• No results found

The determinants of demand for redistribution and how they vary over time

N/A
N/A
Protected

Academic year: 2021

Share "The determinants of demand for redistribution and how they vary over time"

Copied!
42
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

THE DETERMINANTS OF DEMAND FOR REDISTRIBUTION AND HOW THEY VARY OVER TIME

Valentijn van Spijker 6126987

19th of February, 2015

Master Thesis

Supervisor: Maurice Bun

Public Economics

(2)

2

Contents

1. Introduction ... 1 2. Current theories ... 4 2.1. Self-interest ... 4 2.2. Public values ... 6 2.3. Social rivalry ... 8 3. Existing literature ... 9 3.1. Empirical findings ... 9 4. Preliminary expectations ... 12 5. Methodology ... 15 5.1. Data ... 16 5.1.1. Variable selection ... 18 5.1.2. Data manipulation ... 21 5.2. Model ... 21 6. Results ... 25

6.1. Testing the three theories ... 27

6.2. Variation in effects over time ... 32

7. Conclusion ... 35

References ... 37

Appendix A: Fluctuations of dependent variable means over time... 39

(3)

1

1. Introduction

Income and wealth inequality has long been a major concern for society. Yet it wasn’t until the industrial revolution and the rise of the working class that the responsibility to deal with this issue was appointed to governments, at least to some degree. Ever since then people’s demand for redistribution has been a key determinant in their voting behavior, and govern-ments have started undertaking measures to reduce inequality.

These measures can take several forms. They can range from direct money transfers from ‘rich’ to ‘poor’ to the public provision of private goods such as schooling, health care, etc. The importance of this issue to both governments and the public is reflected by the size of the social-protection expenditures in most European countries, often exceeding 30% of GDP (Eurostat). The sheer magnitude of the cost to society and the effort that is put into reducing inequality makes it crucial to know what drives people’s demand for redistribution.

For this thesis I use multiple cross sections of survey data to measure what affects people’s demand for redistribution. Traditionally, politico-economic models on demand for redistri-bution have been based on the assumption that people support redistriredistri-bution for self-inter-ested reasons (Corneo and Grüner, 2002). For example, people might favor redistribution based on the net effect it has on their economic situation, to insure for the risk of losing their job, etc. However, self-interest is not the only factor that could motivate people to support redistribution schemes.

Nowadays, authors generally agree self-interest alone does not accurately explain demand for redistribution observed in reality and typically provide two alternatives to this theory (Corneo and Grüner, 2002; Boarini and Le Clainche, 2009). First, people base their support for redistribution on public values they hold. Altruism, for instance, is a common human value which could account for a large share of the demand for redistribution, in which case

(4)

2

we would assume demand for redistribution to be positively correlated with poverty levels or income inequality. Similarly, people’s support for redistribution could depend on their views with respect to the role of luck versus effort when it comes to people’s income. Some people might be more concerned about who actually qualifies to receive social protection benefits than about the costs. These public values or views can be unique to every individual, and are formed and shaped by people’s personal experiences in life

Second, people might be guided by what in literature is generally called social rivalry (Cor-neo and Grüner, 2002). Individual demand for redistribution could be motivated by the effect it has on their social status, prejudice about recipients or contributors of the redistribution scheme or other reasons that are based on their social benefit. Different individuals can have different motivations for their demand for redistribution, and these do not have to be mutually exclusive.

In a static society, that is, a society where people’s situation does not change over time, one would expect demand for redistribution to remain stable. However, societies are not static, and changes or events that occur over time can translate into changes in demand for redistri-bution through different mechanisms. Based on the three previously described theories, peo-ple can have different motives for their support or opposition for redistribution (Corneo and Grüner, 2002). While altruism is a value common to most human beings, different peoples have different levels of inequality of poverty they consider acceptable. A high income person might oppose redistribution for self-interested reasons when both income inequality levels and poverty rates are low, and support redistribution when they are high. Similarly, a person who previously supported redistribution for altruistic reasons might suddenly oppose it if she feels people who benefit from it are free-riding on those who contribute to society.

While there is a large body of literature on what affects people’s demand for redistribution, to the best of my knowledge all empirical studies have been conducted using single

(5)

cross-3

sectional survey data and hence the time varying character of the effects have not been ana-lyzed. One major flaw with using a single cross section is that it assumes that people’s public views are exogenous and stable over time, when it is very likely that these can change as individuals are faced with new, perhaps unexpected, experiences in life. So while this ap-proach can provide conclusive evidence on which factors help determine demand for redis-tribution, it cannot provide information on how different circumstances affect people’s deci-sion making.

In this thesis I perform an analysis similar to those conducted in previous studies, but using multiple cross-sectional surveys taken at different points in time. Specifically, the dataset consists of six survey waves taken every two years between 2002 and 2012 and across vari-ous European countries. This makes it possible to study if the effects of different parameters relevant to the above theories vary over time, especially in light of the recent economic tur-moil Europe has experienced caused by the global financial crisis in 2008 and the European debt crisis that followed shortly after. The purpose of this thesis is twofold. First, I perform a fixed effect regression using pseudo panel data to provide further evidence on the effects of the theories for demand for redistribution. Second, using multiple cross sections I analyze whether those effects are stable or vary over time. The results show that the latter is indeed the case.

In section 2 I discuss in detail what currently are the most agreed on theories for what deter-mines demand for redistribution. Section 3 provides an overview of the most relevant empir-ical studies and the evidence they provide. In section 4 the expected results are explained and through which mechanisms they are expected to take place. In the methodology (section 5) the data used is discussed in detail, the benefits of using panel data are reviewed and the model is constructed. In section 6 the results of the analysis are discussed and finally in sec-tion 7 I provide concluding remarks.

(6)

4

2. Current theories

In this section I will provide an in depth description of what over the years have become the three most widely used and agreed upon theories for what determines people’s demand for redistribution. As we will see later (section 3), all three models can predict individuals’ de-mand for redistribution and are not mutually exclusive.

In most classical politico-economic models individual demand for redistribution depended only on the pecuniary effect that the redistribution scheme in question would have on her net income (Roberts, 1977). In other words, an individual would only be in favor of redistribution if she were a net benefiter, and would oppose it only is she were a net contributor. This behavior is generally described as the ‘self-interest effect’.

This self-interest theory however, has proven unsuccessful at reproducing some results in real-life data, which leads to believe that there are other factors that – at least in part – deter-mine demand for redistribution. In an effort to help explain the evidence economists have traditionally offered two other theories: the public values effect and the social rivalry effect. In the first an individual’s demand for redistribution is not affected by the net impact on her income but rather by preconceived notions about the necessity and fairness of doing so. In the second individuals look at the effect the proposed redistribution scheme has on her social- rather than economic wealth. The latter arises from discriminatory behavior between social classes and ethnicities. Below, each of these models is elaborated on further.

2.1.

Self-interest

Traditionally this effect has been – and to some degree still remains – the most commonly used. The self-interest theory assumes that people base their demand for redistribution rely-ing only on initial endowment with outcome oriented, self-interested preferences.

