Spatial Inequality and Attitudes Towards Taxation: The Case of Sub-Saharan Africa by Agence Française de Développement

Agence française de développement

Papiers de recherche

AFD Research Papers

Les Papiers de Recherche de l’AFD ont pour but de diffuser rapidement les résultats de travaux en cours. Ils s’adressent principalement aux chercheurs, aux étudiants et au monde académique. Ils couvrent l’ensemble des sujets de travail de l’AFD : analyse économique, théorie économique, analyse des politiques publiques, sciences de l’ingénieur, sociologie, géographie et anthropologie. Une publication dans les Papiers de Recherche de l’AFD n’en exclut aucune autre.

AFD Research Papers are intended to rapidly disseminate findings of ongoing work and mainly target researchers, students and the wider academic community. They cover the full range of AFD work, including: economic analysis, economic theory, policy analysis, engineering sciences, sociology, geography and anthropology. AFD Research Papers and other publications are not mutually exclusive.

Les opinions exprimées dans ce papier sont celles de son (ses) auteur(s) et ne reflètent pas nécessairement celles de l’AFD. Ce document est publié sous l’entière responsabilité de son (ses) auteur(s).

The opinions expressed in this paper are those of the author(s) and do not necessarily reflect the position of AFD. It is therefore published under the sole responsibility of its author(s).

Spatial Inequality and Attitudes Towards Taxation: The Case of Sub-Saharan Africa

Lisa Chauvet Siyavash Eslami Laure Pasquier-Doumer DIAL, LEDa, IRD, Université ParisDauphine, PSL Research University, Paris, France Marin Ferry DIAL, LEDa, IRD, Université ParisDauphine, PSL Research University, Paris, France Université Paris-Est Marne-LaVallée, ERUDITE (EA 437), Marne la Vallée, France

Abstract This paper investigates the relationship between spatial inequality and attitudes towards tax compliance in the context of sub-Saharan African countries. Individuals' attitude toward tax is measured Using round six of the Afrobarometer geocoded data, and spatial inequality measured in terms of a Gini computed over light intensity. Results suggest that spatial inequality is positively associated with tax attitude decision. However, this association depends on the size of the area over which Gini is computed. We provide results for various zones around individuals and find that inequality in the immediate surrounding of individuals has an effect on their attitudes towards taxation. In order to deal with endogeneity problem we further use an instrumental variable approach, making use of an agglomeration model to predict the value of light at the year of the survey. These results are consistent with the OLS identification. We also examine the potential non-linearities in this relationship and find that, when facing high inequality, individuals in the middle of wealth distribution have unfavorable attitudes towards taxation. Finally we find in order for inequality to increase support for taxation, trust in the government should be high.

Keywords Inequality, tax attitude, tax compliance, local development, corruption

JEL Classification O12, O15, O23, O55, P35

Acknowledgments This document has been prepared with the financial assistance of the European Union. The opinions expressed herein must in no way be considered to reflect the official position of the European Union or that of the AFD. The authors declare that they have no relevant or material financial interests that relate to the research described in this paper.

Original version: English

Accepted: January 2020

Introduction Fiscal policies are an important way for

Tax compliance has been widely

income distribution and to provide

effort to collect taxes. Besides the

developing countries to reach a fair

essential public services to the whole

population. But widespread tax evasion makes the mobilization of domestic

revenue challenging for most African

countries. Although these countries have

witnessed an increase of their tax-to-GDP ratio over the past decade, the average level of taxation remains very low at

around 16 percent of gross domestic

product in 2016. The Addis Ababa Action

Agenda highlights how crucial mobilizing domestic resources in developing

countries is to the financing of the

sustainable development goals (SDGs). Yet, while the existing literature has

deeply investigated the economic factors influencing domestic resource

mobilization, the current state of knowledge on what shapes tax

compliance in developing countries is still at a pretty early stage, despite a buoyant

applied economic literature over the past thirty years. In light of the tenth

sustainable development goalâ&#x20AC;&#x2122;s

achievement, which precisely focus on inequality reduction, we study to what

extent the current level of inequality in

developing countries impacts domestic

recognized as crucial in the governmentâ&#x20AC;&#x2122;s traditional drivers of differences in tax

performance between industrialized and developing countries such as the tax base or state capacity in lawsâ&#x20AC;&#x2122;

enforcement, tax compliance can explain most of the tax gap between these two categories of countries. Indeed, while

OECD countries are also exposed to tax evasion, non-compliance behaviors

remain concentrated at the top of the income and capital distributions.

Because of a larger informal sector,

numerous loopholes in tax system, tax

base erosion and profit shifting strategies (UNECA, 2019), developing countries on

the other side, and African countries in particular, suffer from a much more

widespread tax avoidance which affects, yet in a different manner, all individuals along the income distribution. Tax

compliance thus represents an important tool for policy makers. If well addressed, it might consist in a significant (if not the most important) lever for domestic

resources mobilization that developing

countries desperately need to improve by 2030.

resource mobilization through the lens of

In this paper, we build on a flourishing

focuses on the ways spatial inequality

countries) to investigate the effect of

in 30 sub-Sahara African countries using

taxation measured through questions

national levels in order to capture the

Afrobarometer survey. In the tradition of

tax behavior. More specifically, this paper

literature (mostly focused on OECD

affects attitudes towards tax compliance

spatial inequality on attitudes towards

data dis-aggregated at various sub-

provided by the sixth round of the

contribution of local inequality.

Alesina et al. (2016), spatial inequality is

captured using nightlight data which is

countries, respondents living in a

grid level, and which enables us to

inequality report significantly better

available at an approximately 1 km x 1 km

compute a Gini index over administrative division of each country (example:

province). Moreover, since Afrobarometer provides latitude and longitude for each enumeration area (with 8 to

16 respondents in each enumeration area), we also identify alternative

boundaries around each area and

compute a Gini index for different zones

surrounding more plagued by nightlight attitudes towards taxation i.e. they are more in favor of taxation than

respondents living in less unequal areas.

We interpret this result as an indicative of more demand for redistribution in a

context of larger local inequality, while this does not necessarily translate in higher tax compliance.

(at a radius of 20, 30, 40, 50, and

Our findings are robust to the control for

enumeration areas). This allows us to

IV procedure where Gini indices are

inequality in the immediate surrounding

intensity based on the initial distance of

region, and investigate whether this

pixel. We also challenge our results to a

attitude. The focus on inequality at sub-

suggest that our findings remain

suspicion that such dis-aggregation

issues nor by our definition of attitudes

scale since individual’s perception of

individual characteristics available in

their surrounding environment based on

potential heterogeneity in the effect of

2015). Hence a measure of inequality at

taxation. Results indicate that such effect

reflects the perceived inequality from an

bottom of the national wealth

60 kilometers from the centroid of each

endogeneity, conducted using an

distinguish between the observed

computed on predicted pixels’ light

of an individual and inequality in a larger

each pixel fromits closest enlightened

variation has an effect on individual’s tax

wide set of robustness checks which all

national levels is motivated by the

unaffected by potential sample selection

levels are more relevant than the national

towards taxation. Lastly, we exploit

inequality is more likely to be accurate in

Afrobarometer surveys to investigate

day-to-day observations (Newman et al.,

spatial inequality on attitudes towards

the sub-national level probably better

mainly stems from people ranking at the

individual standpoint.

distribution. We also observe that the

We then estimate the contribution of

is sensitive to trust and perception of

spatial inequality, proxied by nightlight Gini indices, to the within country

variation in attitudes towards taxation on a sample of 30 sub-Saharan African

positive contribution of spatial inequality corruption that Afrobarometer’s

respondents grant to their fiscal and ruling institutions.

countries. Imposing controls for urban

The rest of the paper is organized as

channels affecting tax attitudes, and

difference in various concepts related to

average positive effect of spatial

attitudes towards taxation. It also reviews

attitudes towards taxation. Within

relationship between inequality and

versus rural living areas, other potential

follows. Section 2 aims at establishing the

individual characteristics, we find an

tax compliance such as tax morale and

inequality in 30-60km buffer areas on

the existing literature about the

attitudes towards taxation. Section 3

Section 4 presents the results,

specification suggested to assess the

channels, and challenges the robustness

discusses the data and the empirical contribution of spatial inequality to

within-country variation in tax attitudes.

investigates the potential transmission

of our main findings. Section 5 concludes.

Introduction

Fiscal policies are an important way for developing countries to reach a fair income distribution and to provide essential public services to the whole population. But widespread tax evasion makes the mobilization of domestic revenue challenging for most African countries. Although these countries have witnessed an increase of their tax-to-GDP ratio over the past decade, the average level of taxation remains very low at around 16 percent of gross domestic product in 2016. The Addis Ababa Action Agenda highlights how crucial mobilizing domestic resources in developing countries is to the financing of the sustainable development goals (SDGs). Yet, while the existing literature has deeply investigated the economic factors influencing domestic resource mobilization, the current state of knowledge on what shapes tax compliance in developing countries is still at a pretty early stage, despite a buoyant applied economic literature over the past thirty years. In light of the tenth sustainable development goal’s achievement, which precisely focus on inequality reduction, we study to what extent the current level of inequality in developing countries impacts domestic resource mobilization through the lens of tax behavior. More specifically, this paper focuses on the ways spatial inequality affects attitudes towards tax compliance in 30 sub-Sahara African countries using data dis-aggregated at various sub-national levels in order to capture the contribution of local inequality. Tax compliance has been widely recognized as crucial in the government’s effort to collect taxes. Besides the traditional drivers of differences in tax performance between industrialized and developing countries such as the tax base or state capacity in laws’ enforcement, tax compliance can explain most of the tax gap between these two categories of countries. Indeed, while OECD countries are also exposed to tax evasion, non-compliance behaviors remain concentrated at the top of the income and capital distributions. Because of a larger informal sector, numerous loopholes in tax system, tax base erosion and profit shifting strategies (UNECA, 2019), developing countries on the other side, and African countries in particular, suffer from a much more widespread tax avoidance which affects, yet in a different manner, all individuals along the income distribution. Tax compliance thus represents an important tool for policy makers. If well addressed, it might consist in a significant (if not the most important) lever for domestic resources mobilization that developing countries desperately need to improve by 2030. In this paper, we build on a flourishing literature (mostly focused on OECD countries) to investigate the effect of spatial inequality on attitudes towards taxation measured through questions provided by the sixth round of the Afrobarometer survey. In the tradition of Alesina et al. (2016), spatial inequality is captured using nightlight data which is available at an approximately 1km × 1km grid level, and which enables us to compute a Gini index over administrative division of each country (ex.: province). Moreover, since Afrobarometer provides latitude and longitude for each enumeration area (with 8 to 16 respondents in each enumeration area), we also identify alternative boundaries around each area and compute a Gini index for different zones (at a radius of 20, 30, 40, 50, and 60 kilometers from the centroid of each enumeration areas). This allows us to distinguish between the observed inequality in the immediate surrounding of an individual and inequality in a larger region, and investigate whether this variation has an effect on individual’s tax attitude. The focus on inequality at sub-national levels is motivated by the suspicion that such dis-aggregation levels are more relevant than the national scale since individual’s perception of inequality is more likely to be accurate in their surrounding environment based on day-to-day observations (Newman et al., 2015). Hence a measure of inequality at the sub-national level probably better reflects the perceived inequality from an individual standpoint. 2

We then estimate the contribution of spatial inequality, proxied by nightlight Gini indices, to the within country variation in attitudes towards taxation on a sample of 30 sub-Saharan African countries. Imposing controls for urban versus rural living areas, other potential channels affecting tax attitudes, and individual characteristics, we find an average positive effect of spatial inequality in 30-60km buffer areas on attitudes towards taxation. Within countries, respondents living in a surrounding more plagued by nightlight inequality report significantly better attitudes towards taxation i.e. they are more in favor of taxation than respondents living in less unequal areas. We interpret this result as an indicative of more demand for redistribution in a context of larger local inequality, while this does not necessarily translate in higher tax compliance. Our findings are robust to the control for endogeneity, conducted using an IV procedure where Gini indices are computed on predicted pixelsâ&#x20AC;&#x2122; light intensity based on the initial distance of each pixel from its closest enlightened pixel. We also challenge our results to a wide set of robustness checks which all suggest that our findings remain unaffected by potential sample selection issues nor by our definition of attitudes towards taxation. Lastly, we exploit individual characteristics available in Afrobarometer surveys to investigate potential heterogeneity in the effect of spatial inequality on attitudes towards taxation. Results indicate that such effect mainly stems from people ranking at the bottom of the national wealth distribution. We also observe that the positive contribution of spatial inequality is sensitive to trust and perception of corruption that Afrobarometerâ&#x20AC;&#x2122;s respondents grant to their fiscal and ruling institutions. The rest of the paper is organized as follows. Section 2 aims at establishing the difference in various concepts related to tax compliance such as tax morale and attitudes towards taxation. It also review the existing literature about the relationship between inequality and attitudes towards taxation. Section 3 discusses the data and the empirical specification suggested to assess the contribution of spatial inequality to withincountry variation in tax attitudes. Section 4 presents the results, investigates the potential transmission channels, and challenges the robustness of our main findings. Section 5 concludes.

