• No results found

Is Trust Affected by Income Inequality?

N/A
N/A
Protected

Academic year: 2022

Share "Is Trust Affected by Income Inequality?"

Copied!
36
0
0

Loading.... (view fulltext now)

Full text

(1)

Department of Economics Uppsala University

Bachelor’s Thesis Author: Saara Haikonen Supervisor: Che-Yuan Liang Autumn 2020

Is Trust Affected by Income Inequality?

Abstract

Contributing to a growing body of literature on the determinants of trust, I explore how municipal income inequality affects trust at local level (“localized trust”) and at general level (“generalized trust”). Using Swedish population and survey data 1996-2014, I find a negative relationship between a vast majority of inequality and trust variables studied. Even though the association is often statistically significant, the practical effects are nevertheless low. In accordance with previous research, I find that standard level changes in income inequality do not affect municipal levels of generalized trust. However, large increases in municipal poverty rates seem to decrease localized trust, when the rates are based on disposable income.

Inequality in disposable income is more strongly associated with the two forms of trust than pre-tax inequality, suggesting that trust can be influenced by means of income redistribution.

Keywords: Income inequality; Trust; Social capital

(2)

Table of Contents

1. Introduction 3

2. Previous Literature 8

2.1 Effects of Economic Inequality 8

2.2 Determinants of Trust 9

2.3 The Effect of Economic Inequality on Trust 9

3. Data 11

3.1 Income Inequality 11

3.2 Trust 14

4. Empirical Strategy 18

4.1 First-Difference Strategy 19

4.2 Control Variables 21

5. Results 24

5.1 General Income Dispersion 24

5.2 Other Inequality Measures 28

6. Conclusion 32

References 34

(3)

1. Introduction

Economic inequality and trust are fundamental concepts in the social sciences. Interpersonal trust is thought to be one of the main components of social capital (Knack & Keefer 1997), which increases and facilitates cooperation in different social groups, and is thus a foundation of a society. The effects of trust have been extensively studied, and trust is found to be associated with, e.g., higher rates of investment and economic growth1, democratic stability (Uslaner 2002), and less corruption (ibid). Trust is also valuable in itself. Social relationships and psychic stability require trust — constant scepticism about others’ intentions would make it impossible to live in a complicated world (McLeod 2020). Accordingly, at the individual level, trust correlates strongly with happiness (Uslaner 2002).

Economic inequality attracts great attention mainly because of its consequences: one of the main arguments for reducing inequality is inequality’s possible negative effects in areas like politics, crime and health. Among other things, empirical research has established that differences in income inequality are an important factor in explaining cross-country differences in social capital.2 At the micro level, however, the issue remains relatively unexplored and the conclusion unclear. Although inequality seems to better explain differences in trust between countries than within them, different methods to measure economic inequality have given different results (Barone & Mocetti 2016, Gustavsson & Jordahl 2007). Whereas income dispersion itself shows no significant association with trust, there is nonetheless evidence that trust may be negatively affected by income inequality in subgroups like the lower half of the income distribution (ibid.).

Although the question of the exact mechanisms at work is open, economic inequality could affect trust in several ways.3 Inequality by definition increases heterogeneity among individuals, and differences between people seem to generate distrust. Increased inequality could also lead to tensions and increased perception of injustice between people, decreasing trust.4 Some examples of other possible mechanisms are inequality’s effect on migration patterns or human behaviour, which in turn could have an indirect and ambiguous effect on trust. Theoretically, trust in others is both useful and, by definition, uncertain. An optimal amount of trust can thus be seen as the product of the payoff of trust in others and the probability

1 E.g.Knack & Keefer 1997, Dincer & Uslaner 2010, and Zak & Knack 2001. For how historical levels of trust explain current levels of economic development, see Tabellini 2010 and Algan & Cahuc 2010.

2 See e.g. Barone & Mocetti 2016, Knack & Keefer 1997, Bjørnskov 2006, and Rothstein & Uslaner 2005.

3 The mechanisms are discussed especially in Alesina & La Ferrara 2002, and Barone & Mocetti 2016.

4 However, this is not the only possibility. Depending on, e.g., base levels of inequality and economic mobility, increased income inequality could increase perceptions of fairness, and thus trust. For instance, it is sometimes argued that certain differences in income distribution are acceptable to account for differences in productivity.

Previous discussions focus nevertheless exclusively on the positive assocition with perceptions of injustice.

(4)

that other individuals are trustworthy. Correspondingly, changes in heterogeneity or perceptions of injustice could directly affect both the value of trust and the perceived probability that others can be trusted. Homogeneity may even assimilate people’s preferences in itself, and thus automatically eliminate some of the risks related to trust.

Interpersonal trust is often divided in two subgroups: particularized and generalized trust.

Whereas particularized trusters rely only on people they know personally, generalized trust is the ability to trust people in general. The latter form of trust has been the focus in a vast majority of previous studies, and at the heart of the definition of social capital. Even though particularized trust is often economically less interesting than generalized trust, the latter is thought to be based on the former as generalized trusters by definition trust both those they know and those they do not (Newton et al. 2018). Consequently, virtually all generalized trusters are even particularized trusters (ibid.). Further, between the two main forms of trust there are myriad other types of interpersonal trust, which are focused on people neither completely known nor unknown. Some examples of these possible forms are trust in people of the same nationality, religion, or origin.

The purpose of this thesis is to investigate if income inequality affects trust in general (generalized trust) and at a local level (localized trust) in Sweden. Generalized trust is interesting because of its vast effects on society and economic phenomena, and it is the form studied by all important studies on income inequality and trust. Localized trust is trust in people in a local area. Leigh (2006) finds that local income inequality has a greater effect on localized than generalized trust, and I test the hypothesis at municipality level. Previous cross-country studies, the study using several inequality measures (Gustavsson & Jordahl 2007), and the majority of possible mechanisms all predict the effect of income inequality on trust to be negative. However, as previous studies at the micro level have not found any clear association between Gini coefficient and trust, I do not expect income dispersion to be the leading factor.

I measure inequality by using eighteen different measures, half of which are based on pre-tax incomes and half on disposable incomes. The intention is to catch different aspects of changes in income inequality, and to get indications of policies that can be used to increase trust. First, I investigate if trust is affected by income dispersion. This is the most general of the inequality measures and shows the standard deviation of municipal incomes. All previous studies have measured income dispersion, but with a Gini coefficient, which seems to be unrelated to generalized trust. The study by Gustavsson and Jordahl (2007) also measures inequality as the ratio of the lowest incomes to median or the highest incomes, and the ratio of the highest incomes to median incomes. This thesis distinguishes between the incomes of the highest/lowest and the two highest/lowest income deciles, relating all these incomes to median incomes. These distinctions show whether trust is affected more by changes in relative

(5)

poverty/affluence among the very richest/poorest, or in a broader group of high- or low-income earners. Further, I study relative incomes and the shares of the poor and affluent separately.