(7)

5

Two models can predict results that fit this description: the median voter model and the social insurance model (Moene and Wallerstein, 1996). The median voter model of income taxation assumes that only those who expect to benefit from a certain redistribution scheme will sup-port it. Assuming linear taxation with exogenous pre-tax income it means that only people whose income is below the mean income will be in favor, while those who are above it will be against (Roberts, 1977). As a consequence, the location of the median voter’s income with respect to the society’s mean income will determine whether a majority supports of opposes redistribution (Moene and Wallerstein, 1996).

The social insurance model states that redistribution works as a social insurance device where those individuals whose incomes are hit by a negative idiosyncratic shock are reimbursed (Moene and Wallerstein, 1996). This type of demand can arise because in society, every individual is subjected to an uninsurable risk of lifetime income decline (Sinn, 1995). While unavoidable, individuals could influence this risk by the level of personal effort each of them exerts. Following on these assumptions those who contribute to the redistribution scheme are referred to as the ‘lucky’ ones while those who receive transfers – or are bailed out – are referred to as the ‘unlucky’ ones.

This social insurance scheme however – like with most insurance schemes – creates a breach for moral hazard. People’s behavior could be altered in a way that ‘negatively’ affects soci-ety, either by not making much of an effort or by taking excessive risks (Sinn, 1995). The link between ex-ante risk taking behavior by individuals and redistributive taxation has been studied by many authors (Eaton and Rosen, 1980; Varian, 1980; Rochet, 1991; Sinn, 1995). Similarly, the tax imposed on those who contribute to the social insurance could act as a disincentive for effort which could in turn have a negative effect on their lifetime income (Piketty, 1995; Sinn, 1995; Barr, 1992).

(8)

6

2.2.

Public values

The public values theory has often been presented as an alternative to the self-interest theory. While some people may indeed form their opinions on redistribution based on what best suits their pecuniary situation, this model argues that others might have very different reasons to support or oppose redistribution. Some individuals might be unconditional altruist. These individuals’ demand for redistribution will not be affected by their own economic situation or that of the recipient (Walster et al., 1978; Deutsch, 1985; Fong, 2001); nor will their belief on whether recipients deserve it or not.

Others, on the other hand, might relate demand for redistribution to their views on the self-determination of income. Those who believe that people’s incomes depend on luck or cor-ruption could likely be more in favor of redistribution than those who believe that rather effort and talent determine income (Corneo and Grüner, 2002). Two explanations can be given for this. The first is grounded on ethics and states that agents may believe that people whose income is the result of personal merits such as hard work, talent, etc. are more entitled to their income. However, if agents believe that a certain person’s income stems from luck they view government intervention to reduce income inequality more justifiable (Arrow, 1963). The second explanation is based on efficiency. If agents believe income is related to hard work they might expect the incentive costs of redistribution to be high. This may lead them to oppose redistribution since the aggregate income might be negatively affected and hence be ‘counterproductive’ for society.

Piketty (1995) explains with a model how life experiences with respect to dynastic income mobility can affect people’s view on the perceived ‘fairness’ of income and their views on the rewards of effort. This can lead to a self-fulfilling prophesy where those with higher incomes believe in self-determination and those with lower incomes do not (Piketty, 1995).

(9)

7

While the real reasons for achievement may never be known, different people are likely to have several different views on the matter. What is clear, however, is that levels of effort, talent or luck are not observed. This informational uncertainty creates a breach for suspicions about the reasons for economic achievements and moral hazard, leading to undermine sup-port for redistribution (Sinn, 1995). Suppose a person believes income is determined by the level of effort she exerts. To this person, people with low incomes and who would qualify for governmental support are merely suffering the consequences of low levels of effort. In their views a redistribution scheme would only reduce the incentive for these people to work and hence deteriorating society’s aggregate income. However, if the same person experiences a negative shock to her income due to what she considers no fault of her own, her self-deter-mination believes might change making her appreciate the necessity for governmental sup-port.

More recently, studies have focused more on the reciprocal character of income redistribution (Boarini and Le Clainche, 2009; Bowles and Gintis, 2000; Fong et al., 2006). Wealthy people may be in favor of supporting the poorer economically if they know that those who benefit from their contribution have to show that they are willing to cooperate. A tax payer who suspects some benefiters from the redistribution scheme are doing so by voluntarily being unemployed might withdraw her support for redistribution (Bowles and Gintis, 2000). If an agent strongly believes that in order to deserve public resources one needs to provide effort, said agent will oppose all redistribution schemes where reciprocity in not required (Bowles and Gintis, 2000; Fong et al., 2006). Acting along reciprocity standards means expecting redistribution schemes to work in a fair manner where all those involved show their willing-ness to cooperate. The reciprocity view states that when a redistribution scheme is a fair division of advantages and burdens it will be easier to defend by individuals independently of whether they are net ‘payers’ or ‘receivers’.

(10)

8

2.3.

Social rivalry

The social rivalry theory states that an agent’s demand for redistribution depends on the effect it will have on their ‘social wealth’. This social wealth can refer to any number of subjective factors that can affect their wellbeing that is intangible, and is often based on prejudice and discrimination. These can include prejudice towards social classes, ethnicities, religions, marital status, etc. (Cole et al., 1992).

These views can affect redistribution through different mechanisms. For instance, by shifting resources from rich to poor, the government could increase the probability of lucky upwardly mobile individuals to meet or even replace unlucky downwardly mobile individuals in certain neighborhoods or activities. Take for example a proposed redistribution scheme not in the form of a financial transfer from rich to poor but instead the provision of a public good like education. Often these forms of redistribution not only provide the desired effect they were designed for: in this case education attainable to everyone; but also create an opportunity for discrimination. The provision of public education could mean that lower income children will likely make use of them while higher income children still have the possibility to opt for (more expensive) private schools. If a certain agent would rather not have his kids going to the same school as a certain (lower income) this form of redistribution provides just that, making her likely to support it.

Another possibility is that people use income redistribution as a tool to prevent undesirable social situations. For instance, some may view poverty as a large cause of crime. Others might want to avoid the unemployed from marginalizing socially (Piven and Cloward, 1971).

(11)

9

3. Existing literature

People’s demand for redistribution and its key determinants have long been the subject of economic literature. Similarly, many researchers have used empirical data to test the theories. In this section I will briefly describe some of the most relevant previously conducted studies and their findings. In this section I briefly describe the most relevant empirical studies fol-lowed by the evidence they provide.

Recent empirical research on the subject has typically been conducted along some common lines. Most make use of country specific data (Fong 2001; Boarini & Le Clainche 2009) while others use international data (Corneo & Grüner 2002; Luttmer & Singhal 2008) and in all cases it consists of a single cross-sectional survey. In her paper, Fong (2001) measures the effect of social beliefs controlling for the self-interest effect. She does this by means of a probit regression estimating individual support for redistribution based on their self-determi-nation beliefs and a self-interest measure. She first performs this on her whole sample, and then on subsamples by categorizing people according to their income. Corneo & Grüner (2009) and Boarini and Le Clainche (2009) also measure the effects of self-interest and social values but add a third one; the social rivalry and reciprocity respectively.

3.1.