Inequality and attitudes towards taxation

2.1

Attitudes towards taxation, tax compliance, and tax morale: a conceptual framework

Tax compliance is not a straightforward concept. It is complex and shaped by many factors relating to various economic, sociological and political features. The literature has distinguished different tax compliance determinants that can be classified in five broad categories (Fjeldstad et al., 2012; Ali et al., 2014; Lutmer and Singhal, 2014): The first one refers to the seminal work by Allingham and Sandmo (1972) and resorts to a theoretical and economic framework to identify pecuniary determinants of tax compliance. Considering the expected utility of the taxpayer, the authorsâ&#x20AC;&#x2122; model suggests that tax evasion directly stems from an individual trade-off between the costs and benefits associated with evading taxes. More specifically, they represent the individual utility of the tax evader as a function negatively correlated with how easy it is to detect fraud and how costly it is for individuals to be caught (i.e. the magnitude of the penalty associated with tax fraud). The contribution of this pecuniary factor to tax compliance, often coined as the economic deterrence channel, has been recently investigated in the context of African countries by (Fjeldstad et al., 2012; Ali et al., 2014)

which both find that easiness in avoiding taxes is indeed negatively correlated with tax compliance behavior. The economic deterrence mechanism prevailed for a while as the main drivers of tax compliance before the literature starts, in the early 2000s, investigating the role played by non pecuniary motivations. Non pecuniary factors encompass a large set of intrinsic, reciprocity, and social motivations that shape individual attitudes towards taxation and that are usually gathered under the concept of tax morale. The intrinsic motivations refers to the taxpayer’s inner perception of how right it is to contribute to the national tax mobilization effort. Yet, intrinsic motivation to comply with the tax system can be directly related to the individual perception of fairness of the tax system (Besley et al., 2019) which in turn depends on the taxpayer’s reciprocity expectations. The reciprocity channel, referred as the fiscal exchange mechanism, states that tax compliance should be higher when people perceive the benefits they can get out of paying taxes, notably regarding public good provision (Cowell and Gordon, 1988; Moore, 1998, 2004; D’Arcy, 2011). Intuitively, good infrastructure and public services such as health, education, or the police should be associated with more attitudes in favor of taxation (i.e. higher tax morale), without reinforcing coercion (Bodea and LeBas, 2013). This has also been evidenced by Fjeldstad et al. (2012) and Ali et al. (2014) in the context of some African states. Building on the social influence theory, some studies (both in OECD and developing countries) have simultaneously investigated the contribution of norms, social and cultural influence on tax behaviour. In a different context, Banerjee (1992) shows that social influences might directly affect individual economic decision by fostering herd behavior leading people to favour others’ information over their own. In this vein and regarding our particular focus, individuals’ tax attitudes has indeed been shown to be influenced by the average behaviour among the taxpayer’s reference group (i.e. people comply when then perceive that their peers comply as well (Andreoni et al., 1998)). Simultaneously, tax evasion has also been found to be determined by long-run cultural factors which remain more difficult to inflect than taxpayers’ risk aversion (Cummings et al., 2009). Still related with the influence of social factors, the literature has then highlighted the contribution of taxpayer’s treatment perception by fiscal authorities, particularly with respect to treatment endured by other taxpayers. Tax compliance seems to improve when individuals perceive that they - and/or the group they belong to - benefit from a fair treatment from the state, compared to other groups in the society (i.e. when their comparative treatment by fiscal authorities is rather fair (McKerchar and Evans, 2009)). Lastly, and in a pretty straightforward manner, people are also expected to report more pronounced noncompliant behavior when they do not grant much confidence in their public institutions. The relationship between tax compliance and trust in the government, or in fiscal department, is thus expected to be positive (Kirchler et al., 2008). Often labelled as the political legitimacy channel, Fjeldstad et al. (2012) and Ali et al. (2014) have emphasized, thanks to Afrobarometer data, that trust was indeed positively correlated with tax compliance in the context of some African countries. Related to this last point is the role played by non-state actors which may decrease tax compliance when substituting for the state in the provision of public goods. If informed about the presence of non-state actors in their surrounding, taxpayers might believe that taxation is therefore less needed to improve their standards of living thanks to the presence of alternative, and individually less costly, financing sources (Ali et al., 2014). In this paper, we choose not to focus on the restricted concept of tax compliance. Since our analysis mobilizes survey data, self-assessed attitudes towards taxation are likely to be biased. Afrobarometer surveys

consist in self-declared assessments of the quality and perception of public institutions that are collected during face-to-face interviews with Afrobarometer questioners. Although Afrobarometer is a non-state organisation, people surveyed might still perceive that their responses could be reported to public institutions. Hence, on sensitive questions such as the frequency of tax avoidance, people might under-report their noncompliance behavior by fear of retaliation hence giving rise to an attenuation bias. However, looking at other sensitive questions such as trust in institutions, Calvo et al. (2019) evidence that there is no systematic attenuation bias in responses of Afrobarometer respondents. In addition, they could also simply be ashamed of declaring that they do not pay they taxes (and do not contribute to the national effort in raising the country financing capacities) to someone they do not know. These limitations lead us to consider a more encompassing concept that we label as attitudes towards taxation. This concept, somehow, captures the pure tax compliance feature (do you pay taxes or not?) as well as non pecuniary factors explaining tax compliance i.e. the tax morale. Yet, in our framework, tax morale is not limited to the sole intrinsic motivation to pay taxes and also considers the individual perception of the necessity to pay taxes (because of reciprocity or social influence), even this does not imply that the respondent should bear the tax burden, or part of it. Section 3 on data discusses the way we compute the variable capturing attitudes towards taxation.

2.2

To what extent can inequality shape attitudes towards taxation?

Our analysis seeks to understand the link between inequality and attitude toward tax compliance. This relationship has not yet been formalized, but there is a very rich literature analyzing the effect of inequality on tax evasion. A large strand of the behavioral research suggests that within country growing disparity contributes to accentuated taxpayer stress and thus increases the propensity to evade taxation. According to some studies, such negative effect between inequality and tax evasion should mostly occur at both ends of the income distribution given the reduced visibility of transaction among low and high income earners. While richest individuals should be more prone to enter tax-optimization strategies in a context of high inequality (the causal effect of this relationship playing both ways), poorer individual might also consider to avoid taxation. This would be facilitated by a large prevalence of the informal sector to which they most often relate to (especially in the context of low-income countries and sub-Sahara African countries). Despite the absence of study investigating such relationship in the context of African countries, others produced results supporting the negative correlation between inequality and tax compliance. In line with the social actor model, Williams and Krasniqi (2017) examine the intrinsic motivation to pay taxes and investigate the individual and national heterogeneity in tax morale. On a sample of 35 Eurasian countries and using the 2010 Life in Transition Survey they show that age, marital status, education do alter the intrinsic motivation to pay taxes and highlight a negative correlation with income inequality. They also provide evidence for the modernisation hypothesis, according to which tax morale is higher in more developed countries with stronger legal systems and less corruption. GerstenblĂźth et al. (2008) find a similar result using the Latinobarometro (2005) dataset on a cross-section of Latin American countries. Their results suggest that the level of inequality is negatively associated with tax compliance. Finally, Bloomquist (2003) use data on the United States for the 1947 to 1999 period and also find a negative relationship between inequality and tax compliance (proxied with salary under-reporting) Yet, given the original nature of taxes, which aims at financing public goods and services, one might 5

think of tax compliance as a proxy of demand for redistribution. Considering tax compliance through the lens of demand for redistribution leads us to rely on another and flourishing literature investigating the effect of inequality on demand for redistribution. The question of whether inequality affect votes for redistributive policies appeared in the 1970s and suggests that inequality can be an important determinant of tax compliance. The median voter model predicts a positive effect of inequality on pressure for redistribution; by widening the gap between mean and median income, inequality increases the median voter’s benefit from redistribution (Romer, 1975; Meltzer and Richard, 1981). As put by Kenworthy and McCall (2008) : "The lower the median relative to the mean, the more the median income person or household is likely to benefit from government redistribution, in the sense that the transfers she receives will exceed her share of the tax burden." This mechanism relates to the fiscal exchange argument in which the prospects of benefit from public good provision increase tax compliance. The literature seeking to test the link between inequality and the demand for redistribution has been abundant for more than twenty years and yet no consensus has emerged. Some papers have found support for the median voter model prediction that inequality is positively associated with a higher demand for redistribution.1 Gründler and Köllner (2017) provide evidence consistent with the median voter model, but they find that the effect of inequality on redistribution is stronger when using perceived inequality measures rather than objective measures. In addition, they find that the link is more tenuous in developing countries because of weaker political rights. Yet, the median voter model cannot explain the differences observed between the United States and Europe, the latter being characterized by lower inequality and a higher redistribution than the United States (Benabou, 1996). This observation has given rise to new theoretical models introducing political structures in which elites have a greater political weight (Benabou, 2000) or considering the prospects for social mobility as more decisive in the demand for redistribution than position in income distribution (Piketty, 1995; Benabou and Ok, 2001). Heterogeneous preferences for redistribution according to social or identity groups may also explain the failure of the median voter model to explain cross-country differences. Alesina et al. (2001), for example, explain why the United States is both more unequal and less redistributive than Europe by introducing the concept of “reciprocal altruism” whereby voters are reluctant to redistribute money to the poor if the poor are considered responsible for their poverty situation or if they belong to a different ethnic or identity group. This hypothesis of “reciprocal altruism” is justified by numerous studies, mainly conducted in the US context, showing that individuals are all the more favorable to redistribution when the poorest resembles them (Alesina et al., 1999; Luttmer, 2001). Other studies also highlight a negative relationship between inequality and redistribution. Kenworthy and McCall (2007) adopt a long-run intra-country approach by analyzing whether the rise in inequality during the 1980s and 1990s led to more redistribution. Based on the study of 8 OECD countries, they show that the median voter model does not hold. This result is similar to that of Georgiadis and Manning (2012) for the United Kingdom. Islam et al. (2018) reach the same conclusion by studying the link between inequality and redistribution over a long period (1870-2011) in 21 OECD countries. In addition, a recent and abundant literature showing the negative effect of ethnic inequalities on the provision of public goods has also contributed to challenging the neoclassical model of the median voter (Alesina et al., 1999; Alesina 1 See for example Milanovic (2000) on a sample of 24 developed or transition countries; Kevins et al. (2018) for 22 European countries; Kerr (2014), Corcoran and Evans (2010), or Boustan et al. (2013) considering regional variations in the United States.

and Ferrara, 2005). Although the notion of demand for redistribution is different from that of tax compliance, there is a strong link between the two concepts. But while a high demand for redistribution requires tax compliance, a high tax compliance does not necessarily imply a strong demand for redistribution, since the fiscal exchange theory is only one of the many mechanisms that explains tax compliance (Fjeldstad et al., 2012; Ali et al., 2014; Lutmer and Singhal, 2014).

Model and Data

3.1

Identification Strategy

The empirical specification we suggest to implement makes use of cross-section data and aims at estimating attitudes towards taxation at the individual level, imposing controls for different channels mentioned in previous section. We include as predictor an objective inequality measure computed at different levels of disaggregation, hence capturing our relationship of interest between inequality and tax attitude. The empirical specification hence takes the following form: T AXAT Ti,z,c = αc + βIN EQU ALIT Yz,c + θYz,c + γCHAN N ELSi,z,c + δXi,z,c + εi,z,c

(1)

where z is a circular zone with a radius ∈ {20km, 30km, 40km, 50km, 60km}. T AXAT Ti,z,c denotes the attitudes towards taxation of the individual i surveyed in the zone z of country c. IN EQU ALIT Yz represents inequality measured at the the zone z (so in a given country c) which could go beyond subdivision and country boundaries. Yz is a vector of variable capturing characteristics of the zone z around individuals and encompasses the size of the population (in logarithm), and the level of nightlight intensity per capita. CHAN N ELSi,z,c is a vector including variables at the individual-level controlling for the various channels through which attitudes towards taxation might be affected. Xi,z,c comprises individual characteristics that might explain why some people tend to display more pro-tax behaviors than others, such as age, education, employment status. Lastly, αc denotes a set of country fixed-effects, accounting for the contribution of time invariant country-level characteristics. Note that our model also considers urban/rural fixed effects which are not reported in equation 1 in order to keep variables’ subscript short. In this specification, attitudes towards taxation is hence measured at individual level whereas the main explanatory variable is measured at the Zone-level which would be the same for individuals in an enumeration area (depending on the size of the enumeration are). In order to attenuate the potential bias as a result of this difference, we follow literature by clustering the standard errors at the zone-level Abadie et al. (2017).

