Tax Morale, Aversion to Ethnic Diversity, and Decentralization∗ Alessandro Belmonte † Roberto Dell’Anno D´ esir´ ee Teobaldelli §

‡

Abstract This paper analyzes theoretically and empirically the relationship between individuals’ aversion to ethnic diversity, the degree of fiscal and political decentralization, and tax morale. We present a model showing how higher degrees of individuals’ aversion to ethnic diversity may reduce tax morale and why this effects may be smaller in decentralized political and fiscal systems. We test these results by using individual data from the World Value Survey and several measures of decentralization. Our estimates robustly confirm that higher degrees of individuals’ aversion to ethnic diversity are associated to lower tax morale and that this correlation is smaller or null in decentralized systems.

Keywords: Tax morale, aversion to ethnic diversity, decentralization, ethnic fragmentation. JEL Classification: J15, H26, H73

∗

We are grateful to Davide Ticchi for valuable discussions. We also thank two anonymous referees, the participants at the XXVIII meeting of the Societ`a Italiana di Economia Pubblica and at the 2017 meeting of the European Public Choice Society for useful comments and suggestions. † IMT School for Advanced Studies Lucca, e-mail: [email protected] ‡ University of Salerno, e-mail: [email protected] § University of Urbino, e-mail: [email protected]

1

1

Introduction

There is a large theoretical and empirical literature showing that investments in state fiscal capacity, namely economic institutions for tax compliance, are a key feature of economic development (see Besley and Persson, 2013, for a review). At the same time, a widely-accepted literature remarks that tax compliance cannot be fully explained by the level of enforcement or the tax rates and, after a long period in which non-pecuniary motivations have been neglected (Andreoni et al., 1998), tax morale has turned into a key issue among the determinants of tax compliance (Torgler, 2007). Tax morale is usually defined as a moral obligation or an intrinsic motivation to pay taxes, and it can be considered as “an umbrella term capturing non-pecuniary motivations for tax compliance as well as factors that fall outside the standard, expected utility framework” (Luttmer and Singhal, 2014, p. 150).1 A variety of aspects are expected to influence tax morale and the literature has indeed provided evidence on different channels and on the existence of substantial cross-country variation. A strand of literature has emphasized the individual’s relationship with the state, and, specifically, the perception individuals form about the government or the fairness of the tax schedule (e.g. Feld and Frey, 2002; Hofmann et al. 2008; Besley et al. 2015). Other works have shown how culture affects tax compliance in a given country— thus controlling for the same institutional context and environment (e.g. Halla, 2012; DeBacker et al., 2015; Kountouris and Remoundou, 2013). Similarly, Torgler (2006) and Lago-Pe˜ nas and Lago-Pe˜ nas (2010) have estimated positive correlations between beliefs, such as national pride or religiosity, and the country’s level of tax morale. On the other hand, several historical examples bring support in favor of alternative institutional and demographic mechanisms, such as fiscal and political decentralization (Torgler, 2005; G¨ uth et al., 2005; Torgler et al., 2010) and the degree of ethnic diversity of the country population (see Li, 2010, for a comprehensive overview). For example, the reforms implemented over the last three decades of the last century in Belgium have established a unique form of a federal state composed by three regions with segregated political power; along with the Brussels-Capital Region, in which both the French and the Flemish community have historically lived side by side, the Flemish and the Walloon Regions have been established around the two main linguistic communities(see Witte et al., 2009, for an historical excursus of the recent Belgian events). Similar examples in Germany, Switzerland, Canada, or Spain, or in developing countries (e.g., Brazil, India, Indonesia, Bosnia-Herzegovina, Nigeria) show how various forms of federalism 1

See also Alm and Torgler (2006) and Dell’Anno (2009) for similar definitions.

2

and decentralization have been used to prevent ethnic conflicts.2 Our work contributes to this literature by focusing on the individuals’ attitude toward other ethnic groups and its interaction with institutional arrangements as main drivers of tax compliance. Specifically, we analyze, theoretically and empirically, the impact of individuals’ aversion to ethnic diversity on intrinsic motivations to comply with taxes, and test whether and to what extent a decentralized political and fiscal system affects this relationship. We find evidence that a negative attitude toward ethnic diversity reduces tax morale in centralized political systems, while it does not have a statistical significant impact in decentralized states. We also obtain that the negative effect of aversion to ethnic diversity on tax morale is lower in ethnically fragmented communities than in homogeneous countries. Our analysis relies on the idea that individuals more averted to ethnic diversity are also more reluctant to contribute to the provision of public goods that benefits other ethnic groups and, therefore, display a lower tax morale. In decentralized systems, local autonomy ensures that taxes collected in one jurisdiction are spent primarily for the provision of that jurisdiction’s public goods. As single jurisdictions are generally more ethnically homogeneous than the whole country, citizens feel more attached to their community, and their aversion towards other ethnic groups is therefore likely to have a lower impact on their degree of tax morale. We propose a simple framework where individuals are asked to pay taxes to provide public goods that benefit the entire community. All agents have the same income as we do not address the extrinsic motivational components of tax morale.3 We assume that agents incur in a non-pecuniary extra cost when their wealth is used to subsidize the provision of a public good that benefits also members of other ethnic groups. This psychological cost is what we interpret as the loss aversion to ethnic diversity and captures the idea that the taxpayers’ intrinsic motivation to comply with the law is related to the diversity in the ethnic composition of their community. We consider the case where local communities are perfectly homogeneous and, therefore, the risk of contributing to an ethnically fragmented community can be fully insured by implementing a decentralization reform that transfers the powers of taxation and expenditure to local authorities. In a decentralized country, with no within-region ethnic 2

Lijphart (1977) argues that federalism is one of the most important types ofpower-sharing arrangements for divided societies. See on this also Tranchant (2007). 3 We also neglect other sources capable of increasing reciprocity between tax payers and the tax administrator such as the quality of the public good, the efficiency in managing its provision, or the perceptions about the fairness of the tax system.

3

fragmentation, psychological costs are absent and the individuals’ utility loss is equal to the tax paid to finance the public goods provision. The perceived risk is then fully covered after paying the cost of establishing a decentralized system—which we interpret as the risk premium—and it tells us how much the individuals are willing to pay for not contributing to the unfeeling social community; this is what we define the ethnic diversity aversion component of the tax cheating. Our model leads to the conclusion that when taxpayers are loss averted toward ethnic diversity and the ethnic composition follows territorial boundaries, a decentralized fiscal system boils down the negative impact of risk aversion. On the contrary, when ethnic fragmentation is high, but the between regions variation is negligible, decentralization is an ineffective and costly reform. This cost increases with ethnic fragmentation making tax morale relatively higher. We test these predictions using microdata from the World Value Survey (WVS) that collects respondents’ ethnic diversity aversion and their propensity to cheat on taxes as well as their individual characteristics. We combine this dataset with several measures of fiscal decentralization from Fan et al. (2009) in order to decompose the variation in tax morale at individual-, religions-within-country- and country-level using a linear mixed model. This strategy allows us to estimate the impact of country variant variables, such as decentralization, GDP per capita, and other sources of fragmentation, whilst providing robust within groups estimates within religions-within-country and withincountry estimations that cut down eventual sources of omitted variable bias. Our most preferred estimation reports a decreasing of about 0.03 in tax morale in response of a one-scale increasing in ethnic aversion in centralized countries. Consistently with our theory, we find no significant effect of ethnic aversion on tax morale in decentralized countries—even controlling for several sources of diversity, such as language, religion and ethnic diversity. This result is robust to several alternative specifications, to supplementary decentralization measures, to the inclusion of a wide set of controls, and to testing for alternative, competitive mechanisms. We also tackle potential endogeneity issues going from tax morale to ethnic aversion and to decentralization using geographical area to instrument decentralization and several other instrumental variables to select exogenous variation in ethnic aversion. In both cases, we find no substantial change of our baseline estimation. Finally, we explore possible heterogeneous effects of ethnic fragmentation on the relation between tax morale and ethnic aversion and find that, as ethnic fractionalization increases, the negative effect of ethnic aversion on tax morale in centralized countries weakens; the estimated effect of ethnic aversion on tax morale is instead equal to zero for any degree of ethnic fractionalization in decentralized countries, 4

as in the baseline estimation. This paper contributes to the recently growing literature on the effects of the joint interaction between culture and institutions on economic outcomes. Ticchi et al. (2013) emphasize the effects of the two-way complementarities between political regimes and political culture on regimes transitions and democratic consolidation, while Gorodnichenko and Roland (2017) provide evidence consistent with two-way causality between the individualism-collectivism dimension of culture and long-run growth (see also Alesina and Giuliano, 2015, for a review of the literature between culture and institutions). We contribute to this strand of the literature by showing that a country’s institutional feature, here represented by the degree of fiscal decentralization, affects the impact of the individuals’ attitude toward ethnic diversity on the intrinsic motivation to pay taxes. Our work complements the studies analyzing the impact of ethnic diversity on tax morale which point to a negative correlation between ethnic diversity and tax compliance. Li (2010) motivates this result as a consequence of intergroup discrimination, i.e. people favoring policies that offer beneficial treatment to their own ethnic communities and withdraw support for other groups. Conversely, Tusicisny (2014) obtains that trust in government moderates the negative correlation between ethnic fractionalization and tax morale, especially among ethnic minorities.4 All these works are related to the strand of the literature suggesting the detrimental impact of ethnic fractionalization on public sector performance and economic outcomes (see Alesina and La Ferrara, 2005, for a review). For example, Alesina et al. (1999) provide evidence that the share of spending on productive public goods in U.S. urban areas is inversely related to the ethnic fragmentation of the communities. Luttmer (2001) finds that people increase their support for welfare spending as the share of local recipients from their own racial group rises. Alesina et al. (2001) show an inverse relationship between the size of government redistributive spending and the country’s ethnic fragmentation.5 With respect to the fields of research just described, our analysis innovates by pointing out that the country’s degree of ethnic fractionalization does not necessarily affect 4

Alesina et al. (2003) find a negative, but not statistically significant relationship between ethnic fractionalization and tax compliance. Lago-Pe˜ nas and Lago-Pe˜ nas (2010) corroborates this finding, showing that ethnic-linguistic fractionalization is significantly negatively associated with tax morale in European countries. 5 Alesina and La Ferrara (2005) refer to Tajfel et al.’s (1971) Social Identity Theory to explain which psychological mechanisms may link ethnic heterogeneity to the economic choices. According to this approach, once people have categorized themselves as part of a group, they then tend to compare their group with other groups. Consequently, individuals may attribute positive utility to the well-being of members of their own group, and negative utility to that of members of other social groups.

5

negatively the individuals’ intrinsic motivation to pay taxes. We argue that it is not ethnic diversity per se that matters in explaining the individuals’ tax morale; rather, it is the individual’s attitude towards ethnic diversity that matters. Moreover, our work contributes to the strand of the literature that recognizes federalism or fiscal decentralization as a useful mechanism to improve the efficiency of fiscal policies and public goods provision in highly ethnically fragmented societies (e.g., Lijphart, 1977) by clarifying that, at least for what concerns the individuals’ attitude to comply with taxes, decentralization is optimal only when individuals’ ethnic aversion is high. In other words, our theory makes clear that also in societies ethnically fragmented, there is no need to decentralize and bear the correspondent costs (unless there are other reasons for such a choice) when the individuals’ attitude towards ethnic diversity is not negative. Our results are also in line with the empirical literature that highlights how decentralized fiscal systems are characterized by a higher tax morale. For instance, Torgler et al. (2010), using Swiss data, provide evidence that there is a strong and positive correlation between local autonomy, direct democracy and tax morale, whereas Torgler and Werner (2005) find evidence that a higher fiscal autonomy leads to a higher tax compliance in Germany. Martinez-Vazquez and Torgler (2009) show that intrinsic motivation to pay taxes rose in Spain as a consequence of the trend towards an enhanced fiscal federalism. Other studies focus instead on interregional aspects of tax morale. Torgler and Schneider (2007) find that tax morale exhibits regional differences in Switzerland and Spain, whereas Lago-Pe˜ nas and Lago-Pe˜ nas (2010) provide evidence for European countries that tax morale is lower in rich federal regions than rich regions in unitary states, and attribute this result to the higher visibility of interregional transfers in decentralized countries. G¨ uth et al. (2005) contribute to this strand of the literature by providing experimental evidence that tax morale is higher in a decentralized tax system.6 Our paper contributes to this strand of the literature by emphasizing that fiscal and political decentralization may also have an indirect effect on tax morale; specifically, in the case we analyze, the degree of decentralization affects the relationship between individual’s attitude towards ethnic diversity and tax morale. The paper is organized as follows. Section 2 presents a simple model and discusses its comparative statics. Section 3 describes the data employed, the empirical methodology, and the estimation results. It also discusses and addresses potential endogeneity 6

All these findings follow the view that more extensive possibilities for direct political participation lead to higher intrinsic motivation to pay taxes, since taxpayers feel more obliged to be honest when they participate in decision-making through a system of direct democracy (see on this point, Pommerehne and Weck-Hannemann, 1996; Torgler, 2005; Schwarz, 2011).

6

issues. Section 4 analyzes the direct impact of ethnic fractionalization on the relationship between tax morale and ethnic aversion. Section 5 concludes with a summary and a discussion of the main results. Various robustness checks of the empirical analysis are reported in the Online Appendix.

2

Theoretical framework

In this section we highlight the main mechanisms of our empirical analysis presenting a simple theoretical framework linking tax compliance, decentralization, and ethnic fragmentation where citizens are averted to ethnic diversity. We focus, for simplicity, on two regions, A and B, populated by citizens that have same preferences on wealth. All agents have the same income (normalized to 1) and pay taxes τ to provide public goods that will benefit the entire community and remunerate the tax administrator. The provision is efficient and the tax administrator is not selfish. The two regions have same size but they can be ethnically different, i.e. populated by different ethnic groups—none of them excludable from the provision of the public good. Let us denote as φb ∈ [0, 1] the rate of fractionalization between the two regions and as φw ∈ [0, 1] the rate of fractionalization within each region.7 When φb = 0 the ethnic composition of the population in region A is exactly the same as in the region B. The higher is φb the more diverse A and B are in terms of ethnicity. Likewise, when φw = 0, the population in each region is ethnically homogeneous, and when φw = 1 each citizen belongs to a different ethnic group.8 The key assumption of the model is that agents incur in a non-pecuniary extra cost when his or her wealth is used to subsidize a good that also benefits a non-member of his or her ethnic group. This cost is equal to a fraction λ of taxes τ and defines the extent to which agents are loss averted to ethnic diversity (see on loss aversion Koszegi and Rabin, 2006, 2009). When the subsidy goes toward members of their own group no additional costs are perceived. While living in a centralized state, people compare their welfare under an alternative decentralized system. The establishment of this system has an intuitive benefit: it prevents tax payers from region A to contribute to the welfare of those populating region B, and vice versa. So, when contribution across ethnic groups is particularly conflicting, because of a high degree of ethnic aversion λ, decentralization can be used 7

Since we are not interested in non-symmetric equilibria, we also assume that the composition within w w each region is the same in A and B, that is φw A = φB . We therefore drop the subscript in φ . 8 w b On the other end, if φ = 0 and φ = 0, then all the citizens come from the same ethnic group.

