Belief in Fairness, Political Institutions and Welfare Spending

Transcription

1 National University of Singapore From the SelectedWorks of Jiwei QIAN 2017 Belief in Fairness, Political Institutions and Welfare Spending Jiwei Qian eduardo araral, National University of Singapore Available at:

2 Belief in Fairness, Political Institutions and Welfare Spending Jiwei Qian a and Eduardo Araral b Abstract This paper seeks to understand how welfare spending is affected by beliefs about fairness. The political science literature points to the median voter theorem and the effects of political institutions. The economics literature argues that the difference between American and European welfare states can be explained by differences in subjective beliefs. The evolutionary psychology literature suggests that reciprocity is hardwired and universal. We add to the literature in three ways. First, we bring together the political science, psychology and economics literature by testing the relationship between welfare spending and beliefs about fairness controlling for the effects of political institutions. Second, the literature is mostly limited to mature liberal democracies (OECD) while we study both OECD and developing countries. Finally, we provide a nuanced distinction between two types of welfare expenditures which are often ignored in the literature - those that equalize outcomes and those that provide social insurance. We find that, consistent with the median voter hypothesis, a stronger belief about fairness has a statistically strong, positive, robust and significant relationship with government health expenditure, controlling for political institutions. We also find that belief about fairness is closely associated with expenditure on social insurance (e.g. health expenditure) but not with equalization spending such as education. Keywords: Health expenditure, Education expenditure, Belief, Median Voter, Political Institution, Welfare spending. a East Asian Institute, National University of Singapore and corresponding author jiwei.qian@nus.edu.sg b Lee Kuan Yew School of Public Policy, National University of Singapore sppaej@nus.edu.sg 1

3 Introduction This paper seeks to understand how welfare expenditure is affected by voters beliefs about fairness. The economics literature points to the importance of beliefs. For example, it is argued that much of the difference between the American and European welfare states can be explained by differences in subjective beliefs (Alesina, et al., 2001). The political science literature, on the other hand, points to the median voter theorem and suggest that the effects are mediated by political institutions such as voting rules, competition as well as the degree of fiscal and political decentralization (Persson and Tabellini, 2000). The evolutionary psychology literature suggests that belief about fairness is hardwired and universal (Cosmides and Tooby, 2005; Petersen, 2012). We add to the literature in several ways. First, we bring together the literature on politics, psychology and economics. We do this by testing the relationship between education/health policies and beliefs about fairness while controlling for the mediating effects of political institutions. Second, most studies are limited to OECD countries while we study both OECD and developing countries. Third, we control for a larger number of political institutions (e.g. political system, electoral rules, degree of decentralization, social trust, etc.) than is currently used in the literature. Fourth, the literature does not make a nuanced distinction between two types of welfare spending policies. We do so by differentiating between those that aims to equalize outcomes such as education policies and those that provide for social insurance. Our findings suggest that a strong belief about fairness has a statistically strong, positive, robust and significant effect on government spending on health, 2

4 controlling for the effects of political institutions. We also find that belief about fairness is closely associated with expenditure on social insurance (e.g. health expenditure) but not with equalization spending such as education. We conclude that belief about fairness can partly explain variations in certain types of welfare policies and that the effects are mediated by political institutions. Overall, our findings are consistent with the findings from the politics, economics and psychology literatures on the effects of belief about fairness. The rest of the paper is arranged as follows. In the next two sections, we discuss the theoretical background of our hypothesis and review the related literature. Then, we discuss the mechanism how the median voter s belief about fairness can have an impact on welfare spending. Data and methodology employed by this study are presented in the following section. We then show the results of benchmark regression about government health expenditure and comparisons with private health expenditure and government education expenditure. After the section of robustness checks, we conclude with the limitations and implications of the study. Belief about fairness and welfare spending We define belief about fairness in terms of the degree to which voters believe that success is due to luck or connection rather than effort. This definition follows that of the economics literature (Alesina and Giuliano, 2010). and the evolutionary psychology literature (Cosmides and Toby, 2006). In the recent economics of fairness literature in particular, success is regarded as fair if achieved through one s effort and not through luck or connection (Alesina and Angeletos, 2005). Belief about fairness is associated with welfare spending. There are two possible functions of welfare spending policies that are related to belief about fairness 3