This is relevant as the shares of high- and low-income earners might affect trust differently than their relative incomes. Even the policy implications are partly different. To change poverty or affluence rates, the focus should be in decreasing/increasing the share of people under/above a certain income cut-off, whereas relative incomes can be changed by affecting the earnings of those with the lowest or the highest incomes. Although the possible policies probably overlap, this is still a contribution in respect to studies measuring only income dispersion.

This thesis concerns Sweden during the period from 1996 to 2014. The use of a relatively small and homogeneous country reduces the number of possible omitted variables compared to cross- country studies, or to studies concerning large countries like the USA. It also limits measurement errors caused by, e.g., cultural or institutional differences. Sweden is an interesting case when it comes to both economic inequality and trust, as it is one of the most egalitarian and trusting societies in the world (Therborn 2020, Holmberg & Rothstein 2015).

As will be shown later, Swedish trust levels remain relatively stable over the time period under investigation. However, having once been the country with the lowest level of income inequality, Sweden has since the 1980s experienced one of the largest increases in economic inequality in the Western world (Therborn 2020). Even this apparent change speaks for the use of Sweden. An additional level of variation in income inequality and trust comes with municipalities. Sweden is divided into 290 municipalities, all of which have their own local policies and, importantly, municipal tax rates. The large number of municipalities expands the sample size to more than 5500 observations, contributing thus to an increased precision.

Whereas both income inequality and trust vary between municipalities, it is not evident that the municipality level would be the most suitable for analysing the relationship between inequality and trust. It could for instance be argued that inequality at both national and international levels affect trust more than municipal inequality levels. However, even if national and international levels had a greater effect on trust, they still do not vary between municipalities, and thus affect all Swedes equally. Variations in municipal characteristics such as size, geography and segregation may affect the perceptions of inequality, and it is thus not obvious that a given level of inequality would have the same effect in all municipalities. To address this threat, I will later control the analyses for municipal population size. It is also possible that inequality in a local area would be more visible than municipal inequality. Despite this, it is unlikely that anyone would be completely indifferent to municipal inequality, which is probably reflected in local institutions, people and policies.

I use data from national SOM surveys and GeoSweden database. Trust is measured with an annual survey in which respondents indicate on a 0-10 scale how much they trust people in

(6)

general and in their local area. Data on economic inequality comes from a register containing the incomes of the entire Swedish population. I will study the time period between 1996 and 2014, and all the 290 municipalities are included in the analysis. The time period of 19 years and the use of municipal level contribute to previous research, which is often conducted at a more general level during a shorter period of time. Previous studies have mainly examined the effect of Gini coefficients on generalized trust. My study measures economic inequality in several new ways, and extends the analysis to effects on localized trust. Finally, I investigate inequality in pre-tax and disposable incomes separately. That is interesting especially from a policy perspective — taxes and subsidies can be directly affected by economic policy. For instance, if the effect of disposable income inequality on trust is larger than that of pre-tax inequality, economic policy could be used to build trust through redistribution of incomes.

Accordingly, if the effect of pre-tax inequality is larger, policies should change people’s earning capacity. This could be done by, e.g., directing more financial support to some groups of children, or raising the educational achievement of some type of youth.

The analysis is based on the first-difference method, which eliminates time-independent underlying factors using the change in both trust and economic inequality. Thus, as the institutional and cultural factors explaining some of the differences between municipal levels of income inequality and trust are eliminated, the change in trust is ideally caused only by changes in inequality. I study the effects while controlling for the base year's inequality and trust levels, as they may affect how inequality and trust change. The main regressions are based on more than 4000 observations of five-year changes. The municipal units used are notably smaller than those in the majority of previous studies, and enable a more precise analysis of the effect. Moreover, I control my models for several individual characteristics such as mean income, age, and foreign background. When finally adding the control for changes in the municipal population size, identification should rely on the income changes amongst residents only.

This study finds no significant effects of general income dispersion on generalized trust. The result is in line with previous research. However, income dispersion is weakly negatively associated with localized trust. With the exception of income dispersion, all inequality variables have a statistically significant negative association with generalized trust. Growing poverty and affluence rates decrease generalized trust more than changes in relative incomes.

In practice these effects are nevertheless of negligible size, and have economic relevance only if several inequality variables change parallelly. The association between income inequality and localized trust is quite similar, with the exception that the estimates are rarely statistically significant. Increase in poverty rates, based on disposable income, decreases localized trust three times more than generalized trust, and could in some cases be economically significant.

Taken together, inequality in disposable income is more negatively associated with trust than

(7)

inequality in pre-tax income, suggesting that income redistribution can have a role in building trust.

The thesis is structured as follows. Section 2 presents previous research on effects of economic inequality, determinants of trust, and the effect the former has on the latter. Section 3 describes the data I use, and Section 4 outlines the empirical strategy. The results are presented in Section 5, and discussed further in Section 6, which concludes the thesis.

(8)

2. Previous Literature

In this section, I will first present previous research on the effects of economic inequality.

Inequality’s effects on trust are discussed in Section 2.3. Before that, I will briefly summarize the literature on other determinants of trust.

2.1 Effects of Economic Inequality

Economic inequality is negatively associated with happiness and group participation (e.g.

Neckerman & Torche 2007, Uslaner 2002). Income inequality is found to decrease happiness especially amongst the poorest (Alesina et al. 2004). The effect is stronger in Europe than in the USA, and strongest among leftists, indicating that aversity to inequality intensifies the effect (ibid.). Another mechanism suggested is that although relative poverty does negatively affect happiness, individuals do not get happier from being richer than their peers (Van Praag

& Ferrer-i-Carbonell 2011). Previous studies show that the negative impact of inequality on group participation is strong, and that the effect is particularly significant for the both objectively and subjectively wealthy (Alesina & La Ferrara 2000, La Ferrara 2002). It is hypothized that the effect is due to the fact that people seem to prefer interacting with those who are similar to themselves, while income inequality increases social heterogeneity (ibid.).

Perhaps unsurprisingly, economic inequality correlates negatively with democratic values and civic norms (Knack & Keefer 1997, Rothstein & Uslaner 2005). Although it would be plausible, income inequality seems nevertheless to have no effect on redistributive preferences or polarization of political attitudes (Neckerman & Torche 2007, Bosancianu 2017).