7

to cool down regional fiscal disputes. However, the literature has spelled out several drawbacks from establishing a fiscal decentralized system. We distinguish between two types of economic costs. A fixed cost, π, which is related to the duplication of bureaucracy and political decentralized institutions, or to a potential reduction in the economies of scale (compared to a centralized system), or, yet, to externalities of public goods produced at local levels that are not internalized.9 And a variable cost, φw π, which moves upward with the rate of fractionalization within each region. The idea is that though decentralization allows welfare gains related to the possibility of tailoring the provision of public goods and services to the particular preferences of local constituencies (see Bolton and Roland, 1997; Oates 1999), these gains must be lower the higher is φw , that is the more heterogeneous the regions are within their borders. The net cost of decentralization is therefore given by Ω(π, φw ) = (1 + φw )π.

(1)

The timing of actions is as follow: 1. Tax payers receive income and pay taxes. 2. φb and φw are observed. Accordingly, tax payers decide whether implementing a reform that establishes a decentralized system that costs Ω(π, φw ). 3. In either the two fiscal regimes tax payers form their willingness to cheat on taxes. Intrinsic motivations for complying with taxes are formed after the degree of ethnic diversity in the state is realized and the decentralization reform implemented or not. We split the analysis in two cases. To clarify our mechanism, we first present in Section 2.1 the analysis of the special case where φw = 0 so that the two regions are ethnically homogeneous—that is, populated by agents of the same ethnicity—and only between region variation is at play. Then, in Section 2.2, we consider the more general case where the fraction of within-region ethnic fragmentation varies between 0 and 1. 9

For instance, Oates (1972) suggests that in the provision of regional public goods, sub-central governments may find it hard to coordinate to internalize inter-jurisdictional externalities, or to exploit economies of scale. Many public services, in fact, imply decreasing average costs with growing scale of their delivery, because of better use of fixed assets (Stiglitz, 1989). Moreover, decentralization may also suffers from higher administrative and compliance costs due to insufficient administrative capacity of sub-national governments, lack of expertise and resources.

8

2.1

Homogeneous regions

In the simplest scenario, regions A and B are ethnically homogeneous but they can differ to each other. More precisely, they are each populated by two different ethnic groups with probability φb and by the same group with probability 1 − φb . Note that this is the case in which the welfare gains from decentralizing are maximal. Since the rate of fractionalization within each region is null (φw = 0) the net cost of decentralization reduces to Ω(π, 0) = π. At stage 1, people pay taxes and then observe the degree of ethnic diversity in the country. Since the state is organized as a centralized system, taxpayers perceived a psychological loss equal to λτ when φb > 0. When φb = 0 no loss is perceived. The representative agent i’s utility function, with i = {A, B}, after paying taxes τ , is therefore given by a convex combination between the two occurrences: UiC = φb (1 − τ − λτ ) + (1 − φb )(1 − τ ) = 1 − (1 + φb λ)τ.

(2)

Note that when λ is equal to zero taxpayers are not loss averted and do not incur in any extra costs, regardless of the fractionalization of the country. When λ is high, ethnic diversity generates substantial welfare losses. We interpret λ as the degree of loss aversion to ethnic diversity. Alternatively taxpayers can ask the tax administrator to decentralize the fiscal system and render the public good provided in region A excludable for the inhabitants of region B, and vice versa. This reform costs π but cuts φb to zero. Once the reform is implemented, region A does not care about the population in region B anymore and region B does not care about A likewise. Therefore the utility of the agent i after paying taxes in a decentralized country is equal to: UiD = 1 − τ − π.

(3)

In a decentralized country, with no within-region ethnic fragmentation, psychological costs are absent and people just pay taxes τ for benefiting the public good. The risk of contributing to a fragmented community is then fully covered after paying the cost of establishing a decentralized system. π is then the risk premium adverse agents pay to insure their risk—in this case ethnic diversity. It tells us how much people would be willing to pay for not contributing to the unfeeling social community and it is what we interpret as the ethnic diversity aversion component of the tax cheating.

9

One can easily get the optimal amount that in equilibrium tax payers are willing to pay to insure such a risk, equalizing (2) to (3) as follows: π ∗ = φb λτ.

(4)

When π ≤ π ∗ people are in favor of implementing a reform that establishes a decentralized system.

2.2

Heterogeneous regions

In this section, we generalize the analysis allowing for another source of ethnic diversity that makes regions A and B ethnically different within themselves. Specifically, we allow φw to range between 0 and 1. After paying taxes τ and observing the different components of ethnic diversity in the country, each representative agent i compares his or her utility level under the status quo and an alternative decentralized fiscal system. In a centralized system, the utility of each agent i after paying taxes is given by the following convex combination between the utility obtained under an ethnically fragmented constituency and that obtained in the scenario in which people are all members of the same group: UiC = (φw + φb )(1 − τ − λτ ) + (1 − φw − φb )(1 − τ ) = 1 − (1 + (φw + φb )λ)τ.

(5)

In a decentralized system, now, aversion to ethnic diversity also matters as long as ethnic fragmentation within each region is positive, i.e. φw > 0. Although, the between variation would be fully insured—not affecting tax payers decisions anymore— the implementation of the reform must then take into account the fact that (i) it does not fully insure the risk and (ii) it does not permit to achive all the welfare gains associated to decentralization. In other words, under a decentralized fiscal regime with φw > 0, each citizen bears two costs: a reduction in utility from being loss averted, φw λτ , and the net cost of decentralization, Ω(π, φw ) = (1 + φw )π. The utility after the establishment of decentralization is then given by: UiD = φw (1 − τ − λτ ) + (1 − φw )(1 − τ ) − (1 + φw )π = 1 − (1 + φw λ)τ − (1 + φw )π.

(6)

In equilibrium, agents will cover the risk establishing a reform that costs π ∗ : φb π = λτ. (7) 1 + φw As before, when the cost of the reform is low enough, i.e. lower that π ∗ , people will be in favor of transiting to a decentralized system. ∗

10

2.3

Tax morale formation and its determinants

After the decision of establishing a decentralization reform the agents form their intrinsic motivations for complying with taxes. If the reform has not been implemented, their attitude to cheat on taxes is π ∗ given in equation (7). If established, the reform cuts φb to zero. The agents’ ethnic diversity aversion component of the tax cheating is then equal to zero—not affected by λ, although φw might be positive. The idea is that φw cannot be insured, so people would not be willing to pay any dollars instead of contributing to the unfeeling community. We summarize the model predictions in the following proposition, discussing the effects of exogenous variations of our parameters of interest, λ, φw , and φb , on tax cheating. Proposition 1 The following comparative statics results hold: 1. Aversion to ethnic diversity positively affects the willingness to cheat on taxes: ∂π ∗ (λ, φw , φb ) > 0. ∂λ

(8)

2. In a decentralized system the incentive in cheating on taxes is smaller than in a centralized system. 3. Ethnic diversity affects the marginal effect of loss aversion to ethnic diversity on the willingness to cheat on taxes ambiguously, depending on the source of fragmentation: ∂ 2 π ∗ (λ, φw , φb ) ∂ 2 π ∗ (λ, φw , φb ) ≥ 0, ≤ 0. (9) ∂λ∂φb ∂λ∂φw The first simplest prediction of the model clearly comes out from the assumption of agents being reluctant in contributing to a system with more than one ethnic group. The loss aversion to ethnic diversity exerts an impact on tax morale and it is, consistently with previous researches (e.g., Li, 2010; Tusicisny, 2014), negatively related. At the same time, our second result is novel in the literature and tells us that such negative impact could be boiled down once implementing a decentralized fiscal reform. If tax payers concerns only regard ethnic fragmentation, as in our framework, eliminating one big source of ethnic diversity could be an effective policy to increase the country’s tax morale. Within region fragmentation is still at play, but it could not be insured through a decentralization reform and therefore, by this logic, taxpayers would not be willing to pay a dollar to cover it. So, tax cheating does not vary with the within ethnic diversity when differences between the two regions are cut down through the reform. 11

π ∗ (λ, φb , φw ) φb τ 1+φw

φb τ 1+φw0

0

1

λ

Figure 1: A decrease in tax cheating, π ∗ (λ, φb , φw ), when the with-region ethnic frag0 mentation increases from φw to φw . There is a last result that our model accounts for: countries with a more ethnically diverse population may or may not experience an intensification of the impact of the aversion to ethnic diversity on the willingness to cheat on taxes. The first effect is obtained when the regions through which decentralization has to be implemented are ethnically diverse to each other, i.e. when φb is high, while the second effect is more likely to be obtained when the regions through which decentralization has to be implemented are ethnically diverse inside their borders, i.e. when φw is high. In the first scenario there are strong forces that lead individuals to cheat on taxes so that a decentralization reform becomes relatively less costly. In the second, there are strong forces that blend the community; this moves tax morale up and renders the cost of decentralization relatively high (i.e., the cost function Ω moves up). The second scenario is illustrated in Figure 1 that draws the equilibrium levels of tax cheating, π ∗ (λ, φb , φw ), for different values of ethnic aversion, λ. The graph illustrates what happens to this relationship immediately after an increase in the within-region 0 ethnic fragmentation from φw to φw . Recall that an increase in φw makes decentralization a less effective tool to insure the loss aversion towards ethnic diversity. Hence, as φw increases tax payers would be less prone to cover the cost of a decentralization reform and, for the same degree of aversion to ethnic diversity λ, less inclined to cheat on taxes (or a higher tax morale). On the contrary, a rise in φb broadens the negative effect of aversion to ethnic diversity since it makes people even more willing to pay for not contributing to the fiscal system of a ethnically fragmented country. There are no previous works showing evidence on the sign and magnitude of these two links. Our

12

cross-country analysis brings support in favor of the second effect. This framework also suits for studying the potential effects of idiosyncratic versus aggregate shocks to the ethnic composition of the country’s population. An incoming migration flow that specifically targets a region (and not the rest of the country) moves up both φw and φb , generating an ambiguous effect on tax morale. An aggregate migration shock to the country’s population, by contrast, will move up φw that in turn reduces φb . Tax morale is therefore expected to increase. In the remainder of the paper we empirically test these three predictions.

3

Empirical analysis

3.1

Data description

Our research objective is to model individual tax morale as a function of ethnic aversion and decentralization. To this purpose we combine both individual-level and countrylevel data described below that potentially may influence subjective heterogeneity in tax morale, and correlate with ethnic aversion. Details on data description are reported in the Online Appendix, whereas summary statistics are presented in Table A1 in Appendix A. 3.1.1

Individual-level data

We use individual data from the 2005 wave of the World Value Survey—the only providing information upon the individual attitude towards ethnic diversity. Our sample contains more than 30 thousand individuals distributed in 44 countries. Our dependent variable is tax morale (T axM or). It relies on the question about noncompliance attitudes and assesses the extent to which respondents think cheating on taxes is justifiable, when an opportunity is available. In particular, the index is based on the answer, in a 1 to 10 scale, to the following question: “Cheating on taxes, if you have a chance, is: 1 = never justifiable, 10 = always justifiable.” Hence, the lower the score, the higher the tax morale. We rescaled the index, so that higher values of our variable correspond to a higher tax morale of the individual.10 We use the individual attitude toward ethnic diversity (AED) as the main explanatory variable of interest. It comes from the answer, on a scale from 1 to 10, to the following question: “Turning to the question of ethnic diversity, with which of the fol10

The WVS question we use is labeled F116. We simply rescaled it so as T axM or = 11 − F116.

13

lowing views do you agree? Ethnic diversity erodes a country’s unity (scale 1); Ethnic diversity enriches my life (scale 10).” This index has been rescaled so that higher values correspond to individuals with higher aversion to ethnic diversity.11 Following the literature on tax morale determinants (e.g., Torgler and Schaltegger, 2005; Torgler and Schneider, 2007; Lago-Pe˜ nas and Lago-Pe˜ nas, 2010), we additionally employ a group of indicators which account for individual socio-demographic characteristics: gender, age, marital status, education, social class, income, and size of town. - Sex is a dummy variable taking value 1 for male and 2 for female. - Age is a variable taking values from 15 to 98 years. - Marital status: we employ six dummy variables, each for one of the following categories: married; living together as married; divorced; separated; widowed; single/never married; divorced, separated or widow. - Education: we employ eight dummy variables, each for one of the following categories: inadequately completed elementary education; completed (compulsory) elementary education; incomplete secondary school; complete secondary school; incomplete secondary; complete secondary; some university without degree; university with degree. - Social class: we control for five dummies, one for each social class category (working, lower, lower middle, upper middle, upper class). - Income: we use dummy variables for each of the ten income classes. - Size of town: we employ eight dummy variables, each for category of the size of town. This literature has shown that tax compliance tends to be higher among older people, women, married people.12 Moreover, people with a higher income have shown to be more likely to cheat on taxes (Lago-Pe˜ nas and Lago-Pe˜ nas, 2010). The effect of education is instead ambiguous. While more educated individuals are found to be better aware of the benefits related to the public goods provision, they are also more critical about 11

Even in this case, we simply rescaled the WVS question G032 so as AED = 11 − G032. Orviska and Hudson (2003) using sample survey data from randomly chosen group of people, provide evidence that evasion is more common among the young and men, and is condoned by a large proportion of the population particularly ready to take advantage of someone else’s evasion. 12

14

the state’s fiscal policy (Torgler and Schaltegger, 2005; Torgler and Schneider, 2007). Finally, the inclusion of the size of town as control is motivated by the fact that people living in smaller towns develop a greater sense of community and social ethics. This is also in line with the Olson’s (1965) thesis that free riding is more likely to emerge in large groups. Alongside individual characteristics, we also control in the most demanding specification for some level of social capital of the respondents to WVS questionnaire. Specifically, we look at trust your neighborhood (item G007 18 B) and confidence in government (item E069 11). Higher values of these two indexes stand for individuals with higher levels of social capital and confidence in political institutions. 3.1.2

Country-level data

Decentralization measures. The measures of decentralization employed follow those proposed in literature (Rodden, 2004; Treisman, 2008). Whereas defining a suitable index to compare the degree of decentralization among countries is a particularly hard task—since decentralization is a multidimensional concept that encompasses political, administrative and fiscal dimensions, as well as cultural and geographical traits—two main approaches have been employed in the literature to measure the degree of country decentralization. The first kind of measures looks at the political and administrative dimensions of the public decision-making process. In order to account for this aspect, we use two dichotomous variables, f ederal provided by Treisman (2007) and autonomy from Fan et al. (2009). f ederal is a dummy variable on whether the country is federal or not (where 1 denotes the federal state) in the mid-1990s, according to the classification provided by a leading expert of federalism (Elazar, 1995).13 autonomy is a dummy calculated on whether the constitution assigned at least one policy area exclusively to subnational governments or gave subnational governments exclusive authority to legislate on matters not constitutionally assigned to any level. Although both indexes are close to various classics definitions of federalism,14 the 13

According to Elazar’s definition (1995), federal political systems are those in which a general government is constituted by a group of two or more constituent governments which have very substantial reserved or protected powers within the common whole. Following the author’s suggestion, federalism should be understood as constitutionalized power-sharing through systems that combine self-rule and shared rule. 14 As reported in Fan et al. (2009: 17), in Riker’s definition, a federal constitution has (at least) two levels of government governing the same land and people; each level of government has “at least one area

15

variable autonomy represents a broader category, since assigns the value 1 “also to countries that devolve decision-making rights to certain selected regions but not to others” (Fan et al., 2009: 17). The variable f ederal, instead, seems to adopt a stricter criterion: it relies on the primary characteristic of a federal state—a constitutionally guaranteed division of power between central and regional governments (Lijphart, 1984), but defines as federal “those states whose constitutions endow subnational governments with residual authority to decide on matters not explicitly assigned to the central government” (Treisman, 2008: 30). A second set of proxies of decentralization focuses on fiscal decentralization and it usually consists of a ratio between the expenditure (or revenue) of subnational government and the total government expenditure (or revenue) at the national level. We use three indexes drawn from the IMF – Government Finance Statistics and reported in the World Bank’s database of Fiscal Decentralization Indicators. For all of them, we calculate their average values over the period 1972-2000. These are subnational expenditures as a percentage of total expenditures (snete7200), subnational revenues as a percentage of total revenues (snrtr7200), and intergovernmental transfers as a share of sub-national expenditures (vi7200). The latest index, defined as vertical imbalance, measures the degree to which subnational governments rely on central government revenues to support their expenditures. Higher values of this variable correspond to lower degree of fiscal decentralization. To get consistent interpretation of our results, we also rescale vi7200 so that higher values correspond to higher degree of fiscal decentralization, as this means that a lower share of local expenditure comes from central government transfers.15 Bordignon et al. (2013) argues that vertical fiscal imbalances—i.e., higher shares of transfers in the local government budgets—is meant to capture the divergence between own revenues and expenditure at the local level. Higher levels of this index are usually shown to be associated with poorer local governments’ performance. A synthetic index of decentralization. We favor the hypothesis that a federal of action in which it is autonomous”; and this autonomy must be guaranteed in the constitution (Riker 1964: 11). This is similar to Dahl’s definition of federalism as “a system in which some matters are exclusively within the competence of certain local units-cantons, states, provinces-and are constitutionally beyond the scope of the authority of the national government; and where certain other matters are constitutionally outside the scope of the authority of the smaller units” (Dahl, 1986; quoted in Stepan, 2001: 318). P2000 P2000 15 As a result, our measure of vertical imbalance is given by maxk ( T1 t=1972 xkt ) − T1 t=1972 xkt , where t is the year and k is the country. T is the number of years over which the country average is computed. The maximum run over the cross-country distributions.