5 according to the economics and political science literature. The first function of welfare spending is equalization. If one perceives that outcomes from the market are not fair in which luck or connection determine who are rewarded, one may prefer the government to intervene by transferring income from the rich to the poor. Conversely, if one believes that success is achieved by one s effort rather than luck or connection, market mechanism will compensate one s effort and demand for welfare spending will be lower. Empirical studies find that if people believe that success is the result of hard work, they would be more averse to welfare spending (Fong, 2001; Alesina and La Ferrara, 2005). The second function of welfare spending is to provide social insurance. If one believes that success is less likely to depend on effort, and that it is largely the result of luck or connection, one is likely to demand a wider coverage of social insurance to protect people in case an adverse event occurs (Iverson and Soskice, 2001; Moene and Wallerstein, 2001). Providing social insurance therefore is a significant part of government welfare spending. The evolutionary political psychology literature offers a different take on the relationship between people s opinions on welfare policies with their belief about fairness (Petersen, 2012; Petersen, 2015; Gilens, 2000; Oorschot, 2000; Sniderman et al., 1991). The literature suggests that people categorize welfare recipients either lazy or unlucky. Those they regard as unlucky deserve welfare but not the lazy ones. It suggests that this categorization evolved overtime similar to mechanisms that regulate small-scale help giving in ancient societies -those that deserve help will be given help. Petersen calls this the deservingness heuristic and suggests that this heuristic is also what voters use to think about large scale welfare programs. 4

6 Political institutions and welfare spending There are other important determinants for welfare spending apart from beliefs about fairness. In particular, political institutions are critical for welfare spending decisions since these policies are made via political processes. Indeed, the literature has postulated a variety of political institutions associated with welfare expenditure, policies and outcomes. These include the system of government and political traditions, government ideology, political commitment, decentralization, public participation, policy design, fiscal decentralization, party competition, political timing and coalitions, and regulatory capture, among others (Häusermann et al., 2013; Marmor, 1983; Miller, 2005; Navarro et al., 2006; Reich, 1996). From the literature on comparative political economy, government expenditure in a country under parliamentary system is higher compared with a presidential system because of the costs associated with organizing and maintaining a ruling coalition (Persson and Tabellini, 2003). Studies have shown that the competitive elections in the legislative and executive branches of government have statistically positive and significant effect on welfare policies and outcomes (e.g. (Persson and Tabellini, 2003). However, it is suggested that in some developing countries with competitive electoral systems, politicians find it difficult to make credible promises about service provision such as health care because it is hard for voters to assess improvements and to hold politicians accountable (Keefer and Khemani, 2003). Electoral rules matter also. Under the proportional electoral system, house seats are allocated by the share of votes. On the contrary, under plurality electoral 5

7 system, only the candidate with the majority of votes will be elected. Public expenditure is supposed to be higher under the proportional system since transfers to a broader coalition are necessary to win an election (Persson and Tabellini, 2003). The impacts of decentralization on provision of public services and outcomes are mixed and subject to on-going debates. On one hand, Robalino, Picazo and Voetberg, using panel data of rich and poor countries, argue that fiscal decentralization has positive impacts on health policies and outcomes (Robalino, et al., 2011). In contrast, Bossert and Beauvais, employing three case studies, argue otherwise (Bossert and Beauvais, 2002). Amounts of welfare spending may also vary with regime types (i.e. democratic vs. nondemocratic). Compared to democratic regimes, nondemocratic regimes may spend less for welfare since they have to accommodate special interests groups rather than voters in general (Mares and Carnes, 2009). Model of Welfare Spending In this section, we propose a mechanism of welfare spending to show how the median voter s belief about fairness, political institutions and welfare spending are linked with each other. We assume that people s preferences over welfare spending are single-peaked when applying the median voter theorem. We illustrate this mechanism in Figure 1. First, welfare spending is subject to various political institutions (i.e. arrow 1). Second, the median voter s belief about fairness may have an indirect effect on government s welfare spending via political institutions (i.e. arrow 2). Different median voters beliefs may change political institutions such as its degree of competitiveness. For example, if the median voter 6

8 prefers low welfare spending and low tax, which is associated with the belief about fairness(alesina and La Ferrara, 2005), it would be less competitive in the policy space if only the right wing party proposes low tax rate and low spending and other parties propose high spending. However, it will be more competitive if the median voter prefers a tax rate located between the level proposed by left wing and right wing party. Political Institutions Political System Electoral rule Political competition Fiscal decentralization Policy Outcome Government welfare Expenditure (e.g. government health expenditure or education expenditure) The impact of belief over policies depends on which function a redistributive policy fulfills Belief /Ideology (i.e. social insurance/equalization) The Median Voter s Belief about fairness Figure 1: The Model of Beliefs, Political Institutions, and Welfare Policies 7

