Policy Research Working Paper 9507 Trickle Down Tax Morale A Cross Country Survey Experiment Jonathan Mellon Tiago Peixoto Fredrik M Sjoberg Varun Gauri Governance Global Practice January 2021 Policy Research Working Paper 9507 Abstract Studies have encouraged pro-social behavior by experimen- pro-social behavior by groups believed not to contribute tally manipulating people’s views of what others like them their fair share, such as rich people, should be effective tend to do (descriptive norms). These studies positively because it will reduce the subject’s perception that they change behaviors, including charitable giving, littering, are being taken advantage of when they pay taxes. These organ donation, and tax compliance. This paper argues that theories are tested in an online experiment in Kenya, Aus- these results may be explained by a tendency to reciprocate tralia, the United States, the Philippines, and South Africa. positive actions and avoid being taken advantage of. The The findings show that the descriptive norms treatment is descriptive norm account predicts that positively describing ineffective, while the rich people treatment significantly the behavior of ordinary people will be most effective at increases tax morale, supporting reciprocity theory. The increasing citizens’ willingness to pay taxes, and messages findings suggest that tax agencies may increase tax com- describing the behavior of other groups should be less effec- pliance by visibly tackling tax avoidance among groups tive. However, reciprocity theory suggests that highlighting believed to avoid taxes, such as rich citizens. This paper is a product of the Governance 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 authors may be contacted at tpeixoto@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 Trickle Down Tax Morale: A Cross-Country Survey Experiment Jonathan Mellon, Tiago Peixoto, Fredrik M Sjoberg, and Varun Gauri Keywords: TAX MORALE, CROSS-NATIONAL EXPERIMENT, RETRIBUTIVE JUSTICE, TAX EVASION, TAX AVOIDANCE Classifications: H26 - Tax Evasion and Avoidance; H41- Public Goods; H50 - Government expenditures General Introduction In uncertain situations, the best available action may be to copy the behavior of those around you in the hope that they have good reasons for what they are doing. This tendency to imitate has been a highly successful problem-solving strategy for humans (Subiaul 2016). However, this imitation tendency opens up the possibility of influencing behavior by changing people’s perceptions of what those around them are doing. This perception of what others are doing is referred to as a descriptive norm. Treatments attempting to change descriptive norms have been used in more than 34 experiments designed to encourage pro-social behaviors (John, Sanders, and Wang 2015) including organ donation, charitable giving (Shang and Croson 2009; Croson and Shang 2013), reducing littering (Cialdini, Reno, and Kallgren 1990), and—most importantly for this paper—tax paying (Coleman 1996; Del Carpio 2014; Hallsworth et al. 2017; Kettle et al. 2016; John and Blume 2018). Descriptive norm tax experiments generally work by telling citizens that other citizens are paying taxes at high rates. The experiment then measures whether citizens who receive this treatment have higher willingness to pay tax or actually pay higher taxes. The results of these studies have often been positive. Experimental work in the United Kingdom by Hallsworth et al. (2017) found that informing late taxpayers that they were in a small minority who had not paid their tax on time increased the rate of payment by around 1 percentage point compared with a control letter, at essentially zero cost. Similarly, the UK Behavioural Insights Team found that a descriptive norms message increased the rate of people paying their tax debts by 15 percentage points compared to a control letter (Behavioural Insights Team 2012). However, some other tax experiments have found weak or even negative effects of descriptive norms (John and Blume 2018; Del Carpio 2014; Coleman 1996). The positive results from these descriptive norm tax experiments are generally attributed to the tendency of people to copy the behavior of those around them. The descriptive norm treatment “provides information about what is ‘normal’ in a novel or ambiguous