16

political structure is the proper proxy to use in this context because federalism is viewed as a way to manage conflict in ethnically divided societies, especially in countries where such divisions are more pronounced (Treisman, 2008). Federations often emerged with the role of balancing the competing and conflicting demands for autonomy and unity in such countries, on the basis that political recognition of cultural and ethnic pluralism helps to reduce ethnic tensions and conflicts, thus being an important instrument of nation building. Moreover, as argued in Teobaldelli (2011) and Dell’Anno and Teobaldelli (2015), federalism—where the constitution guarantees subnational governments the power to autonomously rule and legislate—is meant to be a process of governmental decentralization in which the devolution of resources goes hand in hand with the transfer of political responsibility, thereby strengthening subnational governments in terms of accountability and good governance. However, there is a key issue concerning the use of this measure. It has been argued that the advantages gained by f ederal actually stem from fiscal decentralization— namely, the devolution of expenditures and revenue-raising power. Thus, federalism may be an imperfect measure because “there can be both centralized and decentralized federations and, similarly, centralized and decentralized unitary states” (Lijphart, 1984: 176). Lijphart also emphasizes that “federalism and decentralization tend to go together ” (especially considering OECD countries) and the same pattern is found by Fisman and Gatti (2002). Moreover, the measure of fiscal decentralization often used in the literature, i.e., the subnational share of total government expenditures (or revenues), is not immune from criticism, either. Its most serious limitation is a possibly weak correspondence between budgetary items and actual decision making. If the budgets of local governments are actually mandated from above, then greater decentralization need not correspond to autonomy in expenditure allocation (Fisman and Gatti, 2002; Panizza, 1999). In order to work, fiscal decentralization requires a sufficient degree of local financial autonomy. Since the indexes, singularly considered, may suffer of such limitations, we propose a new dichotomous variable (Dec) based on the cited above five indexes. This new index aggregates different dimensions of political and financial autonomy into one decentralization dummy variable taking value of 1 if the majority of the decentralization variables considered indicates that the country is decentralized and 0 otherwise. In particular, the aggregation of the five decentralization indexes is made as follows. Each index is rescaled taking the value 1 if the country is decentralized and −1 if it is centralized; in the case of the three continues indexes (subnational expenditures, subnational revenues and vertical imbalance), the country is considered decentralized (resp. centralized), and 17

therefore the rescaled index takes value 1 (resp. −1), when the original index indicates that the country is more decentralized (resp. centralized) than the median. Then, our new decentralization variable (Dec) is obtained from the sum of the five rescaled indexes; if the score obtained from the sum of such indexes is negative, the country will be classified as centralized and our decentralization index will take value 0, while our new index takes value 1 indicating that the country is decentralized when the sum of the five rescaled indexes is positive. Let us index each decentralization measure as s and country as k. In symbols, our index is calculated as follows:  1 if P5 Ds > 0 k s=1 Deck = (10) 0 otherwise where the rescaled indexes Dks from the dummy variables f ederal and autonomy used to realize our synthetic index are obtained as follows:  1 if Sk = 1 s Dk = (11) −1 if Sk = 0; while the three new indexes drawn from the continuous variables (snete7200, snrtr7200 and vi7200) are as follows:  1 if Sk > median(S) Dks = (12) −1 otherwise. In constructing this variable, our final goal is to preserve the sample size, accepting the trade off to lose information of continuous variable on fiscal decentralization to gain a broader definition of overall decentralization.16 According to the Dec index, among the 43 countries of our sample, we count 13 decentralized states and 30 centralized countries.17 Table 1 reports the correlation indexes among the decentralization measures discussed above. Our synthetic index (in column 1), as expected, is significantly correlated 16

In the Online Appendix we construct an alternative synthetic measure using the principal component analysis. Our results are robust to this alternative definition of fiscal decentralization. See Section 4 of the Online Appendix. 17 According to our index, the decentralized states are, respectively: Argentina, Australia, Brazil, Switzerland, Spain, Germany, India, Moldova, Mexico, Malaysia, Sweden, United States, South Africa. See Table A2 in Appendix for a complete list of country and decentralization status.

18

Table 1: Correlations between decentralization measures.

Dec autonomy federal subn. exp. subn. rev. vert. imbalance Notes:

∗

p < 0.10,

∗∗

Dec

auton.

federal

1.000 0.313∗∗∗ 0.627∗∗∗ 0.557∗∗∗ 0.581∗∗∗ 0.256∗∗

1.000 0.508∗∗∗ 0.143 0.097 −0.038

1.000 0.421∗∗∗ 0.408∗∗∗ 0.050

p < 0.05,

∗∗∗

subnat. expend.

1.000 0.948∗∗∗ −0.059

subnat. vertical rev. imb.

1.000 0.161

1.000

p < 0.01

with all the other indexes. The measure reported in the last row is not correlated with any other proxies of decentralization used here, and our interpretation for that is that vertical imbalance captures a complementary dimension of fiscal decentralization—that is, the share of local expenditure coming from the central government. f ederal and autonomy are highly correlated to each other, though federal is also closer to the two indexes measuring the fraction of subnational economic activity—expenditure and revenues. Country-level control variables. We also control for some characteristics of the country. The total population and the GDP per capita, to account for the level of economic development, based on purchasing power parity (PPP) in 2011, are taken from the World Development Indicators (WDI). The measures of ethnic fractionalization (ethnic), language fractionalization (language) and religious fractionalization (religion) come from Alesina et al. (2003); all three indexes range from 0 to 1 with higher values denoting a higher degree of fractionalization. On the whole, our countries sample features an average ethnic fractionalization value of 0.36, that however hides a big heterogeneity (SD = 0.24). Similarly, the average language and religion fractionalization values are 0.32 and 0.43, respectively. A large literature on the impact of ethnic fractionalization on government activities indicates that ethnic and linguistic fractionalization are associated with negative outcomes in terms of both economic output and the quality of governments (see, for instance, Alesina et al., 2003; La Porta et al., 1999). According to these theories that look at the taxpayers preferences (culture) and their extrinsic motivation, public goods provision should be less efficient in divided societies, and this may lower tax morale. While in line with these previous findings, our mechanism is substantially different as it underlines the effect of ethnic fragmentation on tax compliance when 19

agents are loss averted to ethnic diversity. Moreover, we employ two indexes taken from the Worldwide Governance Indicators (WGI) to control for the quality of institutions that prove to be positively related to the quality and the extent of fiscal policies, which may affect, in turn, the external motivation associated to tax morale. The first one is the government effectiveness index, capturing the perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policies. The other one is the control of corruption index capturing the perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as “capture” of the state by elites and private interests. We also control for the burden of taxation that might reduce the extrinsic motivation of comply with taxes and/or exacerbate the aversion toward other ethnic groups. To do so we include the percentage of tax revenues over the GDP, from the WDI database. In our sample the average weight of tax revenue on the GDP is about 17%, but it also accounts for a substantial variation (sd = 6.70). For controlling for confounding political economy mechanisms (see Section 4.4) we include two measures of the strength of local parties from Enikolopov and Zhuravskaya (2007). Specifically, we use the average of the ages of the first government party, second government party, and 1st opposition party between 1972 and 2000, in thousands of years, and the probability that two members of parliament picked at random from among the government parties will be of different parties. Finally, we use country’s legal origin. In our sample, 34% of the countries have French origin and 23% have English origin.

3.2

Empirical strategy

The simplest way to study how the effect of aversion towards ethnic diversity on tax morale varies across differently decentralized systems (or in accordance with the degree of ethnic fragmentation) would be estimating a cross-country model. And this was the strategy of much of the works in the field.18 The advantage of working with a little complex model is however outweighed by the loss of information about tax compliance at individual level. Our data strikingly suggest that tax morale is mainly a purely individual behavior. Table 2 in fact reports the estimated variance components of tax morale in a null model at the country, σu2k , and religion-within-country level, σu2jk , as well as the intraclass 18

See for instance Lee (2010) and citations there in.

20

correlation ρ at either the two levels.19 In either cases ρ is smaller than 0.1 indicating that grouping counts for only a small part of the variation in tax morale—at most 10 per cent of the total variation in tax morale. In order to capture variations induced by individual specific characteristics and by institutional and cultural or social norms in the voluntary compliance with tax laws, we then depart from the existing literature by using a linear mixed model. Though tax morale is a purely individual behavior, its variation across individuals is affected by numerous policies pursued by national tax authorities—either in terms of penalty imposed to detected evasion or programs or public campaigns that are aimed to change the attitudes toward tax evasion—or by specific institutional setting of the country. Cultural norms that intrinsically prescribe individual contribution to the community are also expected to affect tax morale. We capture these different sources of variation in factors other than pecuniary using a three-level nested model. Specifically, we model tax morale as an attitude that varies at individual level but that also substantially depends on religion specific features (level 2) and country specific characteristics (level 3). Each individual i then is assigned to a religion group j within a country k. Random effects then operate at both the country and the religion-within-country levels.20 The role of religion groups varying within countries is meant to capture cultural diversity that might explain different behaviors across individuals in a country.21 Our estimations therefore account for the variation in tax morale within such a groups but also account for that induced by a contemporaneous co-movement of variables at the country level. As our theory predicts in fact the tax morale/ethnic aversion nexus lines up or down according to each country k and steepens according to the degree of country decentralization. To model the relationship among tax morale, ethnic aversion, and a decentralized organization of the state we therefore estimate the following benchmark 19

The null model is specified so as T axM orijk = α + ujk + uk + εijk , where ujk is the random component at the religion-within-country level and uk is that operating at country level. 20 Lago-Pe˜ nas and Lago-Pe˜ nas (2010) is the only relevant exception we are aware of. The two authors estimate a multilevel model where tax morale is modeled exploiting variation within-regions (at individual level) and between-regions. They do not include religion groups random effect, though, that in our view accounts for a significant part of variation in tax morale. 21 The role of culture—with their set of informal norms—on tax morale has been recently emphasized by a set of recent works, such as Halla (2012), DeBacker, Heim, and Tran (2015), and Kountouris and Remoundou (2013).

21

Table 2: Between-groups variations in tax morale and intraclass correlations estimated in a null model with n individuals nested within country k and religion-within-country jk.

Groups

N of groups

Avg. n obs. per group

Country Religion

Nk = 56 Njk = 469

1368.70 163.40

σ2

ρ

0.456 0.095 0.071 0.015

regression: T axM orijk = α + β1 AEDijk + β2 AEDijk × Deck + β3 Deck

(13)

+ Xijk γ + ujk + uk + εijk . Here, Deck indicates a decentralization measure that can be either a dummy or a continuous variable discussed in Section 3.1.2. In Xijk we gather the surveyed individual characteristics introduced in Section 3.1.1—such as gender, age, marital status, education, income, and the size of the town where the survey respondent resides. ujk is a random effect that we introduce to model cultural specific variations in tax morale within a country k. Those specific to each country k are modeled by including uk . Finally, εijk is the idiosyncratic residual that captures the unmodeled component in the tax morale behavior. Since we want to distinguish the effect of aversion to ethnic diversity on tax morale in countries with various degree of decentralization, we add up in equation (13) the interaction term between a measure of decentralization and the individual aversion toward ethnic diversity. Our testable hypotheses imply a negative sign of β1 and a positive sign of β2 . The first parameter, in fact, captures the effect of ethnic aversion on tax morale— and we expect that, within a country and within a nested religious group, more averted individuals toward ethnic diversity think cheating on taxes is somehow justifiable. A positive sign of β2 indicates that the negative effect of ethnic aversion on tax morale is less pronounced in decentralized countries.

3.3

Results

Table 3 reports the estimated parameters in our regression benchmark (13). Columns differ in terms of the decentralization measure adopted. In column (1) we use the synthetic index discussed in Section 3.1.2. We then use (in order) f ederal, autonomy, the 22

Table 3: Tax morale, decentralization, and ethnic aversion. (1) Deck is Dec

(2) Deck is federal

(3) Deck is autonomy

(4) Deck is subn.exp.

(5) (6) Deck is Deck is subn.rev. vert.imb.