3.2

Individual Level Data

The main source of data for this study comes from the 6th round of Afrobarometer surveys conducted in 2014-15. The surveys include a series of questions about attitudes toward taxation that are comparable across more than 30 African countries. As previously explained the use of declarative measures regarding tax compliance might be thought of as debatable since they are likely to suffer from an over-reporting bias. As such we use different dependent variables related to attitudes towards taxation in order to alleviate 7

the potential attenuation bias and increase the reliability of our dependent variable. Hence our dependent variable is a composite measure constructed from questions related to various dimensions of tax morale. These questions are reported in Table 1. Since the scale varies for each question, we first re-scale each variable on a scale from 0 to 1 and then take a simple average. The composite dependent variable is then labelled TAX_ATTITUDE, observed for each Afrobarometer respondents i and is referred as either attitudes towards taxation (that we interchangeably labelled as tax attitudes in what follows). We will further use 3 other dependent variables as robustness checks. Descriptive statistics in Table A1 suggest that, among our sample (respondents pooled across 30 sub-Sahara African countries), individuals report an attitude that seems relatively in favor of taxation with a mean of 0.68. Table 1 â&#x20AC;&#x201C; Measuring Attitudes Towards Taxation: Dependentsâ&#x20AC;&#x2122; Composition Variable

Question

Scale

Rescaled to

NOT_PAYING_TAXi,z,c

Wrong or right not to pay taxes

1-3

0-1

PEOPLE_MUST_PAY_TAXi,z,c

Authorities always have the right to make people pay tax.

1-5

0-1

CITIZEN_MUST_PAY_TAXi,z,c

Citizens must pay taxes vs. no need to tax the people

1-5

0-1

CITIZEN_PAY_TAX_IN_DEMOCi,z,c

Good citizens pay tax in democracy

1-5

0-1

PAY_TAX_INCREASE_HEALTHi,z,c

Pay more taxes to increase health spending

1-5

0-1

Regarding the imposed controls intended to capture the channels through which tax attitudes might be affected CHAN N ELSi,z,c , we use various questions from the Afrobarometer survey, each of which relating to one of the mechanism explained in the introduction. For the economic deterrence mechanism, we consider the difficulty to avoid taxes (DIF F _AV OID_T AXi,z,c ) as well as the perception of individuals about how often people committing fraud can go unpunished (P EOP LE_U N P U N ISHEDi,z,c ) as measures of being caught when not complying. In order to capture the contribution of the fiscal exchange theory we use two questions reflecting the difficulty of getting public services such as medical and police services (DIF F _OBT AIN _M EDICi,z,c and DIF F _OBT AIN _P OLICEi,z,c ). To capture the comparative treatment channel, we consider a question assessing the individual perception of relative poverty which reflects how good the respondent grades his living conditions with respect to the others (IN EQ_P ERCEP T IONi,z,c ). Lastly, in order to control for the effect of political legitimacy on tax compliance we add two variables reporting the level of trust in the tax department as well as in the elected president (T RU ST _T AX_DEP.i,z,c and T RU ST _P RESIDEN Ti,z,c ). The model controls for individual characteristics that may correlate with their willingness to pay taxes. We control for age, AGEi,z,c , education, EDU CAT IONi,z,c , and employment status, EM P LOY Ti,z,c . Afrobarometer surveys also provide questions regarding the type of goods each respondent owns as well as indications regarding their accommodation. Using various individual characteristics we run a principal component analysis in order to get an individual wealth index (W EALT Hi,z,c ) that is added to our specification

as an individual control. More details about these coding of explanatory variables are provided in Table A3 in the Appendix. Table A1 in the Appendix provide summary statistics for variables used in the study (these used in the equation 1 as well as variables mobilized further in the paper).

3.3

Inequality Using Nightlight Data

Turning to our variable of interest and in order to take into account spatial inequality across different sub-national levels we construct a Gini index based on nightlights per capita which is a strong proxy of development at national and subnational level (Henderson et al., 2012; Michalopoulos and Papaioannou, 2013; Beyer et al., 2018). These studies used lights data from older generation of satellites which covered the period between 1992 to 2013. However, since Afrobarometer survey we use were conducted in 2014 and 2015, we make use of the new generation of light data (Visible Infrared Imaging Radiometer Suite (VIIRS)). The main advantage of this new set of light data is no saturation problem.2 . The VIIRS data however (used in this study), has not been used as much as the DMSP-OLS light data. Chen and Nordhaus (2015) shows that it predicts development better than the previous set. In our dataset, lights per capita at the country level has a correlation coefficient of 0.88 with GDP per capita at PPP constant dollar which is significant at 1 percent level. A recent study shows that, relative to VIIRS data, spatial inequality measured from DMSP-OLS is noisy and underestimated (Gibson et al., 2019). Elvidge et al. (2012) was, to our knowledge, the first to use a Gini constructed from light per capita to examine the distribution of income and wealth. Authors however, conclude that this type of Gini is best described as “spatial extend of public good provision”, rather than individual’s income distribution. Nevertheless, the very recent literature dealing with spatial inequality measured from light density, conceptualizes this Gini as income and/or wealth inequality (Alesina et al., 2016) (Bluhm and Wong, 2017). Which interpretation of spatial inequality corresponds better to the light Gini needs more exploration. In order to compute the Gini that represents our measure of inequality (IN EQU ALIT Yz,c ), we first retrieve latitude and longitude of each enumeration area from Afrobarometer to Geo-localize each respondent and further create a buffer zone around each individual for different radii (20,30, 40, 50, to 60 kilometers). We then impose a 1km × 1km fishnet on light data and extract the value of light for each pixel within this buffer zone. We then divide it by population from Gridded Population of World (for International Earth Science Information Network , CIESIN) leading us to obtain the light per capita at the 1km×1km pixel-level. Since we now have the value of light and the population for each pixel we then are able to rank inhabitants of the buffer zone according to the value of light their are exposed to. Using the distribution of inhabitants with the distribution of light within the buffer zone we then compute the Gini coefficient for each buffer zone (z) in each country c using the classic Gini formula: 2

n P

pLIGHT pcp,z,c

p=0

IN EQU ALIT Yz,c = 1 − 2 ×

n−1 P

− LIGHT pcp,z,c

n+1 n

p=0

where LIGHT pcp,z,c is the value of light per capita for pixel p of the buffer zone z in the country c (with p pixels being sorted in ascending order — i.e. LIGHT pcp,z,c ≤ LIGHT pcp+1,z,c ). Lastly, we match each 2 For

a review of the older generation light data and its limits see Doll (2008)

Figure 1 â&#x20AC;&#x201C; Buffer Zone with radius of 50KM

variable computed at the buffer zone-level z (IN EQU ALIT Yz,c , P OP U LAT IONz,c , LIGHT pcz,c 3 ) with Afrobarometerâ&#x20AC;&#x2122;s respondents considered in this study. Figure 1 shows the 50km buffer zone as an example. We further compute Gini indices at the first administrative division-level of each country using the same approach as for buffer zones. Figure A1 in Appendix shows an example of spatial inequality for a few ADM1 regions in South Africa. Table A1 in the Appendix provides descriptive statistics for variables at the buffer zone-level.

3.4

Dealing with endogeneity

Our model specification represented with equation 1 gives rise to several issues that might bias the identification of a causal effect running from inequality to attitudes towards taxation. Unobserved heterogeneity at the country-level such as differences in tax systems or historical context that leads to significant level of inequality are likely to be captured by the inclusion of country fixed-effects. Similarly, although higher population density is often associated with better tax compliance and more pro-tax behaviors, structural gaps in attitudes towards taxation between urban and rural individuals are also accounted for with the inclusion of urban fixed-effects. Yet, reverse causality cannot be controlled for with inclusion of fixed-effects and is therefore likely to bias the identification of a proper causal effect of inequality on attitudes towards taxation. Theoretically, when inequality increases, citizens demand more redistribution which requires higher taxes. In the other way around, and in order to reduce inequality, governments need to implement policies which usually need a strong fiscal capacity, which in itself needs higher taxation. However, we think the difference in the 3 The

last two variable being the aggregation of pixel-level value of population and light per capita.

observation scale between attitudes towards taxation (at the individual-level) and inequality (at the zonelevel) helps in attenuating reverse causality (to some extent). It seems indeed quite unlikely that the tax attitude of a sole individual in a given zone affects the provision of public goods and therefore the level of inequality observed in this area. Nevertheless, capturing inequality at the more disaggregated level (i.e. within smaller buffer zones) can lead to consider very few individuals across the area on which light inequality is measured. Therefore one might think that better and more equal availability of light within a restricted area is potentially influenced by attitude of a given individual with respect to taxation, although it would imply a strong decentralization process where sub-regional provision of electricity is financed by local taxes (which is, to our knowledge, not much generalized in the context of most sub-Sahara African countries). Consequently, and despite the various levels of observations and a large set of controls, estimating equation 1 with OLS estimators does not prevent results being affected by unobserved time-variant confounding factors at the zone-level. In order to deal with endogeneity issue, we rely on an instrumental variable strategy. We construct our instrument, by first dividing the sample of pixels in the year 2000 to lit and non-lit pixels. After identifying the closest lit pixel, n, for every non-lit pixel p, we compute the distance between the two, LN DIST AN CEp,n,2000 . Then using an agglomeration model we predict the value of nightlight for the year of survey.4 We call this the zero-stage estimation which takes the following form: 2 LIGHTp = ζLN DIST AN CEp,n,2000 + λLN DIST AN CEp,n,2000 + φLN P OPp,2000 + ADM 1p + εp

(2)

where LIGHTp is the the observed value of light for pixel p at the year of the survey (2014 or 2015 depending on the country), LN DIST AN CEp,n,2000 is the natural logarithm of the distance between pixel p and the closest lit pixel to it (n) in the year 2000, and LN P OPp,2000 measures the population (in logarithm) of pixel p in year 2000. The model also includes square of distance, and first administrative subdivision fixed-effects. Results of the zero-stage estimation are reported in Table S.A1 in the supplementary appendix. The result display a nonlinear relation between light intensity and the logarithm of distance: pixels that are far from lit-areas in 2000 have a high probability of not being lit in 2014/15, unless these are more isolated from any lit-areas. Population in 2000 is also positively correlated with light value in 2014/15. ˆ pcp ) and Finally we compute a Gini over these predicted lights (i.e. P REDICT ED_LIGHTp (LIGHT use this “predicted gini” (P REDICT ED_IN EQz,c ) as the instrument for inequality at the year of survey IN EQz,c .5 This instrument (which is computed for our various buffer zones as well as the first administrative subdivisions) can thus be defined as: 2 P REDICT ED_IN EQz,c =

n P

iP REDICT ED_LIGHT pcp,z,c

p=0

n−1 P

− P REDICT ED_LIGHT pcp,z,c

n+1 n

p=0

4 Afrobarometer survey was conducted in 2014/2015 and in the baseline estimation we measure IN EQU ALIT Y for the z same years using VIIRS nightlight data. However, VIIRS is only available after 2012. So in order to construct the instrument we used nightlight data from DMSP-OLS, which is the older generation of light data. 5 Note that the zero stage model reports an explanatory power of 30% which is high enough to be relevant and low enough for allowing a different variability in Gini instrument as compared to the one of the actual Gini indices in 2014-2015.

The original model includes the level of light per capita averaged around the individuals, LIGHT pcz , which, as inequality, may be endogenous to the attitude of individuals towards taxes. Therefore, we use lights predicted by an agglomeration model and re-aggregated on various zones z around the individuals, P REDICT ED_LIGHT pcz , to instrument LIGHT pcz .6

Results

4.1

Baseline results

Before discussing the results of the estimations for various zones around the individuals, we first examine the relationship between tax attitude and inequality measured at ADM1 level (first administrative division). Table 2 below reports results of the OLS estimation between inequality measured at first administrative division-level (ADM1-level) and tax attitude, as well as the contribution of all other channels likely to affect attitudes towards taxation. The various channels are sequentially added to the specification in order to verify their inherent effect that might vanish when controlling for all of mechanisms in the same regression. In all regressions standard errors are clustered at the ADM1 level. In Table 2 all of the channels follow the expected signs identified in the literature. When it is perceived that people could get away with tax evasion, and that there is not enough deterrence, their attitude is less likely to be in favor of taxation, albeit this relation is not statistically significant. When access to public goods such as medical services and police is difficult, it discourages individuals to support taxation; when government does not respect the fiscal exchange mechanism, neither do the citizens. Looking at the relationship from another angle, this could suggest that a significant increase in prospects for public goods and services might trigger people to improve their attitude towards tax compliance, though the causal relationship might also run in the opposite direction. Further, the individual perception of her group treatment by the government is negatively and significantly associated with lower tax morale. Additionally when individuals perceive their living conditions to be less favorable than that of others with whom they associate, they do not support taxation. This relates, to some extent, to the comparative treatment mechanism discussed above which states that perception of unfavorable treatment towards oneâ&#x20AC;&#x2122;s reference group adversely affects tax compliance. And finally trust in political institutions and confidence in the tax department are associated with more tax compliance, as suggested by political legitimacy channel. Overall, the results are in line with Ali et al. (2014) and seem to support the theoretical mechanisms determining tax compliance for our sample of African countries. Not reported in the table but included are individual characteristics such as age, education and employment. In line with previous studies in context of Africa (Fjeldstad et al., 2012; Ali et al., 2014), these characteristics suggest that older, more educated and employed people tend to be more in favor of taxation. While most of channels added separately or all together (in column 6) display the expected sign, estimations show that spatial inequality measured at the ADM1 level does not have a direct association with individual attitude towards taxes. However, in some countries first administrative divisions (ADM1) are very vast. The average ADM1 surface area in our sample is around thirty thousands square kilometers (three times that of a zone with a radius of 60 kilometers). As such, spatial inequality observed at this level might 6 when

LIGHT pcz is assumed to be exogenous, the results are similar and lead to the same conclusions.

not be perceived by the individual. Table 2 – Inequality and Attitude towards Tax Compliance: ADM1 level (1)

(2)

(3)

(4)

(5)

(6)

Econ. deterrence

Fiscal Exch.