Crime rates and economic inequality are positively correlated within and between countries (Fajnzylber et al. 2002, Neckerman & Torche 2007). The association with violent crime is especially strong (Neckerman & Torche 2007). However, not all effects of economic inequality are negative. Inequality is associated with higher educational attainment, as economic inequality is associated with both higher returns of schooling and higher public spending for education (Neckerman & Torche 2007). However, individual differences in schooling widen as inequality increases (ibid.). At cross-country level, there is evidence that income inequality is positively related to a stronger work ethic and to an increased labor supply (Corneo & Neher 2014, Bell & Freeman 2001). This could be explained by stronger work incentives as the economic difference between different levels of productivity increases (ibid.). The effects of economic inequality on health and economic growth are widely studied, but remain unclear.

The association between health and inequality remains statistically insignificant, even though theoretical literature suggests a negative relationship (Leigh et al. 2011, Neckerman & Torche 2007). Accordingly, empirical literature has not reached any consensus on the effect of

(9)

inequality on economic growth, and the studies report remarkably different results (Voitchovsky 2011).

2.2 Determinants of Trust

Studies show that the basis for trust is created already in early childhood, and that while later experiences shape it, their effect is not radical (e.g. Uslaner 2002). In addition, some of the trust is genetically heritable (Cawvey et al. 2018), or affected by cultural roots (e.g. Bjørnskov 2006). Trust levels are therefore quite stable over time. However, both localized and generalized trust can be affected by several factors. At individual level, being succesful in life strongly correlates with trust. Trusters are often healthier, better educated, and wealthier than those that do not trust (Newton et al. 2018). Accordingly, recent traumatic experiences and belonging to a historically discriminated group are negatively associated with trust (Alesina &

La Ferrara 2002). Education increases trust at both micro and macro level (Knack & Keefer 1997). Further, Tabellini (2010) finds that low regional trust levels are partly explained by historically high illiteracy rates. Perhaps surprisingly, group participation seems not to generate trust (Knack & Keefer 1997, Newton et al. 2018). This can be explained by the fact that voluntary associations are often socially homogeneous, and do not increase generalized trust by building bridges between different people (Newton et al. 2018).

Good institutions have a great role in creating trust. People trust more if the government cannot act arbitrarily, and politicians are responsive to the electorate (Knack & Keefer 1997, Newton et al. 2018). Institutional mechanisms that ensure that people are accountable for each other maintain trustworthy behavior, increasing trust in others (ibid.). Economic freedom and security of property rights seem to have a large positive effect on trust (Berggren & Jordahl 2006). Weak institutions predict low trust levels in the future (Tabellini 2010), and the effect of institutions seems to be transmitted from parents to children even if the children live under different institutions themselves (Ljunge 2014). Another important determinant of trust is ethnic heterogeneity, which at regional level seems to decrease trust more than income inequality (e.g. Alesina & La Ferrara 2002, Coffé & Geys 2006, Knack & Keefer 1997).

However, the negative impact on trust seems not to be due to ethnic-cultural differences, but due to diversity in terms of nationalities (Coffé & Geys 2006). Social polarization decreases trust (Knack & Keefer 1997), and religious hierarchy is negatively associated with trust, although religious upbringing seems to raise the level of trust (Guiso et al. 2006).

2.3 The Effect of Economic Inequality on Trust

The majority of previous studies on economic inequality and trust has been cross-country comparisons. At cross-country level, all studies conclude that economic inequality is clearly the most important explanatory factor for trust differences (e.g. Barone & Mocetti 2016, Knack

& Keefer 1997, Bjørnskov 2006, Rothstein & Uslaner 2005). This is in contrast to the studies

(10)

within countries, which have not found any statistically significant effects of income inequality on trust, concluding that the variation in trust is mainly explained by ethnic heterogeneity (Alesina & La Ferrara 2002, Leigh 2006, Gustavsson & Jordahl 2007). There are two main explanations for why the results between regional and cross-country levels are so different (Barone & Mocetti 2016). First, the results in cross-country comparisons are probably affected by a greater number of cultural, institutional or social factors explaining some of the effect inequality seems to have on trust. Second, data at regional level may have too little variation to present significant changes in the two variables, leading to insignificant results. A vast majority of previous studies measure inequality solely with Gini coefficient, and limit their analyzes on generalized trust.

There are three important studies at regional level. Gustavsson and Jordahl (2007) investigate the question with Swedish panel and cross-sectional data at county level. They measure the effect of income inequality on generalized trust through four different inequality measures (Gini, P90-10, P50-10 and P90-50), and find a moderate effect5 as income differences among people in the bottom half of the income distribution increase. They also make a distinction between pre-tax and disposable income inequality, but do not find any consistent differences between them. Individuals participating in the panel study have been interviewed twice: at the beginning and at the end of the time period. The major problem of the study is its small panel consisting of only 680 observations, half of which concern 1994 and the rest 1998.

Alesina and La Ferrara (2002) analyze income inequality’s effect on generalized trust in the United States at the level of metropolitan areas. They measure inequality using a Gini coefficient that is based on family income. The study uses data from the time period between 1974 and 1994, and has more than 7,200 observations. While the results suggest that income inequality has a negative association with generalized trust, the coefficients are however statistically insignificant. Moreover, trust levels seem to be determined mainly by ethnic heterogeneity. Finally, Leigh (2006) uses cross-sectional data with 6,500 neighborhood-level observations in Australia. Inequality measured with Gini has no significant effect on neither generalized nor localized trust, while linguistic heterogeneity has a statistically significant negative association with the two forms of trust. However, income inequality at local level has slightly more negative association with localized trust than generalized trust, even though this correlation cannot be seen as causal.

5 According to their estimates for disposable incomes, an increase of one standard deviation in the 50/10 quotient is predicted to decrease generalized trust with 0.6 points on a 0-10 scale. When basing the inequality measure on pre-tax income, the effect halves.

(11)

3. Data

I use data for the 290 Swedish municipalities from the time period between 1996 and 2014.

Data comes from two sources: the national SOM surveys and the GeoSweden database. The two databases provide data at individual level, and link the individuals to their home municipalities, using the borders of 2014.