Empirical findings

The empirical results on the social insurance model have not always been conclusive. On the one hand (Boarini and Le Clainche, 2009) finds that the risk coefficients when testing for its effect on demand for redistribution never significantly differ from zero in any of the income classes. When testing for the effect of individuals’ own perceived exposure to risk on demand for redistribution Boarini and Le Clainche (2009) found that the coefficient stays relatively constant independently on whether they perceive it as being high or low. Similarly, Fong (2001) found that ‘income is a surprisingly poor predictor of redistributive beliefs’. She found

(12)

10

that the difference in preference for redistribution between incomes of $150,000 or more and $10,000 or less were smaller than between the three measures of belief on self-determination. On the other hand, Corneo and Grüner (2002) used answers to the question ‘would you gain or lose if government increased income redistribution?’ as an independent variable to test for its relation on demand for redistribution. The results showed that those who expected to gain from it significantly were more in favor of increasing redistribution than those who expected to loose from it. In the same paper they find that demand for redistribution is negatively correlated with their relative income and is one of the major determinants of demand for redistribution. However, Corneo and Grüner’s work (2002) focused not only on the self-interest model, but rather on the combined effect of a variety of models.

Evidence from the American public opinion surveys show that many redistribution schemes designed to benefit the ‘have-nots’ are largely supported by those who would not directly benefit from them (Boarini and Le Clainche, 2009). In Fong’s study she finds that a sizeable fraction of Americans whose household incomes are higher than $150,000 and are expecting to be upwardly mobile within the next five years are in favor of reducing inequality and helping the poor. Within said group, 24% said the government should ‘redistribute wealth by heavily taxing the rich’ and 67% said that ‘their government in Washington DC should make every possible effort to improve the social and economic position of the poor’. In contrast, of those surveyed with an annual family income below $10,000 more than one third opposed redistributing income by heavily taxing the rich and 21% said that it was their own responsi-bility rather than the government’s to help themselves to improve their position.

In literature a lot of evidence exists that confirms the public values models. In his paper, Piketty (1995) finds that the variable for people’s beliefs with respect to self-determination has a strong explanatory power of individual demand for redistribution. Those who believe that luck rather than effort determine high income do indeed show higher support for

(13)

redis-11

tribution. The same holds when beliefs about the ‘role of family wealth for achieving indi-vidual success’ are used as a variable. These results are confirmed by many studies (Fong, 2001; Boarini and Le Clainche, 2009).

Next to people’s beliefs about self-determination, reciprocity also appears to have a large effect. First of all, Boarini and Le Clainche (2009) find that 87% of the surveyed people in France promote a redistribution scheme that requires a counterpart from those who receive transfers. Gintis and Bowles (2000) find using survey data from the United States that people are more concerned about the conditions that determine which recipients should benefit than about how much it will cost them. Gilens (1999) goes on to state that people are more con-cerned with the question ‘who deserves what’ that ‘who gets what’.

While all these studies show significant results for reciprocity, their ‘direction’ is somewhat ambiguous. Boarini and Le Clainche (2009) classify respondents into groups depending on their views on self-determination and whether they require a counterpart for redistribution. Those who believe in self-determination and would not require a counterpart for the reception of public funding they call the ‘individualists’, and their evidence shows that this group is the least likely to support redistribution. On the opposite side of the scale are those who belief income depends only on luck or cheating and do not require a counterpart for redistribution either – which the call they ‘unconditional altruists’ – this group has the highest probability of being in favor or redistribution. Between these two groups in terms of demand for redis-tribution are those who do require a counterpart, or ‘homo reciprocans’. Within this group their support for redistribution is affected by their self-determination beliefs in a similar way to the previous ones.

When testing the social rivalry model Corneo and Grüner (2002) find evidence that confirms their expectations. Using an elaborate system to account for people’s subjective prestige score within society and grouping the respondent according to income they find that those

(14)

12

who find themselves in an income level with high prestige levels are less in favor of redistri-bution than those with lower prestige levels. Similarly, they find that the size of the ‘prestige gap’ between their income group and the subsequent income groups also affects people’s attitudes toward redistribution. Demand for redistribution decreases as the prestige gap be-tween an individual and his subsequent lower income group decreases. The opposite is true for the gap in prestige between an individual and the subsequent higher income group. Boarini and Le Clainche (2009) find that especially in a time when European governments are less generous in terms of social spending people are willing to go to great lengths to try and keep their job rather than receive unemployment. Although this might only apply to those who are still integrated in the labor market, it does – at least in part – disprove the moral hazard argument.

4. Preliminary expectations

As we saw in the previous sections, changes in society can affect demand for redistribution through different mechanisms. I devote this section to explaining which mechanisms these could be, and what the expected effect is on demand for redistribution. I start by describing how I expect each model to behave in a single cross-sectional survey (with no time variation) followed by a description on how effects could change over time.

As we will see in section 5, I analyze the three theories that authors most commonly agree upon in literature. To test for the self-interest effect, variables containing information on the individual’s economic situation are included. In the case of my thesis, these are the respond-ent’s income and a dummy variable defining if the majority of his household’s income orig-inates from labor (rather than social benefits, or savings). In the self-interest model, income is expected to have a negative effect on demand for redistribution since people who earn more generally benefit less from redistribution, as it usually occurs from ‘rich’ to ‘poor’.

(15)

13

Further, people for whom the majority of household income stems from labor rather than social benefits are expected to have a lower demand for redistribution. This could be either because people with jobs are not generally the receivers of social benefits or because efforts to reduce inequality are usually financed through some sort of tax, which in turn could dis-incentivize effort.

To test the public values effect I use two variables provided by the survey. First, people are asked if it is important in life to be rich. Second, they are asked if it is important in life to be successful and that people recognize your achievements. While these two variables do not directly reflect the self-determination beliefs of respondents it is unlikely that a person who believes money and success are the fruit of luck rather than effort would answer positively to both questions. In the regression both variables are expected to have negative coefficients, since people who believe that effort rather than luck determines income are likely to perceive low income people as uncooperative.

Finally, four variables are used to measure the social rivalry effects. These contain infor-mation about the respondents’ views toward immigration, their feelings of general safety and the government’s role when it comes to providing it. The effect of people’s beliefs on whether immigrants are good or bad for the economy could be ambiguous. On the one hand, people who believe immigration is bad for the economy could have lower demand for redistribution. This because immigrants are usually overrepresented on the lower side of the income scale, meaning they are more likely to benefit from income redistribution. On the other hand how-ever, while people who view immigration as a positive phenomenon for the economy might be more willing to share their social protection scheme with immigrants, their views are likely to be paired with expectations of the immigrants’ participation in the labor market. Given the negative effort incentive redistribution schemes impose, people with these views could likely oppose redistribution for efficiency reasons. The two safety variables reflect the respondents’ satisfaction with the levels of safety in her country. While it does not provide information on

(16)

14

whether they see redistribution schemes as a solution, it is unlikely that people who feel perfectly safe will want to increase redistribution efforts to reduce crime. However, the op-posite could also be the case. People might perceive crime as uncooperative behavior and therefore begrudge increasing social benefits. Finally, for the last variable respondents ex-press to what extent they feel it’s the government’s responsibility to insure safety. While agreeing with this fact does not necessarily mean the government should do so by increasing redistribution, people who view redistribution as a solution will likely see safety as a task for the government. Therefore demand for redistribution is expected to be positively correlated with this.