situation” (Bobek, Hageman, and Kelliher 2013). However, these results also have an alternate explanation. Taxpaying is the archetypal example of a public goods game. Many people in society benefit from government spending, but they generally do so regardless of how much they personally contributed towards this spending in taxation. This leads to a free-rider problem, where less is contributed towards the public good than is optimal because it is in no one’s personal interest to contribute. If citizens do contribute and others do not, they risk being the sucker who pays for everyone else’s benefit (Jasay 1993). Research has shown that people deeply dislike being “the sucker” (Schnake 1991) and will under-contribute if they perceive that others are not doing their fair share. It is therefore possible that what appears to be citizens imitating descriptive norms may instead be citizens believing that they are not being taken advantage of by other citizens and that they can safely contribute without being exploited by free-riders. While these two explanations cannot be distinguished in the standard descriptive norm setup, the two theories do have different predictions for some cases. Descriptive norm theory says that a treatment will be most likely to work when “the source of reference is similar to us” (Bobek, Hageman, and Kelliher 2013). In other words, priming the subject’s group in the descriptive norm will be the most effective. However, the free-rider explanation would argue that changing beliefs about groups who are generally believed to exploit the tax system would be most effective, even if the subject is not a member of these groups. The obvious group to use to distinguish these two effects is rich people. There is substantial reason for citizens to believe that the rich in society are not paying their full tax burden. The Panama Papers revealed the extent to which wealthy citizens around the world have been using various legal and extralegal methods to avoid paying tax in their home countries (Harding 2016). Similarly, companies are able to use various accounting mechanisms and legal structures to pay rates well below the formal rates (for instance the average corporate tax rate actually paid by US companies was 12.1% of profits in 2011 compared to a headline rate of 39.1% (Paletta 2011)). The mechanism behind social norms treatments is important to understand because it affects what tax authorities should prioritize in terms of messaging and enforcement. If descriptive norms are the relevant mechanism, then focusing enforcement efforts on average taxpayers will be most effective because it will change the descriptive norms that are relevant to the most people. However, if the free-rider mechanism is the important one, it suggests that tax authorities should instead focus their enforcement efforts on visibly cracking down on tax evasion by groups who are most seen as exploiting the system, such as the richest taxpayers (Motel 2015). We compare three different social norm treatments focusing on the tax paying behavior of people generally, rich people and companies. On the reference group account, only the people treatment should be effective, while the avoiding being a sucker account would predict that all treatments could be effective but especially the rich people and companies treatments. We ran this survey in five English speaking countries covering a variety of development and tax compliance levels: Kenya, Australia, the United States, the Philippines, and South Africa. Our results support the free-rider interpretation of social norm interventions. We find large effects of priming rich people’s tax compliance (our treatments all suggest that tax compliance is relatively high) on two measures of tax morale. However, we find no evidence of an effect of priming ordinary people’s tax compliance and only a weak non-significant effect of priming corporate tax compliance. Study design We split respondents into four experimental groups: three treatment groups and a control group. The first treatment gave a standard descriptive norms message: “Last year most [country] citizens paid the tax they owed. It is uncommon for people to cheat on their taxes.” To reinforce this message (and implicitly get respondents to endorse it), we asked “Do you think it is a good thing that most people are paying their