9 Third, anticipating the median voters belief about fairness, forward looking politicians choose welfare spending policies to provide equalization and social insurance for voters to win their support (i.e. arrow 3). This avenue of causal relation is what the median voter theorem literature has proposed: welfare spending policies have to meet the expectation of the median voter s preference for welfare, independent of the effects of political institutions. For example, according to a recent study on European welfare states, although Germany, Netherlands and the United Kingdom have different electoral rules, political system and degree of fiscal decentralization, they all experienced a very similar sequence of political and policy changes in terms of work-family policies such as lengthening parental and maternity leave (Morgan, 2013). It is argued that these changes are results of increasing significance of working women in these countries as voting constituencies (Morgan, 2013). This is an example to illustrate that the changing preference of the median voters (e.g. with the growing significance of working women) has a direct impact on welfare spending, despite variations in political institutions. For non-democratic regimes, while welfare expenditure is more likely to be targeted for special interests to maintain political stability, governments may still want to spend budget for welfare to accommodate the belief of masses (Acemoglu and Robinson, 2006). In this case, the median voter s belief about fairness may be interpreted as a proxy measurement for masses belief about fairness. One more issue about this arrow 3 is that the median voter s belief about fairness may have heterogeneous effects over different welfare policies. In particular, 8

10 equalization (i.e. equalizing between the rich and poor) and providing social insurance (i.e. supporting people hit by adverse event) are two functions that government welfare policies fulfill. If people s belief about fairness is more closely associated with one function of welfare policies than the other, the median voter and politicians are more likely to respond to welfare spending policies addressing that function (i.e. either equalization or social insurances). Similarly, forward-looking politicians may not spend enough effort on welfare policies which do not serve that particular function, to accommodate the median voter s belief. In this paper, we examine the significance of arrow 3 in Figure 1 regarding the effect of the belief about fairness on welfare spending policies. Note here that the mechanism in this paper does not suffer from the individualistic fallacy (Przeworski and Teune, 1970), which refers to the case imputing the macro outcomes with the aggregate belief of individual members. According to the mechanism discussed in this section, government welfare expenditure is associated with individual level beliefs via the political processes such as voting in which the median voter s preference is pivotal. In other words, we do not draw macro level conclusions directly from individual level findings. Instead, the individual level belief is translated into the macro-level variable such as government welfare spending only via the political processes (i.e. the median voter theorem). Data & Methodology Data for the belief about fairness are taken from the World Values Survey while data for political institutions are taken from Database of Political Institutions coded by the World Bank economist Philip Keefer (Keefer, 2010). Education, health expenditure 9

11 and GDP data are obtained from the World Bank. Data for Gini coefficient and the share of people with age 65 years or older in the total population are collected from World Bank Database as well as the CIA World Factbook. We collect government expenditure as a share of GDP from Index of Economic Freedom compiled by the Heritage Foundation. Data for political institutions include indexes for different political systems, electoral rules, degree of political competition and degree of fiscal decentralization. For the belief data, we pool three waves of World Values Survey (WVS) between 1989 and After combining these two data sources, we have 62 country-level observations in total (vs. 29 countries studied in a recent research, most of which are OECD countries, see Alesina and Angeletos, 2005). These 62 countries account for 60% of the total world population and over 76% of the total world GDP in The names of these countries are listed in Table 1. 10

12 OECD Non-OECD 1 Australia 1 Albania * 21 India 2 Canada 2 Argentina 22 Indonesia 3 Chile 3 Armenia 23 Iran (Islamic Republic of) 4 Czech Republic 4 Azerbaijan 24 Latvia 5 Estonia 5 Bangladesh 25 Lithuania 6 Finland 6 Belarus 26 Macedonia 7 France 7 Bosnia and Herzegovina * 27 Malaysia 8 Germany 8 Brazil 28 Mali 9 Italy 9 Bulgaria 29 Nigeria * 10 Japan 10 Burkina Faso 30 Pakistan 11 Mexico 11 Colombia 31 Peru 12 Netherlands 12 Croatia 32 Philippines 13 New Zealand 13 Cyprus 33 Republic of Moldova 14 Poland 14 Dominican Republic 34 Romania 15 Slovakia 15 Egypt * 35 Russian Federation* 16 Spain 16 El Salvador * 36 Rwanda 17 Sweden 17 Ethiopia 37 South Africa * 18 Switzerland 18 Georgia 38 Ukraine * 19 Turkey * 19 Ghana 39 Uruguay * 20 United Kingdom 20 Hungary 40 Venezuela * 21 United States of America 41 Zambia * *observations missing for the case of the education expenditure Table 1: List of 62 countries in our sample The independent variable in our paper the belief about fairness - is measured by the mean value of responses for each country for the following two identical but opposite statements in the WVS. Responses are coded on a scale between 1 and 10, 11