Next to the mechanisms that affect redistribution within the models, in this thesis I test whether the predictive power of the models can vary over time. Perhaps by shifting people who previously demanded the level of redistribution that best suited their economic interest to being motivated by altruism. Or some event in a person’s life made her change her mind about social values. I expect economic climate can play a crucial role in this, since this affects a large share of the population’s economic situation. If we assume most people share altruis-tic values but differ in the level of poverty or inequality they consider acceptable, a recession could affect their demand for redistribution. For example, a sudden rise in these levels could mean that for many people their threshold level has been reached, changing their motivation for supporting redistribution. Similarly, if a recession is perceived by people as an exogenous shock, people who were previously convinced of the self-determination of income might change their views in light of the increase in unemployment.

Another possible reason for a change in the predictive power of the models is the successful-ness of previous government efforts to reduce inequality. If inequality levels have decreased it is likely that people who previously demanded redistribution for altruistic reason will be satisfied with the result, and therefor start demanding redistribution in a manner that more

(17)

15

benefits their own interest. Therefore, demand for redistribution is expected to be positively correlated with levels of inequality and negatively with social protection expenses.

Conversely, it could very well be that a negative shock to people’s income and risk of unem-ployment has the opposite effect on demand for redistribution. If the crisis makes people go from living comfortably to worrying about their income, they might feel they need to worry about themselves before worrying about society’s less fortunate. The analysis performed in this thesis uses data acquired in six surveys taken every two years between 2002 and 2012. The second half of that period was one of great economic turmoil, with large drops in GDP, rises in unemployment and budget deficits that often demanded large cuts in government spending. While the expected sign of the effect seems a priori ambiguous, a significant shift in people’s motivation for demanding income redistribution is the expected conclusion of the analysis.

5. Methodology

I start this section by describing the data used in this thesis and provide some descriptive statistics. Then, I go over the identification of the relevant variables for the analysis and dis-cuss them in detail, followed by a report of the steps that were taken to arrive at the final dataset, stating how the variables were manipulated and what data from other sources was added to the dataset. Later, I go over the reasons of using multiple versus a single cross sections and the benefits and drawbacks of using pseudo panel data. Finally, I discuss the model used for the analysis.

(18)

16

Graph 1: Sample’s mean demand for redistribution over time by country

Note: The country means are indexed based on their mean demand for redistribution in year 2002

5.1.

Data

The models are analyzed using data provided by the first (and to date only) six waves of the European Social Survey (the ESS: freely available on www.europeansocialsurvey.org). These cross-sectional surveys have been performed every two years since 2002 in over 30 mainly European countries and across a pool of randomly selected participants. In other words, participants of earlier survey rounds are not asked to take place in a every recurrent survey. Any recurrence of a participant in more than one survey round would be purely co-incidental. With over 500 variables the surveys cover a wide array of topics, some of which are very relevant for this thesis. Among them, there is information on the respondents’ de-

0,85 0,9 0,95 1 1,05 1,1 1,15 1,2 2002 2004 2006 2008 2010 2012 BE CH DE DK ES FI FR GB HU IE NL NO PL PT

(19)

17

Graph 2: Country’s average demand for redistribution versus Gini coefficient

Note: For both the countries’ demand for redistribution the averages were taken of all survey rounds. For the Gini indexes

the average coefficients were calculated based on data for the same periods.

The line displays the OLS estimated effect, with a slope of 0.1093 and an R-squared of 0.0631.

mand for redistribution, some variables on their socio-economics background, their income and source thereof and some variables indicating their social believes and concerns.

Country specific variables which are not provided by the ESS surveys are included in the data set. From the Eurostat database real country GDP, unemployment levels and a measure of income inequality (Gini index) are added. This is also the case for government levels of social expenditure. BE CH DE DK ES FI FR GB HU IE NL NO PL PT SE SI .2 .2 5 .3 .3 5 Ave ra g e G in i .5 .6 .7 .8

(20)

18

5.1.1. Variable selection

Dependent variable

For the dependent variable of this analysis respondents rate the degree to which they agree with the following statement: ‘The government should do more to reduce income inequal-ity’1. Graph 1 shows average demand for redistribution by country and how it changes over

time. As can be seen, some countries show a clear positive trend (such as Germany, Switzer-land, Hungary, etc.) while other countries show the opposite (Norway and France).

Graph 2 depicts a scatter plot of the countries’ average demand for redistribution between 2002 and 2012 and their average Gini coefficient in the same period. The plotted line repre-sents the positive yet insignificant (coefficient value of 0.1093 with a standard error of 0.1125) linear relation there exists a priori between the levels of income inequality and de-mand for redistribution. However, it should be noted that while the Gini does contain valua-ble information on the distribution of income, measuring inequality is not straightforward and it should therefore be interpreted with some care. Similarly, the used measure only takes into account income, leaving out the distribution of wealth in society.

Explanatory variables

The ESS surveys provide data that can be useful in estimating the effects of self-interest, public values and social rivalry on demand for redistribution. For the self-interest effect, two variables are relevant: household income and main source of household income. While all six rounds of surveys provide data on the respondents’ household income, the data collection method changed midway. In the first three waves of the ESS surveys (between 2002 and

1 The literal question respondents are asked is: “Using this card, please say to what extent you agree or disa-gree with each of the following statement: The government should take measures to reduce differences in income levels”. Respondents can answer by giving a value between 1 and 5 (1 for ‘agree strongly’, 3 for ‘nei-ther agree nor disagree’ and 5 for ‘strongly disagree’).

(21)

19

2006), respondents are asked to indicate their household income by choosing one of ten in-come categories that are equal across all countries. From round four onwards (between 2008 and 2012) these ten categories are country specific and each represents a tenth-percentile of said country’s household income. To unify these variables into a single one that is compara-ble across time and country, all income data is converted into euros when necessary using the average exchange rate of the period during which the surveys were being taken (provided by ESS documentation). Then, the resulting amount is divided by the resulting nominal av-erage income per capita in the respondent’s country in the year they were surveyed. Addi-tionally, since the data refers to household income, it is corrected for number of people living in the household in a way similar to Boarini and Le Clainche (2009)2.

Respondents are asked to state what the main source of their household income is. Here they can choose from a selection of possible answers3. Based on their answer, I create three dummy variables depending on whether their main source of income is from labour, from social benefits or others.

Measuring people’s social values is no mean feat. Any attempt at doing so will deliver highly abstract and above all incomplete results. However, for this thesis I use variables similar to

2 To correct for economies of scale within households I use an equivalence scale that is similar that the one used for the provision of income support in France adjusted for purchasing power parity. Here the coeffi-cients are 1 for a one person household, 1.5 for a two person household and 0.3 is added for every addi-tional member of the household. For a household of five people this would mean that the total income is divided by (1 + 0.5 + 0.3 + 0.3 + 0.3) = 2.4.

3 Literal question is: “Please consider the income of all household members and any income which may be received by the household as a whole. What is the main source of income in your household?” Respondents answered by choosing one of the following values:

1 Wages or salaries

2 Income from self-employment (excluding farming) 3 Income from farming

4 Pensions

5 Unemployment/redundancy benefit 6 Any other social benefits or grants 7 Income from investments, savings etc. 8 Income from other sources

(22)

20

the ones used by other authors (Corneo & Grüner (2002), etc.) that have produced significant results in the past.