3.1 Income Inequality

Income inequality can be measured in several ways. The most frequently used measure is the Gini coefficient, which estimates the income dispersion in a population. However, one of its limitations is that the coefficient may be the same for units that have equal levels of income but different income distributions. To overcome this limitation, income inequality can also be measured as the share of poor or rich of the population, or the grade of poverty or affluence in different income deciles. In addition, income itself can be defined in many different ways. For instance, income can be measured before or after taxes and subsidies, or at different levels, e.g.

individually or at household level. All of these measures show slightly different aspects of income inequality, and can thus be thought to have different effects on the dependent variable.

The economic inequality measures are based on the GeoSweden database and its data on the entire Swedish population's incomes.6 The inclusion of the entire population makes data by definition representative within Sweden, and minimizes measurement errors. In contrast to survey-based data, an official register avoids problems such as under- or overestimated values or otherwise incorrect answers, which could make the data inaccurate. The inequality measures are based on individual incomes, which are not sensitive to variations in household size between households or over time. Both pre-tax and disposable income are included.

Furthermore, the measures are constructed using individuals between the ages of 25 and 59, since they form the main group in the labor market. Finally, individual values are used to derive 290 municipal values for every year. All inequality measures will thus have nineteen values for each municipality during the sample period, giving 5510 observations per variable.

6 The data is made available by Che-Yuan Liang. See Hu & Liang 2020.

(12)

Figure 1. Pre-tax and disposable income dispersion 1996-2014

Average standard deviation in municipal income 1996-2014. Blue (dark): pre-tax measure. Orange: disposable.

Dashed lines indicate the confidence intervals in values between the municipalities.

General income dispersion is measured by the standard deviation of CPI-adjusted log income.

Pre-tax and disposable income are analyzed separately in order to distinguish the effect of redistribution. Pre-tax income is mainly based on salary and capital income, but as disposable income is the income after taxes and possible subsidies, it can be directly affected by economic policy. As Figure 1 shows, the average standard deviation in disposable income has grown significantly over the entire time period, from 1.4 to 2.1. While the trend in pre-tax income fluctuates, even this measure lands on a slightly higher level than at the beginning of the time period — from around 4.0 to 4.2. The difference between the two measures is reduced so that the standard deviation in disposable income in 2014 is about half of that in pre-tax income.

Moving to the disposable income measure, the differences between municipal values grow slightly over the years: the standard deviation increases from 0.31 to 0.41, which approximates the standard deviation of the pre-tax income measure during the entire time period.

The following eight measures in Figure 2 show income inequality as a share of the poor and rich of the municipal population. The measures show municipal rates of residents whose incomes relative to the municipal median incomes are below or above normalized national income decile cut-offs. The poverty and affluence cut-offs are thus common to all municipalities and years7, but the poverty and affluence rates vary. The poverty rate measure shows the share of municipal residents whose relative incomes are below the national second decile, and the affluence rate the share with relative incomes above the eight decile. To get a more nuanced analysis, I use additional measures for those below the first decile and above the ninth decile. Furthermore, pre-tax and disposable income are analyzed separately.

7 Inequality measures are constructed using data from years 1991-2014.

(13)

Figure 2. Poverty and affluence rates 1996-2014

On the left: share of people below the normalized 2nd decile in solid lines and below the 1st decile in dashed lines.

On the right: the share above normalized 8th and 9th deciles in solid and dashed lines, respectively. Blue (dark):

pre-tax income, and orange: disposable income. Average municipal values 1996-2014.

As Figure 2 shows, the disposable income measures display a clear pattern: both the poverty and affluence rates have grown over time. This growth is somewhat stronger under the 2nd and above the 8th decile than amongst the very poorest or richest, but all variables follow the trend.

Pre-tax income measures fluctuate, but land on the approximately same level as in the beginning of the time period. The differences between the two measures indicate that the increased relative shares of poor and affluent are explained by changes in the redistribution.

Figure 3. Relative incomes in the lowest and highest income deciles 1996-2014

On the left: relative share of municipal median income for individuals below the normalized 2nd decile in solid lines and below the 1st decile in dashed lines. On the right: the income share for those above the normalized 8nd and 9nd deciles in solid respectively dashed line. Blue (dark): pre-tax income, and orange: disposable income.

Average municipal values 1996-2014.

(14)

The last eight measures in Figure 3 show development of relative income in lowest and highest income deciles, again separately for pre-tax income and disposable income. The poverty measures show how much individuals under the municipal 2nd and 1st income deciles earn in relation to the municipal median income, and the affluence measures show the same for individuals above the 8th and 9th income deciles. The 1st and 9th deciles are included to see the effect of the income changes specifically amongst those who earn the least and the most.

Figure 3 above shows that relative disposable income in the lowest deciles has fallen sharply during the last ten years, while the highest disposable incomes show a positive trend. Pre-tax income has varied in all deciles, but lands on the same level, which shows that the difference between disposable and pre-tax incomes has decreased. Compared with the highest deciles, this difference is two times as large in the poorest deciles. Under the 1st and 2nd deciles, the difference between relative disposable and pre-tax incomes is more than thirty percentage points, the difference above the 8th decile being less than a third of that.

3.2 Trust

As trust is based on a subjective experience not fully reflected in any visible characteristics, it cannot be measured directly or precisely. The ambiguity of the definition and the phenomenon itself makes trust difficult to capture compared to more established numerical measures such as income inequality. Individuals who are asked about their trust can interpret both the concept of trust and group definitions such as people in general or people in the area differently. The answers can be affected by respondents’ perceptions of the typical trust level, which in turn may vary between individuals due to social, individual, and cultural factors. The wording and the number of answer alternatives can also affect answers. Finally, individuals may be inclined to overestimate or underestimate their trust level depending on, for example, norms or identity.

There are empirical studies on measuring trust. The most common questions like Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?8, seem to be linked to the respondents' own trustworthiness instead of trust in other people (Glaeser et al. 2000). The same study argues that the best way to measure trust is instead to ask about past behavior in situations that have required the ability to trust others. Previous studies have often tried to avoid the problem by giving readers an expressed freedom to interpret the study's way of measuring trust as they wish. In addition, Alesina and La Ferrara (2000) argue that answers on the trust question cannot indicate trustworthiness, as they strongly correlate with individual characteristics such as gender and ethnic background.

In this thesis, I choose to measure trust with similar questions as in previous studies for practical

8 GSS, National Opinion Research Center's General Social Survey in the USA. One of the questions studied by Glaeser et al.

(15)

reasons: I have a limited amount of possible trust related questions, none of which have a clear behavioral connection.