13 with 10 indicating the strongest belief in luck and connections as determinants of success. Statement 1: In the long run, hard work usually brings a better life. Statement 2: Hard work doesn t generally bring success it is more a matter of luck and connections. Reponses to above WVS statements are used to show the association between the belief about fairness and welfare spending in the earlier literature (e.g., Alesina and Angeletos, 2005; Alesina et al., 2001). In this paper, we use cross-sectional Ordinary Least Squares (OLS) regression to find the association between the median voter s belief about fairness and government welfare expenditure. How to identify the effect of the median voter s belief about fairness on welfare spending is the key methodological concern for our study. We are faced with three major methodological issues in identifying this relation. First, the reverse causality problem may arise given the possibility that direction of causation may be from welfare spending policy to the median voter s belief (i.e. a welfare spending fosters an entitlement belief among voters). For example, Alesina and Angeletos suggest that the median voter s belief about fairness leads to multiple equilibria in the level of taxation and welfare spending (Alesina and Angeletos, 2005). Their model implies that a citizen anticipates a higher tax rate and his efforts (work ethics) in turn will be lower if much of his income will be subject to a high tax rate. This lower level of efforts then justifies his belief about fairness. Also, Lindbeck and Nyberg argue that welfare weakens the work ethics of people in welfare states (Lindbeck and Nyberg, 2006). 12

14 In addition, there is another possibility of reverse causality. Variances in economic and social conditions (i.e. demand for healthcare and education), which are associated with different amounts of government welfare expenditures, may lead to different preferences for welfare. We address the issue of reverse causality in two ways. First, we use predetermined values for control and independent variables in our regression analysis to address the concern of reverse causality. In other words, given the timing of dependent variables, we choose values of the control and independent variables from the earlier period. Second, we compare our benchmark regression results with results of models regressing private health expenditure as a share of GDP on beliefs. Here, private health expenditure s share in GDP is considered to be a proxy measurement indicating demand for healthcare. If private health expenditure s share in GDP is not correlated with the median voter s beliefs about fairness, we can exclude the possibility that different levels of demand of healthcare lead to different beliefs about fairness. Second, omitted variable bias is another concern. Types of political regimes and types of health financing institutions are not included in the data analysis. Types of political regimes matter since authoritarian regimes may respond to elites belief rather than voters belief. We address the issue for the types of political regimes by using Polity IV index in the robustness check section. Regarding to the types of health financing institutions, government health expenditure can be a function of different institutions used to organize health systems such as national health system (e.g. in UK or Italy), social insurance system (e.g. in Germany or Japan), or private insurance oriented health financing system (e.g. in Netherland). However, data for different institutions of health financing is not available for this research. 13

15 Also, apart from belief about fairness, people s perceptions about social trust or government responsibility may have impacts on the government social expenditure. People with higher social trust may be more likely to vote for higher tax rate and higher public expenditure (Algan et al., 2016; Bergh and Bjørnskov, 2014; Daniele and Geys, 2015). Voters believing that government is responsible for people s wellbeing are more likely to prefer a larger amount of government health expenditure. In the robustness check section, we address this issue by adding variables measuring social trust and perceptions for government responsibility as control variables. Third, measurement error is another concern. The mean values of responses to the statements in the WVS for a given country are nevertheless time variant (see Bisin and Verdier, 2000, 2017) and the responses collected in the earlier waves may not be a good proxy for the median voter s belief in the later period. In the robustness check section, we address this issue by using the latest wave instead of all waves of responses in the WVS as the measurement for the belief about fairness. We acknowledge that a cross-sectional study such as this may not be able to fully identify the causal relation between the median voter s belief and public expenditure outcomes. Both reverse causality and omitted variable bias issues can be raised in this context. Such causal relation between government health expenditure and the belief about fairness can only be identified from both time series variation and cross sectional variations, which will be left to future research. However, we believe that identifying the correlation between the median voter s belief and welfare policies is a significant first step understanding the effects of belief and institutions in welfare policies. 14

16 Variable Definition and Coding Our variables consist of 1) outcome variables (welfare spending represented by government education/health expenditure as a share of GDP and total expenditure); 2) ideology variable: the median voter s belief about fairness; 3) political institution variables including type of political system, competitiveness of elections, electoral rules, and fiscal and political decentralization, etc.; 4) control variables including GDP per capita, Gini coefficient, share of people aged 65 and over in total population, and share of under-14 years old in total population in We used these control variables for a number of reasons. First, government spending for welfare depends on its fiscal capacity which in turn depends on a country s income level (i.e. GDP per capita) via taxation. Also, people in countries with different income levels may have different demand for health care and education. GDP per capita then is a proxy measurement for both fiscal capacity and demand for health care/education in a country. Second, income Gini coefficient is included since the level of welfare spending is expected to be positively associated with the degree of income inequality, i.e. theoretically we expect higher welfare spending in countries with higher income inequalities (Meltzer and Richard, 1981). Third, the share of people aged 65 years old and over is a proxy variable to show the demand of health service. The higher the share of elderly population, the demand of healthcare services is likely to be higher. For instance, it is estimated that health expenditure for people over 65 years old usually is three to five times to that of other age groups in OECD countries (Casey et al., 2003). Similarly, a higher share of 15