Two variables provided by the ESS surveys contain information on the respondent’s public values. For each, they are asked to what extend they believe it’s important to be rich and successful respectively4. Graph A.1 (appendix A) shows the variation in the respondents’

average response over time. Note how the average response varies over time. At first glance these changes suggest that public values are not fixed over time and hence support this thesis’ approach of using multiple cross-sections to analyze their effect on demand for redistribution. The social rivalry theory states that people view redistribution as a tool to target undesirable social phenomena such as general safety, immigration, etc. Therefore, I use variables that depict the respondents’ sentiment toward those issues. In the survey, respondents are asked to what extend they feel unsafe in the dark and how important they consider it is to feel safe56. Additionally, they are asked if it is important to a have strong government that insures safety7. This last variable should carry information on whether the respondent believes it’s the gov-ernment’s responsibility to insure safety. Finally, respondents also state whether they believe immigrants are good or bad for the economy8. As can be seen in graph A.2 (appendix A)

4 Questions respondents are asked are: “It is important to her/him to be rich. She/he wants to have a lot of money and expensive things” and “Being very successful is important to her/him. She/he hopes people will recognise her/his achievements”. Respondents can answer with a number between 1 and 6 (1 being ‘very much like me’ and 6 ‘not like me at all’).

5 Questions respondents are asked: “It is important to her/him to live in secure surroundings. She/he avoids anything that might endanger her/his safety” Respondents answer like in footnote 4.

6Question: “How safe do you - or would you - feel walking alone in this area after dark? Do - or would - you feel…” Answers range from 1 to 4 (1 being ‘very safe’ and 4 ‘very unsafe’).

7 Question: “It is important to her/him that the government ensures her/his safety against all threats. She/he wants the state to be strong so it can defend its citizens” Answers like in footnote 4.

8 Question: “Would you say it is generally bad or good for [respondent’s country]'s economy that people come to live here from other countries?” Respondent answers a value from 0 to 10 (0 for “bad” and 10 for “good”).

(23)

21

people’s believes are not stable over time, suggesting once again that their effect on redistri-bution should be measured using time varying data.

5.1.2. Data manipulation

The six waves are combined to form a set of repeated cross-sections. Only countries which have participated in all six rounds are included. The resulting panel contains six rounds cov-ering 16 countries (out of an original 36) with a total of more than 160,000 observations9. Data entries where respondents refused to answer, answered that they did not know or simply did not answer are all treated as missing entries. For simplicity I rescale most of the ordinal variables to a value between zero and one, where zero means the strongly disagree and one they strongly agree.

Finally, for the panel data analysis (see section 5.2) a pseudo panel is created by grouping the respondents into cohorts. These cohorts can be based on characteristics that are unique to each cohort and does not vary over time or on characteristics that vary over time but not over cohorts. For this thesis the cohorts are based on country, gender and date of birth10. Once specified, the data is ‘collapsed’ into a pseudo panel, a process by which the cohort’s variable means are taken as a value. In the resulting dataset each cohort is treated as an observation taking the mean value of the whole cohort for each variable.

5.2.

Model

The purpose of this thesis is twofold. On the one hand I provide additional evidence on the determinants of demand for income redistribution. On the other hand, I analyze whether the effects of the determinants of demand for redistribution vary over time. In contrast to recent

9 Slovenia in 2012 was the country with the least amount of observation with 1,257.

10 Respondents are categorized into four age categories based on four age quartiles. The thresholds are: be-fore or in 1946, between 1947 and 1961, between 1962 and 1975 and after 1975.

(24)

22

literature, I use survey data that was collected in six waves over a period of twelve years rather that a single cross section in time.

Like we saw in section 5.1.1, for the dependent variable respondents are asked to rate to what degree they think the government should take measures to reduce differences in income lev-els. The nature of the question makes comparing answers from one country to another very difficult. In doing so the results could be inaccurate for two reasons. On the one hand, the fact that the question asks whether the respondent believes the government (of their country of residence) should do more to reduce income inequality (in their country of residence) means that respondents’ answers will depend on a number of factors that are specific to their country; such as levels on income inequality, government efforts to reduce income inequality, etc. On the other hand, cultural differences between countries might lead to inherent differ-ences in answering behavior independently of the respondents’ country’s initial endowment. Therefore, countries’ demand for redistribution should only be compared to itself over time, rather than compared to other countries’.

I begin my analysis by performing an OLS regression on a pooled dataset containing all six survey rounds. Like previously mentioned (section 5.1) ESS survey respondents are not fol-lowed over time. Instead, they are randomly selected for each new survey round. The design of the first model is comparable to others previously used in literature and looks as follows:

𝑌𝑖 = 𝛼 + 𝛽′1𝑋1𝑖+ 𝛽′2𝑋2𝑖+ 𝛽′3𝑋3𝑖+ 𝛿′𝑊𝑖 + 𝜀𝑖 (1)

where 𝑌𝑖 is the demand for redistribution of a random person in a random year or survey round. Variables 𝑋1𝑖, 𝑋2𝑖 and 𝑋3𝑖 are vectors of the respondents’ self-interest, public values and social rivalry variables respectively. With 𝑊𝑖 I control for country, time (year in which

(25)

23

the survey was taken) and country trends11. Furthermore, 𝛼 is a constant, 𝛽′

1 to 𝛽′3 and 𝛿 are

a transposed vector of the coefficients of the explanatory variables and controls respectively. Finally, 𝜀𝑖 is an error term. The results of this analysis are shown in table 1.

Since the individuals of each survey round are not followed over time and thus unique, the method used for model (1) cannot account for individual specific unobserved variables, which could in turn result in biased estimates. While it is possible to estimate individual fixed-effects using repeated cross-sections, this approach would be consistent only if the in-dividual fixed effects are uncorrelated with any of the explanatory variables, a condition which is hard to test (Verbeek, 2008). In contrast, when panel data is available, this can be solved by using a fixed-effect estimator which is treated as a fixed unknown parameter and estimating a within-regression. In the case of this thesis, the fixed effects of main concern are country specific, and therefore very likely to be correlated with some explanatory varia-bles such as levels of unemployment, gross domestic product and inequality indexes, all of which are unique to a given country.

When cohort averages are measured using a large number of individual observations, the same method can be used on a pseudo panel dataset (Verbeek, 2008). Unfortunately, there is no literature on the required cohort size. Empirical studies have varied greatly in the average cohort size used, ranging between 190 and 500 initial observations (Verbeek, 2008). Deve-reux (2007) recently argued that possibly a size of 2000 or more was needed. In any case, it generally holds that the smaller the relative magnitude of the measurement errors with respect to the within cohort variance, the smaller the bias of the pseudo panel analysis (Verbeek, 2008).