AEDijk

-0.030∗∗∗ (0.005)

-0.027∗∗∗ (0.005)

-0.026∗∗∗ (0.005)

-0.039∗∗∗ (0.009)

-0.033∗∗∗ (0.009)

-0.040∗∗ (0.017)

AEDijk × Deck

0.033∗∗∗ (0.010)

0.026∗∗ (0.011)

0.028∗∗ (0.011)

0.001∗∗∗ (0.000)

0.001∗∗ (0.000)

0.000 (0.000)

Deck

-0.355∗∗ (0.173)

-0.464∗ (0.246)

-0.144 (0.213)

-0.010∗ (0.005)

-0.009 (0.006)

0.002 (0.004)

N Njk (Nk ) 2 Rijk log − likelihood

30,647 219(32) 0.016 -64,740

30,647 219(32) 0.015 -64,742

29,250 205(30) 0.016 -61,741

25,524 189(27) 0.018 -55,014

24,940 182(26) 0.019 -53,671

25,076 185(25) 0.018 -54,088

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity, while Deck stands for one of the six measures of decentralization. All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

subnational expenditures and revenues. In the last column Deck is vertical imbalance. All the columns control the effect of ethnic aversion, and the mediated effect through decentralization, on tax morale using the country ethnic, language, and religion fragmentation indexes. They also include individual controls. The controls are not reported in the table due to a size constraint. Table 3 also reports informations on the number of observations in each nested group, on the fraction of explained variability across in2 22 dividuals in their surveyed level of tax morale within a religion-within-country, Rijk , and on the maximum of the log-likelihood function estimated with the data, reported in the last row as a goodness-of-fit approximation. Data robustly reveal that individuals more averted to ethnic diversity are, on average, significantly less willing to comply with tax payment (βˆ 1 < 0) in all the columns presented in Table 3. Overall we find intrinsic motivations in tax compliance to be negatively affected by own belief about diversity even controlling for disparate sources of diversity. All specifications in our regression, in fact, look at countries with same level of ethnic, religion, and language fractionalization. Among the survey respondents with higher levels of ethnic aversion those living in variance at level m under the larger model 2 In each level m = {ijk, jk, k}, Rm = 1 − unexplained unexplained variance at level m under the null model . See also Xu (2003) and Gelman and Pardoe (2006) on R2 and linear mixed models. 22

23

Marginal Effects of Ethnic Aversion on Tax Morale

Dec 0

0 -.04

-.02

0

.02

0 -.04

p90

-.02

0

.02

-.04

p90

-.02

0

.02

p90

p75

p50

p25

vertical imbalance

p75 subnational revenues

subnational expenditures

1 autonomy

1 federal

1

p75

p50 p25 p10

p10 -.06

-.04

-.02

0

.02

p50

p25

p10 -.04

-.02

0

.02

.04

-.05 -.04 -.03 -.02 -.01

0

Figure 2: Marginal effects of one unit of ethnic aversion on tax morale for different values of decentralization. x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of each of the three continuous measures of decentralization from the bottom to the top. decentralized countries show higher desire to comply with the law (βˆ 2 > 0). Centralized states (with Deck equals to zero or small), on the contrary, offer to individuals more averted to ethnic diversity less motivations to pay taxes. This result casts light on the role of institutional features in influencing reciprocity and dimensions typically intrinsic to one person’s self-image, pride, altruism toward others, honesty, or yet fulfillment of civic duties. In Figure 2 we additionally report the marginal effects on tax morale of ethnic aversion for different values of decentralization. Specifically, Figure 2 is organized in six panels, one for each measure of decentralization employed in Table 3. Each graph plot point estimations of the marginal effects (on the x-axis) for different quantiles of the distribution of each measure of decentralization. brackets draw 95% confidence interval. For the dummy variables we only have two values, 0-1, whereas for the continuous ones we report marginal effects in correspondence of the 10th, the 25th, the 50th, the 75th,

24

and the 90th quantile from the bottom to the top of the y-axis. All-in-all, marginal effects of ethnic aversion on tax morale are never statistically different from zero for highly decentralized countries (the top of each panel). Climbing down the distribution of each decentralization measure turns the effect of ethnic aversion on tax morale negative. In terms of magnitude, these effects range from −0.030 and −0.026 when we consider dichotomous measures of decentralization (Dec, f ederal, and autonomy in the top panels). When we use continuous variables of decentralization— such as subnational expenditures and revenues—this range broadens to −0.032/ − 0.040, but results qualitatively do not change.23

3.4

Robustness analysis

In this section we discuss potential issues threatening our results from potential confounding mechanisms. We then consider several strategies to employ for addressing these concerns. Finally, we present an alternative specification with a wide set of controls.24 Our results bring evidence in favor of decentralization as a major country-level factor affecting individual tax morale. Other country-level factors are however expected to confound the effect of ethnic aversion on tax morale in decentralized and centralized states. For example, people more averted to ethnic diversity could report a lower degree of tax morale in decentralized states because those states are overall wealthier. The heterogeneous effect we have documented could then relies on income and wellbeing and not on that particular institutional feature we try to capture here. Following a similar lead, one may think that the burden of taxation can affect the willingness of tax payers to contribute and, at the same time, take to extremes people aversion on the ethnic diversity. GDP per capita level and the level of taxation are then our first main concern. Other potentially omitted country-level variables that are generally considered in the field literature are corruption and some measures of government effectiveness that could affect the extrinsic motivations of complying with taxes. Confounding political economy mechanisms can be at play if we do not control for 23

Coefficients are much larger when one runs a log-linear model, as we do in Section 3 of the Online Appendix. In this alternative specification, we find that when ethnic aversion increases by one point the magnitude of variation in tax morale goes from a 1.2% decrease, when decentralization is autonomy or federal, to a 2% decrease when we measure decentralization employing vertical imbalance. 24 As a robustness, we additionally run the same exercise we presented in Table 3 using other measures of decentralization from Fan et al. (2009). These results are presented in Section 6 of the Online Appendix.

25

the strength of political parties at the local level. Riker (1964), for example, theorizes that local political parties can have the incentive to establish a decentralized system in order to gain further private power. Therefore, one might expect that the stronger are local parties the higher is this competitive effect. To rule out this possibility we use two different measures: the age of main parties and a measure of fractionalization of governing parties. At the individual level, the most threatening concern is given by social capital that might move simultaneously tax morale and aversion to ethnic diversity. We therefore add up two measures of trust from the WVS to our regression model (13): trust your neighborhood (item G007 18 B) and confidence in government (item E069 11). In Table A3 in Appendix we present our estimations of regression (13), conditional on the logarithm of the GDP per capita, the WGI indices of corruption and government effectiveness as well as the logarithm of the population size in country k, the level of tax revenues, fractionalization of government parties and the average parties ages. Finally we include countries legal origin dummies to provide a coherent cross-country comparison and the two measures of social capital discussed above.25 This exercise leaves things substantially unchanged. Although in all the columns the magnitude of the coefficient βˆ 2 slightly reduces they remain firmly statistically significant up to 5%. On top of that, the marginal effects, drawn in Figure A1 in the Appendix, trace quasi-identical patterns of Figure 2. Therefore, we can confidently confirm our hypotheses.

3.5

Endogeneity issues

One of the main sources of variation we use on the right hand side of equation (13) is fiscal decentralization, Deck . However, fiscal decentralization might be endogenous, that is pre-determined by levels of tax morale. For example, it could be that people with fewer intrinsic motivation to comply with taxes are in favor of a reform establishing a decentralized regime to reduce the exposure to shock at the ethnic composition. If this is the case we would not be able to disentangle the effect going from fiscal decentralization to tax morale from that going from tax morale to fiscal decentralization. 25

To avoid to losing part of our information due to missing values, we impute missing values for a given additional control using its sample means. Moreover, to take into account the different “type” of information we employ a set of dummy variables (one for each control) that take on value of 1 if the observation has been replaced. Nonetheless, our results are robust to the specification with no imputation of the missing values.

26

In this section, we propose an instrumental variable approach to dealing with it. Specifically, we use geographical area of countries and its interaction with aversion to ethnic diversity to instrument decentralization and its interaction with AED. We therefore follow a previous literature that extensively made use of geographic area for selecting exogenous variation of fiscal decentralization (Panizza, 1999; Arzaghi and Henderson, 2005; Enikolopov and Zhuravskaya, 2007; Canavire-Bacarreza et al., 2016; MartinezVazquez et al., 2016). The idea behind this strategy is that coordination cost increases with size, leading larger countries to adopt a decentralized system. In Table A4 in Appendix we report our estimations. It is organized as follows: Panel A replicates our baseline estimation of column 1 of Table 3 for easiness of comparison in which we used our synthetic measure of decentralization. Panel B presents the estimation of the IV mixed model (i.e., the second stage). These estimations are both highly significant and similar to the ones reported in Panel A. However, once we remove endogenous variation from our decentralization measure we find that coefficients are slightly smaller for centralized countries (βˆ 1 = −0.026) and slightly bigger for decentralized ones (βˆ 2 = 0.050). In Panel C we present the estimations from the two first stage; in column 3 we use Areak and its interaction with aversion to ethnic diversity to explain AEDijk × Deck , in column 4 Areak is used to select variation of our decentralization measure. Both the two first stage are robust as indicated by the Wald χ2 Statistics that are very high. The two instruments are therefore very strong. Finally, Panel E reports the reduced form estimation which shows how Areak alone does not explain much of the tax morale variation. We therefore conclude that endogeneity bias are not such to modify our results.26

4

Ethnic aversion and tax morale: the role of ethnic fractionalization

Our theoretical and empirical analysis have shown that individuals’ negative view about “others” potentially undermines their intrinsic motivation to comply with taxes and therefore to contribute to the provision of public goods in (ethnically) heterogeneous 26

In the Online Appendix we also propose two additional IV strategies using a measure of risk aversion and attitudes toward ius sanguinis to select exogenous variation of attitudes toward ethnic diversity. We discuss them in Section 2.1 and 2.2 of the Online Appendix. On top of that, in Section 2.3 we pull all together presenting two full-fledged IV models, one with risk aversion and geographic area as instruments (see Table B5) and one with ius sanguinis and geographic area as instruments (see Table B6).

27

Aversion to Ethnic Diversity (Avg.) 4 8 6

JOR

EGY

IND

GHA THA

2

BGR GEO MDA HUN DEUSVN ROM MAR KOR CYP POL ETH JPN ITA NOR AUS PER FIN ZMB USA CHL ARG MYS ADO BRA MEX TTO SWE URY MLIBFA IDN CHN VNM

0

.2

.4 .6 Ethnic Fragmentation

.8

1

Figure 3: Countries scatter plot according to the average value of aversion to ethnic diversity (in the y-axis) and ethnic fragmentation (in the x-axis). communities. In particular, we have explored the effect of ethnic aversion on tax morale, mediated by the country’s degree of decentralization, controlling for the level of ethnic fractionalization (among many other country- and individual-level variables). Another related question is whether individuals averted to ethnic diversity are more or less prone to comply with taxes in more ethnically fractionalized societies. This question is particularly interesting as one might expect that individuals’ attitude toward ethnic diversity is related to the degree of ethnic fractionalization in the country. However, as it can be observed from Figure 3, there does not seem to be a particular pattern between the countries’ average degree of aversion to ethnic diversity (reported on the vertical axis) and the index of ethnic fractionalization (on the horizontal axis). For example, India is one of the countries where individuals reveal the highest degree of aversion toward other ethnicities while its degree of ethnic fractionalization is at intermediate levels. Our theoretical framework does not allow us to give an unambiguous answer to the question addressed here as its predictions about the effect of an increase in ethnic fractionalization on the (negative) impact of the aversion to ethnic diversity on tax morale depend on the source of heterogeneity, an information that we do not have. If the higher ethnic fractionalization is generated by the increase in ethnic diversity between regions (i.e., by a higher φb ), then our model predicts a higher negative effect of ethnic aversion on tax morale. Conversely, the negative effect of ethnic aversion on tax morale 28

is reduced by a higher degree of fractionalization when the latter is generated by a larger within region component (i.e., by a higher φw ). While the data at our disposal do not provide any information about the sources of ethnic fractionalization and, therefore, do not allow us to test the predictions of our model, we think that understanding how ethnic fractionalization affects the impact of ethnic aversion in centralized and decentralized countries is an interesting issue. Therefore, we add to our benchmark regression the following three terms: (i) the interaction between aversion to ethnic diversity and ethnic fractionalization, (ii) the interaction between the aversion to ethnic diversity, ethnic fractionalization and decentralization, (iii) the interaction between ethnic fractionalization and decentralization, and estimate the following saturated regression model: T axM orijk = α + β1 AEDijk + β2 AEDijk × Deck + β3 Deck + β4 EthF ractk + θ1 AEDijk × EthF ractk + θ2 AEDijk × EthF ractk × Deck

(14)

+ θ3 EthF ractk × Deck + Xijk γ + ujk + uk + εijk . As the marginal effect of the aversion to ethnic diversity on tax morale is now given by the following expression ∂T axM orijk = β1 + β2 Deck + (θ1 + θ2 Deck ) × EthF ractk , ∂AEDijk

(15)

θ1 measures the effect of ethnic aversion on tax morale in centralized countries for the part affected by ethnic fractionalization, whereas θ2 represents the same effect in decentralized countries. A positive (negative) sign of the estimated parameter ˆθ1 would bring support in favor of the hypothesis that a spread in ethnic fractionalization reduces (increases) the negative impact of ethnic aversion on tax morale in centralized countries. A positive (negative) sign of the estimated parameter ˆθ2 would support the hypothesis that higher degrees of ethnic fractionalization rises (reduces) the negative impact of ethnic aversion on tax morale in more decentralized countries. The estimated coefficients of Equation (14) are reported in Table 4 where each column refers to the measure of decentralization employed in the estimation, whereas Figure A2 shows the estimated marginal effects of aversion to ethnic diversity on tax morale using (15).27 Given the quite high number of interaction terms, the interpretation of the results is difficult and, therefore, for simplicity, we shall focus on column 1 where our synthetic dichotomous measure of decentralization is employed. 27

The marginal effects of aversion to ethnic diversity on tax morale estimated through Equation (14) using our principal component of decentralization are additionally presented in Section 4 of the Online Appendix.

29

Table 4: Tax morale, decentralization, ethnic aversion, and ethnic fractionalization. (1) Deck is Dec

(2) Deck is federal

(3) Deck is autonomy

-0.046∗∗∗ (0.011)

-0.050∗∗∗ (0.010)

-0.036∗∗∗ (0.010)

-0.048∗∗ (0.019)

-0.042∗∗ (0.018)

0.040 (0.044)

AEDijk × Deck

0.042∗ (0.021)

0.089∗∗∗ (0.024)

0.027 (0.024)

-0.000 (0.001)

-0.001 (0.001)

-0.001∗ (0.001)

Deck

0.038 (0.316)

-0.140 (0.422)

0.022 (0.426)

0.005 (0.009)

0.009 (0.011)

0.013∗ (0.007)

EthF ractk

-0.103 (0.566)

-0.378 (0.545)

-0.393 (0.603)

0.185 (0.768)

0.131 (0.766)

1.501 (0.945)

AEDijk × EthF ractk

0.040∗ (0.023)

0.061∗∗∗ (0.023)

0.029 (0.024)

0.007 (0.039)

0.010 (0.037)

-0.144∗ (0.074)

AEDijk × EthF ractk × Deck

-0.021 (0.050)

-0.166∗∗∗ (0.057)

-0.004 (0.052)

0.004∗∗ (0.002)

0.005∗∗ (0.002)

0.003∗∗∗ (0.001)

EthF ractk × Deck

-2.059∗ (1.134)

-1.253 (1.423)

-1.206 (1.422)

-0.075∗∗ (0.034)

-0.096∗∗ (0.043)

-0.044∗∗∗ (0.016)

N Njk (Nk ) 2 Rijk log − likelihood

30,647 219(32) 0.016 -64,740

30,647 219(32) 0.016 -64,738

29,250 205(30) 0.016 -61,741

25,524 189(27) 0.019 -55,009

24,940 182(26) 0.020 -53,666

25,076 185(25) 0.019 -54,084

AEDijk

(4) (5) (6) Deck is Deck is Deck is subn.exp. subn.rev. vert.imb.

