Agree that the gap between the rich and poor was too large Taxes (Progressive) -0.003 0.030* 0.012 0.015 0.011 0.028 (0.02) (0.01) (0.01) (0.01) (0.02) (0.01) Taxes (Not Progressive) -0.029 -0.027* -0.013 -0.035** -0.030 -0.056** (0.02) (0.01) (0.02) (0.01) (0.03) (0.01) Transfers (Progressive) -0.019* 0.009 0.000 -0.008 0.002 -0.007 (0.01) (0.01) (0.01) (0.02) (0.01) (0.02) Transfers (Not Progressive) -0.010 -0.003 -0.009 -0.016 -0.001 0.021 (0.01) (0.00) (0.01) (0.03) (0.00) (0.02) Combined (Progressive) -0.012 0.002 -0.000 0.013 0.010 0.005 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.019 -0.021 0.006 -0.009 -0.014 -0.021 (0.03) (0.01) (0.01) (0.03) (0.03) (0.03) Panel B - Did not agree that the gap between the rich and poor was too large Taxes (Progressive) 0.041 -0.001 0.053 0.018 0.019 0.059 (0.04) (0.04) (0.03) (0.02) (0.04) (0.06) Taxes (Not Progressive) -0.007 -0.011 -0.017 -0.001 -0.022 -0.027 (0.04) (0.02) (0.01) (0.03) (0.03) (0.04) Transfers (Progressive) 0.020 0.007 0.043* 0.006 0.013 0.037 (0.02) (0.01) (0.02) (0.02) (0.03) (0.03) Transfers (Not Progressive) 0.039 0.037 0.052 0.052 -0.016 0.085 (0.06) (0.05) (0.01) (0.04) (0.03) (0.03) Combined (Progressive) 0.028 0.021 0.068** 0.036** 0.018 0.071* (0.03) (0.02) (0.02) (0.01) (0.02) (0.03) Combined (Not Progressive) 0.013 -0.036 0.043 0.084 -0.046 0.042 (0.03) (0.06) (0.02) (0.03) (0.02) (0.02) Note: This table is directly comparable to Tables 2 and 3 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on whether they preferred lower levels of inequality. Respondents’ preferences about the level of inequality in their country are based on Q4. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects based on whether respondents’ preferred lower levels of inequality, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A20 – Heterogeneous effects of the treatments based on whether respondents were working Direct Punishable Important Right to Tax Do not Refuse INDEX b/se b/se b/se b/se b/se b/se Panel A - Respondents stated that they were either an employee or self-employed Taxes (Progressive) 0.008 0.015 0.015* 0.019** 0.012 0.030* (0.01) (0.02) (0.01) (0.00) (0.01) (0.01) Taxes (Not Progressive) -0.038 -0.039 -0.016 -0.014 -0.054* -0.069* (0.03) (0.02) (0.01) (0.01) (0.02) (0.03) Transfers (Progressive) -0.004 0.003 0.013 0.016 0.001 0.012 (0.02) (0.02) (0.01) (0.02) (0.01) (0.03) Transfers (Not Progressive) 0.011 0.029 0.004 -0.003 0.001 0.018 (0.02) (0.01) (0.00) (0.01) (0.03) (0.01) Combined (Progressive) -0.011 0.001 0.014 0.027* 0.003 0.014 (0.02) (0.02) (0.01) (0.01) (0.00) (0.02) Combined (Not Progressive) 0.018** -0.022* 0.041 0.021 -0.008 0.033 (0.00) (0.00) (0.01) (0.03) (0.01) (0.01) Panel B - Respondents stated that they were neither an employee or self-employed Taxes (Progressive) 0.005 0.032 0.032* 0.011 0.011 0.041** (0.02) (0.02) (0.01) (0.02) (0.01) (0.01) Taxes (Not Progressive) -0.005 -0.002 -0.009 -0.041 -0.001 -0.024 (0.02) (0.02) (0.02) (0.02) (0.04) (0.01) Transfers (Progressive) -0.010 0.019 0.019 -0.025 0.006 0.005 (0.01) (0.02) (0.01) (0.01) (0.01) (0.01) Transfers (Not Progressive) -0.014 -0.019 -0.001 -0.000 -0.020 -0.021 (0.01) (0.01) (0.01) (0.05) (0.01) (0.04) Combined (Progressive) 0.013 0.013 0.028** 0.012 0.021 0.037* (0.02) (0.01) (0.01) (0.01) (0.01) (0.01) Combined (Not Progressive) -0.047 -0.025 -0.023 -0.006 -0.041 -0.062 (0.05) (0.02) (0.01) (0.02) (0.02) (0.03) Note: This table is directly comparable to Tables 2 and 3 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper, except the regression analysis is conducted separately for respondents based on whether or not they are working. Respondents’ employment status is based on Q3. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the heterogeneous treatment effects based on whether respondents’ were working or not, where countries are pooled based on whether the tax and/or transfer system is actually progressive. Table A21: Overall effects of the treatments (excluding the fastest 5% and slowest 5% of respondents based on the time taken to complete survey) Direct Punishable Important Right to Tax Do not Refuse INDEX b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Progressive) -0.001 0.020 0.015* 0.021* 0.012 0.027** (0.01) (0.01) (0.00) (0.01) (0.01) (0.01) p-value 0.945 0.274 0.050 0.099 0.357 0.036 Observations 7045 7059 7062 7063 7048 7246 Taxes (Not Progressive) -0.031 -0.028* -0.005 -0.017 -0.034 -0.043** (0.01) (0.01) (0.01) (0.01) (0.02) (0.01) p-value 0.113 0.055 0.615 0.164 0.269 0.035 Observations 6737 6756 6762 6759 6717 6959 Transfers (Progressive) -0.016* 0.006 0.012 0.008 0.001 0.006 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) p-value 0.097 0.695 0.227 0.621 0.896 0.711 Observations 10222 10242 10252 10240 10219 10478 Transfers (Not Progressive) -0.006 -0.001 0.008 -0.000 -0.008 -0.008 (0.01) (0.02) (0.01) (0.03) (0.00) (0.01) p-value 0.553 0.964 0.682 0.993 0.339 0.461 Observations 