taxes?” with the response options yes and no. The other two treatments emphasized the tax paying behavior of groups the respondent would not be expected to belong to but are generally believed to have high rates of tax evasion. Either “Last year most rich [country] citizens paid the tax they owed. It is uncommon for rich people to cheat on their taxes.” or “Last year most [country] companies paid the tax they owed. It is uncommon for companies to cheat on their taxes,” with follow-up questions “Do you think it is a good thing that most rich people are paying their taxes?” and “Do you think it is a good thing that most companies are paying their taxes?” respectively. The experimental groups were compared to a control group where respondents were first shown the statement: “There are many popular search engines in the world. Search engines are used every day by over 1 billion people worldwide.” and then were asked “Many people say they are annoyed by all the advertising on search engines. Does this apply to you?” with the answer options of yes and no available. 1 These treatments allow us to compare the effectiveness of a standard descriptive norms treatment where the reference group is relevant to the citizen, with treatments of the descriptive behavior of salient outgroups. In this study we are interested in two outcomes related to tax morale: 1) a survey measure of tax morale “If a taxpayer does not report all of his income in order to pay less income taxes do you feel it is: Not wrong, A bit wrong, Wrong, and Seriously wrong” and 2) a survey measure of how much respondents would fine tax cheats: “If a taxpayer does not report all of their income in order to pay less income taxes, what percentage of their income should they pay as a penalty? None (0%), 1-10%, 11-20%, More than 20%.” Riwi survey platform We conducted the survey using the Riwi survey platform between 2017-06-21 and 2017-06-29. Riwi recruits respondents by presenting the survey to people who accidentally type an incorrect URL into their browser. Normally these websites will display advertisements to people but Riwi buys temporary space on these pages to display the survey instead. Respondents are first asked their age and gender (with respondents below the age threshold removed from the survey) and their location is determined based on their IP address. Because Riwi respondents are not expecting to take a survey, Riwi surveys have high attrition rates as the survey goes on. As a result, we take care to check for differential attrition and covariate imbalance (see appendix). We found no evidence of either problem in this study. Statistics The social norm study was conducted between 2017-06-22 and 2017-06-29. A total of 14,499 subjects answered Q1 (after the age & gender selector) and 6,698 answered Q8, a completion rate of 46.2%. Descriptive Statistics from the subjects that completed the study are shown in table 1. Table 1 Descriptive statistics Statistic N Mean St. Dev. Min Pctl(25) Pctl(75) Max Age 6,055 32.48 14.20 16 21 40 65 Female 6,055 0.35 0.48 0 0 1 1 Business Owner 6,054 0.06 0.23 0.00 0.00 0.00 1.00 1 An initial positive sign that respondents actually read the messages is that around 70% of respondents answered yes to the question in each treatment condition, whereas only 55% of the control group answered ‘yes’ that they are annoyed by advertising in search engines. Smartphone 6,055 0.56 0.50 0 0 1 1 Completed all Qs 6,055 1.00 0.00 1 1 1 1 Results Treatment effect The analysis is conducted only on respondents who completed the entire set of questions in order to account for differential attrition for different questions. Figure 1 shows the means within each treatment group for the tax morale and tax fine outcomes. The companies pay and rich people pay treatment means are significantly higher than the control group for both dependent variables. The people pay treatment is not significantly higher in either case. The ATE from a linear model (table 2) reveals that the rich people pay treatment increases the preference for tougher fines and increases tax morale. Figure 1 Tax Fine and Tax Morale Mean Responses Across Conditions (Completes Only) higher values indicate more support for fining tax cheats and stronger beliefs that cheating on taxes is wrong In the tax fine model, the rich people pay treatment makes respondents