11 Country trends are controlled for by introducing interaction variables between country dummies and the survey round.

(26)

24

Of course, the use of pseudo panel data also brings some drawbacks. The fact the cohort average is used instead of individual data means that the effects of control variables that are significant otherwise, but do not vary over time are not identified. In addition, many relevant individual data is lost when the cross-sections are converted to a pseudo panel data-set. Below, model (2) shows the same specification as model (1) only adapted to pseudo panel data:

𝑌̅𝑐𝑡 = 𝛼 + 𝛽′1𝑋̅̅̅1𝑐𝑡+ 𝛽′2𝑋̅̅̅2𝑐𝑡+ 𝛽′3𝑋̅̅̅3𝑐𝑡+ 𝛿′𝑊̅𝑐𝑡+ 𝜃𝑐 + 𝜀̅𝑐𝑡,

𝑐 = 1, … , 𝐶; 𝑡 = 1, … ,6

(2)

where 𝑐 represents the cohort and 𝑡 the time or round of the survey and 𝜃𝑐 represent the cohort fixed effects. The macron, or bar, on top of the parameters specifies that the cohort mean is used. As with model (1), in some specifications next to fixed country effects, time effects and country trends are included. The results of this analysis are shown in table 2. Next to making the estimation of cohort fixed-effects possible, using pseudo panel data can provide information on the level of serial correlation within a model since it allows the esti-mation of a dynamic model (Stock and Watson, 2003). Without pseudo panel data, the lack of historical information for every single individual interviewed makes it impossible to per-form a dynamic model with a lagged dependent variable using pooled repeated cross-sec-tions. For this thesis it can be especially useful since many of the social values and moral views of respondents are unobserved and hard to quantify. Model (3) shows the dynamic model that will be tested:

𝑌̅𝑐𝑡 = 𝛾𝑌̅𝑐,𝑡−1+ 𝛼 + 𝛽′1𝑋̅̅̅1𝑐𝑡+ 𝛽′2𝑋̅̅̅2𝑐𝑡+ 𝛽′3𝑋̅̅̅3𝑐𝑡+ 𝛿′𝑊̅𝑐𝑡+ 𝜃𝑐 + 𝜀̅𝑐𝑡, 𝑐 = 1, … , 𝐶; 𝑡 = 1, … ,6

(27)

25

where 𝛾 measures the effect of the cohorts’ demand for redistribution in the previous period, represented by 𝑌̅𝑐,𝑡−1. In other words, 𝛾 provides information on the time it takes for changes in the explanatory variables to have an effect on demand for redistribution. All other param-eters represent the same as in model (2). Again, in some specifications next to fixed country effects, time effects and country or cohort trends are included. The results of this regression can be seen in table 3.

Finally, to test whether the effects of self-interest, public values and social rivalry vary over time, I perform a revised version of model (1) where the explanatory variables are replaced by their interaction with year or round dummies. This specification takes the following form:

𝑌𝑖 = 𝛼 + 𝛽′1𝑡𝑋1𝑖𝑡+ 𝛽′2𝑡𝑋2𝑖𝑡+ 𝛽′3𝑡𝑋3𝑖𝑡+ 𝛿′𝑊𝑖+ 𝜀𝑖, 𝑡 = 1, … ,6

(4)

where 𝑋1𝑖𝑡, 𝑋2𝑖𝑡 and 𝑋3𝑖𝑡 represent the vectors of explanatory variables for self-interest, pub-lic values and social rivalry interacted with round or year dummies. This means that the es-timated coefficients 𝛽1𝑡, 𝛽2𝑡 and 𝛽3𝑡 show the effects of the explanatory variable for each year the survey was taken. The rest of the parameters are the same as in model (1). Again, the model controls for country, survey round and country trends. The results of this analysis can be seen in table 4.

6. Results

In this section the results of the analysis are discussed. First, the results of testing the effects of the three previously discussed theories for the determination of demand for redistribution are discussed. This is followed by the results of testing for the time varying character of the effects.

(28)

26

Table 1: Effect of the different models on demand for redistribution

Demand for redistribution (x100) (1) (2) (3) (4) (5) (6) (7)

Self-interest:

- Individual income (% GDP) -7.42 -6.48 -5.95 -5.97 -6.02 (1.37) (1.19) (1.04) (0.98) (0.96)

- Majority of income from labor? -3.68 -1.52 -1.57 -1.55 -1.59 (0.79) (0.64) (0.59) (0.59) (0.59) Public values: - It is important to be rich -6.28 -7.52 -8.46 -8.45 -8.40 (2.26) (2.01) (1.53) (1.54) (1.56) - It is important to be successful -2.86 -5.46 -4.52 -4.55 -4.56 (1.56) (1.58) (0.93) (0.92) (0.91) Social rivalry:

- Immigrants are good for the economy

-5.57 -3.36 -2.70 -2.69 -2.66 (2.06) (2.16) (1.32) (1.36) (1.40) - You feel unsafe when walking

in the dark

5.45 4.08 3.06 3.10 3.13 (1.78) (1.74) (0.66) (0.71) (0.69) - Important to live in safe and

secure surroundings

5.70 6.00 3.43 3.42 3.35 (1.60) (1.45) (0.88) (0.87) (0.88) - Important government is

strong and ensures safety

11.8 13.3 10.1 10.1 9.99 (2.30) (2.43) (1.32) (1.31) (1.31) Constant 77.1 74.0 58.6 67.4 70.5 70.3 70.2 (1.44) (1.70) (3.58) (3.59) (1.14) (1.28) (1.35) Controlled for country - - - - Yes Yes Yes Controlled for survey round - - - Yes Yes Control for country trends - - - Yes

N 138998 173176 165680 128209 128209 128209 128209 R-sq 0.0302 0.0060 0.0311 0.0636 0.1271 0.1273 0.1313

Note: OLS regressions with demand for redistribution as dependent variable and the variables on the left column as

inde-pendent variables as shown in the different specifications (1 to 7) using all six survey cross-sections pooled together. Standard errors clustered at country level.

If specified, controlled for countries and survey rounds using dummy variables and for country trends using interaction variables of country dummy and survey round.

(29)

27

6.1.

Testing the three theories

In table 1 the measured effects of the three different theories by means of OLS regressions on all six cross sections are displayed. In columns 1 to 3 each theory is tested separately. In columns 4 to 7 the effects are measured simultaneously, and some control variables are in-cluded. The respondents’ income shows, as expected, a relatively large and significant neg-ative effect on demand for redistribution (with a p-value under 0.01). This effect remains stable in all configurations of the regression used in table 1, increasing the legitimacy of the estimated coefficients. Similarly, the source of the respondents’ income also has the expected sign. When measured with income only, as in column 1 of table 1, a person whose majority of income stems from labor will on average have a 3.7 point decrease in demand for redistri-bution on a scale from 0 to 100 with a significance below 0.01. The magnitude of this effect drops to about half when the variables for the other two theories are introduced, but stays constant when controls for country, round and country trends are added (with significance below 5%).

The public values variables also show significant effects. in column 2 of table 1 the variable for people’s views on the importance of being successful shows no significant effect when controlling only for the public values effect. However, when the effects of all three theories are measured simultaneously and controls are added the measured coefficient becomes larger and significant. Respondents’ beliefs on the importance of being rich show, perhaps unsur-prisingly, quite large negative effects on demand for redistribution in all regression specifi-cations of table 1.