Comp. Treatment

Pol. Legit.

All Channels

-0.005 (0.031)

-0.013 (0.031)

-0.006 (0.031)

-0.010 (0.030)

Dependent: TAXATTi,z,c

INEQUALITYz

-0.007 (0.031)

DIFF._AVOID_TAXi,z,c PEOPLE_UNPUNISHEDi,z,c

0.001 (0.003)

0.003 (0.003)

-0.007∗∗∗ (0.002)

-0.005∗∗ (0.002)

DIFF._OBTAIN_MEDICi,z,c

-0.015∗∗∗ (0.003)

-0.009∗∗∗ (0.002)

DIFF._OBTAIN_POLICEi,z,c

-0.012∗∗∗ (0.003)

-0.005 (0.003)

POVERTY_RELATIVEi,z,c

-0.012∗∗∗ (0.002)

-0.007∗∗∗ (0.002)

ETHNIC_UNFAIRi,z,c

-0.018∗∗∗ (0.002)

-0.011∗∗∗ (0.002)

TRUST_TAX_DEP.i,z,c

0.037∗∗∗ (0.003)

0.036∗∗∗ (0.003)

TRUST_PRESIDENTi,z,c

0.009∗∗∗ (0.002)

0.006∗∗∗ (0.002)

LIGHTpcADM 1

-0.001 (0.001)

-0.000 (0.001)

-0.001 (0.001)

0.100

0.101

0.105

0.110

0.144

0.151

Country FE Urabn/Rural FE Indiv. Controls Cluster Level

Yes Yes Yes ADM1

N. Obs. N. Countries N. Subdivisions

37552 362 30

Notes: Ordinary least squares (OLS) regressions on tax attitudes for 2014/2015 cross-section of 30 Sub-Saharan African countries. In columns (1) through (6) IN EQU ALIT Y z is a Gini computed over first administrative divisions of each country. All estimations control for individual characteristics (age, education, employment and wealth), and include country and urban/rural fixed effects with robust standard errors clustered at ADM1 level (reported in parentheses). ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

In order to explore this suspicion we have computed multiple Gini for smaller areas, from a radius of 20 to 60 kilometers surrounding each individual. Table 3 contains the result of these estimations. Each column reports the OLS estimated coefficients of Equation 1 for one of the zones, for which IN EQU ALIT Yz is computed over a circular zone with the radius of 20, 30, 40, 50 or 60 kilometers. All estimations control for the various channels mentioned in the introduction, as well as country and urban/rural fixed effect and heteroskedasticity with robust standard errors clustered at same level as the variable of interest, IN EQU ALIT Yz . 13

Table 3 – Inequality and Attitude towards Tax Compliance: Zone Level

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

INEQUALITYz

0.034∗∗∗ (0.011)

0.031∗∗∗ (0.012)

0.025∗∗ (0.012)

0.015 (0.013)

0.016 (0.013)

-0.005 (0.030)

DIFF._AVOID_TAXi,z,c

0.003∗ (0.002)

0.003 (0.002)

0.003∗ (0.002)

0.003 (0.003)

PEOPLE_UNPUNISHEDi,z,c

-0.006∗∗∗ (0.001)

-0.005∗∗∗ (0.001)

-0.005∗∗ (0.002)

DIFF._OBTAIN_MEDICi,z,c

-0.009∗∗∗ (0.002)

DIFF._OBTAIN_POLICEi,z,c

-0.005∗ (0.002)

-0.005 (0.003)

POVERTY_RELATIVEi,z,c

-0.007∗∗∗ (0.001)

-0.007∗∗∗ (0.002)

ETHNIC_UNFAIRi,z,c

-0.011∗∗∗ (0.002)

TRUST_TAX_DEP.i,z,c

0.036∗∗∗ (0.002)

0.036∗∗∗ (0.003)

TRUST_PRESIDENTi,z,c

0.006∗∗∗ (0.001)

0.005∗∗∗ (0.001)

0.006∗∗∗ (0.002)

LNPOPz

-0.001 (0.001)

-0.002 (0.002)

-0.007 (0.005)

LIGHTpcz

0.003 (0.002)

0.002 (0.002)

0.001 (0.001)

0.002 (0.003)

-0.003 (0.002)

0.151

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

37552 4370 362 30

37547 4369 362 30

37552 4370 362 30

37552 362 362 30

Dependent: TAXATTi,z,c

R2 Country FE Urabn/Rural FE Indiv. Controls Cluster Level N. N. N. N.

Obs. Cluster ADM1 Countries

Notes: Ordinary least squares (OLS) regressions on tax attitudes for 2014/2015 cross-section of 30 Sub-Saharan African countries. In columns (1) through (5) IN EQU ALIT Y z is a Gini computed over different circular zones at the radius identified on the top of each column; from 20 to 60 Kilometers. In column(6) IN EQU ALIT Y z is measured at first administrative division (ADM1) level of the countries. All estimations control for individual characteristics (age, education, employment and wealth), and include country and urban/rural fixed effects with robust standard errors clustered at the level over which inequality is computed (reported in parentheses). ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

A review of Table 3 suggests that our suspicion is correct and, contrary to ADM1-level, inequality measured at smaller areas does influence the attitudes of tax payers. The estimated coefficients in Column (1) to (3) show positive and statistically significant association between inequality and attitudes toward tax compliance. However, when estimated on OLS this association fades away for larger zones with radius of 50 and 60 kilometers and ADM1 level. Nonetheless, since Table 3 only contains OLS estimations we do not consider this as evidence for lack of relationship between large-area inequality and tax compliance. Indeed OLS coefficients could be biased downward for example if the zone suffers a shock that could increase inequality and at the same time decrease the willingness of individuals to pay taxes. In order to deal with this problem we implement an instrumental variable approached as explained in section 3.1. Table 4 reproduces the result of OLS estimations (Panel A) and presents the results of the IV estimations of Equation 1 (Panel B) and their first steps (Panel C). Panel C of Table 4 contains the first-stage estimation results which show that predicted inequality increases todayâ&#x20AC;&#x2122;s inequality and the first-stage post-estimation statistics such as F-stat confirms that the constructed instrument is relatively strong. Regarding second-step IV results in Table 4, Panel B, we find that the results are in line with the OLS estimations: higher inequality increases the willingness of individuals to pay taxes. Moreover, the IV results show that the OLS coefficients are underestimated and suggest larger bias for higher radiusr; for 20 kilometers buffer zone the IV estimated coefficient of inequality is two times that of OLS whereas for 40 kilometers it is four and a half times larger. This may be explained by the fact that the zero-stage estimation, which is an agglomeration model, performs better in predicting inequality on a small radius than on a large one, hence threatening the exclusion restrictions on larger zones. Finally, Table 4 shows that the inequality measured at ADM1 level (Column (6)) is not statistically significant, either in OLS or in IV. We interpret these results as evidence for the hypothesis that inequality at the immediate surrounding of citizens have an impact on their attitudes towards taxation, whereas when individual does not perceive inequality, even if larger, it does not affect their attitudes.

4.2

Robustness checks

As a robust test, Table S.A2 in the supplementary appendix reproduces the result of Equation 1 using ADM1 as the level of clustering for all the specifications, which is much larger than the zones. Imposing robust-standard clustering at a larger level helps in accounting for potential spatial correlation in individual attitudes towards taxation between people leaving in the same subdivision (while such spatial correlation is much likely to occurs in neighbourhood or in zones smallest than sub-division). Both IV and OLS estimations correspond to the main results in Table 4. Although we loose some precision as it is expected with a cluster at a higher aggregation level. Using the standardized coefficients (beta coefficients) we find that a one standard deviation increase in inequality for 60-kilometer radius, increases tax attitude by 0.28 of its standard deviation which corresponds to a 8% increase in the average tax attitude score in our sample. In order to challenge the robustness of the result, we also estimate Equation 1 using alternative dependent variables. As mentioned before declarative tax-related measures might suffer from over-reporting bias. Using the same questions in Table 1 we construct 3 alternative composite dependent variables. Table A5 in the Appendix reports the results of OLS and IV estimation using inequality measured over a radius of 30 kilometers. Both OLS and IV results support the main results. Moreover, all three alternative dependents have larger magnitudes which is to be expected since in construction of the main dependent variable our 15

goal was to compute a conservative measure of attitudes towards tax compliance. Table 4 – Inequality adn Attitudes towrds Tax Compliance: Different Zones

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.034∗∗∗ (0.011)

0.031∗∗∗ (0.012)

0.025∗∗ (0.012)

0.015 (0.013)

0.016 (0.013)

-0.005 (0.030)

0.003 (0.002)

0.002 (0.002)

0.001 (0.001)

0.002 (0.003)

-0.003 (0.002)

0.151

0.056∗ (0.032)

0.099∗∗ (0.044)

0.114∗∗ (0.050)

0.116∗∗ (0.057)

0.135∗∗ (0.066)

-0.077 (0.114)

0.004 (0.002)

0.002 (0.002)

0.001 (0.001)

0.005 (0.004)

-0.003 (0.002)

0.074

0.072

0.071

0.070

0.073

0.276∗∗∗ (0.026)

0.232∗∗∗ (0.029)

0.222∗∗∗ (0.027)

0.207∗∗∗ (0.025)

0.184∗∗∗ (0.021)

0.194∗∗∗ (0.053)

0.000 (0.001)

0.001∗ (0.001)

0.000 (0.001)

0.001∗∗ (0.001)

0.007∗∗∗ (0.002)

-0.004 (0.005)

0.00 57

0.00 33

0.00 34

0.00 39

0.00 7

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

37552 4370 362 30

37547 4369 362 30

37552 4370 362 30

37552 362 362 30

Notes: Panel A presents the Ordinary least squares (OLS) regressions on tax attitudes for 2014/2015 cross-section of 30 Sub-Saharan African countries (corresponding to the previous table). In columns (1) through (5) IN EQU ALIT Y z is a Gini computed over different circular zones at the radius identified on the top of each column; from 20 to 60 Kilometers. In column(6) IN EQU ALIT Y z is measured at first administrative division (ADM1) level of the countries. Panel B and C show, respectively, the second and first stage of the instrumental variable approach. IN EQU ALIT Y z is instrumented by a P REDICT ED_IN EQz,c based on an agglomeration model using the pixels with light in 2000. All estimations in all the panels control for all the mechanisms explained in the previous section as well as individual characteristics (age, education, employment and wealth), and include country and urban/rural fixed effects with robust standard errors clustered at the level over which inequality is computed (reported in parentheses). ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

We then remove individuals in buffer zones near to their country boarders (in a range of alternately 10, 20, 16

and 30km). This helps us to mitigate the issue of buffer zones overlapping some others from a neighbouring country. Inequality in a buffer zone that straddles two countries might have differentiated effect on tax attitudes since the individual demand for redistribution (and hence tax attitudes) depends partly on the perception that taxpayers grant to their government and home country’s institutions. Results from tables S.A3 to S.A8 in the supplementary appendix suggest that baseline results are mostly driven by people living away from national boarders. Indeed when restricting the sample to Afrobarometer’s respondents living in the 10 up to 30km near the boarder, the effect of inequality on tax attitudes looses in significance. This indirectly highlights the potential undefined effect of inequality on attitudes towards taxation when zone-level inequality is defined and constructed over two countries. Lastly we re-run equation 1 removing one country (and alternately one frist administration subdivision - ADM1) in order to ensure that the effect we observed in Tables 3 and 4 is a genuine average effect and not only driven by observations from one given country or one particular ADM1. Results are not reported in the appendices but support our previous findings and are available upon request to the authors. In the remainder of the paper we follow the same pattern for all tables (Panel A OLS, Panel B second-stage, Panel C first stage), and only discuss them where there is a significant change.

4.3

Heterogeneity of the effect of Inequality

Previous results suggest a linear positive relationship between inequality and tax attitudes. In the following we examine the presence of non-linearities in this relationship. We first examine whether wealth groups and the position of an individual in the wealth distribution changes their attitude towards taxation when facing inequality. We then explore whether quality of institutions influences how inequality affects citizens attitude towards taxes. 4.3.1

Does inequality affect wealth groups differently?