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity, while Deck stands for one of the six measures of decentralization. EthF ractk is the index of ethnic fractionalization. All the columns include individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

The first row of Table 4 reports the estimated coefficient βˆ 1 which represents the “direct” effect of the aversion to ethnic diversity on tax morale. In column 1 such estimate is –0.046 and it is significant at 1% level; hence, it is higher than the baseline estimate equal to –0.030 reported in column 1 of Table 3. The second row of the table shows the estimated coefficient βˆ 2 which measures the effect of the aversion to ethnic diversity, moderated by levels of decentralization, on tax morale; this coefficient is positive and significant, and similar in size to βˆ 1 and to the estimated coefficient βˆ 2 of the baseline specification reported in Table 3. These estimates suggest that in centralized

30

Ethnic Fragmentation

Ethnic Fragmentation

Marginal Effects of Ethnic Aversion on Tax Morale p90 p75 p50 p25 p10

p90 p75 p50 p25 p10 -.04 -.02 0 .02 .04 Decentralized Countries (CI 95%)

Ethnic Fragmentation

Ethnic Fragmentation

-.06 -.04 -.02 0 Centralized Countries (CI 95%)

p90 p75 p50 p25 p10 -.06 -.04 -.02 0 Centralized Countries (CI 90%)

p90 p75 p50 p25 p10 -.04 -.02 0 .02 .04 Decentralized Countries (CI 90%)

Figure 4: Marginal effects of ethnic aversion on tax morale for different values of ethnic fractionalization in centralized (Dec = 0) and decentralized countries (Dec = 1). The uppermost panels report confidence bands at 95%, while the lowermost panels report confidence bands at 90%. countries with no ethnic fractionalization (our baseline category) the negative effect of the aversion to ethnic diversity on tax morale are now larger. From Table 4 we also observe that the estimated coefficient ˆθ1 , which represents the effect of the aversion to ethnic diversity moderated by levels of ethnic fragmentation on tax morale, is positive and its magnitude is close to βˆ 1 . This suggests that ethnic aversion might have limited effects on tax morale in centralized countries with extremely high levels of ethnic fractionalization. The estimated coefficient ˆθ2 instead is not statistically different from zero and this suggests that ethnic fractionalization does not affect the relation between ethnic aversion and tax morale in decentralized countries.28 Figure 4 shows the estimated marginal effects of ethnic aversion on tax morale for different values of ethnic fractionalization, using our synthetic measure of decentralization It is also worth noting that the estimated coefficient ˆθ3 is negative and this means that tax morale is lower in decentralized countries with a substantial diversity in ethnic composition. 28

31

in column 1. The two uppermost panels report confidence bands at the 95% level, while the two lowermost panels report confidence bands at the 90% level. In the two leftmost panels we report the marginal effects in centralized countries, while in the two rightmost ones we draw those estimated in decentralized countries. Note that the marginal effects computed at the median level of ethnic fractionalization (p50) are the same as those presented in the first panel of Figure 2. We observe that, as ethnic fractionalization increases, the negative effect of ethnic aversion on tax morale in centralized countries lowers; at the highest levels of ethnic fractionalization in our sample, the estimated effect of ethnic aversion on tax morale remains negative but it looses precision and becomes not statistically significant at 95% level of confidence, while it remains statistically significant at 90% level of confidence. The estimated effect of ethnic aversion on tax morale is instead equal to zero for any degree of ethnic fractionalization in decentralized countries, as in the baseline estimation.29 In sum, the results presented here confirm those of the baseline estimations reported in the previous section. The only additional result is that the size of the negative effect of ethnic aversion on tax morale in centralized countries may be weakened by higher degrees of country’s fractionalization. Our dataset does not contain information about the geographical distribution of ethnic diversity and, therefore, it does not allow us to understand whether the explanation provided by our model (i.e., that a large part of the spread in ethnic fractionalization is due to a larger within region component φw which in turn generates a moderating effect on the link between ethnic aversion and tax morale) is at the heart of such result. We are also not aware about the existence of microdata with information at local level that would help us to test our theory; this would probably also go beyond the scope of this paper. Hence, we leave to future research the understanding of the exact mechanisms explaining why more fractionalization may reduce the negative impact of ethnic aversion on tax morale in centralized countries.

5

Conclusions

In this paper we analyzed the effects of the individuals’ aversion to ethnic diversity on tax morale under different institutional frameworks. Our work is based on the idea that individuals who are averted to ethnic diversity are more reluctant to contribute to 29

In Section 6 of the Online Appendix we additionally report the marginal effects of ethnic aversion on tax morale for different values of ethnic fractionalization using the remaining two dichotomous measures of decentralization, federal (Figure B7) and autonomy (Figure B8), we have employed throughout the paper.

32

the provision of public goods which can benefit other (ethnic) groups. In decentralized countries, the individuals’ welfare losses associated with the financing of public goods benefiting other ethnic groups is reduced because the provision of public goods and services is made by jurisdictions characterized by communities more homogeneous than the whole country, which increases the individuals’ intrinsic motivation to pay taxes. We presented a simple model showing the mechanisms at work and then tested the main predictions of our theory using microdata from the World Value Survey and various measures of fiscal decentralization. Our analysis led to two main results. First, a negative attitude toward ethnic diversity reduces tax morale in centralized political systems, while it does not have a statistical significant effect in decentralized ones. Second, the negative effect of individuals’ ethnic aversion on tax morale is lower in homogenous countries than in ethnically fragmented states. In terms of normative results, the question under what conditions the negative effects of individuals’ aversion to ethnic diversity on tax morale may be reduced has policy relevance as many developing countries are ethnically fragmented and composed by groups with a negative attitude towards other ethnicities. This paper shows that a proper choice of the institutional setting (in this case, more fiscal decentralization) might improve tax morale and favors investments in state fiscal capacity.

References [1] Alesina, A., Baqir, R. and Easterly, W. (1999). Public goods and ethnic divisions. The Quarterly Journal of Economics, 114(4), 1243-1284. [2] Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., and Wacziarg, R. (2003). Fractionalization. Journal of Economic Growth, 8(2), 155-194. [3] Alesina, A., Glaeser, E., and Sacerdote, B. (2001). Why doesn’t the U.S. have a European style welfare state? Brookings Papers on Economic Activity, 187-278. [4] Alesina, A., La Ferrara, E. (2005). Ethnic diversity and economic performance. Journal of Economic Literature, 43(3), 762-800. [5] Alesina, A., Giuliano, P. (2015). Culture and Institutions. Journal of Economic Literature, 53(4): 898–944. [6] Alm, J., Torgler, B. (2006). Culture differences and tax morale in the United States and in Europe. Journal of Economic Psychology, 27(2), 224-246. 33

[7] Andreoni, J., Erard, B., Feinstein, J., (1998). Tax compliance. Journal of Economic Literature, 36, 818-860. [8] Arzaghi, M., Henderson, V.J., (2005). Why countries are fiscally decentralizing. Journal of Public Economics, 89(7), 1157-1189. [9] Besley, T., Persson, T. (2013). Taxation and Develoment. In: Auerbach, A., Chetty, R., Feldstein, M., Saez, E. (eds.), Handbook of Public Economics. [10] Besley, T., Jensen, A., Persson, T. (2015). Norms, Enforcement, and Tax Evasion, CEPR Discussion Papers 10372. [11] Bolton P., Roland G. (1997). The Breakup of Nations: A Political Economy Analysis, Quaterly Journal of Economics, 112(4): 1057-1090. [12] Bordignon, M., Gamalerio, M., Turati, G. (2013). Decentralization, vertical fiscal imbalance, and political selection. CESifo Working Paper Series No. 4459. [13] Canavire-Bacarreza, G., Martinez-Vazquez, J., Yedgenov B. (2016). Reexamining the determinants of fiscal decentralization: what is the role of geography?, Journal of Economic Geography, 1–41. DOI:https://doi.org/10.1093/jeg/lbw032 [14] Dahl, R. (1986). Federalism and the democratic process. In Democracy, Identity and Equality. Oslo: Norwegian University Press. [15] DeBacker, J., Heim, B.T., Tran, A. (2015). Importing corruption culture from overseas: Evidence from corporate tax evasion in the United States. Journal of Financial Economics, 117(1), 122-138. [16] Dell’Anno R. (2009). Tax evasion, tax morale and policy maker’s effectiveness, The Journal of Socio-Economics, 38(6), 988-997. [17] Dell’Anno R., Teobaldelli D. (2015). Keeping both corruption and the shadow economy in check: the role of decentralization. International Tax and Public Finance, 22(1), 1-40. [18] Elazar, Daniel J. (1995). From statism to federalism: a paradigm shift. Publius, 25(2), 5-18. [19] Enikolopov, R., Zhuravskaya, E. (2007). Decentralization and political institutions. Journal of Public Economics, 91, 2261-2290. 34

[20] Fan, S. C., Lin, C., Treisman, D. (2009). Political decentralization and corruption: evidence from around the world. Journal of Public Economics, 93, 14-34. [21] Feld, L., Frey, B. (2002). Trust breeds trust: how taxpayers are treated. Economics of Governance, 3, 87-99. [22] Fisman, R., Gatti, R. (2002). Decentralization and corruption: evidence across countries. Journal of Public Economics, 83, 325-345. [23] Frey, B., Torgler, B. (2007). Tax morale and conditional cooperation. Journal of Comparative Economics, 35(1), 136-159. [24] Gelman A. and Pardoe L. (2006). Bayesian Measures of Explained Variance and Pooling in Multilevel (Hierarchical) Models. Technometrics, 48(2), 241-251. [25] Gorodnichenko, Y. and Roland, G. (2017). Culture, Institutions and the Wealth of Nations. Review of Economics and Statistics, 99(3), 402-416. [26] G¨ uth, W., Levati, V. and Sausgruber, R. (2005). Tax morale and (de-)centralization: an experimental study. Public Choice, 125(1), 171-188. [27] Halla, M. (2012). Tax morale and compliance behavior: first evidence on a causal link. The B.E. Journal of Economic Analysis & Policy: Advances, 12(1): 1-27. [28] Hofmann, E., Hoelzl, E., and Kirchler, E. (2008). Preconditions of Voluntary Tax Compliance: Knowledge and Evaluation of Taxation, Norms, Fairness, and Motivation to Cooperate. Journal of Psychology, 216(4): 209-217. [29] Koszegi, B., Rabin, M. (2006). A model of reference-dependent preferences. The Quarterly Journal of Economics, 121(4), 1133-1165. [30] Koszegi, B., Rabin, M. (2009). Reference-dependent consumption plans. American Economic Review, 99(3), 909-936. [31] Kountouris, Y., and Remoundou, K. (2013). Is There a Cultural Component in Tax Morale? Evidence from Immigrants in Europe. Journal of Economic Behavior & Organization, 96: 104-119. [32] Lago-Pe˜ nas, I., Lago-Pe˜ nas, S. (2010). The determinants of tax morale in comparative perspective: Evidence from European countries. European Journal of Political Economy, 26, 441-453. 35

[33] La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R. (1999). The quality of government. Journal of Law, Economics and Organization, 15(1), 222-279. [34] Li, X.S. (2010). Social identities, ethnic diversity, and tax morale. Public Finance Review, 38(2), 146-177. [35] Lijphart, A. (1977). Democracy in Plural Societies: A Comparative Exploration. New Haven: Yale University Press. [36] Lijphart, A. (1984). Democracies: patterns of majoritarian and consensus government in twenty-one countries. New Haven/London: Yale University Press. [37] Luttmer, E. (2001). Group loyalty and the taste for redistribution. Journal of Political Economy, 109(3), 500-528. [38] Luttmer, Erzo F. P., Singhal M. (2014). Tax morale. Journal of Economic Perspectives, 28(4), 149-68. [39] Martinez-Vazquez, J., Torgler, B. (2009). The evolution of tax morale in modern Spain. Journal of Economic Issues, 43(1), 1-28. [40] Martinez-Vazquez, J., Lago-Pe˜ nas, S., Sacchi, A. (2016). The impact of fiscal decentralization: a survey. Journal of Economic Surveys. DOI: 10.1111/joes.12182. [41] Oates, W.E. (1972). Fiscal federalism. NY: Harcourt Brace Jovanovich. [42] Oates, W. E. (1999). An Essay on Fiscal Federalism. Journal of Economic Literature, 37(3), 1120–1149. [43] Olson, M. (1965). The logic of collective action. Public goods and the theory of groups. Cambridge: Harvard University Press. [44] Orviska, M., Hudson, J. (2003). Tax evasion, civic duty and the law abiding citizen. European Journal of Political Economy, 19, 83-102. [45] Panizza, U. (1999). On the determinants of fiscal centralization: theory and evidence. Journal of Public Economics, 74, 97-139. [46] Pommerehne, W., Weck-Hannemann, H. (1996). Tax rates, tax administration and income tax evasion in Switzerland. Public Choice, 88, 161-170.

36

[47] Riker, W.S. (1964). Federalism. Origin, operation, significance. Boston: Little Brown. [48] Rodden, J. (2004). Comparative federalism and decentralization. Comparative Politics, 481-500. [49] Schwarz, P. (2011). The impact of federalism on tax morale: evidence from Spain. Available at (http://www.webmeets.com/SAEe/2011/prog/viewpaper.asp?pid= 188) [50] Stepan, A. 2001. Arguing Comparative Politics, New York: Oxford University Press. [51] Stiglitz, J.E. (1989). Economics of the public sector. New York: Norton. [52] Tajfel, H., Billig, M., Bundy, R.P., Flament, C. (1971). Social categorization and intergroup behaviour. European Journal of Social Psychology, 1(2), 149-178. [53] Teobaldelli, D. (2011). Federalism and the shadow economy. Public Choice, 146(34), 269-289. [54] Ticchi, D., Verdier, T., Vindigni, A. (2013). Democracy, Dictatorship and the Cultural Transmission of Political Values. IZA Discussion Papers No. 7441. [55] Torgler, B., Schneider, F.G. (2007). What shapes attitudes toward paying taxes? Evidence from multicultural European countries. Social Science Quarterly, 88(2), 443-470. [56] Torgler, B. (2006). The importance of faith: tax morale and religiosity. Journal of Economic Behavior & Organization, 61, 81-109. [57] Torgler, B. (2007). Tax compliance and tax morale: a theoretical and empirical analysis. Cheltenham, UK: Edward Elgar Publishing. [58] Torgler, B. (2005). Tax morale and direct democracy. European Journal of Political Economy, 21, 525-531. [59] Torgler, B., Werner, J. (2005). Fiscal autonomy and tax morale: evidence from Germany. Public Finance and Management, 5(4), 460-485. [60] Torgler, B., Schaltegger, C.A. (2005). Tax morale and fiscal policy. CREMA Working Paper Series 2005-30, Center for Research in Economics, Management and the Arts (CREMA). 37

[61] Torgler, B., Schneider, F., Schaltegger, C.A. (2010). Local autonomy, tax morale and the shadow economy. Public Choice, 144(1-2), 293-321. [62] Tranchant, J.P. (2007). Decentralization and Ethnic Conflict: The Role of Empowerment. MPRA Paper No. 3713, M¨ unchen: University Library of Munich. Available at: http://mpra.ub.unimuenchen.de/3713/1/MPRA paper 3713.pdf [63] Treisman, D. (2007). What have we learned about the causes of corruption from ten years of cross-national empirical research? Annual Review of Political Science, 10, 211-244. [64] Treisman, D. (2008). Decentralization dataset, 2008. Available at author’s webpage. [65] Tusicisny, A. (2014). Is ethnic diversity a poverty trap? A complex relationship between ethnicity, trust, and tax morale. European University Institute MWP, 2014/09. http://hdl.handle.net/1814/31613 [66] Witte, E., Craeybeckx, J., Meynen, A. (2009). Political History of Belgium: From 1830 Onwards. Academic and Scientific Publishers. [67] Xu (2003). Measuring explained variation in linear mixed effects models. Statistics in Medicine, 22(22), 3527-3541.

38

A

Appendix Table A1: Summary statistics.