significantly more likely to endorse more robust tax enforcement than the people pay treatment. This difference is only significant at the 10% level in the tax morale model. Overall, however, our evidence strongly suggests that the rich people pay treatment is more effective than the more general people pay treatment when considering the evidence across both measures. Tax morale Tax fine (Intercept) 2.64*** 2.38*** (0.03) (0.03) Treatment People pay 0.06 0.01 (0.04) (0.04) Rich people pay 0.14*** 0.30*** (0.04) (0.04) Companies pay 0.09* 0.10* (0.04) (0.04) R2 0.00 0.01 Adj. R2 0.00 0.01 Num. obs. 6054 6054 RMSE 1.00 1.04 *** p < 0.001, ** p < 0.01, * p < 0.05 Table 2 Pooled Treatment Effects for Tax Morale Dependent Variable (OLS). Positive coefficients indicate a stronger belief that cheating on taxes is wrong. The manipulation checks were not statistically significant in this study and the effect sizes were substantively small, although the coefficients are in the right direction. This does not disprove that the treatment worked, but it does leave open the possibility that our treatment could work through mechanisms other than the one we proposed. The full manipulation check results are shown in the appendix. Although there was substantial attrition in this survey, models of retention and balance do not show any problems in this study (see appendix for details), with no detectable differences in retention or balance across the experimental groups. We can also look at how consistent these results are across each country. Table 3 shows the treatment effects on tax morale within each country. We get positive and significant results of the rich people pay treatment in three of the five countries and a non-significant and positive result in South Africa. The one outlier in the study is the United States where there is a small non- significant negative effect of the rich people pay treatment. The people pay treatment comes out as significant and positive in the Philippines but not in any other country context. The companies pay treatment shows a significant and positive effect in Kenya but not in any other country. Again, the most consistent evidence is for the rich people pay treatment, with more varied results for the other treatments. US PH ZA AU KE (Intercept) 2.74*** 2.50*** 2.58*** 2.65*** 2.68*** (0.05) (0.06) (0.08) (0.06) (0.05) Treatment People pay -0.09 0.19* 0.12 0.13 0.04 (0.07) (0.08) (0.11) (0.08) (0.08) Rich people pay -0.03 0.20* 0.19 0.26** 0.17* (0.07) (0.08) (0.11) (0.08) (0.08) Companies pay -0.13 0.14 0.22 0.10 0.21** (0.07) (0.08) (0.11) (0.08) (0.07) 2 R 0.00 0.01 0.01 0.01 0.01 Adj. R2 0.00 0.00 0.00 0.01 0.01 Num. obs. 1516 1516 715 972 1335 RMSE 0.99 1.06 1.05 0.92 0.98 *** p < 0.001, ** p < 0.01, * p < 0.05 Table 3 Treatment Effects by Country for Tax Morale Dependent Variable (OLS). Positive coefficients indicate a stronger belief that cheating on taxes is wrong. For the tax fine measure, the results are also reasonably consistent. Table 4 shows that four of the five countries show a positive and significant effect of the rich people pay treatment on the tax fine outcome measure. Australia does not show a significant effect, but the coefficient is relatively large and in the expected direction. The people pay treatment does not show a significant effect in any country. US PH ZA AU KE (Intercept) 2.31*** 2.30*** 2.32*** 2.52*** 2.48*** (0.05) (0.06) (0.07) (0.07) (0.06) Treatment People pay 0.00 0.14 0.11 -0.06 -0.09 (0.07) (0.08) (0.10) (0.09) (0.08) Rich people pay 0.20** 0.41*** 0.33** 0.26** 0.33*** (0.07) (0.08) (0.11) (0.09) (0.08) Companies pay -0.00 0.17* 0.19 0.07 0.11 (0.07) (0.08) (0.11) (0.09) (0.08) 2 R 0.01 0.02 0.01 0.01 0.02 Adj. R2 0.01 0.02 0.01 0.01 0.02 Num. obs. 1516 1516 715 972 1335 RMSE 1.02 1.09 1.01 1.02 1.03 *** p < 0.001, ** p < 0.01, * p < 0.05 Table 4 Treatment Effects by Country for Tax Fine Dependent Variable (OLS). Positive coefficients indicate more support for fining tax cheats. Other possible mechanisms? Our theory for the larger effects of the rich people pay in a descriptive norms experiment is based on the ideas of reciprocity and avoiding being taken advantage of. Reciprocity can be positive or negative. On one account these results could be driven by positive reciprocity: responding to a positive action by someone else with a