Data on trust comes from the SOM surveys, which have been conducted annually by the University of Gothenburg from 1986. The survey's focus is on Swedes' opinions and habits in areas such as politics, consumption and values. The respondents are randomly chosen across the whole country, and from 1992 onwards the selection is based on all people living in Sweden regardless of citizenship. The respondents are between the ages of 15 and 85. Surveys are sent home to the selected respondents, and the participation is voluntary. The yearly sample size has grown from approximately 2,800 to 13,600 individuals during the nineteen years studied, which increases the statistical significance but does not in itself change representativeness. The response rate fell from about 60 percent in 1996 to just over 50 percent in 2014. This could be problematic if the reduction was systematic, i.e., the individuals who did not participate in the survey were on average different from those who participated. The problem is particularly relevant when studying trust, as people who do not trust others can be assumed to have less incentive to contribute to social surveys. Other things equal, this predicts that the trust levels on average increase if the response rate decreases. Further, an increased income inequality could affect the incentives and possibilities of some groups of people to participate in surveys, and they can on average have different trust levels than those who are affected less by inequality change. However, as the response rate clearly does not affect income inequality, this does not threaten the identification of a causal effect. The bias in answers can be limited by the choice of method and control variables.

Figure 4. Generalized and localized trust over time

The average trust levels in SOM-surveys. 1996-2014: generalized trust in grey, and 2007-2014: localized trust in red (dark).

(16)

This thesis focuses on two forms of trust. Generalized trust is measured with the statement Generally, people can be trusted and localized trust with People can be trusted in the area where you live. The answer is given as a number on a scale from 0 to 10, where ten is the highest value. The question on generalized trust has been asked since 1996 and that of localized trust from 2007. There are 73,868 observations for generalized and 22,171 for localized trust, excluding blank answers and double crosses. The latter has even relatively fewer observations, since the question was asked in only about half of the surveys 2007-2014. In contrast, the question on generalized trust has been included in all surveys during the time period. The response rates are thus similar. Figure 4 shows that despite small fluctuations, average generalized trust has been about 6.4 throughout the time period, while localized trust has varied around 7.1. The trends are similar, and the last measurement points attain approximately the same values as at the beginning of the time periods: 6.6 and 7.3. The difference between the lowest and highest annual values is in both cases less than 0.4 units.

Figure 5 shows the average municipal trust levels during the time periods studied. The variation in trust can be considered to be relatively low: the values of both generalized and localized trust in most cases between six and eight on a 0-10 scale. The localized trust values are generally about one unit higher than the general ones, and only three municipalities out of 290 have on average lower localized than generalized trust levels. The maps do not present any clear geographical trends; if anything, both higher localized trust and lower generalized trust seem to be concentrated in municipalities with a low population density.

(17)

Figure 5. Average municipal trust

Average municipal trust levels in Sweden during the time periods studied. On the left: generalized trust (average between 1996 and 2014). On the right localized trust (average between 2007 and 2014).

(18)

4. Empirical Strategy

In this section I discuss how I study the association between income inequality and trust. I first present my empirical method, moving then to control variables used in the final regression model. As was seen in the previous section, income inequality has on average increased over time while trust levels fluctuate without any clear trend. In Figure 6 below I plot the municipal changes in income dispersion and generalized trust during the sample period. There is indeed a weak negative association between changes in income inequality and trust, the correlation being stronger for disposable than pre-tax income. However, this correlation could be explained by other variables or reverse causality, of which I consider the former to be the major challenge. It is unlikely that trust would to any great extent affect economic inequality within municipalities, as they share similar laws and institutions. In addition, I use the first-difference strategy eliminating variables that are constant over time by focusing on the change between two annual values. However, as there can still be several time-varying factors threatening a causal interpretation of possible effects, I try to minimize the problem by using control variables and time fixed effects in my regression models.

Figure 6. Income dispersion and generalized trust changes by municipality 1996-2014

Municipal changes 1996-2014. Each dot represents one municipality. On the left: changes in municipal levels of pre-tax income dispersion and generalized trust. On the right: changes in disposable income dispersion and generalized trust. Changes in income dispersion is on the horizontal axis, and changes in trust on the vertical.

Estimated slope coefficients are -0.04 and -0.07. The dots are population weighted.

(19)

4.1 First-Difference Strategy

When it comes to trust and economic inequality, it is hardly possible to account for everything that affects the two variables, despite the focus on a relatively homogeneous country. As these omitted variables could threaten the identification of the effect of income inequality on trust, I address the problem by using the first-difference strategy. By taking the difference between two time periods, the method eliminates underlying variables that are constant over time. I then use the changes in the variables during the time period to isolate the effect inequality has on trust. However, if trust or income inequality changes slowly, the method can lead to insignificant results. The problem with small annual changes can be reduced by extending the time interval, which, however, reduces the number of first-differences and thus statistical significance.

As changes in trust and inequality are at least partly dependent on common trends over time, even the base regression model will control for time fixed effects. First-difference analyzes are thus based on the following regression model

𝛥𝑇𝑦𝑚 = 𝛽𝛥𝑂𝑦𝑚 + 𝑌𝑦 + 𝑢𝑦𝑚 (1)

where ΔT is the change in generalized or localized trust, ΔO is the change in income inequality, y is the index for the base year, m is a given municipality, Y is time fixed effects, and u is the error term.

I estimate effects with ordinary least squares. I use robust standard errors as the error term in the regression model can be assumed to vary at different levels of income inequality, i.e., there is heteroscedasticity. Furthermore, as standard errors can be assumed to correlate within municipalities, I cluster them at the municipal level.

I will now discuss the optimal first-difference length. In Figure 7, I use the regression model above to illustrate changes in the difference length. Using pre-tax income dispersion as the independent and generalized trust as the dependent variable, the statistical significance increases sharply when y is at least two years, and estimated values stabilize. However, when it comes to estimating the effect on localized trust, point estimates vary greatly regardless of the length of the time period, even though the confidence intervals decrease somewhat.

(20)

Figur 7. First-difference estimates, by difference length

Having pre-tax income dispersion as the independent variable, the figure shows estimated coefficients by difference length (on the vertical axis). Dependent variabels: generalized trust (on the left), and localized trust (on the right). Dashed lines show 95% confidence intervals.

I aim to determine the lenght of the time interval so that the levels of trust and inequality have enough time to change, while at the same time maximizing the amount of observations. The length of the time interval is in particular limited by localized trust, which has observations for only eight years. A low number of observations decreases statistical significance, and leads also to more uncertain results than when the sample size is large. Consequently, I will use a difference length of five years for all my inequality measures. This gives 14 base years and 4060 observations for generalized trust, and three base years and 870 observations for localized trust. Even though the sample size of localized trust is limited, it nevertheless exceeds that of, for instance, Gustavsson and Jordahl (680 observations). As discussed before, reverse causality is an unlikely threat to causality as the Swedish municipalities are quite similar when it comes to institutions and other possible channels. In addition, using a difference length of only five years limits the threat of trust levels affecting inequality through politics. Imposing and implementing political decisions takes time, and trust can thus hardly lead to any great political changes during a time period as short as five years.