Individual-level variables Tax Morale Ethnic Aversion Sex Age Marital Status Education Social Class Scales of Income Size of Town Trust Trust in Govt. Country-level variables Dec Autonomy Federal Subnat. Expenditure Subnat. Revenues Vertical Imbalance Ethnic Fract. Language Fract. Religion Fract. GDP per capita (log) Population (log) Corruption Index Gov. Effectiveness Index Tax Revenue (% gdp) Party Age Fract. of Govt. Parties English Leg. Or. French Leg. Or. German Leg. Or. Socialist Leg. Or. Scandinavian Leg. Or.

mean

sd

8.72 4.45 1.51 41.66 2.71 4.68 3.33 4.77 4.75 2.92 2.48

2.20 2.74 0.50 16.38 2.19 2.23 0.98 2.22 2.44 0.79 0.92

0.30 0.28 0.23 24.45 19.58 62.26 0.36 0.32 0.43 9.52 16.97 0.38 0.45 17.48 0.03 0.26 0.23 0.34 0.14 0.23 0.07

0.46 0.00 1.00 0.45 0.00 1.00 0.43 0.00 1.00 16.18 2.14 54.84 14.58 1.60 51.48 21.21 0.00 92.17 0.24 0.00 0.74 0.27 0.00 0.84 0.24 0.00 0.82 1.03 6.98 11.05 1.83 11.34 21.01 1.02 -1.12 2.59 0.92 -0.91 2.13 6.70 8.63 41.64 0.03 0.00 0.14 0.25 0.00 0.74 0.42 0.00 1.00 0.48 0.00 1.00 0.35 0.00 1.00 0.42 0.00 1.00 0.25 0.00 1.00

39

min

max

1.00 10.00 1.00 10.00 1.00 2.00 15.00 98.00 1.00 6.00 1.00 8.00 1.00 5.00 1.00 10.00 1.00 8.00 1.00 4.00 1.00 4.00

obs. 55,365 55,365 55,338 55,224 55,245 50,978 51,968 51,618 42,006 53,924 51,250 43 40 43 36 35 31 38 38 38 42 44 45 45 38 41 43 44 44 44 44 44

Marginal Effects of Ethnic Aversion on Tax Morale

Dec 0

0 -.04

-.02

0

.02

0 -.04

p90

-.02

0

.02

-.03 -.02 -.01

p90

0

.01

.02

p90

p75

p50

p25

vertical imbalance

p75 subnational revenues

subnational expenditures

1 autonomy

1 federal

1

p75

p50 p25 p10

p10 -.06 -.04 -.02

0

.02 .04

p50

p25

p10 -.04

-.02

0

.02

.04

-.05 -.04 -.03 -.02 -.01

0

Figure A1: Marginal effects of aversion to ethnic diversity on tax morale to a change in decentralization (full set of controls). x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of each of the three continuous measures of decentralization from the bottom to the top.

40

Marginal Effects of Ethnic Aversion on Tax Morale

Dec 0

0 -.04

-.02

0

.02

0 -.04

p90

-.02

0

.02

-.04

p90

-.02

0

.02

p90

p75

p50

p25

vertical imbalance

p75 subnational revenues

subnational expenditures

1 autonomy

1 federal

1

p75

p50 p25 p10

p10 -.06 -.04 -.02

0

.02 .04

p50

p25

p10 -.05

0

.05

-.04 -.03 -.02 -.01

0

.01

Figure A2: Marginal effects of one unit of ethnic aversion on tax morale for different values of decentralization using Equation (14) and holding ethnic fragmentation fixed. x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of each of the three continuous measures of decentralization from the bottom to the top.

41

Table A2: Decentralization measures by country. Dec autonomy

Argentina Australia Burkina Faso Bulgaria Brazil Switzerland Chile China Cyprus Germany Egypt Spain Ethiopia Finland Georgia Ghana Hungary Indonesia India Italy Jordan Japan Korea Morocco Moldova Mexico Mali Malaysia Norway Poland Romania Serbia Slovenia Sweden Thailand Trinidad and Tobago Turkey Taiwan, China Ukraine Uruguay United States Vietnam South Africa Zambia

ARG AUS BFA BGR BRA CHE CHL CHN CYP DEU EGY ESP ETH FIN GEO GHA HUN IDN IND ITA JOR JPN KOR MAR MDA MEX MLI MYS NOR POL ROM SRB SVN SWE THA TTO TUR TWN UKR URY USA VNM ZAF ZMB

1 1 0 0 1 1 0 0 0 1 0 1 0 0 0 0 0 0 1 0 0 0 0 0 1 1 0 1 0 0 0 0 1 0 0 0 0 0 0 1 0 1 0

f ederal

1 0 0 1 1 0 0 1 0 0 1 1 0 0 0 0 0 1 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1

0 0 1 0

42

subnational subnational expenditures revenues 32.605483 27.540157

vertical imbalance

1 1 0 0 1 1 0 0 0 1 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 1 0 0 0

37.968482 41.190838 3.4735385 18.290861 34.239581 51.888997 6.5822997 54.843189 2.1380194 42.092175

14.288232 25.492137 45.912849 5.0424785 51.481553 2.8578643 35.578183

53.71882 76.06923 65.43143 68.21504 75.32173 71.96277 57.83129 91.17384 76.01637

19.987468 2.2754571 38.102237

12.252808 2.8987799 31.235549

57.71594 92.1683 66.61002

21.631288 11.890149 45.473825 21.18715 5.8557305 43.457615 31.334226 5.8066361 27.875636 21.146239

12.279565 3.0805095 33.051234 8.1266166 7.4937871 38.123117 14.804028 4.1232391 23.565682 20.251953

46.28557 20.152 60.507 32.72783

19.060285 34.602032 22.835237 14.322803

15.951196 24.435636 15.978326 9.8204822

78.3403 65.82964 59.0459 57.52032

0 0 0 0 0 0 0 0 1 0 0 0

11.089915 37.746384 10.517253 4.8329806 50.706748

9.058923 32.234827 5.8969731 1.6047783 49.890498

75.44896 75.33686 49.57942 0

9.2968712 44.622662

9.2577896 40.079655

91.47559 65.56709

26.733982 5.131685

13.525037 5.6210283

39.8257 70.82536

28.82753 71.73647 88.84158

Table A3: Tax morale, decentralization, and ethnic aversion. Additional controls. (1) Deck is Dec

(2) Deck is federal

(3) Deck is autonomy

(4) Deck is subn.exp.

(5) (6) Deck is Deck is subn.rev. vert.imb.

AEDijk

-0.027∗∗∗ (0.005)

-0.023∗∗∗ (0.005)

-0.022∗∗∗ (0.005)

-0.035∗∗∗ (0.009)

-0.029∗∗∗ (0.009)

-0.039∗∗ (0.017)

AEDijk × Deck

0.033∗∗∗ (0.010)

0.025∗∗ (0.011)

0.027∗∗ (0.011)

0.001∗∗∗ (0.000)

0.001∗∗ (0.000)

0.000 (0.000)

Deck

-1.173∗∗∗ (0.362)

-1.683∗∗∗ (0.520)

-0.889∗ (0.462)

-0.065∗∗∗ (0.019)

-0.071∗∗∗ (0.023)

-0.022∗∗ (0.009)

N Njk (Nk ) 2 Rijk log − likelihood

30,647 219(32) 0.020 -64,679

30,647 219(32) 0.020 -64,682

29,250 205(30) 0.020 -61,687

25,524 189(27) 0.023 -54,959

24,940 182(26) 0.024 -53,619

25,076 185(25) 0.023 -54,038

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity, while Deck stands for one of the six measures of decentralization. All the columns include ethnic, language, religion fragmentation, the logarithm of the gdp per capita and of the population, tax revenues, an index of corruption and the government effectiveness index as well as an index of fractionalization of government parties and the average parties ages are also included. Legal origins dummies and individuals controls are also included as well as dummies that take on value of 1 if the observation has been replaced by the sample mean. In addition to Table 3, individual controls include a measure of trust and of confidence about the government. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Marginal effects are reported in Figure A1.

43

Table A4: IV estimation. (1) Tax Morale

(2) Tax Morale

(3) AEDijk × Deck

(4) Deck

(5) Tax Morale

Panel A – Mixed Model AEDijk -0.030∗∗∗ (0.005) AEDijk × Deck

0.033∗∗∗ (0.010)

Deck

-0.355∗∗ (0.173)

Panel B – IV Mixed Model AEDijk

-0.026∗∗∗ (0.0045)

AEDijk × Deck

0.050∗∗∗ (0.012)

Deck

-0.229 (0.411)

Panel C – First Stage AEDijk × Areak

0.000∗∗∗ (0.000)

Areak

-0.000 (0.000)

0.000∗∗∗ (0.000)

Panel D – Reduced Form AEDijk

-0.019∗∗∗ (0.005)

AEDijk × Areak

-0.000 (0.000)

Areak

-0.000 (0.000)

Observations Wald χ2 Statistics

30,647

30,647

30,647 25,983.02

30,647 379.82

30,647

Notes: Dependent variable is tax morale in columns 1, 2, and 5. AEDijk is the individual degree of aversion to ethnic diversity, while Deck is the synthetic measure of decentralization. Areak is the size in km2 of the geographical area of the country. All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

44

Tax Morale, Aversion to Ethnic Diversity, and Decentralization by Alessandro Belmonte, Roberto Dell’Anno, and D´esir´ee Teobaldelli

Online Appendix

1

Trust

In this section we report some robustness of our empirical analysis. One of the main concern is that some individual unobservables are correlated with the error term, therefore making our estimates of β1 and β2 biased. For instance, social capital might be correlated with both ethnic aversion and tax morale, so that we cannot identify the effect of the first variable over the second one. Here, we show how our estimates are robust to the inclusion of a wide set of measures of trust, from the 2005 WVS wave. Specifically, we include trust your neighborhood (item G007 18 B), trust people you know personally (item G007 33 B), trust people you meet for the first time (item G007 34 B), trust people of another religion (item G007 35 B), trust people of another nationality (item G007 36 B) in our baseline analysis (Equation 13 of the paper). Finally, we also employ confidence in government (item E069 11) as a measure of trust toward institutions. All these measures have been first rescaled, so as higher values are taken to mean respondents with higher levels of trust, and then replaced with the sample mean in case of missing values, to hold the sample constant across different specifications1 In Table B1 we report estimations of our baseline model described in Equation (13) of the paper, using as a decentralization variable our most preferred synthetic measure. In column 1, we reproduce for easiness of comparison the same estimation that we have reported in column 1 of Table 3 in the paper, i.e., without any measure of trust. The six reminder columns include as control each of the six measures of social capital as well as each of six dummies that take on value of 1 if the observation as been replaced by the sample mean. Moreover, in Table B2 we replicate the entire Table 3 in the paper controlling for two additional measures of social capital: trust your neighborhood and confidence in government. Our estimations do not change substantially and in terms of marginal effects (Figure B1) we find identical patterns than those presented in Figure 1 in the paper. Overall, these results suggest that the effect of ethnic aversion on tax morale presented in Table 3 of the paper is not affected by the omission of variables related to the individual level of trusts toward other persons or institutions. 1

Missing values are about 1% of the sample in column 1 of Table 3 for trust your neighborhood (item G007 18 B), the measure with the lowest number of missing values, and about 6.5% for confidence in government (item E069 11), the measure with the highest number of missing values.

1

2

Additional Endogeneity Issues

Our analysis might be threatened by a potential reverse causality going from tax morale to ethnic aversion or from tax morale to decentralization. The idea is that individuals with higher intrinsic motivations to comply with taxes can also form positive attitudes towards the “other,” or that lower levels of tax morale might favor the adoption of reforms that bring the country to establish a decentralized fiscal regime. We have discussed the second potential channel using an IV strategy that employs geographic variables to select exogenous variation of our measures of decentralization (see Section 4.4.3 in the paper). In this section we discuss two candidate instruments that can be employed to address the first concern. At the best of our knowledge there is no previous attempt in the literature to employ an IV strategy to explain how ethnic aversion causes tax morale. Finally, we combine the two strategies with a fully identified model.

2.1

Risk Aversion

The first instrument we use to instrument individual ethnic aversion is risk aversion. The idea behind the use of this instrument comes out from our theory. Specifically, we use the degree of risk aversion, and its interaction with decentralization, to channel that part of variation in aversion to ethnic diversity that is exogenous to tax morale. In order to understand the causal mechanism recall that the main assumption of our theoretical model is that tax payers incur in a non-pecuniary extra cost when his or her wealth is used to subsidize a good that also benefits a non-member of his or her ethnic group. Because of this behavioral component (captured by the degree of loss aversion to ethnic diversity, λ), tax payers would be willing to insure such a risk by paying a cost Ω(π, φw ) that establishes a decentralized fiscal system in the country. Relevance of the IV. Key to this mechanism is the variation in the levels of λ. When tax payers bear a high degree of loss aversion to ethnic diversity (i.e., high λ) their tax morale moves down. Conversely, if λ is low the ethnic aversion/tax morale nexus becomes less tight. Arguably, λ is very likely to co-move with individual, general attitude in taking risk, so that a measure of individual risk aversion is likely to be a good candidate to be explain part of AED variation. Exclusion restriction. Our exclusion restriction requires that the only channel through which risk aversion affects tax morale is via λ. That is to say, that risk aversion has to be directly unrelated with tax morale. To see that, recall that the notion of tax morale is very specific, and it has been defined as the intrinsic motivation for complying with taxes, a motivation that falls outside the standard, expected utility framework (see, for example, Luttmer and Singhal, 2014). Moreover, it has been documented that in an environment without tax enforcement, individual contribution is not affected by its own risk aversion (see Dwenger et al., 2014), which is rather likely to arise when individuals decide to evade taxes under the threat of being charged by a fine. We therefore believe that risk aversion is a plausible candidate for instrumenting the effect of AED on tax morale. To do so, we measure risk aversion constructing a dummy which takes value 1 if the respondent answers 2 to the item C009 of the 2005 wave of the WVS: “Regardless of whether you’re actually looking for a job, which one would you, personally, place first if you were looking for a job? 1. A good income, 2. A safe job with no risk, 3. Working with people you like, 4. Doing an important job, 5. Do something for community” Overall, 34.56% of the total responds “a safe job with no risk” to question

2

C009. In Table B3 we present our estimations. In panel A we reproduce our baseline results presented in the paper (column 1 of Table 3) for comparison. In panel B we report our two stage least square estimations using our measure of risk aversion, and its interaction with our synthetic index of decentralization, as an instrument. Coefficients are very much alike than those estimated in panel A. Specifically, a one point increase in aversion to ethnic diversity predicts a 0.029 decreases in tax morale. Our second prediction is confirmed as well: in decentralized systems, ethnic aversion does not explain significant variation in tax morale. In panel C and D, we present coefficients estimations from the first stage. Overall, coefficients are very significant as well as the Wald χ2 test that are very high. Taken together, using a two stage least square estimation, these results suggest that our estimations presented in column 1 of Table 3 are not bias by endogeneity or omission of unobservables.