This section investigates non-linearity in the relationship between inequality and tax attitude with regard to the position of the respondents in the wealth distribution. Both inequality and taxes affect various wealth groups differently. Rich and poor individuals face different types and rates of tax ratios. Many poor individuals are not eligible for formal taxes, and yet they might be more in favor of taxation whereas the situation for those higher in wealth groups could be different. Moreover, theoretically the impact of inequality on the poor is expected to be larger than on the rich. As such we anticipate a heterogeneous effect of inequality on tax attitude. In order to investigate such non-linearity we re-run the previous model extended with a dummy variable capturing the wealth quintiles of each respondent and its interaction with the Gini indicator. The empirical specification can thus be written as: T AXAT Ti,z,c = αc + βIN EQU ALIT Yz,c +

4 X

φj Qji,z,c +

4 X

ψj Qji,z,c × IN EQU ALIT Yz,c + εi,z,c (3)

j=1

where Qji,z,c is a dummy variable indicating the quintile to which the respondent i in the zone z of country c belongs. We use the wealth index constructed from information in Afrobarometer surveys to compute the different quintiles. We also chose to define the quintiles at the national level rather than the subdivision 17

level in order to have enough individuals in each quintile. Although not reported in the equation 3, this new specification still controls for contribution of other channels discussed in section 3.1, CHAN N ELSi,z,c , as well as individual controls, Xi,z,c , and country and urban/rural fixed effects. The standard errors are still clustered at the zone z level. In IV estimations, we used the interaction of the quintiles with the instrument of inequality as the instrument for the interaction terms. Table 5 reports the results of OLS and IV estimations of 3 for different zones. One can notice that while the inequality is positive and statistically significant (in line with the previous tables), the only interaction term that is statistically significant is with the third quintile, which has a negative sign. In Column (3) onward in Panel B, the interaction with the third quintile is now positive, albeit not significantly different from zero. But, we now observe that the effect of the sole inequality (so in interaction with quintiles) is still positive but drastically loose in precision and significance. It seems that the overall positive effect identified priory is mostly driven Afrobarometer’s respondents at the bottom of the wealth distribution i.e. those belonging to the first two quintiles. Indeed the coefficient associated with interaction terms are significantly positive for buffer zones of 40 up to 60 km and as compared to respondents from the fifth quintile. In addition joint significance with inequality in level (not interacted) is marginally significant (around 11%) which suggest that the relationship of interest is at play for a particular segment of the population only, that is to say the poorer ones. This let us think, that in line we predictions of the median voter theorem, the 40% bottom of the wealth distribution is prone to support higher redistribution (through attitudes more favourable towards taxation. when inequality in terms of public good provision are significant. For median voter theorem to be at play here, one factor is democratic condition in these countries; we expect in countries that enjoy a higher level of quality of democracy to observe a stronger effect. Table A6 shows the results of interaction with respondent’s satisfaction with the democracy in the country and the respondent’s perception of the extend of the democracy. For both measures of quality of democracy the interaction is positive and statistically significant suggesting that inequality’s effect on tax attitude increases with quality of the democracy. We then also investigate whether inequality has a different effect on tax attitudes depending on the distance of Afrobarometers’ respondents to the highest light in their own buffer zone. We expect respondents far away to be more likely to report attitudes in favor of taxation when being exposed to higher inequality since they would represent those left behind in terms of public goods provision. Results of Table A7 in the Appendix suggest first that the sole distance is negatively associated with attitudes towards taxation since far away respondent are more subject to display less rosy attitudes towards taxation than those closer to the highest light in their zone. Indeed people having few light in their restricted neighbourhood but close from intense light can still benefit from electricity provision if their work or one of their activity is located near the highest zone light. Conversely people really far away are constraint by the geographic distance which reduces their light access. Their need of public good provision is therefore important so do their demand for redistribution, as suggest the results. We also test interaction with the economic distance defined as the difference in light intensity with the highest light in the zone. We also suggest a second measure of economic distance which is divided by the geographical distance and that we coined weighted economic distance. Yet, results of tables S.A9 and S.A10 in the supplementary appendix do not emphasize an effect as similar as when we take the simple straight distance with highest light.

Table 5 – Inequality and Attitudes towards Tax Compliance w/r to Wealth Groups (1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.046∗∗ (0.021)

0.043∗∗ (0.022)

0.032 (0.022)

0.018 (0.022)

0.024 (0.022)

-0.030 (0.036)

q1=1 X INEQUALITYz

0.003 (0.027)

0.004 (0.027)

0.017 (0.027)

0.019 (0.027)

0.013 (0.027)

0.063 (0.042)

q2=1 X INEQUALITYz

-0.008 (0.025)

-0.000 (0.025)

0.002 (0.025)

0.004 (0.025)

0.001 (0.025)

0.072∗∗ (0.035)

q3=1 X INEQUALITYz

-0.053∗∗ (0.023)

-0.057∗∗ (0.024)

-0.058∗∗ (0.023)

-0.046∗ (0.024)

-0.057∗∗ (0.024)

-0.022 (0.033)

q4=1 X INEQUALITYz

0.002 (0.023)

-0.007 (0.024)

0.007 (0.023)

0.010 (0.023)

0.007 (0.023)

0.027 (0.030)

Joint Sig. P-value (q3)

0.059

0.052

0.040

0.104

0.038

0.356

R2 Panel B: IV - Second Stage INEQUALITYz

0.152

0.151

0.152

0.050 (0.041)

0.080 (0.052)

0.072 (0.058)

0.063 (0.066)

0.080 (0.071)

-0.148 (0.117)

q1=1 X INEQUALITYz

0.026 (0.043)

0.032 (0.042)

0.069∗ (0.041)

0.096∗∗ (0.042)

0.105∗∗ (0.041)

0.150 (0.098)

q2=1 X INEQUALITYz

0.023 (0.038)

0.042 (0.037)

0.061 (0.037)

0.078∗∗ (0.038)

0.088∗∗ (0.037)

0.157∗∗ (0.074)

q3=1 X INEQUALITYz

-0.034 (0.037)

-0.010 (0.037)

0.026 (0.037)

0.042 (0.037)

0.050 (0.036)

0.086 (0.065)

q4=1 X INEQUALITYz

0.016 (0.037)

0.019 (0.036)

0.038 (0.035)

0.043 (0.034)

0.057∗ (0.033)

0.095 (0.059)

Joint Sig. P-value (q3)

0.453

0.274

0.185

0.156

0.116

0.257

R2 Panel C: IV - First Stage PREDICTED_INEQz

0.075

0.073

0.071

0.070

0.069

0.074

0.275∗∗∗ (0.029)

0.216∗∗∗ (0.032)

0.205∗∗∗ (0.031)

0.193∗∗∗ (0.028)

0.181∗∗∗ (0.024)

0.222∗∗∗ (0.058)

q1=1 X PREDICTED_INEQz

-0.017 (0.021)

0.005 (0.022)

0.011 (0.019)

0.007 (0.018)

-0.007 (0.017)

-0.077∗∗ (0.035)

q2=1 X PREDICTED_INEQz

0.012 (0.019)

0.033∗ (0.019)

0.035∗∗ (0.017)

0.024 (0.016)

0.007 (0.015)

-0.032 (0.032)

q3=1 X PREDICTED_INEQz

-0.000 (0.016)

0.014 (0.016)

0.010 (0.014)

0.008 (0.014)

-0.004 (0.014)

-0.020 (0.025)

q4=1 X PREDICTED_INEQz

0.007 (0.014)

0.024∗ (0.014)

0.028∗∗ (0.012)

0.030∗∗ (0.012)

0.020∗ (0.011)

-0.021 (0.018)

0.00 19

0.00 11

0.00 13

0.00 2

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

37552 4370 362 30

37547 4369 362 30

37552 4370 362 30

37552 362 362 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes: Panel A shows the OLS estimation of Equation 3 in19 which we examine whether inequality affects wealth groups differently. Panel B and C provide, respectively, the second and first stage of the instrumental variable approach. The p-value of joint significance of IN EQU ALIT Yz,c and its interaction term is reported at the bottom of each panel. All estimations control for all of the mechanisms explained in the previous section as well as individual characteristics (age, education, employment and wealth), and include country and urban/rural fixed effects with robust standard errors clustered at the level over which inequality is computed (reported in parentheses). ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

4.3.2

Sensitivity to trust in institutions

We further examine whether the institutional environment influences the positive impact of inequality on individual’s attitude towards tax. In line with the other channels that have been highlighted in the existing literature, the inclusion of an interaction term between inequality and trust in institutions seek to capture the differential effect of inequality with respect to the political legitimacy channel discussed in section 2. Building on our prior findings we expected Afrobarometers’ respondents to display attitudes in favor of taxation when inequality in their surrounding is important, even more when they trust their public institutions and economic policy-makers. Indeed, in trusty public and political environment people would be more inclined to ask for redistribution while in a context of corrupt and less reliable politics or public administration people will know that deploying pro-tax attitudes in order to help government finance public goods provision in order to reduce inequality would be useless. Table 6 shows the result of interaction terms between inequality and trust in (columns (1) to (5)). As in previous tables, all models control for a set of country and urban/rural fixed-effects and control variables at the individual and country-level. Trust in tax department and in the president are present in all models since they were part of the baseline estimation model in Table 3). We also report the p-value of the joint significant test for inequality and the corresponding interaction term. Note that all trust variables are coded in order for the higher values to show more confidence. Surprisingly in all models both OLS and IV estimated coefficients of inequality are not statistically significant whereas the interaction terms are positive and statistically significant (besides trust in elected local officials in column (4)). However, the joint significance test shows that in all estimations these variables are jointly significant. These results suggests that inequality would result in more support for taxation only if citizens trust their government. One point of concern is the negative sign of the interacted variable (trust variables) when it enters the model in level which is significant in 4 of the estimation. Although the coefficients are jointly significant, this suggests that for certain levels of inequality, marginal effect of trust is zero (including at mean). This issue needs further assessment of each individual trust variable to identify the critical thresholds under/above which marginal effects are significant. In the same vein, we run the same estimates as these reported in Table 6 but using corruption perception variable instead of trust. As we expected, results of table 7 report a negative coefficient for the interaction term between inequality and corruption variables. Almost all types of corruption perception besides — surprisingly— the one that people display regarding their tax department are dampening the positive effect of inequality on attitudes towards taxation. As for results in table 6 corruption variables in level report positive coefficient, hence suggesting that perception of a large prevalence of corruption in public institutions is positively correlated with attitudes in favor of taxation.

Table 6 – Interaction with Institutional Environment: Trust (OLS & IV) 30km (1) Trust Tax Dep.

(2) Trust President

(3) Trust Parlia.

(4) Trust Elec. Loc.

(5) Trust Rull.

0.000 (0.019)

0.012 (0.022)

-0.000 (0.021)

0.008 (0.019)

-0.022 (0.021)

INEQUALITYz X VARi,z,c

0.013∗∗ (0.006)

0.007 (0.007)

0.012∗ (0.006)

0.009 (0.006)

0.021∗∗∗ (0.007)

VARi,z,c

0.028∗∗∗ (0.004)

0.001 (0.004)

-0.010∗∗ (0.005)

-0.009∗∗ (0.004)

-0.015∗∗∗ (0.004)

0.003

0.015

0.004

0.009

0.000

0.151

0.157

0.003 (0.046)

-0.038 (0.050)

-0.003 (0.047)

0.047 (0.044)

-0.088∗ (0.052)

INEQUALITYz X VARi,z,c

0.026∗∗∗ (0.010)

0.037∗∗∗ (0.011)

0.027∗∗ (0.011)

0.007 (0.010)

0.051∗∗∗ (0.012)

VARi,z,c

0.020∗∗∗ (0.007)

-0.018∗∗ (0.007)

-0.020∗∗∗ (0.007)

-0.008 (0.007)

-0.034∗∗∗ (0.008)

0.003

0.000

0.005

0.150

0.000

0.074

0.073

0.074

0.013∗∗∗ (0.004)

0.014∗∗∗ (0.004)

0.005 (0.004)

-0.001 (0.004)

0.012∗∗ (0.005)

0.235∗∗∗ (0.027)

0.228∗∗∗ (0.028)

0.252∗∗∗ (0.027)

0.270∗∗∗ (0.027)

0.216∗∗∗ (0.028)

0.00 40

0.00 41

0.00 34

Yes Yes Yes ZoneID

37552 4370 362 30

37549 4370 362 30

35362 4182 345 28

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz

Joint Sig. P-value R

Panel B: IV - Second Stage INEQUALITYz

Joint Sig. P-value 2

R Panel C: IV - First Stage PREDICTED_INEQz X VARi,z,c PREDICTED_INEQz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters Subdivisions Countries

Notes: Panel A, B and C provide, respectively, the OLS, second stage and first stage estimation of instrumental approach. The p-value of joint significance of IN EQU ALIT Yz,c and its interaction term with each trust variable, is reported at the bottom of each panel. All estimations control for all of the mechanisms explained in the previous section as well as individual characteristics (age, education, employment and wealth), and include country and urban/rural fixed effects with robust standard errors clustered at the level over which inequality is computed (reported in parentheses). ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 7 – Interaction with Institutional Environment: Corruption (OLS & IV) 30km (1) Corr. Tax Dep.

(2) Corr. President

(3) Corr. Gov. Off.

(4) Corr cop.

(5) Corr Courts

(6) Handle Corr.