17 under-14 years old in total population is associated with a larger share of public education expenditure to be spent on primary and secondary schools. Values of all dependent variables are calculated based on 2010 data. The reason to use 2010 data is that we have less missing values of observations for these dependent variables in 2010 compared to those of other years. For control and independent variables including ideology and political institution variables, as we mentioned earlier, we use values before 2010 such that they can be treated as predetermined when the values of the outcome variables are generated. Table 2 summarizes the list of variables, their definition and coding. Note here that following the literature (Keefer, 2007), the variable poli_com measures the competitiveness of elections in the executive and legislative branch of government (i.e. poli_com =1 when both electoral competition for both legislative positions and executive positions are high enough (i.e. with score 7). This variable is also an important measurement to distinguish democratic and nondemocratic regimes. For example, in our database, over 67% of nondemocratic regimes score no more than 3 out of 7 regarding to electoral competition for both legislative positions and executive positions in over 4,900 observations between 1946 and 2008 (Svolik, 2012: 36-7). In other words, the regime types (democratic and nondemocratic) are being taken into account for regression exercises in this paper by adding these two variables. Also, in the robustness check section, we use the indexes in Policy IV dataset to exclude the observations of authoritarian regime to check the validity of our results. 16

18 Variables Coding Note Outcome Variables GHE_ share_gdp GHE_share_Fiscal_Expenditure Government Health expenditure as a share of GDP, (%) in 2010 Government Health expenditure as a share of total government expenditure, (%) in 2010 Source: World Bank database Source: World Bank database PHE_share_GDP GEE_share_Fiscal_Expenditure GEE_share_GDP Ideology Variables Belief trust Gov_resp Private Health Expenditure as a share of GDP, (%) in 2010, Government Education Expenditure as a share of total government expenditure, (%) in 2010 Government Education Expenditure as a share of GDP, (%) in 2010 The value locates between 1 and 10. The higher the value, the more people appreciate the redistribution. This variable is defined by the mean value of answers for the question in the world value survey (pooling three waves between 1989 and 2008). There are 15, 45 and 42 countrylevel observations in three waves (89-93, 94-98,05-08). There is one wave of responses in 33 countries. There are two and three waves of responses in 19 and 10 countries respectively. The value is mean value of answers to the following question in WVS. Generally speaking, would you say that most people can be trusted (1) or that you need to be very careful in dealing with people (2)?. The answer is coded as either 1 or 2. The value is mean value of answers to the following question in WVS. the government (or people) should take more responsibility to ensure responsibility to that everyone is provided for. The answer locates between 1 and 10. Source: World Bank database Source: World Bank database Source: World Bank database Source: World value Survey Source: World value Survey Government responsibility Source: World value Survey 17

19 Political Institutions Variables System Parliamentary (3), Assemblyelected President (2), Presidential (1) Measuring political system. source: Database of Political Institutions Poli_com Pol_Decent_Muni Pol_Decent_State Fiscal_decent Majority Control Variables Poli_com=1 if both Eiec and Liec equal to 7* 0 if neither local executive nor local legislature are locally elected; 1 if the executive is appointed, but the legislature elected; 2 if they are both locally elected. 0 if neither State executive nor State legislature are locally elected; 1 if the executive is appointed, but the legislature elected; 2 if they are both locally elected. 1 if the state/provinces have authority over taxing, spending, or legislating. 0 otherwise Majority= 1 if most seats are plurality, Majority= 0 if most seats are proportional. Gini gini coefficient in 2009 GDP per capita GDP per capita in year 2010 Measuring political competition. source: Database of Political Institutions Measuring political decentralization. source: Database of Political Institutions Measuring political decentralization. source: Database of Political Institutions Measuring fiscal decentralization. source: Database of Political Institutions Electoral rules for the house seats, source: Database of Political Institutions Source: World Bank database Source: World Bank database Share of people over 65 years old Source: World Bank Age65 total population in 2010 (%) database * Poli_com =1 when both two following two variables Liec and Eiec, which are defined below, equal to 7: Liec: Intensity of electoral competition for legislative positions, average of ordinal ranking from 0 (non-existent) to 7 (highly competitive) from 1975 to Ordinal ranking: No legislature=1; Unelected legislature=2; Elected, 1 candidate=3; 1 party, multiple candidates=4; multiple parties are legal but only one party won seats=5; multiple parties DID win seats but the largest party received more than 75% of the seats=6; largest party got less than 75%=7 Eiec: Intensity of electoral competition for executive positions, average of ordinal ranking from 0 (non-existent) to 7 (highly competitive) from 1975 to Ordinal ranking: No executive =1; Unelected executive =2;Elected, 1 candidate=3; 1 party, multiple candidates=4; multiple parties are legal but only one party won the executive position=5; an executive elected in a competition with multiple parties and received more than 75% of the votes=6; the candidate got less than 75%=7 Table 2: Summary of variables and coding 18