2.2

Ius Sanguinis

In this subsection we bring further evidence in favor of the causal link going from ethnic aversion to tax morale. We do that by employing an additional instrument in alternative to risk aversion: individual attitudes toward the ius sanguinis. The ius sanguinis (in English, right of blood) is a Roman Law principle of nationality according to which citizenship is not determined by place of birth but because the new born has at least one parent who is citizen of the state. It opposes to ius soli (in English, right of soil) which states that new born in the territory of a given state are automatically eligible to receive the state’s nationality. Relevance of the IV. Attitudes toward citizenship requirements are very likely to explain individual attitudes toward migrants and, more generally, toward the “other.” The idea is that those people in favor of the ius sanguinis are also in favor of setting up barriers that discourage incoming migrant flow (as their offspring will not take the arrival country citizenship), and should be more averted to ethnic diversity as well. A measure of individual attitudes toward ius sanguinis, capturing this mechanism, should therefore be an eligible candidate to explain variation of AED. Exclusion restriction. Moreover, tax morale should be unrelated with attitude towards nationality law. Beliefs over ius sanguinis and ius soli are related to the ideological and political view of the world of the respondent (right wing versus left wing) and should be totally uncorrelated with the intrinsic motivation of paying taxes if not through our mechanism passing by ethnic aversion. With this idea in mind we therefore propose an IV strategy through which we instrument AED with attitude toward ius sanguinis. To measure these attitudes we construct a dummy variable taking value 1 if the respondent answers 1 or 2 to the question G028 of the 2005 wave of the WVS: “In your opinion, how important should the following be as requirements for somebody seeking citizenship of your country? 1. Very important, 2. Rather important, 3. Not important.” Overall, 62.90% of the total responds either “very important” or “rather important” to question G028. In Table B4 we present our estimations, which is organized in exactly the same way of Table B3. Therefore, panel A reproduces the baseline analysis that we have presented in the paper (column 1 of Table 3). In panel B we report our two stage least square estimations using our measure of attitude toward ius sanguinis as an instrument of AED and its interaction with decentralization for instrumenting AEDijk ×Deck . Also in this case, 3

coefficients are very closed to those estimated in panel A. Specifically, a one point increase in aversion to ethnic diversity predicts a 0.029 decreases in tax morale. The coefficient of the interaction term slightly increases, meaning that decentralized countries have on average a population with a 0.037 more on the 0-10 scale of tax morale. Finally, in panel C and D, we present the first stage estimations. Also in this case, our coefficients are very significant as well as the Wald χ2 test that are very high, meaning that our instrument is capable of explain a significant fraction of variation of AED. Once again, we can conclude that our estimations presented in column 1 of Table 3 are not bias by endogeneity or omission of unobservables.

2.3

Fully Identified Models

Finally, we run two fully identified models. In the first one we use risk aversion and Area, the instrument we use in the paper to address endogeneity issues arising from tax morale to decentralization, to identify the effect of ethnic aversion in centralized and decentralized countries. In the second one we use ius sanguinis and Area. The results related to the first exercise are presented in Table B5, which is organized in the following way: as before Panel A and Panel B report column 1 of Table 3 and the two stage least squares estimation. The comparison between the two columns shows quasiidentical figures. Panels C, D, and E present the three first stage estimations. All together, they report very high Wald χ2 statistics which exclude any potential weakness of our two instruments. Finally, Panel F presents reduced form estimations. The results obtained from employed ius sanguinis and Area simultaneously as instruments are presented in Table B6. The organization of Table B6 is identical of that of Table B5. Also in this case the comparison between column 1, that reports column 1 of Table 3, and column 2, that reports the two stage least squares estimation, shows similar figures in magnitude. The coefficient of the aversion to ethnic diversity only slightly reduced, while its interaction with decentralization is quasi-identical. As illustrated by the high Wald χ2 statistics, the first stages (Panels D, E, and F) are all very robust and the instruments not weak. Finally, Panel F presents reduced form estimations.

3

A Log-Lin Estimation

In this section, we present an alternative empirical specification of that analyzed in the paper (see equation 13 of the paper). The new specification takes into account the scarce individual variation in tax morale and, more importantly, the negative skewed shape of its distribution. Indeed, according to the WVS, individuals mostly respond that cheating on taxes is never justifiable. More precisely, we count 60.56% individuals responding 10 (“cheating on taxes is never justifiable”), 12.04% responding 9, up to just 1.98% declaring that cheating on taxes is always justifiable (taxmorale = 1). As Figure B2 shows, tax morale distribution is negative skewed (skewness is −1.94) and its mean is close to 10 (the mean is 8.74). To account for its asymmetric distribution we proceed in different steps. We first take a logarithmic transformation of the WVS item F116 (cheating on taxes). We then rescale it so as higher values are taken to mean less inclination to cheat on taxes—i.e., tax morale. The new tax morale measure plotted on a log-scale shows now a substantial variation (see Figure B3). Finally, we estimate a log-linear multilevel model, presented in Table B7. Recall that,

4

for small values of the coefficient, 100 ∗ β1 can be interpreted as the percentage increase in tax morale when ethnic aversion increases by one point. All the estimates reported in the first row of Table B7 feature a sizable effect of aversion to ethnic diversity on tax morale. Specifically, when ethnic aversion increases by one point the magnitude of variation in tax morale goes from a 1.2% decrease, when decentralization is autonomy or f ederal, to a 2% decrease when we measure decentralization employing vertical imbalance. Looking at the two ends of the distribution of aversion to ethnic diversity, this also means that an individual with the highest degree of aversion (i.e., AED = 10) has an estimated degree of tax morale which is 12% smaller, in the most conservative model, or 20% smaller, in the most favorable estimation, than the same peer with AED = 0. It is worth to mention, finally, that our linear-linear multilevel estimates are similar in magnitude to those estimated in prior contributions (e.g., Lago-Pe˜ nas and Lago-Pe˜ nas, 2010; and Halla and Shneider, 2014) that have investigated the main determinants of tax morale. For example, Table 3 at p. 450 in Lago-Pe˜ nas and Lago-Pe˜ nas (2010) reports an estimated coefficient of the variable Satisfaction with democracy (defined in a 0-10 scale, as our index of individuals’ attitude towards ethnic diversity) that ranges from 0.032 to 0.046 (i.e., the same size of our coefficient of ethnic aversion); the variables Religion and Trust in politicians (both defined in a 0-10 scale) have estimated coefficients respectively ranging from 0.027 to 0.034 and from 0.010 to 0.175; again, quite similar to our estimated coefficients of ethnic aversion.

4

A Principal Component Analysis of Fiscal Decentralization

In this section we propose an alternative technique to synthesize the five measures of decentralization that we extensively employed in the paper. Specifically, we use an orthogonal transformation to convert the five measure of decentralization in a principal component, P Ck . However, since our panel dataset is not balanced we first address the problem of joint missing values. We use two different, though similar in principal, strategies. For the continuous measures of decentralization (the subnational expenditures as a percentage of total expenditures, the subnational revenues as a percentage of total revenues, and the intergovernmental transfers as a share of sub-national expenditures) we use the “hotdeck ” strategy described in Rubin and Schenker (1986). Specifically, we use information within each 11 strata P5 of scountries defined by a stratification of the cross-country distribution according to s=1 Dk . Recall that Dks is a variable equals to 1 if the country k is a decentralized one according to the index s (see its definition introduced in equation 11 and 12 of the paper). Otherwise it takes value P5 −1. We present the country distribution of strata in Figure B4. When s=1 Dks = 5 all the five measures agree that country k is a decentralized P one. We have 3 out of 163 countries collecting a score of 5. On the other end, when 5s=1 Dks = −5 all the five measures agree that country k is a centralized one. We have 11 out of 163 countries collecting a score of −5. Within each stratum we are therefore confident that countries are consistently judged as decentralized or not. P The hotdeck procedure replaces missing lines in each stratum with same score 5s=1 Dks by lines sampled from the residual complete lines in the same stratum, using the approximate Bayesian bootstrap method of Rubin and Schenker (1986). As for the 0-1 variables, f ederal and autonomy, we simply proceed by replacing their missing values using our synthetic measure, Deck . 33 replacements are done for autonomy,

5

none for f ederal which has the same observations of Deck . We select the principal component with the highest eigenvalue (2.44). The pairwise correlation between the P Ck and the other measures is as follows:

P Ck autonk f ederalk snete7200k snrtr7200k vi7200k ∗

p < 0.10,

∗∗

pc1 1.000 0.585∗∗∗ 0.731∗∗∗ 0.882∗∗∗ 0.848∗∗∗ -0.245∗∗∗ p < 0.05,

∗∗∗

p < 0.01

In Table B8 we present our results for the two main analyzes in the paper: our baseline specification of equation (13) and that of equation (14). In Figure B5 we also plot the marginal effects for different indexes of the P Ck distribution.

5

Other measures of decentralization

In this section, as a further robustness check, we replicate our results using additional measures of decentralization from Fan et al. (2009), that are less common in the literature. The measures we include here reflect the administrative, electoral and personnel dimensions of decentralization, and do not explicitly capture our mechanism, mainly related to the devolution of decision-making about regulation or taxation to subnational levels. More precisely, we use here the following alternative measures: • tier that coded administrations with a state executive body that is funded from the public budget, has authority over several public services, and with territorial jurisdiction; • bottier that codifies the number of bottom level administrative units; • sizebot that collect information on the average size (in squared kilometers) of the bottom level administrative units; • botel is a variable that takes on 1 if executives at bottom tier are directly elected; • secel is a variable that takes on 1 if executives at second lowest tier are directly elected; • subgemp measures the non-central government employment as % of total government employment. Summary statistics are reported in Table B9. In Table B10 we present our estimations using regression (13) in the paper. All these alternative specifications bring further evidence in favor of the negative link between aversion to ethnic diversity and tax morale. All the six columns report negative and significant estimations for β1 . Looking at the marginal effects, presented in Figure B6, one can see that our second prediction is robust to five out of six of these alternative measures. Marginal effects of aversion to ethnic diversity on tax 6

morale move from the negative sign rightward when tier, bottier, botel, secel, and subgemp increase. When we use variation of sizebot our prediction is not confirmed. But this is not surprise: sizebot, in fact, measures the average size of the bottom level administrative units—something very distant from our mechanism.

6

Marginal effects of ethnic aversion on tax morale for different degree of ethnic diversity

In Figures B7 and B8, we illustrate the marginal effects of ethnic aversion on tax morale for different values of ethnic fractionalization, using two additional dichotomous measures of decentralization: federal (column 2 of Table 4) and autonomy (column 3 of Table 4). The two figures are constructed in exactly the same way as Figure 4 in the paper, where we employ our synthetic measure of decentralization to split the country sample. The leftmost panel reports the marginal effects in centralized countries, while the rightmost in decentralized countries. The marginal effects computed at the median level of ethnic fractionalization (p50) are those presented in the second and third panel of Figure 2, respectively. From the two exercises, one may see that an increase in ethnic fractionalization weakens the negative effect of ethnic aversion on tax morale in centralized countries, as it was the case when we used the synthetic measure of decentralization. In decentralized countries, we do find as well that the effect of ethnic aversion on tax morale does not vary statistically with co-movements of ethnic fractionalization. This additional evidence confirms our findings in Figure 4 in the paper.

7

Data Description

Data sources are described in Table B11.

References [1] Dwenger N., Kleven H., Rasul I., Rincke J. (2016). Extrinsic and Intrinsic Motivations for Tax Compliance: Evidence from a Field Experiment in Germany. American Economic Journal: Economic Policy, 8(3), 203-32. [2] Halla, M. and Schneider F.G. (2014). Taxes and Benefits:Two Options to Cheat on the State. Oxford Bulletin of Economics and Statistics, 76(3), 411-431. [3] Lago-Pe˜ nas, I., Lago-Pe˜ nas, S. (2010). The determinants of tax morale in comparative perspective: Evidence from European countries. European Journal of Political Economy, 26, 441-453. [4] Rubin D.B., Schenker N. (1986). Multiple Imputation for Interval Estimation from Simple Random Samples with Ignorable Nonresponse. Journal of the American Statistical Association , 81, 366-374.

7

Table B1: Tax morale, decentralization, ethnic aversion, and trust. AEDijk AEDijk × Deck Deck

(1)

(2)

(3)

(4)

(5)

(6)

(7)

-0.030∗∗∗

-0.029∗∗∗

-0.029∗∗∗

-0.030∗∗∗

-0.027∗∗∗

-0.026∗∗∗

(0.005)

(0.005)

(0.005)

(0.005)

(0.005)

(0.005)

-0.028∗∗∗ (0.005)

0.033∗∗∗

0.034∗∗∗

0.033∗∗∗

0.033∗∗∗

0.035∗∗∗

(0.010)

(0.010)

(0.010)

(0.010)

(0.010)

0.035∗∗∗ (0.010)

0.032∗∗∗ (0.010)

-0.355∗∗ (0.173)

-0.382∗∗ (0.169)

-0.407∗∗ (0.170)

-0.368∗∗ (0.173)

-0.408∗∗ (0.172)

-0.411∗∗ (0.173)

-0.327∗ (0.170)

0.139∗∗∗ (0.016)

trust

0.175∗∗∗ (0.017)

trust personally

0.027∗ (0.016)

trust first time

0.124∗∗∗ (0.016)

trust other rel.

0.133∗∗∗ (0.016)

trust other nat.

0.117∗∗∗ (0.015)

trust gov. N Njk (Nk ) 2 Rijk

30647 219(32) 0.016

30647 219(32) 0.018

30647 219(32) 0.019

30647 219(32) 0.016

30647 219(32) 0.018

30647 219(32) 0.018

30647 219(32) 0.019

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity, while Deck is our synthetic measure of decentralization. The other explanatory variables are, in order: trust your neighborhood (item G007 18 B), trust people you know personally (item G007 33 B), trust people you meet for the first time (item G007 34 B), trust people of another religion (item G007 35 B), trust people of another nationality (item G007 36 B) and confidence in government (item E069 11). All the columns include ethnic, language, and religion fragmentation as well as individuals controls and dummies that take on value of 1 if the observation has been replaced by the sample mean. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

8

Table B2: Tax morale, decentralization, ethnic aversion, and trust. (1)

(2)

(3)

(4)

(5)

(6)

AEDijk

-0.027∗∗∗ (0.005)

-0.023∗∗∗ (0.005)

-0.022∗∗∗ (0.005)

-0.036∗∗∗ (0.009)

-0.030∗∗∗ (0.009)

-0.039∗∗ (0.017)

AEDijk × Deck

0.033∗∗∗ (0.010)

Deck

-0.353∗∗ (0.167)

AEDijk × f ederalk

0.025∗∗ (0.011)

f ederalk

-0.432∗ (0.236)

AEDijk × autonomyk

0.027∗∗ (0.011)

autonomyk

-0.545 (0.333)

AEDijk × subn.expk

0.001∗∗∗ (0.000)

subn.expk

-0.012∗∗ (0.005)

AEDijk × subn.revk

0.001∗∗ (0.000)

subn.revk

-0.012∗∗ (0.006)

AEDijk × vert.imbk

0.000 (0.000)

vert.imbk

0.001 (0.005)

trust

0.124∗∗∗ (0.016)

0.123∗∗∗ (0.016)

0.123∗∗∗ (0.017)

0.120∗∗∗ (0.018)

0.117∗∗∗ (0.018)

0.116∗∗∗ (0.018)

trust gov.

0.102∗∗∗ (0.015)

0.101∗∗∗ (0.015)

0.098∗∗∗ (0.015)

0.111∗∗∗ (0.016)

0.114∗∗∗ (0.017)

0.112∗∗∗ (0.017)

N Njk (Nk ) 2 Rijk

30647 219(32) 0.020

30647 219(32) 0.020

29250 205(30) 0.021

25524 189(27) 0.022

24940 182(26) 0.023

25076 185(25) 0.022

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity, trust is trust your neighborhood (item G007 18 B) and trust gov. is confidence in government (item E069 11). All the columns include ethnic, language, and religion fragmentation as well as individuals controls and dummies that take on value of 1 if the observation has been replaced by the sample mean. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

9

Marginal Effects of Ethnic Aversion on Tax Morale

Dec 0

0 -.04

-.02

0

.02

0 -.04

p90

-.02

0

.02

-.03 -.02 -.01

p90

0

.01

.02

p90

p75

p50

p25

vertical imbalance

p75 subnational revenues

subnational expenditures

1 autonomy

1 federal

1

p75

p50 p25 p10

p10 -.06 -.04 -.02

0

.02 .04

p50

p25

p10 -.04

-.02

0

.02

.04

-.05 -.04 -.03 -.02 -.01

0

0

.2

Fraction

.4

.6

Figure B1: Marginal effects of aversion to ethnic diversity on tax morale to a change in decentralization. Further individual controls of social capital. x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of each of the three continuous measures of decentralization from the bottom to the top.