0.028 (0.020)

0.051∗∗∗ (0.018)

0.039∗ (0.021)

0.053∗∗ (0.024)

0.031 (0.020)

0.020 (0.019)

0.002 (0.008)

-0.010 (0.007)

-0.003 (0.008)

-0.009 (0.008)

0.000 (0.008)

0.006 (0.007)

-0.018∗∗∗ (0.005)

-0.003 (0.005)

-0.005 (0.005)

-0.001 (0.005)

-0.010∗∗ (0.005)

0.005 (0.005)

Joint Sig. P-value

0.022

0.008

0.024

0.018

0.023

0.018

R2 Panel B: IV - Second Stage INEQUALITYz

0.155

0.152

0.153

0.152

0.077 (0.049)

0.166∗∗∗ (0.045)

0.171∗∗∗ (0.051)

0.179∗∗∗ (0.055)

0.131∗∗∗ (0.048)

0.029 (0.045)

INEQUALITYz X VARi,z,c

-0.004 (0.013)

-0.046∗∗∗ (0.012)

-0.044∗∗∗ (0.013)

-0.044∗∗∗ (0.014)

-0.028∗∗ (0.013)

0.020∗ (0.012)

VARi,z,c

-0.014∗ (0.008)

0.019∗∗ (0.008)

0.020∗∗ (0.008)

0.021∗∗ (0.009)

0.007 (0.008)

-0.004 (0.008)

Joint Sig. P-value

0.175

0.000

0.001

0.002

0.022

R2 Panel C: IV - First Stage PREDICTED_INEQz X VARi,z,c

0.078

0.074

0.073

0.074

0.075

0.011∗ (0.006)

0.002 (0.005)

0.010∗ (0.005)

0.009∗ (0.005)

0.007 (0.007)

0.008∗ (0.005)

0.242∗∗∗ (0.033)

0.262∗∗∗ (0.031)

0.245∗∗∗ (0.031)

0.244∗∗∗ (0.031)

0.251∗∗∗ (0.034)

0.249∗∗∗ (0.024)

0.00 37

0.00 39

0.00 37

0.00 41

Yes Yes Yes ZoneID

37534 4370 362 30

37542 4370 362 30

37537 4370 362 30

37541 4370 362 30

37533 4370 362 30

37546 4370 362 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz INEQUALITYz X VARi,z,c VARi,z,c

PREDICTED_INEQz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters Subdivisions Countries

Notes: Panel A, B and C provide, respectively, the OLS, second stage and first stage estimation of instrumental approach. The p-value of joint significance of IN EQU ALIT Yz,c and its interaction term with each corruption variable, is reported at the bottom of each panel. All estimations control for all of the mechanisms explained in the previous section as well as individual characteristics (age, education, employment and wealth), and include country and urban/rural fixed effects with robust standard errors clustered at the level over which inequality is computed (reported in parentheses). ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Conclusion

This paper examines the relationship between inequality and attitudes towards taxation in 30 African countries. Using Geo-coded data from round 6 of Afrobarometer survey, we construct a composite measure of attitude towards taxation at the individual level. Further using light intensity data from VIIRS we compute a Gini index over alternative zones surrounding each individual. The results show that when facing high levels of inequality, citizens are more supportive of taxation. However, this effect only exists when inequality is measured at circular areas with radius of 30 to 60 kilometers and the impact fades away at larger areas such as ADM1 level. Although we compute an objective measure of inequality, these results provide evidence for the hypothesis that perceived inequality is a more important factor and that policy makers needs to focus on local inequality alongside aggregated inequality. These findings are robust to instrumental approach as well. We further examine the heterogeneity of inequality with regard wealth groups to which each respondents belongs. We find that individuals in â&#x20AC;&#x153;poorer classâ&#x20AC;? (i.e. the bottom 40% f the weallth distribution) are more supportive of taxation when inequality is high which grant support for the median voter theorem indirectly challenged in this paper. Estimate results also suggest that taxpayers display more attitudes in favor of taxation in reaction to high level of local inequality when they perceive their institutions as relatively more democratic. This provides another support for the median voter mechanisms where in absence of democratic environment, accountability and voice from the government people would be less inclined in asking for more redistribution. In the same vein, we finally examine the effect of institutional environment on relationship between inequality and tax compliance. Consistently with our previous results, we find that inequality raises support for taxes only if citizens trust the government.

References Abadie, A., Athey, S., Imbens, G. W., and Wooldridge, J. (2017). When should you adjust standard errors for clustering? Technical report, National Bureau of Economic Research. Alesina, A., Baqir, R., and Easterly, W. (1999). Public goods and ethnic divisions. The Quarterly Journal of Economics, 114(4):1243–1284. Alesina, A. and Ferrara, E. L. (2005). Ethnic diversity and economic performance. Journal of economic literature, 43(3):762–800. Alesina, A., Glaeser, E., and Sacerdote, B. (2001). Why doesn’t the us have a european-style welfare system? Technical report, National bureau of economic research. Alesina, A., Michalopoulos, S., and Papaioannou, E. (2016). Ethnic inequality. Journal of Political Economy, 124(2):428–488. Ali, M., Fjeldstad, O.-H., and Sjursen, I. H. (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. G. and Sandmo, A. (1972). Income tax evasion: a theoretical analysis. Journal of Public Economics, 1:323–338. Andreoni, J., Erard, B., and Feinstein, J. (1998). Tax compliance. Journal of economic literature, 36(2):818– 860. Banerjee, A. V. (1992). A simple model of herd behavior. The quarterly journal of economics, 107(3):797–817. Benabou, R. (1996). Inequality and growth. NBER macroeconomics annual, 11:11–74. Benabou, R. (2000). Unequal societies: Income distribution and the social contract. American Economic Review, 90(1):96–129. Benabou, R. and Ok, E. A. (2001). Social mobility and the demand for redistribution: the poum hypothesis. The Quarterly Journal of Economics, 116(2):447–487. Besley, T., Jensen, A., and Persson, T. (2019). Norms, enforcement, and tax evasion. Technical report, National Bureau of Economic Research. Beyer, R. C., Chhabra, E., Galdo, V., and Rama, M. (2018). Measuring districts’ monthly economic activity from outer space. Bloomquist, K. M. (2003). Us income inequality and tax evasion: A synthesis. Tax Notes International, 31(4):347–368. Bluhm, R. and Wong, M. H. (2017). Neighborhood disputes? spatial inequalities and civil conflict in africa. Bodea, C. and LeBas, A. (2013). The origins of social contracts: Attitudes towards taxation in urban nigeria.

Boustan, L., Ferreira, F., Winkler, H., and Zolt, E. M. (2013). The effect of rising income inequality on taxation and public expenditures: Evidence from us municipalities and school districts, 1970–2000. Review of Economics and Statistics, 95(4):1291–1302. Calvo, T., Razafindrakoto, M., and Roubaud, F. (2019). Fear of the state in governance surveys? empirical evidence from african countries. World Development, 123:104609. Chen, X. and Nordhaus, W. (2015). A test of the new viirs lights data set: Population and economic output in africa. Remote Sensing, 7(4):4937–4947. Corcoran, S. and Evans, W. N. (2010). Income inequality, the median voter, and the support for public education. Technical report, National Bureau of Economic Research. Cowell, F. A. and Gordon, J. P. (1988). Unwillingness to pay: Tax evasion and public good provision. Journal of Public Economics, 36(3):305–321. Cummings, R. G., Martinez-Vazquez, J., McKee, M., and Torgler, B. (2009). Tax morale affects tax compliance: Evidence from surveys and an artefactual field experiment. Journal of Economic Behavior & Organization, 70(3):447–457. D’Arcy, M. (2011). Why do citizens assent to pay tax? Legitimacy, taxation and the African state. Doll, C. N. (2008). Ciesin thematic guide to night-time light remote sensing and its applications. Center for International Earth Science Information Network of Columbia University, Palisades, NY. Elvidge, C. D., Baugh, K. E., Anderson, S. J., Sutton, P. C., and Ghosh, T. (2012). The night light development index (nldi): a spatially explicit measure of human development from satellite data. Social Geography, 7(1):23–35. Fjeldstad, O.-H., Schulz-Herzenberg, C., and Sjursen, I. H. (2012). People’s views of taxation in africa. for International Earth Science Information Network (CIESIN)-Columbia University, C. (2015). Gridded population of the world, version 4 (gpwv4): Population count. Georgiadis, A. and Manning, A. (2012). Spend it like beckham? inequality and redistribution in the uk, 1983–2004. Public choice, 151(3-4):537–563. Gerstenblüth, M., Melgar, N., Pagano, J. P., and Rossi, M. (2008). Threats in latin american and caribbean countries: how do inequality and the asymmetries of rules affect tax morale? Documento de Trabajo/FCSDE; 14/08. Gibson, J., Olivia, S., Boe-Gibson, G., et al. (2019). A test of dmps and viirs night lights data for estimating gdp and spatial inequality for rural and urban areas. Technical report. Gründler, K. and Köllner, S. (2017). Determinants of governmental redistribution: Income distribution, development levels, and the role of perceptions. Journal of Comparative Economics, 45(4):930–962. Henderson, J. V., Storeygard, A., and Weil, D. N. (2012). Measuring economic growth from outer space. American economic review, 102(2):994–1028. 25

Islam, M. R., Madsen, J. B., and Doucouliagos, H. (2018). Does inequality constrain the power to tax? evidence from the oecd. European Journal of Political Economy, 52:1–17. Kenworthy, L. and McCall, L. (2007). Inequality, public opinion and redistribution. Socio-Economic Review, 6(1):35–68. Kenworthy, L. and McCall, L. (2008). Inequality, public opinion and redistribution. Socio-Economic Review, 6(1):35–68. Kerr, W. R. (2014). Income inequality and social preferences for redistribution and compensation differentials. Journal of Monetary Economics, 66:62–78. Kevins, A., Horn, A., Jensen, C., and Van Kersbergen, K. (2018). Yardsticks of inequality: Preferences for redistribution in advanced democracies. Journal of European Social Policy, page 0958928717753579. Kirchler, E., Hoelzl, E., and Wahl, I. (2008). Enforced versus voluntary tax compliance: The “slippery slope” framework. Journal of Economic Psychology, 29(2):210 – 225. Lutmer, E. F. and Singhal, M. (2014). Tax moral. Journal of Economic Perspectives, 28(4):149–168. Luttmer, E. F. (2001). Group loyalty and the taste for redistribution. Journal of political Economy, 109(3):500–528. McKerchar, M. and Evans, C. (2009). Sustaining growth in developing economies through improved taxpayer compliance: Challenges for policy makers and revenue authorities. eJournal of Tax Research, 7(2):171. Meltzer, A. H. and Richard, S. F. (1981). A rational theory of the size of government. Journal of political Economy, 89(5):914–927. Michalopoulos, S. and Papaioannou, E. (2013). National institutions and subnational development in africa. The Quarterly Journal of Economics, 129(1):151–213. Milanovic, B. (2000). The median-voter hypothesis, income inequality, and income redistribution: an empirical test with the required data. European Journal of Political Economy, 16(3):367–410. Moore, M. (1998). Death without taxes: Democracy, state capacity, and aid dependence in the fourth world. The democratic developmental state: Politics and institutional design, 86. Moore, M. (2004). Revenues, state formation, and the quality of governance in developing countries. International Political Science Review, 25(3):297–319. Newman, B. J., Johnston, C. D., and Lown, P. L. (2015). False consciousness or class awareness? local income inequality, personal economic position, and belief in american meritocracy. American Journal of Political Science, 59(2):326–340. Piketty, T. (1995).

Social mobility and redistributive politics.

The Quarterly journal of economics,

110(3):551–584. Romer, T. (1975). Individual welfare, majority voting, and the properties of a linear income tax. Journal of Public Economics, 4(2):163–185. 26

UNECA (2019). Economic Report on Africa. Fiscal Policy for Financing Sustainable Development in Africa. Technical report, United Nations Economic Commission for Africa (UNECA). Williams, C. C. and Krasniqi, B. (2017). Evaluating the individual-and country-level variations in tax morale: evidence from 35 eurasian countries. Journal of Economic Studies, 44(5):816â&#x20AC;&#x201C;832.

Appendix Table A1 – Summary Statistics

Dependent TAX_ATTITUDEi,z,c (scale) TAXATT_Ai,z,c (0 − 1 score) (3 Criteria) TAXATT_Bi,z,c (0 − 3 scale) (3 Criteria) TAXATT_Ci,z,c (0 − 5 scale) (5 Criteria) Composition of Dependent NOT_PAYING_TAX_(WRONG/RIGHT)z,c PEOPLE_MUST_PAY_TAXz,c CITIZEN_MUST_PAY_TAXz,c PAY_TAX_INCREASE_HEALTHz,c CITIZEN_PAY_TAX_IN_DEMOCz,c Poverty WEALTHi,z,c POVERTY_RELATIVEi,z,c POVERYT_SUBJECT.i,z,c Tax DIFF._KNOW_TAXi,z,c TRUST_TAX_DEP.i,z,c DIFF._AVOID_TAXi,z,c PEOPLE_UNPUNISHEDi,z,c Pub. Goods DIFF._OBTAIN_POLICEi,z,c DIFF._OBTAIN_MEDICi,z,c Institutional TRUST_PRESIDENTi,z,c TRUST_PARLIAMENTi,z,c TRUST_ELECTED_LOCi,z,c TRUST_RULLING_PARTYi,z,c CORRUPTION_TAXi,z,c CORRUPT_COURTSi,z,c CORRUPT_POLICEi,z,c CORRUPT_GOV._OFFi,z,c Individual Characteristics AGEi,z,c EDUCATIONi,z,c EMPLOYTi,z,c ETHNIC_UNFAIRi,z,c N. SUBDIVISIONS N. COUNTRIES

Mean

Std.Dev.

Min

Max

Obs.