20 Hypothesis Based on the preceding review of the literature and the theory that have been postulated, we test the following hypothesis on the relationship between belief about fairness and expenditure on welfare. The less the median voter believes that success is the result of hard work, the more likely that government will spend a higher percentage of GDP and government expenditure on education and health, controlling for the effects of political institutions. Results Table 3 summarizes our descriptive statistics. The standard deviations are relatively large across countries for government health expenditure as a share of fiscal expenditure and GDP. The health expenditure as a share of fiscal expenditure varies from 3.5% to over 20%. The maximum value of belief on the connection and luck matter is over 6 while the minimum value is about 2.3. Table 4 shows the correlation among independent variables. It suggests that multicollinearity is not a big concern since most correlation coefficients are lower than 0.6. We also calculate the variance inflation factors (VIF) for each variable and the highest value of VIF is 5.06, which is much lower than the widely-used threshold value 10. This value suggests that multicollinearity is not a big concern. 19

21 Variable obs Mean Std. Dev. Min Max GHE_share_Fiscal_Expenditure (%) GHE_share_GDP (%) GEE_share_Fiscal_Expenditure (%) GEE_share_GDP (%) PHE_share_GDP (%) belief system poli_com majority fiscal_decent poldecent_muni poldecent_state Gini gdp_per_capita (US$ 10,000) age65 (%) age14(%) trust gov_resp Table 3: Descriptive statistics 20

22 Belief system poli_co m gini gdp_p er_cap ita age65 majority trust belief 1 system poli_com gini gdp_per_cap ita age majority trust gov_resp gov_res p Variable VIF 1/VIF gdp_per_capita age trust gov_resp gini belief poli_com system majority Mean VIF 2.60 Table 4: Correlation Matrix and VIF 21

23 Our benchmark OLS regression results are summarized in Tables 5 and 6 below. Table 5 shows the results of our regressions when the dependent variable is the government health expenditure as a share of GDP or government fiscal expenditure. Table 5, shows that a stronger belief about fairness indeed is associated with higher government health expenditure as a share of GDP as well as government expenditure, conditional on political institutions such as types of political system, fiscal decentralization and the extent of electoral competition. Model (1) in table 5 includes both political institution variables and the median voter s belief about fairness as regressors. We find that the coefficient of the median voter s belief about fairness is statistically significant in model (1), consistent with our theoretical expectations. Model (2) adds the fiscal decentralization variable as well as the interaction terms between fiscal and political decentralization as regressors. These variables are added to control the effect of fiscal decentralization and how it varies with political decentralization. We find that the coefficient of belief about fairness is statistically significant. From model (1) and (2) in table 5, one standard deviation of the belief about fairness above its mean is associated with an increase of government health expenditure by 0.38 to 0.45 percent of GDP 1. The capacity for welfare spending, of course, depends on fiscal capacity. Government health expenditure is subject to state s fiscal capacity when politicians are willing to spend more on health. Taking into account government s fiscal capacity, 1 These numbers are calculated by (stand deviation of the variable belief) times the coefficients. 22

24 models (3) and (4) use the government health expenditure as a share of total government fiscal expenditure. The coefficient of the median voter s belief about fairness is statistically significant in models (3) and (4). These results imply that the government health expenditure is more likely to be adjusted by politicians as a response to accommodate the median voter belief about fairness, compared to other government expenditures. From model (3) and (4) one standard deviation of the belief about fairness above its mean is associated with an increase of government health expenditure by about from 1.25 to 1.35 percent of total government expenditure 2. The coefficient of Gini is positively significant in all models except model (2). Also, GDP per capita is significantly and positively associated with the government health expenditure s share of GDP for all models. The signs for control variables GDP per capita and Gini coefficient are consistent with the result in other studies (e.g. Alesina and Angeletos, 2005). The coefficient of political system is significant in all models except model (2), which implies that in the parliamentary system, the government is likely to spend more on health care compared to the case under the presidential system. The coefficient for competitiveness is significant and negative in model (3) while the rules of election are not statistically significant. It is conceivable that, given the median voter s preference about fairness, absent of competitive elections may provide a government with more discretion for spending. The coefficients of political and fiscal decentralization and electoral rules are not statistically significant. 2 These numbers are calculated by (stand deviation of the variable belief) times the coefficients. 23