Table 1 presents the means and the standard deviations of the inequality variables and their average five-year changes. The standard deviations show that there are great inequality differences between municipalities and years. However, the five-year changes in income inequality are on average relatively small. Disposable income measures display smaller values than pre-tax measures, except for relative incomes in the 1st and 2nd deciles. On average, inequality has grown. The five-year changes are generally positive, and, with the exception of income dispersion, greater in disposable income measures.

(21)

Table 1. Summary statistics for income inequality measures 1996-2014

The panel on the left shows average municipal five-year changes in income inequality measures.

4.2 Control Variables

To identify the causal effect, the error term should not correlate with the dependent variable.

Ideally, income inequality would thus be randomly distributed into the 290 municipalities. As this is of course not the case, the risk of omitted variables is minimized by using control variables. Previous literature shows that trust has a strong positive association with education, and that it is significantly negatively correlated with foreign background and unemployment (Alesina & La Ferrara 2002, Leigh 2006, Gustavsson & Jordahl 2007). Age and income seem to be positively associated with trust (ibid.). All of these variables are associated with municipal incomes and thus income inequality. In addition, previous studies at the municipality level have controlled for ethnic fragmentation and population size. Leigh (2006) finds linguistic fragmentation is the most important determinant of both generalized and localized trust in Australia. In Sweden, Gustavsson and Jordahl (2007) control for ethnic fragmentation and the share of foreign-born, and find that these explain a significant amount of differences in generalized trust. Further, both Gustavsson and Jordahl and Alesina and La Ferrara (2002) include a control for crime, but it does not apparently affect trust levels (see also Uslaner 2002).

However, population mobility, i.e. the share of people who have recently moved, is found to be negatively associated with trust (Leigh 2006).

(22)

I include all following control variables at the base year level. Using an unique value for each municipality and year, the regressions are controlled for municipal gender distribution, share of young and old, average education level, share with a foreign background, share unemployed, average income, and the municipality's population size. I will also control for the five-year changes in the population size. However, changes in other control variables are left uncontrolled as these are possible mechanisms for changes in income inequality. For instance, a change in unemployment rate probably affects poverty rates and income distribution. As discussed in Section 4.1, I use time fixed effects to account for general trust and inequality trends over time. In addition, the regressions are controlled for the base year's inequality and trust levels, as they could affect changes in both inequality and trust. For instance, a high level of income inequality could increase the probability for particularly inequality averse individuals to move away, affecting the levels of inequality and trust in both the old and the new municipalities. A control for base year’s inequality and trust levels probably minimizes this problem, even though ideally I would be able to identify the in- and out-movers themselves.

Finally, all regressions are population weighted.

The GeoSweden database provides three control variables. First, I control for foreign background by including the shares of residents born outside Sweden in Europe and outside Sweden and Europe. I also include the share without a high school diploma and with a university degree to control for education. The shares of residents aged at most 24 years ("young") and at least 60 years (“old”) control for age. In order to account for gender, unemployment and income, I derive average municipal values from answers of the SOM survey. As these values are not register based, they do not necessarily correspond to the actual municipal values. However, the values are calculated in the same way every year in all municipalities, so there should not be any great problems with internal validity. Control variables for gender and unemployment control for the share of women and unemployed of the municipal populations. When it comes to income, individuals are divided into five groups of approximately equal size according to their annual disposable household income. In 2014, those in the first group earned less than 200,000 Swedish crowns per year, while the fifth group earned at least 700,000.

Table 2 shows the means, standard deviations and the number of observations for trust measures and control variables, except the base year inequality level in Table 1.

(23)

Table 2. Summary statistics for trust measures and control variables 1996-2014

The table shows average municipal levels in trust and shares in control variables, and average five-year changes in trust and population size. The values are based on data during 1996-2014, except for localized trust 2007-2014.

Income 1-5 (highest), Education 1 = share without high school degree, Education 3 = share with university degree, Born in Europe = share born in Europe outside Sweden, Young = < 25 years, Old = > 60 years

In Section 5, I estimate regressions with following model

𝛥𝑇𝑦𝑚 = 𝛽1𝛥𝑂𝑦𝑚 + 𝛽2𝑂𝑦𝑚 + 𝛽3𝑇𝑦𝑚+ 𝛽4𝛥𝑃𝑦𝑚 + 𝛽5𝑋𝑦𝑚 + 𝛽6𝑌𝑦 + 𝑢𝑦𝑚 (2) where 𝛥𝑇is the change in generalized or local trust, 𝛥𝑂is the change in inequality measured by any of the 18 measures, y is index for the base year, m is a given municipality, Y is time fixed effects and u is the error term. O is the inequality base year y, T is the level of trust base year y, 𝛥𝑃is the change in population in five years, and X are control variables at base year level.

(24)

5. Results

5.1 General Income Dispersion

I begin with investigating how general income dispersion affects trust. I will first compare different methods to analyze the association between pre-tax income dispersion and generalized trust, moving then to disposable income dispersion. In Table 3 the effect of one unit change in standard deviation of log income (SD) is estimated with seven different models.

It should be noted that the seventh model (column 4, lower panel) is the main specification, and the other models are used for purposes of comparison only. First, the effect is estimated with OLS using absolute levels of pre-tax income dispersion and generalized trust. The estimate displayed in column 1 is based on a simple regression, without using any of the covariates discussed in Section 4.2. All remaining estimates are calculated using the first- difference method. The upper panel in columns 2-4 uses a difference length of one year, and the lower panel the subsequently predominant length of five years. In column 2, no covariates are included. Time fixed effects are added in column 3, making the model similar to Equation 1. Finally, all covariates are used to derive the point estimates in column 4. This final model is similar to Equation 2.