3646 3650 3651 3646 3631 3792 Combined (Progressive) -0.007 -0.002 0.014* 0.021 0.009 0.017 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) p-value 0.634 0.885 0.056 0.110 0.166 0.270 Observations 9962 10001 10004 9996 9955 10245 Combined (Not progressive) -0.017 -0.029 0.008** 0.004 -0.021 -0.027 (0.02) (0.02) (0.00) (0.02) (0.01) (0.01) p-value 0.559 0.319 0.021 0.883 0.238 0.314 Observations 3595 3593 3615 3598 3583 3735 Note: This table is comparable to Table 1 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper. * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly disagree" or "Disagree" and 0 oth- erwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table shows the overall impact of each of the treatments (excluding the fastest 5% and slowest 5% of respondents based on the time taken to complete survey) relative to the control group, where countries are pooled based on whether the tax and/or transfer system is progressive. Table A22: Overall effects of the treatments (without controls) Direct Punishable Important Right to Tax Do not Refuse INDEX b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Progressive) 0.010 0.023 0.026** 0.016 0.014 0.039** (0.01) (0.02) (0.01) (0.01) (0.01) (0.01) p-value 0.535 0.300 0.037 0.172 0.179 0.048 Observations 7605 7605 7605 7605 7605 7605 Taxes (Not Progressive) -0.022 -0.022* -0.013 -0.027 -0.029 -0.048** (0.01) (0.01) (0.02) (0.01) (0.02) (0.01) p-value 0.231 0.078 0.530 0.161 0.327 0.013 Observations 7435 7435 7435 7435 7435 7435 Transfers (Progressive) -0.006 0.010 0.016 -0.002 0.005 0.010 (0.01) (0.01) (0.01) (0.02) (0.01) (0.02) p-value 0.621 0.442 0.278 0.899 0.578 0.660 Observations 11318 11318 11318 11318 11318 11318 Transfers (Not Progressive) 0.001 0.005 0.003 -0.005 -0.009 -0.000 (0.00) (0.01) (0.01) (0.03) (0.01) (0.02) p-value 0.826 0.761 0.815 0.912 0.334 0.996 Observations 3810 3810 3810 3810 3810 3810 Combined (Progressive) 0.000 0.009 0.022* 0.021* 0.013** 0.027* (0.01) (0.01) (0.01) (0.01) (0.00) (0.01) p-value 0.987 0.394 0.059 0.095 0.040 0.080 Observations 11066 11066 11066 11066 11066 11066 Combined (Not progressive) -0.006 -0.023 0.013 0.008 -0.018 -0.005 (0.02) (0.01) (0.01) (0.03) (0.02) (0.02) p-value 0.855 0.252 0.271 0.829 0.478 0.861 Observations 3769 3769 3769 3769 3769 3769 Note: This table is comparable to Table 1 in Section 4 of the paper. This table is based on Equation 7 in Section 3 of the paper (except there are no control variables). * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). This table shows the overall impact of each of the treatments (without controls) relative to the control group, where countries are pooled based on whether the tax and/or transfer system is progressive. Table A23: Leebounds analysis for the taxes treatment Direct Punishable Important Right to Tax Do not Refuse INDEX b/se/p b/se/p b/se/p b/se/p b/se/p b/se/p Taxes (Progressive) Lower bound -0.003 0.007 0.001 -0.005 -0.003 0.011 (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) p-value 0.792 0.519 0.905 0.677 0.774 0.506 Upper bound 0.013 0.016 0.010 0.011 0.009 0.030** (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) p-value 0.279 0.192 0.398 0.236 0.465 0.046 Taxes (Not Progressive) Lower bound -0.028** -0.018* -0.018 -0.007 -0.026** -0.055*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) p-value 0.022 0.098 0.145 0.587 0.017 0.000 Upper bound -0.016 -0.005 -0.007 0.002 -0.013 -0.017 (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) p-value 0.189 0.659 0.544 0.836 0.285 0.288 Note: * p < 0.1, ** p < 0.05, *** p < 0.01. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly disagree" or "Disagree" and 0 oth- erwise). INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This table presents the upper and lower Leebounds (based on Lee (2009)) for the taxes treatment, where countries are pooled based on whether the tax system is progressive. Figure A1: Difference between the Gross and Net GINI index in all developing countries where comparable data exists Note: Countries marked in yellow were included in this study. The year shown in brackets next to each country is the year in which the survey took place that the GINI index is based on. The United States is included as a point of comparison. This figure shows that the difference between the gross (i.e., pre-taxes and government transfers) and net (i.e., post-taxes and government transfers) GINI index is negligible in some countries in this study and far more substantial in others. Source: CEQ, 2021 Figure A2: Share of respondents in each quintile in each country that agreed richer households should pay a higher share of their income in tax than poorer households Note: CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Preferences about progressivity of taxes are based on Q9, which asks respondents whether they think that richer households should pay a higher share of their income in tax than poorer