The effects displayed for the social rivalry variables show varying results. In column 3 of table 1 we see that, on average, people who believe immigrants are good for the economy are less in favour of redistribution (with a significance level below 5%). However, this effect fades as the other theory’s variables are added to the regression as well as controls for coun-try, survey round and country trends. The effect of people’s feelings of safety in the

(30)

28

Table 2: Fixed effects regression measuring effect of the three models

Demand for redistribution (x100) (1) (2) (3) (4) (5)

Self-interest:

- Individual income (% GDP) -3.56 -3.24 -5.64 (1.05) (0.94) (1.60)

- Majority of income from labor? -2.70 -1.51 0.88 (1.25) (1.07) (0.88) Public values: - It is important to be rich -10.4 -12.7 3.92 (5.23) (4.79) (6.23) - It is important to be successful 14.1 7.70 5.54 (5.09) (4.99) (6.38) Social rivalry:

- Immigrants are good for the economy

-4.83 -8.08 11.2 (4.81) (5.11) (6.56) - You feel unsafe when walking

in the dark

-2.23 -0.11 7.42 (3.72) (4.38) (5.83) - Important to live in safe and

secure surroundings

13.3 14.1 7.30 (5.01) (5.08) (6.61) - Important government is strong

and ensures safety

18.4 12.8 16.8 (4.75) (5.14) (8.30)

Gini index 0.069

(0.12)

Log country GDP per capita -10.4

(2.17)

Government Social protection expenditure 0.43 (0.079) Constant 74.0 66.2 50.8 58.4 137.3 (0.99) (2.21) (4.80) (5.10) (23.8) N 720 768 768 720 400 R-sq 0.0521 0.0157 0.0894 0.1355 0.2291

Note: Fixed effects regressions with demand for redistribution as dependent variable and the variables on the left column

as independent variables as shown in the different specifications (1 to 5). Standard errors clustered at cohort level.

Cohorts in all specifications defined by respondent’s country, quartile of year born and gender. Standard errors in parentheses.

(31)

29

Table 3: Effect on demand for redistribution using pseudo panel data and controlling for dynamic effects Demand for redistribution (x100) (1) (2) (3) (4) (5) (6)

Self-interest:

- Individual income (% GDP) -7.91 -9.78 -7.12 -8.68 -5.61 -6.32 (1.18) (1.35) (2.01) (2.18) (1.56) (1.69)

- Majority of income from labor? -1.22 -0.15 -0.45 0.076 -2.46 -2.64 (1.01) (0.98) (0.97) (0.91) (0.84) (1.15) Public values: - It is important to be rich -8.41 -9.49 -2.65 -2.66 1.17 -2.92 (4.48) (4.80) (4.44) (4.86) (4.78) (5.23) - It is important to be successful 8.28 8.94 5.99 4.58 5.46 6.83 (4.71) (5.13) (4.76) (5.17) (4.63) (4.71) Social rivalry:

- Immigrants are good for the economy

-2.89 0.45 -1.36 2.41 0.31 -5.77 (6.09) (6.40) (5.63) (5.81) (6.82) (6.55) - You feel unsafe when walking

in the dark

-1.16 -3.05 4.90 1.61 2.01 2.27 (4.86) (5.03) (5.08) (5.01) (4.71) (5.13) - Important to live in safe and

secure surroundings

10.7 7.65 14.3 10.6 3.56 5.26 (5.55) (5.93) (5.72) (6.37) (5.39) (5.72) - Important government is

strong and ensures safety

19.5 26.0 12.4 19.4 11.6 11.2 (6.15) (6.62) (6.58) (7.17) (6.18) (6.28)

Gini index -0.096 -0.23 -0.11 -0.27 -0.17 -0.16 (0.090) (0.090) (0.094) (0.096) (0.15) (0.14)

Log country GDP per capita -6.25 -10.5 -9.58 -13.9 -9.38 -12.0 (1.77) (2.20) (2.06) (2.16) (3.71) (3.72) Lagged cohort demand for

redistribution

0.0042 -0.013 -0.24 (0.049) (0.050) (0.043) Controlled for survey round - - yes yes yes yes Controlled for country trends - - - - yes yes Constant 121.6 165.5 152.5 201.2 159.8 203.4 (18.4) (22.2) (20.9) (22.9) (37.8) (37.3)

N 600 552 600 552 600 552

R-sq 0.2074 0.2387 0.2354 0.2693 0.4384 0.4962

Note: Fixed effects regressions with demand for redistribution as dependent variable and the variables on the left column

as independent variables as shown in the different specifications (1 to 6). Standard errors clustered at cohort level.

Cohorts in all specifications defined by respondent’s country, quartile of year born and gender.

If specified, controlled for survey rounds using dummy variables and for country trends using interaction variables of country dummy and survey round.

(32)

30

dark have a large and significant positive effect in all configurations of the regressions in table 1. The same is true for variables its ‘important to live in safe and secure surroundings’ and its ‘important government is strong and ensures safety’.

The specifications displayed in table 1 were estimated using pooled regressions on all six cross-sections. In other words, they do not control for individual specific unobserved heter-ogeneity, which could lead to biased estimates. Therefore, in table 2 the same specifications have been estimated but using a fixed effects regression on pseudo panel data. In table 3 the same specification is estimated controlling for lagged demand for redistribution, as well as survey rounds and country trends. The cohorts are specified based on the country in which the respondent took the survey, their gender and year of birth (divided into four groups based on sample quartiles). With a much smaller sample size of between 768 and 400 (depending on regression specification) some results are still relatively large and in some cases signifi-cant. In all specifications of table 2 respondent’s income as a percentage of country GDP per capita has a significant (p-value below 0.01) negative effect on demand for redistribution. In contrast, while the respondents’ source of income shows negative effects for all specifica-tions of the model, only the one measure with self-interest variables exclusively is significant with a p-value below 5% (table 2 column 1). However, since this was originally measured by a dummy variable, the conversion to a pseudo panel dataset will simply show the share of cohort respondents whose main source of income is from labour. Any fluctuations to this value will likely be discrete in comparison to fluctuations in social beliefs variables and other determinants of demand for redistribution. In table 3 dynamic effects are tested by including a one-period lagged demand for redistribution value (in columns 2, 4 and 6). The effect of individual income is significant for all specifications.

Respondents’ views on the importance of being rich show significant negative effects on demand for redistribution in columns 2 and 4 of table 2. However, when country specific

(33)

31

control variables such as Gini index, GDP per capita and government social protection ex-penditure are added the measured effect turns very insignificant and even positive. The cor-relation on the pseudo data panel between the importance for respondents of being rich and the levels of inequality, public spending and country GDP per capita are considerable, being just below -0.4 for each. This could lead to some of the effects of the importance given to being rich to be captured by the other variables. The bad measure for cohorts’ public values becomes even more apparent when in table 3 the variable shows very large fluctuations in the estimated coefficients, which do not significantly differ from zero. The importance of being successful stated by respondents is only significant with a positive effect when the public values effect variables are measured exclusively (column 4). The pattern shown by the effects of both public value variables suggest that the high correlation between the two combined with difficulties that arise when trying to measure such abstract variables do not prove very powerful estimates when working with cohort averages (see appendix B table B.1 for correlations of the explanatory variables in the pseudo panel).