0

2

4

6

8

Tax Morale

Figure B2: Tax Morale distribution.

10

10

Table B3: Instrumental Variable Approach. Risk Aversion. (1)

(2)

(3)

(4)

(5)

Panel A – Mixed Model AEDijk -0.0302∗∗∗ (0.0054) AEDijk × Deck

0.0327∗∗∗ (0.0099)

Panel B – IV Mixed Model AEDijk

-0.0299∗∗∗ (0.0054) 0.0327∗∗∗ (0.0100)

AEDijk × Deck

Panel C – First Stage (dep. var AEDijk ) Risk Aversion

0.0708∗∗ (0.0304)

Panel D – First Stage (dep. var AEDijk × Deck ) Risk Aversion ×Deck

0.3043∗∗∗ (0.0321)

Panel E – Reduced Form Risk Aversion

0.1022∗∗∗ (0.0285)

Risk Aversion ×Deck Observations Wald χ2 Statistics

0.0022 (0.0551) 30,647

30,388

30,388 506.54

30,388 654.15

30,388

Notes: Dependent variable is tax morale in columns 1, 2, and 5. AEDijk is the individual degree of aversion to ethnic diversity, while Deck is our synthetic measure of decentralization. Risk Aversion is defined as a dummy which takes value 1 if the respondent answers 2 to the item C009 of the 2005 wave of the WVS: “Regardless of whether you’re actually looking for a job, which one would you, personally, place first if you were looking for a job? 1. A good income, 2. A safe job with no risk, 3. Working with people you like, 4. Doing an important job, 5. Do something for community.” All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

11

Table B4: Instrumental Variable Approach. Attitude toward ius sanguinis. (1)

(2)

(3)

(4)

(5)

Panel A – Mixed Model AEDijk -0.0302∗∗∗ (0.0054) AEDijk × Deck

0.0327∗∗∗ (0.0099)

Panel B – IV Mixed Model AEDijk

-0.0292∗∗∗ (0.0055) 0.0373∗∗∗ (0.0102)

AEDijk × Deck

Panel C – First Stage (dep. var AEDijk ) Ius Sanguinis

0.5296∗∗∗ (0.0330)

Panel D – First Stage (dep. var AEDijk × Deck ) Ius Sanguinis ×Deck

0.8699∗∗∗ (0.0331)

Panel E – Reduced Form Ius Sanguinis

-0.1076∗∗∗ (0.0316) -0.2070∗∗∗ (0.0583)

Ius Sanguinis ×Deck Observations Wald χ2 Statistics

30,647

30,117

30,117 772.72

30,117 1,304.56

30,117

Notes: Dependent variable is tax morale in columns 1, 2, and 5. AEDijk is the individual degree of aversion to ethnic diversity, while Deck is our synthetic measure of decentralization. Risk Aversion is defined as a dummy variable taking value 1 if the respondent answers 1 or 2 to the question G028 of the 2005 wave of the WVS: “In your opinion, how important should the following be as requirements for somebody seeking citizenship of your country? 1. Very important, 2. Rather important, 3. Not important.” All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

12

Table B5: Instrumental Variable Approach. Full Model (Risk aversion and Area). (1)

(2)

(3)

(4)

(5)

(6)

Panel A – Mixed Model AEDijk -0.0302∗∗∗ (0.0054) AEDijk × Deck

0.0327∗∗∗ (0.0099)

Panel B – IV Mixed Model AEDijk

-0.0304∗∗∗ (0.0055) 0.0342∗∗∗ (0.0100)

AEDijk × Deck

Panel C – First Stage (dep. var AEDijk ) Risk Aversion

0.0707∗∗ (0.0305)

Area

-0.0000 (0.0000)

Panel D – First Stage (dep. var AEDijk × Deck ) Risk Aversion ×Deck

0.0000∗∗∗ (0.0000) 0.0000∗∗ (0.0000)

Area Panel E – First Stage (dep. var Deck ) Area

0.0000∗∗∗ (0.0000)

Panel F – Reduced Form Risk Aversion

0.1011∗∗∗ (0.0283)

Risk Aversion × Area

0.0000 (0.0000)

Area

-0.0000 (0.0000)

Observations Wald χ2 Statistics

30,647

30,388

30,388 506.63

30,388 303.17

30,388 380.15

30,388

Notes: Dependent variable is tax morale in columns 1, 2, and 6. AEDijk is the individual degree of aversion to ethnic diversity, while Deck is our synthetic measure of decentralization. Risk Aversion is defined as a dummy which takes value 1 if the respondent answers 2 to the item C009 of the 2005 wave of the WVS: “Regardless of whether you’re actually looking for a job, which one would you, personally, place first if you were looking for a job? 1. A good income, 2. A safe job with no risk, 3. Working with people you like, 4. Doing an important job, 5. Do something for community.” Areak is the size in km2 of the geographical area of the country. All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

13

Table B6: Instrumental Variable Approach. Full Model (ius sanguinis and Area). (1)

(2)

(3)

(4)

(5)

(6)

Panel A – Mixed Model AEDijk -0.0302∗∗∗ (0.0054) AEDijk × Deck

0.0327∗∗∗ (0.0099)

Panel B – IV Mixed Model AEDijk

-0.0277∗∗∗ (0.0055) 0.0322∗∗∗ (0.0101)

AEDijk × Deck

Panel C – First Stage (dep. var AEDijk ) ius sanguinis

0.5292∗∗∗ (0.0330)

Area

-0.0000 (0.0000)

Panel D – First Stage (dep. var AEDijk × Deck ) ius sanguinis ×Deck

0.0000∗∗∗ (0.0000)

Area

0.0000 (0.0000)

Panel E – First Stage (dep. var Deck ) Area

0.0000∗∗∗ (0.0000)

Panel F – Reduced Form ius sanguinis

-0.1312∗∗∗ (0.0308)

ius sanguinis × Area

-0.0000∗∗ (0.0000)

Area

-0.0000 (0.0000)

Observations Wald χ2 Statistics

30,647

30,117

30,117 772.10

30,117 575.30

30,117 384.98

30,117

Notes: Dependent variable is tax morale in columns 1, 2, and 6. AEDijk is the individual degree of aversion to ethnic diversity, while Deck is our synthetic measure of decentralization. Risk Aversion is defined as a dummy variable taking value 1 if the respondent answers 1 or 2 to the question G028 of the 2005 wave of the WVS: “In your opinion, how important should the following be as requirements for somebody seeking citizenship of your country? 1. Very important, 2. Rather important, 3. Not important.” Areak is the size in km2 of the geographical area of the country. All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

14

.6 .4 Fraction .2 0 -2.5

-2

-1.5 -1 Tax Morale (log)

-.5

0

0

10

Number of countries 20 30

40

50

Figure B3: Tax Morale distribution. Logarithmic transformation.

-5

0 strata

Figure B4: Cross-Country distribution of Strata.

15

5

Table B7: A log-linear multilevel estimation. AEDijk

(1)

(2)

(3)

(4)

(5)

(6)

-0.013∗∗∗ (0.002)

-0.012∗∗∗ (0.002)

-0.012∗∗∗ (0.002)

-0.016∗∗∗ (0.003)

-0.014∗∗∗ (0.003)

-0.020∗∗∗ (0.005)

AEDijk × Deck

0.008∗∗ (0.003)

Deck

-0.139∗∗ (0.060)

AEDijk × f ederalk

0.006∗ (0.004)

f ederalk

-0.148∗ (0.086)

AEDijk × autonomyk

0.008∗∗ (0.004)

autonomyk

-0.017 (0.074)

AEDijk × subn.expk

0.000∗∗ (0.000)

subn.expk

-0.003∗ (0.002)

AEDijk × subn.revk

0.000 (0.000)

subn.revk

-0.003 (0.002)

AEDijk × vert.imbk

0.000∗ (0.000)

vert.imbk

0.001 (0.002)

Observations Njk (Nk ) 2 Rijk

30647 219(32) 0.020

30647 219(32) 0.020

29250 205(30) 0.020

25524 189(27) 0.022

24940 182(26) 0.023

25076 185(25) 0.023

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity. All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

16

Table B8: Principal Component. (1)

(2)

AEDijk

-0.024∗∗∗ (0.005)

-0.040∗∗∗ (0.010)

AEDijk × P Ck

0.007∗∗∗ (0.002)

0.005 (0.006)

P Ck

-0.151∗∗∗ (0.051)

-0.011 (0.076)

-1.038∗ (0.559)

-0.674 (0.493)

EthF ractk AEDijk × EthF ractk

0.041∗ (0.021)

AEDijk × EthF ractk × P Ck

0.005 (0.013)

EthF ractk × P Ck

-0.406 (0.274)

N Njk (Nk ) 2 Rijk

30647 219(32) 0.015

30647 219(32) 0.016

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity, while P Ck stands for the principal component of the five measures of decentralization that we employed in the paper. All the columns include ethnic, language, and religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

Table B9: Summary statistics of alternative measures of decentralization. tier bottier sizebot botel secel subgemp

mean

sd

min

max

obs.

3.60 15519.92 1.33 0.83 0.47 7.23

0.90 40529.66 2.93 0.35 0.49 5.20

2.00 13.00 0.01 0.00 0.00 0.66

6.00 237333.00 13.21 1.00 1.00 18.37

41 38 38 32 30 31

17

Marginal Effects of Ethnic Aversion on Tax Morale p90

Principal Component

Principal Component

p90

p75

p50

p25 p10

p75

p50

p25 p10 -.04

-.02

0

.02

-.06 -.04 -.02

0

.02

Figure B5: Marginal effects of aversion to ethnic diversity on tax morale to a change in decentralization (principal component). x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of the distribution of the principal component of decentralization from the bottom to the top.

Table B10: Tax morale, decentralization, and ethnic aversion. Alternative measures of decentralization. (1) Deck is tier

(2) Deck is bottier

(3) Deck is sizebot

(4) Deck is botel

(5) Deck is secel

(6) Deck is subgemp

AEDijk

-0.108∗∗∗ (0.018)

-0.035∗∗∗ (0.005)

-0.011∗∗ (0.005)

-0.043∗∗∗ (0.010)

-0.047∗∗∗ (0.007)

-0.007 (0.007)

AEDijk × Deck

0.023∗∗∗ (0.004)

0.000∗∗∗ (0.000)

-0.004∗∗∗ (0.002)

0.010 (0.012)

0.001∗∗∗ (0.012)

-0.003∗ (0.002)

Deck

-0.034 (0.097)

-0.000∗∗ (0.000)

-0.010 (0.027)

-0.323 (0.261)

0.013 (0.192)

0.064∗∗ (0.026)

N Njk (Nk ) 2 Rijk log − likelihood

29359 210(31) 0.016 -62,295

27838 198(29) 0.015 -58,997

27838 198(29) 0.016 -59,033

23745 163(23) 0.018 -50,383

21203 146(21) 0.019 -44,793

23461 179(24) 0.018 -53,014

Notes: Dependent variable is tax morale. AEDijk is the individual degree of aversion to ethnic diversity, while Deck stands for one of the six alternative measures of decentralization. All the columns include ethnic, language, religion fragmentation as well as individuals controls. Robust standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

18

Marginal Effects of Ethnic Aversion on Tax Morale p90

sizebot

p90

bottier

tiers

5

4

p75

p75

p50 p25 p10

p50 p25 p10

3 -.06

-.04

-.02

0

.02

-.05 -.04 -.03 -.02 -.01

0

-.03

-.02

-.01

0

p90

p75 subgemp

1

botel

secel

1

0

p50

0 -.06

-.05

-.04

-.03

-.02

-.06

-.05

-.04

-.03

-.02

p25 p10 -.04 -.03 -.02 -.01

0

Figure B6: Marginal effects of aversion to ethnic diversity on tax morale to a change in decentralization (Alternative measures of decentralization). x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of each of each continuous measure of decentralization from the bottom to the top.

19

p90

p90

p75

p75 Ethnic Fragmentation

Ethnic Fragmentation

Marginal Effects of Ethnic Aversion on Tax Morale (Federal)

p50

p50

p25

p25

p10

p10 -.06 -.04 -.02 0 .02 Centralized Countries

-.1 -.05 0 .05 Decentralized Countries

Figure B7: Marginal effects of aversion to ethnic diversity on tax morale to a change in ethnic diversity (Federal). x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of each of each continuous measure of decentralization from the bottom to the top.

20

p90

p90

p75

p75 Ethnic Fragmentation

Ethnic Fragmentation

Marginal Effects of Ethnic Aversion on Tax Morale (Autonomy)

p50

p50

p25

p25

p10

p10 -.05-.04-.03-.02-.01 0 Centralized Countries

-.04 -.02 0 .02 .04 Decentralized Countries

Figure B8: Marginal effects of aversion to ethnic diversity on tax morale to a change in ethnic diversity (Autonomy). x-axis reports the marginal effect. y-axis is the 10th percentile, 25th percentile, 50th percentile, 75th percentile, 90th percentile of each of each continuous measure of decentralization from the bottom to the top.

21

Table B11: Data description Tax Morale

Trust your government

Question F116: Cheating on taxes, if you have a chance, is: 1 = never justifiable, 10 = always justifiable. Question G032: Turning to the question of ethnic diversity, with which of the following views do you agree? Ethnic diversity erodes a country’s unity (scale 1); Ethnic diversity enriches my life (scale 10). Question X001. Question X003. Question X007. Question X025: What is the highest educational level that you have attained? 1 = inadequately completed elementary education, 8 = upper-level tertiary certificate. Question X046: Socio-econonic status of respondent. 1 = Upper class, 5 = Lower class. Question X047: 1 = Lower step, 10 = Higher step. Question X049: 1 = size < 2k, 8 = size > 500k. Question G007 18: 1 = Not trust at all, 4 = Trust completely. Question E069 11: 1 = Not at all, 4 = A great deal.

Dec

Synthetic index of decentralization.

Autonomy

Dummy calculated on whether the constitution assigned at least one policy area exclusively to subnational governments or gave subnational governments exclusive authority to legislate on matters not constitutionally assigned to any level. Dummy variable on whether the country is federal or not.

Ethnic Aversion

Sex Age Marital Status Education

Social Class Scales of Income Size of Town Trust your neighborhood

Federal Subnat. Expenditure Subnat. Revenues Vertical Imbalance

Subnational expenditures as a percentage of total expenditures. Subnational revenues as a percentage of total revenues. Intergovernmental transfers as a share of sub-national expenditures.

Ethnic Fract. Language Fract. Religion Fract.

Index of ethnic fractionalization. Index of language fractionalization. Index of religious fractionalization.

GDP per capita (log)

GDP per capita in 2005 based on purchasing power parity (PPP) in 2011. Total population counted in 2005. Tax Revenues in percentage of GDP, 2005.

Population (log) Tax Revenue (% GDP) Corruption Index

Gov. Effectiveness Index

Party age

Fract. of Govt. parties

Legal Origin

Index on the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as capture of the state by elites and private interests. Index on the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policies. Average of the ages of the first government party, second government party, and 1st opposition party between 1972 and 2000 (thousands of years). Probability that two members of parliament picked at random from among the government parties will be of different parties. Set of dummies indicating the country’s legal origin.

22

2005 World Value Survey wave

Our elaboration Treisman (2007)

IMF – Government Finance Statistics

Alesina et al. (2003)

World Development Indicators

World Government Indicators

Enikolopov and Zhuravskaya (2007)

La Porta et al. (1999)