households. This figure shows the share of respondents in each quintile in each country that agreed richer households should pay a higher share of their income in tax than poorer households. Figure A3: Share of respondents in each quintile in each country that agreed poorer households should receive a higher share of their income in transfers than richer households Note: CO: Colombia. GH: Ghana. ID: Indonesia. JO: Jordan. LK: Sri Lanka. MX: Mexico. TZ: Tanzania. ZA: South Africa. Preferences about progressivity of transfers are based on Q11, which asks respondents whether they think that poorer households should receive a higher share of their income in transfers than richer households. This figure shows the share of respondents in each quintile in each country that agreed poorer households should receive a higher share of their income in transfers than richer households. Figure A4: Overall impact of the tax treatment in each country Note: These results are based on Equation 7 in Section 3 of the paper, however country fixed effects are not included. 90 percent confidence intervals are displayed in this figure. WTP Tax INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. This figure shows the overall impact of the tax treatment in each country. Figure A5: Impact of the tax treatment compared to the transfers and combined treatments Note: The "treatment effect" is based on the regression in Equation 7 with a dummy variable whereby receiving the taxes treatment is coded as 1 and receiving one of the other treatments is coded as 0. Direct: Based on Q14, which asks whether respondents would not pay tax if they knew they would not get caught (variable takes value of 0 if they select "Strongly Agree" or "Agree" and 1 otherwise). Punishable: Based on Q15, which asks respondents their views about people not paying tax (variable takes value of 1 if they select "This is wrong and punishable" and 0 otherwise). Important: Based on Q16, which asks respondents whether it is important for people to pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise). Right to Tax: Based on Q17, which asks respondents whether the government always has a right to make people pay tax (variable takes value of 1 if they select "Strongly Agree" or "Agree" and 0 otherwise).Do not Refuse: Based on Q18, which asks whether people should refuse to pay taxes until they receive more government transfers (variable takes value of 1 if they select "Strongly Disagree" or "Disagree" and 1 otherwise). WTP Tax INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. 90 percent confidence intervals are displayed in this figure. This figure shows the impact of tax treatment compared to the transfers and combined treatments in the six countries for which these treatments were in opposing directions (Ghana, Indonesia, Jordan, Sri Lanka, South Africa and Tanzania). Figure A6: Impact of the tax treatment on Willingness to Pay Tax by quintile Note: Q1: Poorest quintile, based on answer to Q5. Q2: Second poorest quintile, based on answer to Q5. Q3: Middle quintile, based on answer to Q5. Q4: Second richest quintile, based on answer to Q5. Q5: Richest quintile, based on answer to Q5. 90 percent confidence intervals are displayed in this figure. WTP tax INDEX: An unweighted average of the Z-scores of all five outcome variables, oriented so that a higher index means more willingness to pay tax. 90 percent confidence intervals are displayed in this figure. This figure shows the overall impact of the tax treatment on the willingness to pay tax index for each quintile, where countries are pooled based on whether taxes are actually progressive. Q0 (age_group / gender) Q1 (q01_education) Q2 (q02_live) Q3 (q03_employment) Q4 (q04_rich_poor_gap) Q5 (q05_house_income_group) Q6 (q06_perceived_payed_tax) Q7 (q07_perceived_tax_and_transfer) Q8 (q08_rich_pay_more_tax_now) Q9 (q09_rich_should_pay_more_tax) Q10 (q10_poor_incr_transfers_now) Q11 (q11_poor_should_incr_transfers) Q12 (q12_continue_with_survey) Country Specific Experiments (Q13) Q14 (q14_wouldnt_pay_if_not_caught) Q15 (q15_not_paying_taxes) Q16 (q16_important_to_pay_tax) Q17 (q17_gov_right_to_make_ppl_pay) Q18 (q18_refuse_taxes) Q19 (q19_feedback) Taxes treatment - Colombia Recent research* in Colombia shows: Richer households pay a larger share of their income in taxes than Poorer households Decrease in income from taxes Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - Colombia Recent research* in Colombia shows: Poorer households receive a much larger share of their income in government transfers than Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - Colombia Recent research* in Colombia shows: Richer households pay more in taxes than they receive in government transfers, whereas Poorer households receive more in government transfers than they pay in taxes Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes treatment - Ghana Recent research* in Ghana shows: Richer