Table 3. Effects of pre-tax income dispersion on generalized trust

Column 1: simple regression between inequality and trust levels, estimated with OLS. Column 2: first-difference with a difference length of one year in upper panel, and five years in lower panel. Column 3 adds time fixed effects to the FD model. All covariates are included in column 4. Number of observations: 5510 (upper panel) or 4060 (five-year changes in the lower panel). Regressions are weighted by municipal working-age population in the base year. Clustered, robust standard errors in parentheses. *** p<0.01, ** p<0.05

The methods lead to quite different point estimates. Beginning with the model using absolute levels of income distribution, the point estimate is negative and statistically significant at the one percent level. In columns 2-4, the first-difference estimates vary both between and within models. The difference length of one year displays statistically significant and positive estimates, and the final specification in column 4 lands on 0.5. The model seems to be relatively

(25)

robust, the estimates between Equation 1 and 2 (columns 3 and 4) being almost identical.

Moving to the five-year length in the lower panel, the estimates are negative, but smaller than in the panel above. The coefficient increases in size when adding time fixed effects, and halves when introducing additional covariates. The main specification in column 4 displays a coefficient of -0.13. In contrast to the estimates without control variables, this coefficient lacks statistical significance. In addition, substantial relative changes between columns 3 and 4 suggest low robustness.

The results indicate two things. First, the choice of method has an important role in estimating the effect of inequality on trust. Even though pre-tax income dispersion is negatively associated with generalized trust (column 1), the difference length in the first-difference method affects both the size and the direction of the estimated effect. Compared with the difference length of five years, the length of one year uses a larger number of base years, which increases the models’ statistical significance. However, as the annual changes in both inequality and trust are typically small and the effect probably takes some time to be visible, a long difference length probably reflects more truthful effects. Second, pre-tax income dispersion does not seem to have any practical effect on trust, regardless of possible causality. Relating the final estimate of -0.131 to average municipal inequality differences in Table 1, an inequality increase of 0.4 units, about one standard deviation of inequality levels and changes, would be associated with 0.05 units decrease in generalized trust. As trust is measured on a 0-10 scale, this is a decrease of a negligible economic significance, corresponding to less than 1% of the mean municipal trust level.

In Table 4 I analyze the effect of disposable income dispersion on generalized trust. The simple OLS regression between inequality and trust levels is displayed in column 1. Columns 2-4 use the first-difference method to estimate the effect, first without covariates, then with time fixed effects, and finally with all covariates. The difference length of one year is shown in the upper panel, and the lower one shows the five-year length.

(26)

Table 4. Effects of disposable income dispersion on generalized trust

Column 1: simple regression between inequality and trust levels, estimated with OLS. Column 2: first-difference with a difference length of one year in upper panel, and five years in lower panel. Column 3 adds time fixed effects to the FD model. All covariates are included in column 4. Number of observations: 5510 (upper panel) or 4060 (five-year changes in the lower panel). Regressions are weighted by municipal working-age population in the base year. Clustered, robust standard errors in parentheses. * p<0.1

Column 1 shows that there is no correlation between the levels of generalized trust and disposable income dispersion. Accordingly, all remaining estimates are quite small, and in most cases smaller than when estimated with pre-tax income in Table 3. The first-difference estimates in the upper panel are on average more negative than those using the difference length of five years. However, the final estimates in column 4 are quite similar. The only point estimate to be statistically significant (column 3) shows that when using the difference length of five years and time fixed effects, one unit increase in disposable income dispersion is predicted to decrease generalized trust with -0.15 units. The estimate changes to -0.116 when adding control variables and becomes statistically insignificant. This indicates both low robustness and low practical importance. As for the pre-tax variable, an inequality increase of 0.4 units would result in only 0.05 units decrease in generalized trust, if the result could be causally interpreted.

The main model specifications in tables 3 and 4 show that general income dispersion seems not to have any effect on generalized trust. The result is in line with previous studies, which estimate income dispersion with Gini coefficient and find no significant effects. The estimates of -0.131 and -0.116 are near zero and statistically insignificant, and the varying results between different specifications suggest low robustness. If something, income dispersion seems to be weakly negatively associated with generalized trust. However, the magnitude of the coefficients is too marginal for any practical importance. Indeed, to change generalized trust with at least 0.5 units on its 0-10 scale, the standard deviation of log income should change over 3.8 (pre-tax income) or 4.3 units (disposable income). As the average for municipal income dispersion is 3.9 units, income inequality should at least double or be completely

(27)

eliminated to change trust with a half of its standard deviation. As the inequality variable does, on average, hardly change over time (see Table 1), it can thus be concluded that general income dispersion does not in practice affect generalized trust.

As income dispersion is measured locally, inequality would potentially have an effect on trust at local level, even if it lacks an association with generalized trust. To investigate that, I replicate the analysis with localized trust as the dependent variable. I use exclusively my main specification; the first-difference method with a difference length of five years. Beginning with pre-tax income dispersion, I first estimate the effect without any covariates. Time fixed effects are added in column 2, and all control variables in column 3. Columns 4-6 are identical to the three first columns but use disposable income as the independent variable.

Table 5. Effects of income dispersion on localized trust

Changes in the standard deviation of municipal pre-tax and disposable income, with and without time fixed effects and control variables. 870 observations of five-year changes (290 municipalities and 3 base years). Regressions are weighted by municipal working-age population in the base year. Clustered, robust standard errors in parentheses. ** p<0.05, * p<0.1

The magnitude of most of the point estimates shows that the association between income dispersion and localized trust is stronger than that of generalized trust. Pre-tax estimates change remarkably when time fixed effects and control variables are added to the regression model, and the final estimate in column 3 lands on -0.763. The estimate is neither robust nor statistically significant, but the association is nevertheless almost six times stronger than what is the case for generalized trust. Should income inequality increase with one standard deviation, localized trust would thus decrease with 0.31 units. The association becomes still a little stronger in column 6: a change of one standard deviation in disposable income dispersion would change localized trust with 0.34 units, which equals to 16% of the standard deviation of localized trust. The estimates in columns 4 and 5 are statistically significant at the five and ten percent levels, and even though the final estimate is not, it does not percentially differ remarkably from the two other estimates.

The results suggest that although income dispersion is not associated with generalized trust, it has a weak negative association with localized trust. The association is slightly stronger for disposable income than for pre-tax income, suggesting that it is mainly others’ consumption

(28)

capacity that is correlated with localized trust. However, because of the lack of statistical significance, the null hypothesis of no effect cannot be rejected. In addition, it should be noted that the association between income dispersion and trust may not be causally interpreted. I discuss this potential problem at the end of Section 5.2.

5.2 Other Inequality Measures

The results in the previous section show that income dispersion is not associated with generalized trust, but it seems to have a weak association with localized trust. The fact that localized trust is associated with municipal income dispersion whereas generalized trust is not, supports the hypothesis that municipal inequality correlates more with localized than generalized trust. As previous studies have found that other kinds of income inequalities could affect trust more strongly than income dispersion measured with Gini, I will now investigate this separately for generalized and localized trust.