0.68 0.69 1.98 3.17

0.20 0.23 0.95 1.34

0 0 0 0

1 1 3 5

37552 37552 37552 37552

2.50 3.82 2.17 2.89 2.68

0.63 1.18 1.17 1.62 0.61

1 1 1 1 1

3 5 5 5 3

37552 37552 37552 37552 37552

0.03 2.10 2.73

0.98 1.02 1.20

-3 0 1

6 4 5

37552 37552 37401

2.56 2.37 2.75 1.79

1.25 1.18 1.35 1.01

0 0 0 0

4 4 4 4

37516 37552 37552 37552

0.60 1.52

1.17 1.36

0 0

4 4

37552 37552

2.68 2.48 2.43 2.43 2.23 2.19 2.51 2.32

1.18 1.12 1.12 1.16 1.08 1.05 1.03 1.00

0 0 0 0 0 0 0 0

4 4 4 4 4 4 4 4

37552 37549 37549 35362 37534 37533 37541 37537

37.08 3.49 0.40 1.54 362.00 30.00

14.29 2.17 0.49 1.00 0.00 0.00

18 0 0 0 362 30

105 9 1 4 362 30

37552 37552 37552 37552 37552 37552

Table A2 â&#x20AC;&#x201C; Summary Statistics: Nightlight Variables

Inequality INEQUALITY20 INEQUALITY30 INEQUALITY40 INEQUALITY50 INEQUALITY60 INEQUALITYADM 1 Lightpc LIGHTpc20 LIGHTpc50 LIGHTpc30 LIGHTpc40 LIGHTpc60 LIGHTpcADM 1

Mean

Std.Dev.

Min

Max

Obs.

0.62 0.64 0.66 0.68 0.69 0.70

0.19 0.18 0.18 0.18 0.18 0.17

0 0 0 0 0 0

1 1 1 1 1 1

37552 37552 37547 37552 37552 37552

0.03 0.05 0.04 0.05 0.03 0.02

0.31 0.63 0.45 0.62 0.26 0.35

-0 -0 -0 -0 -0 -0

9 12 9 12 10 27

37552 37552 37552 37552 37552 37552

0.47 0.51 0.49 0.50 0.53 0.52

0.24 0.23 0.24 0.23 0.23 0.23

0 0 0 0 0 0

1 1 1 1 1 1

37552 37552 37552 37552 37552 37552

-0.08 -0.10 -0.09 -0.09 -0.02 0.02

2.03 2.21 2.08 2.14 0.80 0.35

-40 -44 -41 -43 -16 -0

5 4 5 4 4 27

37552 37552 37552 37552 37552 37552

Instrument Inequality Predicted PREDICTED_INEQ20 PREDICTED_INEQ50 PREDICTED_INEQ30 PREDICTED_INEQ40 PREDICTED_INEQ60 PREDICTED_INEQADM 1 Lightpc Predicted PREDICTED_LIGHTpc20 PREDICTED_LIGHTpc50 PREDICTED_LIGHTpc30 PREDICTED_LIGHTpc40 PREDICTED_LIGHTpc60 PREDICTED_LIGHTpcADM 1

Figure A1 â&#x20AC;&#x201C; Buffer Zone with radius of 50KM

This Figure is only for illustration purposes. The source of data for this picture is DMSP-OLS light data and not VIIRS which we use in order to compute the inequality measures. DMSP-OLS data is only available until 2013, as a result we use VIIRS data which are available for since 2012. The figure shows part of South Africa, and displays how we compute inequality at ADM1 level. The numbers displayed inside each [red] boundary is the corresponding Gini.

Table A3 – Afrobarometer data - Recoding Variablei,z,c

Description

Initial Coding

Final Coding

VAR:

Corruption: Corruption: Corruption: Corruption:

-1: Missing 0: None 1: Some of them 2: Most of them 3: All of them 9: Don’t know

. : Missing 1: Not at all 2: Just a little 3: Somewhat 4: A lot DK dummy (0/1)

VAR:

Trust Trust Trust Trust Trust Trust

-1: Missing 0: Not at all 1: Just a little 2: Somewhat 3: A lot 9: Don’t know

. : Missing 1: Not at all 2: Just a little 3: Somewhat 4: A lot 0: Don’t know DK dummy (0/1)

CORRUPT_VAR tax officials judges gov. officials police

TRUST_VAR tax department key leader parliament elec. local. gov. ruling party courts of law

DIFF._KNOW_TAX

Difficulty to find out what taxes or fees to pay

1: 2: 3: 4: 9:

Very easy Easy Difficult Very difficult Don’t know

1: Very easy 2: Easy 3: Difficult 4: Very difficult 0: Don’t know DK dummy (0/1)

DIFF._AVOID_TAX

Difficulty to avoid paying taxes

1: 2: 3: 4: 7: 9:

Very easy Easy Difficult Very difficult Don’t have to pay taxes Don’t know

1: Very easy 2: Easy 3: Difficult 4: Very difficult 0: Don’t pay taxes 0: Don’t know TAX_PAYER (0/1) DK dummy (0/1)

DIFF._OBTAIN_MEDIC

Difficulty to obtain medical treatment

0: 1: 2: 3: 4:

No Contact Very Difficult Difficult Easy Very Easy

0: No Contact 4: Very Difficult 3: Diffficult 2: Easy 1: Very Easy ND_MEDIC (0/1)

Table A4 – Afrobarometer data - Recoding (continued) Variablei,s,c,t

Description

Initial Coding

Final Coding

DIFF._OBTAIN_POLICE

Difficulty to obtain help from the police

0: 1: 2: 3: 4:

No Contact Very Difficult Difficult Easy Very Easy

0: No Contact 4: Very Difficult 3: Diffficult 2: Easy 1: Very Easy ND_COP (0/1)

PEOPLE_UNPUNISHED

How often ordinary people unpunished

0: 1: 2: 3: 9:

Never Rarely Often Always Don’t know

1: Never 2: Rarely 3: Often 4: Always 0: Don’t know DK dummy (0/1)

INEQ_PERCEPTION

Your living conditions vs. Others

1: Much Worse 2: Worse 3: Same 4: Better 5: Much Better . : Missing

4: Much Worse 3: Worse 2: Same 1: Better 0: Much Better . : Missing

ETHNIC_UNFAIR

Ethnic unfair

0: Not Applicable 1: Never 2: Sometimes 3: Often 4: Always . : Missing

0: Not Applicable 1: Never 2: Sometimes 3: Often 4: Always . : Missing ETHNIC_APPLIC (0/1)

Table A5 – Different Compositions of Dependent Variable: OLS & IV 30km

Dependent:

Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters Subdivisions Countries

Note: Standard errors in parentheses

∗

(1) TAXATTi,z,c

(2) TAXATT_Ai,z,c

(3) TAXATT_Bi,z,c

(4) TAXATT_Ci,z,c

Baseline

3 Criteria (Score 0-1)

3 Criteria (Score 0-3)

5 Criteria (Score 0-5)

0.031∗∗∗ (0.012)

0.034∗∗ (0.013)

0.145∗∗∗ (0.054)

0.214∗∗∗ (0.079)

0.002 (0.002)

0.001 (0.002)

0.005 (0.008)

0.007 (0.011)

0.151

0.149

0.135

0.145

0.067∗ (0.036)

0.088∗∗ (0.041)

0.325∗ (0.170)

0.483∗ (0.247)

0.004∗∗∗ (0.001)

0.018∗∗∗ (0.006)

0.025∗∗∗ (0.009)

0.074

0.066

0.059

0.068

0.266∗∗∗ (0.024)

0.004∗∗∗ (0.000)

0.00 61

Yes Yes Yes ZoneID

37552 4370 362 30

p < 0.10,

∗∗

p < 0.05,

∗∗∗

p < 0.01

Table A6 â&#x20AC;&#x201C; Interaction with Democratic Condition (OLS & IV) 30km

(1) Extend Democ.

(2) Satisf. Democ.

0.003 (0.018)

0.018 (0.020)

INEQUALITYz X VARi,z,c

0.012** (0.006)

0.004 (0.005)

VARi,z,c

0.001 (0.004)

0.007** (0.004)

0.004

0.025

0.153

0.005 (0.044)

-0.003 (0.046)

0.027*** (0.010)

0.023*** (0.009)

-0.008 (0.006)

-0.005 (0.006)

0.003

0.075

0.007 (0.005)

0.005 (0.004)

0.249*** (0.029)

0.251*** (0.029)

0.00 39

0.00 40

Yes Yes Yes ZoneID

37549 4370 362 30

37547 4370 362 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz

Joint Sig. P-value R

Panel B: IV - Second Stage INEQUALITYz INEQUALITYz X VARi,z,c VARi,z,c Joint Sig. P-value 2

R Panel C: IV - First Stage PREDICTED_INEQz X VARi,z,c PREDICTED_INEQz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters Subdivisions Countries

Note: Standard errors in parentheses * p < 0.10, ** p < 0.05, 34 *** p < 0.01

Table A7 â&#x20AC;&#x201C; Inequality and Attitudes towards Tax Compliance: Interaction with Distance to Highest Light in the Zone (1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

0.004 (0.027)

0.012 (0.030)

0.010 (0.032)

-0.009 (0.032)

-0.020 (0.033)

INEQUALITYz X LNDIST_TO_LIGHTz

0.013 (0.011)

0.008 (0.011)

0.005 (0.010)

0.008 (0.009)

0.011 (0.009)

LNDIST_TO_LIGHTz

-0.007 (0.007)

-0.004 (0.007)

-0.008 (0.007)

-0.011* (0.007)

Joint Sig. P-value

0.007

0.021

0.101

0.321

0.206

R2 Panel B: IV - Second Stage INEQUALITYz

0.150

-0.040 (0.048)

0.021 (0.067)

0.019 (0.071)

-0.010 (0.080)

-0.010 (0.078)

INEQUALITYz X LNDIST_TO_LIGHTz

0.040** (0.018)

0.027 (0.019)

0.031* (0.018)

0.037** (0.018)

0.042*** (0.016)

LNDIST_TO_LIGHTz

-0.022** (0.011)

-0.016 (0.012)

-0.020* (0.012)

-0.026** (0.012)

-0.031*** (0.011)

Joint Sig. P-value

0.045

0.037

0.028

0.027

0.011

R2 Panel C: IV - First Stage PREDICTED_INEQz

0.073

0.072

0.071

0.070

0.069

0.332*** (0.043)

0.215*** (0.051)

0.215*** (0.040)

0.199*** (0.043)

0.207*** (0.041)

PREDICTED_INEQz X LNDIST_TO_LIGHTz

-0.025 (0.015)

0.007 (0.015)

0.003 (0.014)

-0.007 (0.012)

LNDIST_TO_LIGHTz

0.010 (0.007)

0.004 (0.007)

0.002 (0.007)

0.001 (0.007)

0.005 (0.007)

0.00 36

0.00 21

0.00 23

Yes Yes Yes ZoneID

36908 4283 362 30

36903 4282 362 30

36908 4283 362 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes: * p < 0.10, ** p < 0.05, *** p < 0.01

Supplementary Appendix Spatial Inequality and Attitudes Towards Taxation: The Case of Sub-Saharan Africa Table S.A1 – Stage Zero

(1) Dependent: Pixel Light LNDISTANCE_p,pl,2000

-0.573∗∗∗ (0.001)

LNDISTANCE_p,pl,2000 × LNDISTANCE_p,pl,2000

0.105∗∗∗ (0.000)

LNPOP_p,pl,2000

0.298∗∗∗ (0.001)

0.29

Country FE ADM1 FE

Yes Yes

N. Obs. N. Countries N. ADM1

18334032 604 36

Note: Standard errors in parentheses

∗

p < 0.10,

∗∗

p < 0.05,

∗∗∗

p < 0.01

Table S.A2 – Using ADM1 as cluster for all regressions

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.034∗∗∗ (0.013)

0.031∗∗ (0.015)

0.025 (0.017)

0.015 (0.017)

0.016 (0.019)

-0.005 (0.030)

0.003 (0.004)

0.002 (0.003)

0.001 (0.002)

0.002 (0.005)

-0.003 (0.002)

0.151

0.056 (0.046)

0.099∗ (0.059)

0.114∗ (0.066)

0.116 (0.079)

0.135 (0.093)

-0.077 (0.114)

0.004 (0.004)

0.002 (0.003)

0.001 (0.002)

0.005 (0.006)

-0.003 (0.002)

0.074

0.072

0.071

0.070

0.073

0.276∗∗∗ (0.028)

0.232∗∗∗ (0.035)

0.222∗∗∗ (0.036)

0.207∗∗∗ (0.036)

0.184∗∗∗ (0.036)

0.194∗∗∗ (0.053)

0.000 (0.001)

0.001 (0.001)

0.000 (0.001)

0.001 (0.001)

0.007∗∗∗ (0.002)

-0.004 (0.005)

0.00 48

0.00 22

0.00 19

0.00 16

0.00 14

0.00 7

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

37552 4370 362 30

37547 4369 362 30

37552 4370 362 30

37552 362 362 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Note: Robust Standard errors in parentheses

∗

p < 0.10,

∗∗

p < 0.05,

∗∗∗

p < 0.01

Table S.A3 – Inequality and Attitudes towards Tax Compliance: Removing Obs. Close to Border (10km)

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes:

∗

p < 0.10,

∗∗

p < 0.05,

∗∗∗

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.034∗∗∗ (0.013)

0.035∗∗∗ (0.013)

0.028∗∗ (0.014)

0.021 (0.014)

0.018 (0.015)

-0.001 (0.033)