positive action of your own. On this view tax paying by rich people would be a positive action that respondents are reacting to by increasing their willingness to take a positive action in return (paying taxes and supporting a more effective tax collection system). On the other account, these results may be driven by reducing negative reciprocity: responding to a negative action by someone else with a negative action of your own. On this view, respondents do not generally believe that the rich pay their fair share in society (which is perceived as a negative action against them) and respond with a negative action in return (not paying tax and being unwilling to support effective tax collection systems). Our treatment then reduces the beliefs of respondents that rich people are taking a negative action and reduces the negative reciprocal response of respondents. The distinction between these two types of reciprocity is rooted in the motivation. In negative reciprocity the motivation is one of retribution or sanctioning and in positive reciprocity it is driven by a sense of obligation to return favors. While both of these motivations are universal among humans, their relative importance varies across cultures and individuals. Future work could further narrow down the importance of these motivations for tax compliance. The standard descriptive norms account does not seem to fit our results well, because descriptive norms are usually conceived as being about people who are similar to the individual (Larimer et al. 2011), whereas we only see effects when we describe the pro-social behavior of rich people. We can further test the descriptive norm theory by looking at the effect of the company treatment on the subgroup of respondents who own businesses. Descriptive norm theory would predict that we should find the strongest treatment effect for the companies pay and rich people pay conditions among these respondents, because they should identify most closely with that reference group. Table 5 shows OLS models of both outcomes, with the treatments interacted with whether the respondent owns a business. We find some evidence that different taxpayer groups react differently to the treatments. In general, respondents increase their propensity to support higher tax fines when they are presented with information about rich taxpayers’ high compliance. However, we actually see the opposite effect among business owners and there is also a negative effect of the companies pay treatment on this subgroup. No interactions were significant for the tax morale measure. This is the exact opposite of the descriptive norms prediction. The treatment works well for those respondents who are dissimilar to the rich/company reference group, but it does not work for those respondents who are similar. Tax morale Tax fine (Intercept) 2.63*** 2.37*** (0.03) (0.03) Treatment People pay 0.07 0.01 (0.04) (0.04) Rich people pay 0.15*** 0.31*** (0.04) (0.04) Companies pay 0.11** 0.12** (0.04) (0.04) Business owner 0.07 0.14 (0.12) (0.12) People pay * business owner -0.11 0.05 (0.16) (0.17) Rich people pay * business owner -0.25 -0.24 (0.16) (0.17) Companies pay * business owner -0.31 -0.35* (0.16) (0.17) R2 0.00 0.01 Adj. R2 0.00 0.01 Num. obs. 6053 6053 RMSE 1.00 1.04 *** p < 0.001, ** p < 0.01, * p < 0.05 Table 5 Treatment Effects among business owners and non- business owners (OLS) This interaction effect also provides evidence against another alternative explanation: that respondents may see rich people as having better information about the benefits and costs of tax evasion. If the information mechanism was at work, this would be the most relevant to business owners (because the information is most applicable), but instead the companies pay and rich people pay treatment effects are weakest among business owners. A variant of the social norms argument may still be possible however. If people are not in fact most influenced by the behavior of people similar to them, but by role models with higher status, then perhaps people are simply learning their tax paying norms from the highest status members of society such as rich people. This interpretation is not that plausible in this study. Our manipulation check variables ask how common it is for the different groups to cheat on their taxes. We can use the control group to assess how