Of the social rivalry variables only the respondents’ feeling of safety and their beliefs towards the need for a strong government have significant effects. The measured effect of the former loses significance when levels of inequality and country’s GDP per capita are added. While some of the social rivalry effects fluctuate depending on the regression specification, others remain stable in table 3. The variable for feeling safe in the dark does not significantly differ from zero in any specification, and the effect fluctuates from positive to negative. In contrast, the effects of the importance of feeling safe and the importance of having a strong govern-ment remain quite stable in all specifications. The effect of importance to be safe is only significant below 0.5) when survey round is controlled for (column 3), while the importance of a strong government is no longer significant when country trends are introduced (columns 5 and 6). A possible reason for this is that there exist correlation between the country trends and trends in the respondents’ view on the importance for a strong government that ensures safety.

(34)

32

Tables 2 and 3 show that the country GDP per capita has a large and significant negative effect on demand for redistribution. Whether this is a causal relation is highly doubtful, since countries with low inequality tend to have higher GDP’s per capita, and therefore this coef-ficient could be picking up some of the effect on inequality. However, levels of inequality show very small effect on demand for redistribution only significantly different from zero in the dynamic models without country trends (columns 2 and 4). Possible explanations for this are that, like previously mentioned, the Gini index is an imperfect measure of income ine-quality and measures only income. Furthermore, variations in Gini indexes are relatively very small over time, making it likely that a large part of the effect is taken up by the cohort fixed effects (which are specified among other things by country).

Finally, table 3 shows that the estimated coefficient for the demand for redistribution in the survey fluctuates quite a lot. Perhaps surprisingly, in columns 4 and 6 the estimated effect is negative, which would suggest that the high demand for redistribution in the previous period would have a negative effect on this one. However, when a similar regression is run without variables containing information about time such as survey round and country trends (results not included in this thesis), all estimates remain positive. A possible explanation for this is that there is a negative bias in the OLS estimator of the autoregressive coefficient.

6.2.

Variation in effects over time

To test whether the effects of self-interest, public values and social rivalry vary over time, an OLS regression is performed on all six cross sections where the explanatory variables are replaced by their interaction with survey round (or year) dummies. Afterwards, an F-test is used to test whether the estimated coefficients differ significantly over time. The results of this regression are displayed in table 4. In addition to the explanatory variables shown in the table the regression controlled for unemployment level, country, survey round and country trends.

(35)

33

Table 4: Effect on demand for redistribution over time Demand for redistribution (x100)

Interaction variables: 2002 2004 2006 2008 2010 2012 F-test

Self-interest: (p-value)

- Individual income (% GDP) -4.41 -4.76 -6.28 -11.8 -10.5 -8.20 7.96

(0.82) (0.88) (1.30) (1.05) (1.26) (2.05) (0.00)

- Majority of income from labor? -1.26 -1.42 -1.56 -1.23 -1.35 -1.32 0.12

(0.83) (0.95) (0.66) (0.54) (0.69) (0.62) (0.99) Public values: - It is important to be rich -1.96 -1.75 -1.65 -1.69 -1.37 -1.62 1.18 (0.44) (0.34) (0.34) (0.36) (0.31) (0.38) (0.37) - It is important to be successful -0.86 -0.87 -0.91 -0.82 -0.91 -1.07 0.98 (0.28) (0.19) (0.32) (0.20) (0.18) (0.21) (0.46) Social rivalry:

- Immigrants are good for the economy

-0.24 -0.38 -0.29 -0.16 -0.31 -0.094 1.72

(0.22) (0.18) (0.14) (0.15) (0.14) (0.16) (0.19)

- You feel unsafe when walking in the dark

1.39 1.28 0.96 0.86 0.97 0.56 1.72

(0.33) (0.25) (0.48) (0.38) (0.37) (0.25) (0.19)

- Important to live in safe and secure surroundings

0.62 0.88 0.56 0.38 0.71 0.77 2.43

(0.28) (0.25) (0.21) (0.21) (0.28) (0.21) (0.08)

- Important government is strong and ensures safety

1.93 1.55 1.88 2.12 2.20 2.29 4.09 (0.34) (0.30) (0.32) (0.24) (0.31) (0.29) (0.02) Constant 67.3 (3.15) N 128209 R-squared 0.1336

Note: OLS regression with demand for redistribution as dependent variable and the variables on the left column as

inde-pendent variables using all six survey cross-sections pooled together.

Shaded area indicates variable in left column was interacted with survey round or year dummy (as seen on top of the shaded columns).

In the right most column the results of an F-test for equality of coefficients are displayed in italic. Standard errors clustered at country level.

Controlled for levels of unemployment, countries and survey rounds using dummy variables and for country trends using interaction variables of country dummy and survey round.

(36)

34

The self-interest variables show varying effects. Individual income shows negative effects on demand for redistribution for all six survey rounds significant for p-values below 0.01. Additionally, the effects show considerable variation over time; from 2006 onwards the ef-fect more than doubles in magnitude. An f-test with a value of 7.96 means we can reject the hypothesis that income effects are constant over time with a significance level below 0.001. The respondents’ source of income shows relatively constant negative effects over time with relatively constant standard errors. Only in three out of six rounds are the effects significant at 5% level.

Table 4 also shows that the variables measuring the public values effect ‘it is important to be rich’ and ‘it is important to be successful’ have constant negative effects over time. All co-efficients of both variables are significant with a p-value below 0.01 except for the variable ‘it is important to be successful’ in 2002 and 2006, with a p-value below 1% and 5% respec-tively. However, in neither case can the hypothesis be rejected that values are constant over time.

The effects of the statement ‘immigrants are good for the economy’ are negative for all sur-vey rounds, but only significant for 2010 with a p-value below 5%. People’s feelings of ‘safety in the dark’ do have a significant positive effect for all survey rounds except 2006. However, the effect seems to be diminishing over time, and with a significance level below 0.05 we can reject that they are constant over time, proving once again that the effects do vary. The same is true for people who find it important to live in safe and secure surroundings. Table 4 shows a positive and significant effect on demand for redistribution for all years except for 2008. Again, we can reject the hypothesis that effects are constant over time with a significance level below 0.1. Finally, people’s beliefs that the government should be strong and ensure safety have a significant (below 1%) positive effect on demand for redistribution. At face value it seems the effect is increasing over time, and an f-test confirms that, rejecting

Referenties

GERELATEERDE DOCUMENTEN

As I held her in my arms that last night, I tried to imagine what life would be like without her, for at last there had come to me the realization that I loved her-- loved my

It was some time before we could persuade the girl to continue her meal, but at last Bradley prevailed upon her, pointing out that we had come upstream nearly forty miles since

The results showed that new accounting standards have effect on how scale of company affect capital structure, market timing activities exist in Chinese market,

deduced that regularly spaced N-pulse configurations are more stable than any other N-pulse configuration – in fact, irregular patterns typically evolve toward regularity on

The first secondary research question is: “Which variables positively affect the actual demand for public transport services among seniors in the rural area of the municipalities

I do, therefore, agree with Kleerekoper that it is dangerous to arouse the expecta­ tion that we need only continue to study and develop, and we shall, within

This basic qualitative interview strategy offers some possibilities for finding “truth“ and relevance, but in order to dig deeper and in other directions four specific techniques

The present text seems strongly to indicate the territorial restoration of the nation (cf. It will be greatly enlarged and permanently settled. However, we must