0.000 (0.003)

0.000 (0.002)

0.000 (0.001)

0.001 (0.003)

-0.004∗ (0.002)

0.149

0.148

0.020 (0.038)

0.053 (0.044)

0.090∗ (0.050)

0.112∗∗ (0.056)

0.126∗∗ (0.063)

0.007 (0.112)

0.001 (0.003)

0.001 (0.002)

0.000 (0.001)

0.004 (0.004)

-0.004∗ (0.002)

0.070

0.069

0.067

0.066

0.070

0.266∗∗∗ (0.022)

0.260∗∗∗ (0.028)

0.252∗∗∗ (0.026)

0.246∗∗∗ (0.025)

0.227∗∗∗ (0.023)

0.205∗∗∗ (0.055)

0.001 (0.001)

0.002∗∗∗ (0.001)

0.001 (0.001)

0.001∗∗ (0.001)

0.008∗∗∗ (0.002)

-0.003 (0.005)

0.00 71

0.00 43

0.00 45

0.00 49

0.00 48

0.01 7

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

29105 3416 339 30

29100 3415 339 30

29105 3416 339 30

29105 339 339 30

p < 0.01

Table S.A4 – Inequality and Attitudes towards Tax Compliance: Only Obs. Close to Border (10km)

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes:

∗

p < 0.10,

∗∗

p < 0.05,

∗∗∗

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.050∗ (0.028)

0.044 (0.030)

0.033 (0.032)

0.021 (0.032)

0.032 (0.035)

-0.012 (0.040)

0.067 (0.230)

0.012 (0.203)

0.074 (0.223)

0.009 (0.226)

-0.106 (0.243)

-0.066 (0.112)

0.192

0.134∗ (0.077)

0.242 (0.154)

0.153 (0.188)

-0.019 (0.233)

0.046 (0.248)

-0.383∗ (0.200)

-0.179 (0.253)

-0.130 (0.253)

-0.025 (0.749)

-0.231 (0.728)

-0.179 (0.675)

-0.057 (0.192)

0.088

0.081

0.087

0.089

0.090

0.058

0.271∗∗∗ (0.031)

0.145∗∗∗ (0.027)

0.123∗∗∗ (0.030)

0.105∗∗∗ (0.031)

0.100∗∗∗ (0.031)

0.190∗∗ (0.076)

-0.295∗∗ (0.124)

-0.473∗∗∗ (0.153)

-0.377 (0.434)

-0.526 (0.497)

-0.382 (0.567)

0.007 (0.308)

0.00 37

0.00 15

0.00 9

0.00 7

0.00 6

0.03 3

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

8447 954 193 30

8447 193 193 30

p < 0.01

Table S.A5 – Inequality and Attitudes towards Tax Compliance: Removing Obs. Close to Border (20km)

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes:

∗

p < 0.10,

∗∗

p < 0.05,

∗∗∗

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.029∗∗ (0.014)

0.037∗∗∗ (0.014)

0.029∗∗ (0.015)

0.025∗ (0.015)

0.018 (0.016)

0.011 (0.036)

0.001 (0.003)

0.001 (0.002)

0.001 (0.001)

0.002 (0.003)

-0.002 (0.002)

0.143

0.010 (0.042)

0.052 (0.047)

0.110∗∗ (0.056)

0.126∗∗ (0.063)

0.117∗ (0.067)

0.082 (0.116)

0.003 (0.003)

0.001 (0.002)

0.001 (0.001)

0.005 (0.004)

-0.002 (0.003)

0.067

0.064

0.063

0.064

0.065

0.261∗∗∗ (0.020)

0.262∗∗∗ (0.024)

0.246∗∗∗ (0.025)

0.237∗∗∗ (0.025)

0.232∗∗∗ (0.025)

0.199∗∗∗ (0.059)

0.001 (0.001)

0.002∗∗∗ (0.001)

0.001 (0.001)

0.001∗ (0.001)

0.008∗∗∗ (0.002)

-0.005 (0.006)

0.00 88

0.00 58

0.00 50

0.00 45

0.00 44

0.01 6

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

25377 2976 322 28

25377 322 322 28

p < 0.01

Table S.A6 – Inequality and Attitudes towards Tax Compliance: Only Obs. Close to Border (20km)

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.060∗∗∗ (0.021)

0.042∗ (0.024)

0.038 (0.026)

0.016 (0.026)

0.030 (0.028)

-0.026 (0.036)

0.087∗∗ (0.041)

0.084∗∗ (0.034)

0.077∗∗∗ (0.025)

0.064∗∗∗ (0.024)

0.093 (0.089)

-0.048 (0.096)

0.184

0.183

0.182

0.183

0.184

0.108∗ (0.055)

0.168 (0.107)

0.084 (0.116)

0.019 (0.132)

0.063 (0.149)

-0.387∗∗ (0.181)

0.065 (0.062)

0.076 (0.056)

0.067 (0.045)

0.047 (0.048)

0.036 (0.154)

-0.170 (0.158)

0.093

0.088

0.092

0.063

0.291∗∗∗ (0.032)

0.164∗∗∗ (0.028)

0.160∗∗∗ (0.029)

0.154∗∗∗ (0.029)

0.129∗∗∗ (0.028)

0.203∗∗∗ (0.068)

-0.076 (0.056)

-0.170∗∗∗ (0.050)

-0.168∗∗∗ (0.042)

-0.186∗∗∗ (0.047)

-0.276∗∗∗ (0.097)

-0.228 (0.249)

0.00 41

0.00 18

0.00 16

0.00 15

0.00 12

0.01 4

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

12175 1394 240 30

12170 1393 240 30

12175 1394 240 30

12175 240 240 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes:

∗

p < 0.10,

∗∗

p < 0.05,

∗∗∗

p < 0.01

Table S.A7 – Inequality and Attitudes towards Tax Compliance: Removing Obs. Close to Border (30km)

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes:

∗

p < 0.10,

∗∗

p < 0.05,

∗∗∗

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.035∗∗ (0.014)

0.042∗∗∗ (0.015)

0.036∗∗ (0.015)

0.036∗∗ (0.016)

0.025 (0.016)

0.032 (0.037)

0.003 (0.003)

0.002 (0.002)

0.002 (0.001)

0.004 (0.003)

-0.002 (0.002)

0.141

0.140

0.006 (0.041)

0.051 (0.045)

0.109∗∗ (0.055)

0.157∗∗ (0.064)

0.153∗∗ (0.065)

0.107 (0.105)

0.004 (0.003)

0.002 (0.002)

0.001 (0.001)

0.002 (0.001)

0.008∗ (0.004)

-0.002 (0.003)

0.066

0.064

0.061

0.060

0.064

0.274∗∗∗ (0.020)

0.287∗∗∗ (0.023)

0.262∗∗∗ (0.025)

0.248∗∗∗ (0.026)

0.252∗∗∗ (0.027)

0.210∗∗∗ (0.058)

0.001 (0.001)

0.002∗∗∗ (0.001)

0.001 (0.001)

0.001∗ (0.001)

0.008∗∗∗ (0.002)

-0.005 (0.006)

0.00 94

0.00 77

0.00 57

0.00 46

0.00 45

0.01 7

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

22740 2666 307 28

22740 307 307 28

p < 0.01

Table S.A8 â&#x20AC;&#x201C; Inequality and Attitudes towards Tax Compliance: Only Obs. Close to Border (30km)

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz LIGHTpcz R2 Panel B: IV - Second Stage INEQUALITYz LIGHTpcz R2 Panel C: IV - First Stage PREDICTED_INEQz PREDICTED_LIGHTpcz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

(6) ADM1

0.048*** (0.018)

0.039* (0.022)

0.031 (0.024)

0.001 (0.023)

0.017 (0.025)

-0.057 (0.037)

0.067 (0.045)

0.067* (0.037)

0.063** (0.027)

0.049* (0.027)

0.042 (0.104)

-0.057 (0.099)

0.179

0.178

0.179

0.090* (0.049)

0.150 (0.093)

0.083 (0.095)

0.024 (0.111)

0.070 (0.133)

-0.328* (0.173)

0.054 (0.060)

0.071 (0.054)

0.061 (0.044)

0.041 (0.047)

0.024 (0.149)

-0.149 (0.145)

0.090

0.087

0.090

0.075

0.306*** (0.028)

0.169*** (0.026)

0.174*** (0.025)

0.162*** (0.026)

0.130*** (0.025)

0.187*** (0.064)

-0.091 (0.058)

-0.199*** (0.060)

-0.185*** (0.046)

-0.218*** (0.054)

-0.355*** (0.111)

-0.223 (0.256)

0.00 60

0.00 22

0.00 25

0.00 20

0.00 15

0.01 4

Yes Yes Yes ZoneID

Yes Yes Yes ADM1

14812 1704 255 30

14807 1703 255 30

14812 1704 255 30

14812 255 255 30

Notes: * p < 0.10, ** p < 0.05, *** p < 0.01

Table S.A9 â&#x20AC;&#x201C; Inequality and Attitudes towards Tax Compliance: Interaction with Economic Distance to Highest Light in the Zone (1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

0.031** (0.015)

0.029* (0.017)

0.016 (0.019)

0.018 (0.021)

0.025 (0.023)

INEQUALITYz X ECON_DIST_LIGHTz

0.000 (0.000)

-0.000 (0.000)

0.000 (0.000)

-0.000 (0.000)

ECON_DIST_LIGHTz

0.000 (0.000)

Joint Sig. P-value

0.026

0.093

0.357

0.702

0.529

R2 Panel B: IV - Second Stage INEQUALITYz

0.149

0.148

0.149

0.018 (0.042)

0.054 (0.053)

0.048 (0.055)

0.063 (0.061)

0.075 (0.065)

INEQUALITYz X ECON_DIST_LIGHTz

0.001** (0.001)

0.001 (0.001)

ECON_DIST_LIGHTz

-0.001** (0.000)

-0.001* (0.000)

-0.001** (0.000)

-0.001 (0.000)

-0.000 (0.000)

Joint Sig. P-value

0.032

0.043

0.029

0.168

0.207

R2 Panel C: IV - First Stage PREDICTED_INEQz

0.073

0.071

0.070

0.246*** (0.030)

0.231*** (0.038)

0.251*** (0.037)

0.250*** (0.030)

0.246*** (0.031)

-0.000 (0.001)

-0.001* (0.001)

-0.001*** (0.000)

0.001*** (0.000)

0.002*** (0.000)

0.001*** (0.000)

0.00 27

0.00 16

0.00 17

0.00 23

Yes Yes Yes ZoneID

33238 3864 355 30

33233 3863 355 30

33238 3864 355 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz

PREDICTED_INEQz X ECON_DIST_LIGHTz ECON_DIST_LIGHTz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes: * p < 0.10, ** p < 0.05, *** p < 0.01

Table S.A10 â&#x20AC;&#x201C; Inequality and Attitudes towards Tax Compliance: Weighted Economic Distance to Highest Light in the Zone

Interaction with

(1) 20km

(2) 30km

(3) 40km

(4) 50km

(5) 60km

0.038*** (0.013)

0.033** (0.014)

0.025* (0.015)

0.017 (0.015)

0.018 (0.015)

INEQUALITYz X WE_ECON_DIST_LIGHTz

-0.001 (0.002)

-0.002 (0.002)

WE_ECON_DIST_LIGHTz

0.001 (0.001)

0.001 (0.002)

0.002 (0.002)

Joint Sig. P-value

0.015

0.056

0.233

0.403

0.391

R2 Panel B: IV - Second Stage INEQUALITYz

0.150

0.149

0.064 (0.042)

0.106** (0.054)

0.118** (0.059)

0.121* (0.064)

0.137** (0.069)

INEQUALITYz X WE_ECON_DIST_LIGHTz

-0.001 (0.003)

-0.002 (0.003)

-0.004 (0.003)

-0.004 (0.004)

-0.005 (0.004)

WE_ECON_DIST_LIGHTz

0.000 (0.002)

0.001 (0.002)

0.003 (0.002)

0.003 (0.003)

0.004 (0.003)

Joint Sig. P-value

0.314

0.131

0.100

0.112

0.074

R2 Panel C: IV - First Stage PREDICTED_INEQz

0.073

0.072

0.070

0.229*** (0.029)

0.207*** (0.034)

0.204*** (0.032)

0.200*** (0.029)

0.188*** (0.025)

PREDICTED_INEQz X WE_ECON_DIST_LIGHTz

0.001 (0.002)

-0.000 (0.003)

0.001 (0.003)

0.000 (0.003)

-0.000 (0.003)

WE_ECON_DIST_LIGHTz

0.003** (0.001)

0.002 (0.001)

0.001 (0.001)

0.001 (0.002)

0.00 25

0.00 15

0.00 17

0.00 19

0.00 24

Yes Yes Yes ZoneID

32163 3745 354 30

32551 3782 355 30

32715 3802 355 30

32913 3823 355 30

32960 3829 355 30

Dependent: TAXATTi,z,c Panel A: OLS INEQUALITYz

Kleibergen-Paap LM stat (p-value) Kleibergen-Paap F-stat Country FE Urabn/Rural FE All Controls Cluster Level N. N. N. N.

Obs. Clusters ADM1 Countries

Notes: * p < 0.10, ** p < 0.05, *** p < 0.01