reliably these groups were seen as paying their taxes. Among respondents in the control group, rich people are ranked the worst (mean 2.8 on a 1-4 scale), followed by companies (2.7), and then ordinary people 2.5 (all these differences are statistically significant). This does not suggest that rich people are being held up as positive role models to aspire to and is more consistent with the treatment working by slightly improving respondents’ beliefs about the behavior of rich people. A final mechanism that is more difficult to dismiss is that the experiment could be priming certain types of taxpayers rather than changing beliefs about them. Therefore, if people are thinking primarily about rich people, they may be more likely to say that tax paying is important because they think rich people do not pay enough tax. Combined with the relatively weak manipulation check results, this is the most plausible alternative explanation for our findings. Future work could distinguish priming effects by another control treatment that mentions the different groups but does not talk about the levels of tax compliance among the groups. If priming is the mechanism, we would expect that treatments merely priming rich people would work, even without any mention of tax. Conclusions This study provides strong initial evidence that priming citizens about the tax paying behavior of rich people in their society can have strong short-term effects on their stated tax morale (at least in the short term). In fact, this effect is much stronger than the effects of priming respondents about the behavior of ordinary citizens or companies. Our findings imply a trickle-down tax morale mechanism, where efforts to curb elite misuse of the tax system may have beneficial spillovers for the rest of the population’s tax compliance. These results suggest that governments wishing to improve tax morale may want to begin by tackling the issue among the richest sections of society, to remove the impression that the citizens who have done best from society are not paying their fair share. The most obvious limitation to this study is that it uses an attitudinal dependent variable rather than actual behavior. This leaves open the possibility that descriptive norms treatments priming the behavior of the general population have a stronger behavioral effect than our priming of rich people’s tax compliance. Future work should aim to try versions of this treatment in field experiments with behavioral outcome data. Our findings also call into question whether the standard account of how descriptive norms work is correct. Future work should do more to assess whether imitation is really the mechanism behind other findings attributed to descriptive norms, such as in the areas of charitable giving, turnout and organ donation. These studies may also be better explained either by a reciprocity or an aversion to being taken advantage of. Imitation is certainly a deep human tendency, but reciprocity and avoiding exploitation are probably just as deeply rooted in human psychology. References Behavioural Insights Team. 2012. “Applying behavioural insights to reduce fraud, error and debt.” Cabinet Office. https://doi.org/10.1111/0031-806x.00038. Bobek, Donna D., Amy M. Hageman, and Charles F. Kelliher. 2013. “Analyzing the Role of Social Norms in Tax Compliance Behavior.” Journal of Business Ethics 115 (3): 451–68. https://doi.org/10.1007/s10551-012-1390-7. Cialdini, Robert B., Raymond R. Reno, and Carl A. Kallgren. 1990. “A Focus Theory of Normative Conduct: Recycling the Concept of Norms to Reduce Littering in Public Places.” Journal of Personality and Social Psychology 58 (6): 1015–26. https://doi.org/10.1037/0022- 3514.58.6.1015. Coleman, Steve. 1996. “The Minnesota income tax compliance experiment: State tax results.” Minnesota Department of Revenue 4 (April): 528–41. http://mpra.ub.uni-muenchen.de/4827/. Croson, Rachel, and Jen Shang. 2013. “Limits of the effect of social information on the voluntary provision of public goods: Evidence from field experiments.” Economic Inquiry 51 (1): 473–77. https://doi.org/10.1111/j.1465-7295.2012.00468.x. Del Carpio, Lucia. 2014. “Are the Neighbors Cheating? Evidence from a Social Norm Experiment on Property Taxes in Peru.” Unpublished Manuscript, no. November. Hallsworth, Michael, John A. List, Robert D. Metcalfe, and Ivo Vlaev. 