25 GHE_share _GDP GHE_share _GDP GHE_share_Fiscal_ Expenditure GHE_share_Fiscal_ Expenditure (1) (2) (3) (4) Belief 0.466* 0.546* 1.513** 1.627* (0.241) (0.286) (0.703) (0.830) gdp_per_ capita 0.730*** 0.706** 1.623*** 1.793*** (0.212) (0.295) (0.324) (0.490) Gini 4.975** *** 20.83** (2.251) (2.575) (6.422) (8.153) age ** (0.0540) (0.0808) (0.138) (0.182) System 0.461* ** 1.649** (0.236) (0.286) (0.599) (0.793) poli_com * (0.410) (0.597) (1.079) (1.749) majority (0.402) (0.460) (0.965) (1.203) fiscal_dec ent fiscal_dec ent* Pol Decent State fiscal_dec ent* Pol Decent Muni (1.198) (0.373) (0.693) (3.332) (1.235) (1.675) constant (1.700) (1.906) (4.496) (5.589) N adj. R Note: GHE_share_GDP denotes the share between government health expenditure and GDP GHE_share_Fiscal_Expenditure denotes the share between government health expenditure and government expenditure Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01 Table 5: Government Health Expenditure and Belief 24

26 Interestingly, the coefficient of the share of people over 65 years old in total population is not statistically significant in all models but model (1). These results are conceivable if both government expenditure in general and government health expenditure in particular increase with the share of elderly people. Total government expenditure may also increase with the share of elderly people in total population though other avenues (e.g. pension outlay accounts for considerable part of government expenditure and increasing share of elderly people may lead to increasing government expenditure on pensions). In this case, the share of elderly people may not have significant impact on the share of health expenditure in total government expenditure given that both numerator (i.e. government health expenditure) and denominator (i.e. total government expenditure) increase with the share of elderly population. In Table 6, we replace our dependent variable with government education expenditure as a share of government expenditure to check whether the belief about fairness is associated with government education expenditure. From earlier literature (Alesina et al., 2001; Alesina and Angeletos, 2005), the median voter s belief about fairness may be associated with redistributive social spending also, which may include government income support for individuals such as pension, sickness, unemployment, social assistance as well as social services such as health care and family services such as childcare allowance. Public education expenditure, especially government expenditure on compulsory education, also serves some function of redistribution for the development of human capital and it is expected to be associated with the median voter s belief about fairness. 25

27 We collect government education expenditure in 2010 from the World Bank data sets. We end up with 50 observations in our database. The dependent variables are government education expenditure as a share of GDP and government fiscal expenditure. We include both the median voter s belief about fairness and indexes of political institutions as our regressors. Further, we add a control variable, proportion of people in age 65 or older, with the share of under-14 years old in total population in 2010 (variable name: age14). The regression results for government education expenditure are shown in Table 6 above. We find that for all models, the sign of coefficient of the variable Belief is positive. However, in contrast to the case of government health expenditure, the coefficient of the median voter s belief about fairness is not statistically significant for government education expenditure in all models except model (2). The coefficient of the share of population aged 14 or younger is statistically significant in models (3) and (4). This is not surprising because much of government expenditure is mainly targeted for certain age groups in the phase of compulsory education. 26

28 GEE_share_ GDP GEE_share_ GDP GEE_share_Fiscal_Ex penditure GEE_share_Fiscal_Ex penditure (1) (2) (3) (4) Belief ** (0.236) (0.208) (0.706) (0.843) gdp_per_c apita (0.272) (0.257) (0.344) (0.393) gini * (0.548) (0.409) (7.084) (7.477) age ** 0.226** (0.412) (0.384) (0.0779) (0.0902) system (2.957) (2.739) (0.813) (1.000) poli_com (0.0251) (0.0204) (1.378) (1.757) majority ** (0.137) (0.117) (1.126) (1.287) fiscal_dec ent fiscal_dec ent* Pol Decent State fiscal_dec ent* Pol Decent Muni (1.009) (0.415) (0.590) (2.801) (1.104) (1.746) constant 4.095** (1.997) (1.846) (5.881) (7.316) N adj. R Note:GEE/GDP denotes the share between government education expenditure and GDP GEE_share_Fiscal_Expenditure denotes the share between government education expenditure and government expenditure. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01 Table 6: Government Education Expenditure and Belief 27

29 Comparison between health and education expenditure Table 5 and 6 shows some interesting results and implications. It suggests that belief about fairness varies between policies that help equalize outcomes (such as education policies) and policies that provide social insurance (e.g. health expenditure). This result can be explained perhaps by the differences between these two types of welfare policies. Social insurance spending provides support to households and individuals who may be suffer from adverse shocks. The political psychology literature suggests that reciprocity is hardwired in human psychology. Voters expect politicians to help those who deserve to get help in the face of catastrophic events. In contrast, education spending serves to equalize the gap between the rich and the poor by ensuring mobility in the labor market. It gives a leg up to the poor but it does not serve the same purpose as social insurance in helping those in catastrophic events. Perhaps this difference can be explained by the availability heuristic in psychology which suggests that voters rely on immediate examples that come to mind when categorizing recipients of welfare spending. Our conclusion is also consistent with the evolutionary political psychology literature via the deservingness heuristics. The literature suggests that people categorize welfare recipients either as lazy or unlucky. Those they regard as unlucky deserves welfare but not the lazy ones. The literature suggests that this categorization evolved overtime similar to mechanisms that regulate small-scale help giving in ancient societies -those that deserve help will be given help. This categorization is also what voters use to think about large scale welfare programs. 28