Beginning with generalized trust, Table 6 reports estimates for 16 additional inequality measures. Columns 1-4 show how one-unit changes in municipal poverty and affluence rates affect trust. For the poverty rates, columns 1 and 2 use changes in share of residents along first and second normalized national income decile cut-offs, the first decile comprising the poorest residents. Similarly, columns 3 and 4 show changes above the two most affluent deciles.

Columns 5-8 are based on relative income measures, that is, changes in municipal ratios of average decile incomes over the median income. Columns 5 and 6 show the effects of relative income changes among people under the lowest income deciles, and columns 7 and 8 above the eight and ninth deciles. The upper panel shows pre-tax measures, and the lower one disposable income measures.

Table 6. Effects of poverty/affluence rates and relative incomes on generalized trust

Separately for pre-tax and disposable income inequalities: effects of municipal poverty/affluence rates in columns 1-4, and of municipal relative incomes in 5-8. All regressions include year dummies and all control variables mentioned in Section 4.2. 4060 observations of five-year changes (290 municipalities and 14 base years).

(29)

Regressions are weighted by municipal working-age population in the base year. Clustered, robust standard errors in parentheses. *** p<0.01, ** p<0.05

Beginning with poverty rates, municipal shares of residents under the two lowest income deciles seems to have a negative effect on generalized trust. Using disposable income instead of pre-tax income doubles the impact of poverty rates, and the effect is stronger when including only the poorest residents (column 1). The results are similar for the affluence rate. Increase in the municipal shares of the affluent decreases generalized trust. Compared with poverty rate measures, the relative difference between disposable and pre-tax income measures is smaller.

However, the difference between the two income deciles is by contrast greater. Furthermore, the affluence rate has stronger association with generalized trust than what the poverty rate has.

Even though all point estimates are significant at the one percent level, that is not to say that a change in municipal poverty or affluence rates would in practice affect generalized trust. If a variable would increase with one percentage point over the time of five years, a generous assumption for all measures but the poverty rate in disposable income, generalized trust would decrease between 0.02 and 0.07 units on its 0-10 scale. To obtain a change of 0.5 units in trust in five years, the shares of residents under the lowest income deciles or above the most affluent deciles should be doubled.9 This is however very unlikely as the poverty and affluence rates are fairly stable over time.

Moving to relative income measures, the point estimates decrease remarkably, but remain in most cases statistically significant at least at the five percent level. Increase in the relative incomes of those under the lowest deciles is positively associated with generalized trust, whereas a relative income increase above the eight and ninth decile decreases trust. Again, the changes above the eight and ninth income decile affect trust more than changes under the two lowest deciles. Disposable income measures are more tightly associated with generalized trust than pre-tax income measures, and in both cases inequality under second and above eight income deciles affects trust more than that under first or above ninth deciles. In practice the coefficients are nevertheless marginal. One percentage point decrease in the relative incomes of the residents under the two lowest deciles would decrease trust at most 0.01 units (disposable income under the 2nd decile), and a similar increase above the affluent deciles would decrease trust with not more than 0.03 units (disposable income above the 8th decile).

In summary, Table 6 has shown that all inequality measures are statistically significantly associated with generalized trust, and, among the variables, those based on disposable income have the strongest relationship. However, even though trust is decreased by increasing poverty and affluence rates as well as inequality in relative incomes, the effect seems to be of negligible

9Calculated as 0.5/(point estimate). To decrease generalized trust with 0.5 points, poverty rates should increase between 12 and 27 percentage points, and affluence rates between 7 and 19 percentage points.

(30)

size. A typical one percent change in income inequality changes trust with 0-6 percent of its standard deviation, or at most one percent of its mean. It is nevertheless possible that a change in several inequality measures would together affect trust over a longer period of time.

Municipal shares of relatively poor and rich affect generalized trust more than their relative incomes, and the association becomes stronger when moving to the most affluent deciles. In other words, the results indicate that an economic policy changing the share of residents above the most affluent disposable income deciles would influence trust more than redistribution to the poor or changing people’s earnings capacity.

Having investigated the effects of different inequality measures on generalized trust, I turn to their effects on localized trust, which are presented in Table 7. Columns 1 and 2 show

poverty rate, columns 3 and 4 affluence rate, and columns 5-8 relative incomes of both the relatively poor and the rich.

Table 7. Effects of poverty/affluence rates and relative incomes localized trust

Separately for pre-tax and disposable income inequalities: effects of municipal poverty/affluence rates in columns 1-4, and of municipal relative incomes in 5-8. All regressions include year dummies and all control variables mentioned in Section 4.2. 870 observations of five-year changes (290 municipalities and 3 base years).

Regressions are weighted by municipal working-age population in the base year. Clustered, robust standard errors in parentheses. ** p<0.05

Share of residents under the first and second income deciles is negatively associated with localized trust, and point estimates for poverty rates are greater than those in Table 6.

Disposable income inequality affects trust substantially more than pre-tax inequality, and the latter measures are statistically significant at the five percent level. Coefficients for affluence rate are all negative and statistically insignificant, and for the most part lower than poverty rate estimates or results in Table 6. Defining income deciles after pre-tax incomes seems to have greater effect on localized trust than when using disposable income measures. As with generalized trust, one percentage point increase in the inequality measures affects localized trust only marginally, even though poverty rate estimates are clearly greater. When it comes to disposable income measures in columns 1 and 2, an inequality increase of five percentage

References

Related documents

A sentivity analysis was performed to check the robustness of the model. The results of estimation in sentivity analysis are given in the tables 2 and 3. To check the

Starting with examining the validity of Kuznets hypothesis in China’s situation, we separately estimate the effect that China’s rapid economic growth takes on the overall

In last subsection, we used the agriculture and non-agriculture output proportion as the measurement of economic growth level, and used Gini index as the measure of income

This paper proposes a measure of guanxi to capture its multiple dimensions and studies its impact on income inequality, using China household finance survey data.. In

Thus, besides effects such as lower efficiency and, over time, hinder to economic growth, the empirical results in this study support the argument that financial repression has

Secondly, a multilevel regression analysis is performed in order to test if a country´s level of income inequality affects differences (polarization) in attitudes between people

Pareto-efficient marginal income tax rates as a function of the consumption levels (for a log-normal consumption distribution) in equilibrium, based on non-self-centered

KEYWORDS: micro data, panel data, poverty, poverty traps, income distribution, inequality, top incomes, welfare state, household disposable income, income mobility,