2017. “The behavioralist as tax collector: Using natural field experiments to enhance tax compliance.” Journal of Public Economics 148: 14–31. https://doi.org/10.1016/j.jpubeco.2017.02.003. Harding, Luke. 2016. “What are the Panama Papers? A guide to history’s biggest data leak | News | The Guardian.” https://www.theguardian.com/news/2016/apr/03/what-you-need-to- know-about-the-panama-papers. Jasay, Anthony de. 1993. “Taxpayers, Suckers and Free Riders.” Journal of Theoretical Politics 5 (1): 117–25. John, Peter, and Toby Blume. 2018. “How best to nudge taxpayers? The impact of message simplification and descriptive social norms on payment rates in a central London local authority.” Journal of Behavioral Public Administration 1 (1). https://doi.org/10.30636/jbpa.11.10. John, Peter, Michael Sanders, and Jennifer Wang. 2015. “The Use of Descriptive Norms in Public Administration: A Panacea for Improving Citizen Behaviours?” SSRN Electronic Journal, 1–28. https://doi.org/10.2139/ssrn.2514536. Kettle, Stewart, Mateo Hernandez, Simon Ruda, and Michael Sanders. 2016. “Behavioral Interventions in Tax Compliance - Evidence from Guatemala.” World Bank Policy Research, no. June: 1–40. https://ssrn.com/abstract=2811337. Larimer, Mary E, Clayton Neighbors, Joseph W LaBrie, David C Atkins, Melissa A Lewis, Christine M Lee, Jason R Kilmer, et al. 2011. “Descriptive drinking norms: For whom does reference group matter?” Journal of Studies on Alcohol and Drugs 72 (5): 833–43. http://www.ncbi.nlm.nih.gov/pubmed/21906510 http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC3174027. Motel, Seth. 2015. “5 facts on how Americans view taxes.” https://www.pewresearch.org/fact- tank/2015/04/10/5-facts-on-how-americans-view-taxes/. Paletta, Damian. 2011. “Tax Break Pushes Corporate Taxes to Just 12.1% of Profits, Lowest Level in 40 Years - WSJ.” https://www.wsj.com/articles/SB10001424052970204662204577199492233215330. Schnake, Mel. 1991. “Equity in Effort: The "Sucker Effect" in Co-Acting Groups.” Journal of Management 17 (1): 41–55. Shang, Jen, and Rachel Croson. 2009. “A field experiment in charitable contribution: The impact of social information on the voluntary provision of public goods.” Economic Journal 119 (540): 1422–39. https://doi.org/10.1111/j.1468-0297.2009.02267.x. Subiaul, Francys. 2016. “What’s special about human imitation? A comparison with enculturated apes.” Behavioral Sciences 6 (3): 1–26. https://doi.org/10.3390/bs6030013 Appendix Manipulation checks Table A shows the results of the manipulation checks. The results were in the expected direction but not statistically significant and the effect sizes are substantively small. The scale for the dependent variable runs from 1=Very uncommon to 4=Very common, i.e. a negative coefficient suggests an assessment in line with the treatment content (although the effect size for the rich people check is substantively small). People (Q6) Companies (Q7) Rich people (Q8) (Intercept) 2.52*** 2.67*** 2.77*** (0.02) (0.03) (0.03) Treatment People pay -0.05 -0.01 -0.00 (0.03) (0.04) (0.04) Rich people pay -0.02 -0.01 -0.01 (0.03) (0.04) (0.04) Companies pay -0.06 -0.03 0.02 (0.03) (0.04) (0.04) R2 0.00 0.00 0.00 Num. obs. 6640 6687 6698 RMSE 0.98 1.02 1.06 *** p < 0.001, ** p < 0.01, * p < 0.05 Table A Manipulation checks (OLS) Retention and Balance Table B shows how treatment group predicts completion of the Riwi survey. There are no significant differences in attrition across the experimental groups. There are also no significant differences in gender and age (both measured prior to treatment) distributions across experimental groups. There are no indications of imbalance on either of the available covariates. Completion (logit) Answer Q4 (logit) Answer Q5 (logit) Female (logit) Age (OLS) (Intercept) -1.38*** -0.38*** -0.34*** -0.31*** 36.83*** (0.03) (0.02) (0.02) (0.02) (0.18) Treatment People pay 0.02 0.03 -0.00 0.02 -0.03 (0.04) (0.03) (0.03) (0.03) (0.25) Rich people pay 0.02 0.01 -0.01 0.04 0.11 (0.04) (0.03) (0.03) (0.03) (0.25) Companies pay 0.05 -0.01 -0.02 0.03 -0.23 (0.04) (0.03) (0.03) (0.03) (0.25) Log Likelihood -14979.18 -19955.99 -20016.88 -20161.13 Deviance 29958.36 39911.98 40033.76 40322.26 Num. obs. 29525 29525 29525 29525 29525 R2 0.00 RMSE 15.05 *** p < 0.001, ** p < 0.01, * p < 0.05 Table B Retention and balance models