households pay a larger share of their income in taxes than Poorer households Decrease in income from taxes Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - Ghana Recent research* in Ghana shows: Poorer households receive a similar share of their income in government transfers as Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - Ghana Recent research* in Ghana shows: Most households pay more in taxes than they receive in government transfers and Richer households pay more than Poorer households Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes treatment - Indonesia Recent research* in Indonesia shows: Richer households pay a similar share of their income in taxes as Poorer households Decrease in income from taxes Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - Indonesia Recent research* in Indonesia shows: Poorer households receive a larger share of their income in government transfers than Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - Indonesia Recent research* in Indonesia shows: Richer households pay more in taxes than they receive in government transfers, whereas Poorer households receive more in government transfers than they pay in taxes Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes treatment - Jordan Recent research* in Jordan shows: Richer households pay a similar share of their income in taxes as Poorer households Decrease in income from taxes Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - Jordan Recent research* in Jordan shows: Poorer households receive a much larger share of their income in government transfers than Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - Jordan Recent research* in Jordan shows: Richer households pay more in taxes than they receive in government transfers, whereas Poorer households receive more in government transfers than they pay in taxes Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes treatment - Mexico Recent research* in Mexico shows: Richer households pay a larger share of their income in taxes than Poorer households Decrease in income from taxes Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - Mexico Recent research* in Mexico shows: Poorer households receive a much larger share of their income in government transfers than Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - Mexico Recent research* in Mexico shows: Richer households pay more in taxes than they receive in government transfers, whereas Poorer households receive more in government transfers than they pay in taxes Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes treatment – Sri Lanka Recent research* in Sri Lanka shows: Richer households pay a similar share of their income in taxes as Poorer households Decrease in income from taxes Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - Sri Lanka Recent research* in Sri Lanka shows: Poorer households receive a much larger share of their income in government transfers than Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - Sri Lanka Recent research* in Sri Lanka shows: Richer households pay more in taxes than they receive in government transfers, whereas Poorer households receive more in government transfers than they pay in taxes Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes treatment - Tanzania Recent research* in Tanzania shows: Richer households pay a much larger share of their income in taxes than Poorer households 0% Decrease in income from taxes -5% -10% -15% -20% Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - Tanzania Recent research* in Tanzania shows: Poorer households receive a similar share of their income in government transfers as Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - Tanzania Recent research* in Tanzania shows: Most households pay more in taxes than they receive in government transfers and Richer households pay more than Poorer households Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes treatment – South Africa Recent research* in South Africa shows: Poorer households pay a much larger share of their income in taxes than Richer households Decrease in income from taxes Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Transfers treatment - South Africa Recent research* in South Africa shows: Poorer households receive a much larger share of their income in government transfers than Richer households Increase in income from transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations. Taxes and transfers treatment - South Africa Recent research* in South Africa shows: Richer households pay more in taxes than they receive in government transfers, whereas Poorer households receive more in government transfers than they pay in taxes Change in income from taxes and transfers Poorest 2nd Poorest Middle 2nd Richest Richest group group group group group *This information recently became publicly available online through a collaboration between universities, civil society and international organisations.