30 Evolutionary game theory also makes the same conclusion: selfish individuals for their own selfish good will tend to be nice (Axelrod, 1984). This is nothing new. In fact, folk theorems such as the golden rule makes the same point i.e. do unto others what you want others do unto you. It is in one s selfish interest to be fair to others because the same adverse situation can also befall on you. In short, belief about fairness seem to be hardwired and universal. Robustness Checks In this section, we perform several robustness checks on our findings on the impact of belief about fairness on health expenditure. One may argue against our earlier results that the variance in the demand for healthcare may have an effect on the median voter s belief. For example, one who is healthy and thus has lower demand for healthcare is more likely to spend less on health and also is more likely to believe that success is a result of hard work. Hence, the variations of government expenditure may not have been caused by different beliefs but different demand for healthcare. To show that our earlier results are robust to this threat, we replace the dependent variable with the private health expenditure as a share of GDP in all models. If private health expenditure is not correlated with belief, we can infer that health condition or demand for healthcare is not correlated with beliefs about fairness. Also, politicians in authoritarian regimes may respond to elite s interests rather than the median voter s belief. To exclude the possibility that it is the authoritarian regimes which drive the regression results, we have another robustness check by removing three observations from the authoritarian regimes. Following the 29

31 literature, we define authoritarian regimes as countries scoring -6 or below on the Polity IV Index in year Table 7 shows that the coefficient of the belief about fairness is not statistically significant for private health expenditure in models (1) and (2). These results suggest that health condition or demand for health care in a country, for which private health expenditure is a proxy measurement, is not correlated with the median voter s belief about fairness. Therefore, it is unlikely that variations of health conditions or demand for health result in different beliefs about fairness. From this comparison between the government and private health expenditure, we are able to show the reliability of our benchmark results that government health expenditure is statistically significantly associated with the median voter s belief about fairness. In models (3) and (4) of Table 7, after excluding three observations from the authoritarian regimes, the coefficient of the median voter s belief about fairness is statistically significant in all models while the magnitude of the coefficient is similar to the result reported in table 5. 3 See Brancati (2014). These authoritarian regimes include Azerbaijan, Belarus and Iran. 30

32 PHE_share_ GDP PHE_share_ GDP GHE_share_Fiscal_ex penditure (excluding authoritarian regimes) GHE_share_Fiscal_ex penditure (excluding authoritarian regimes) (1) (2) (3) (4) Belief ** 1.963** (0.281) (0.301) (0.619) (0.763) gdp_per_c apita *** 1.698*** (0.174) (0.212) (0.341) (0.545) Gini *** 18.61** (2.253) (2.484) (6.381) (7.952) age (0.0596) (0.0639) (0.144) (0.195) system * ** 1.519* (0.293) (0.352) (0.614) (0.779) poli_com * (0.419) (0.444) (1.119) (1.742) majority (0.368) (0.376) (0.989) (1.232) fiscal_dece nt fiscal_dece nt* Pol Decent State fiscal_dece nt* Pol Decent Muni (0.792) (0.328) (0.438) (3.532) (1.220) (1.716) constant 3.838*** 2.739* * (1.296) (1.580) (4.084) (4.888) N adj. R Note:PHE_share_GDP denotes the share between private health expenditure and GDP GHE_share_Fiscal_Expenditure denotes the share between government health expenditure and government expenditure. Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01 Table 7: Health Expenditure and Belief 31

33 In Table 8, to check the robustness of our results, we also include the variables of social trust and perception of government responsibility as control variables. The median voter with higher social trust or higher expectation on government responsibility for people s welfare may prefer larger welfare expenditure (Bergh and Bjørnskov, 2014; Daniele and Geys, 2015). We use the mean value of responses to the question about social trust and government responsibility in the WVS. The question about social trust variable in WVS is as follows: Generally speaking, would you say that most people can be trusted (1) or that you need to be very careful in dealing with people (2)?. The variable trust denotes the mean value of responses to above question (1 or 2). The question about perceptions of government responsibility in WVS is the government (or people) should take more responsibility to ensure responsibility so that everyone is provided for. The variable gov_resp denotes the mean value of the responses to above question (scale from 1 to 10). We report the results in the Table 8. The results in table 8 show that, after including variables for social trust and government responsibility, the coefficient of the belief about fairness remains statistically significant while coefficients of social trust and government responsibility are not significant in all models. Also, as reported in table 4, correlation among these three perception variables - belief about fairness, social trust and government responsibility - is relatively low (i.e. less than 0.11). 32