the Normativeview

In the normative view of the reference group, an individual evaluates the overall degree of income inequality in the reference group, but without making any comparisons to individuals who are richer or poorer than she is.

As in Section 13.2, regarding the comparative view of the reference group, there is evidence on the normative view of the reference group from both subjective well-being research and experimental analysis.

13.3.1 Inequality and Well-Being: What Do People Say?

We are interested in this chapter, as the title suggests, in individuals’ attitudes toward or opinions about inequality. There are a number of ways in which these can be elicited, including direct questioning, experimental approaches, or inference from observed behaviors. In this subsection, we consider the contribution of “happiness economics,” in which some measure of income inequality is related to the individual’s self-reported well-being. In general, an equation similar to the following is estimated:

In this approach, we collect survey information on the subjective well-being of an indi­vidual i, living in some aggregate area j (where j is often, but not always, a country) at time t. This subjective well-being is related to a vector of standard demographic vari­ables (age, sex, education, labor-force and marital statuses and almost always the indi­vidual’s or the household’s income) through the vector #946;. Of most interest to us here is the conditional correlation (i.e., controlling for all the variables in the vector X) between well-being and the aggregate measure of inequality in area j, Ineq_jt. The esti­mated value of the parameter #947; shows us whether individuals, ceteris paribus, tick up or down their self-reported well-being scores in areas with higher or lower levels of income inequality.

The estimation of an equation like Equation (13.3) allows the “value ofinequality,” as it were, to be inferred from the empirical relationship between the observed inequality around the individual and their reported level of subjective well-being. This latter is most often measured by questions about the individual’s happiness, life, and income satisfac­tion or some other measure of general psychological functioning. Multivariate regres­sions allow not only the sign of the conditional correlation between income inequality and subjective well-being to be established (#947; shown earlier), but also the eco­nomic importance of any relationship that is identified (via the comparison of #947; to some of the estimated #946; coefficients on other variables, such as income or unemployment).

This “happiness” approach to valuing public goods has now appeared a number of times in the subjective well-being literature. Some well-known pieces of work in this respect have considered inflation and unemployment (Di Tella et al., 2001), aircraft noise (Van Praag and Baarsma, 2005), and pollution (Luechinger, 2009), although there are by now many other applications.

Cross-section and panel data allow the happiness or satisfaction of tens or even hun­dreds of thousands of individuals to be measured. It is perhaps easy to get carried away by the sheer number of degrees of freedom here. Except that, as we suggest later, this is largely illusory: Although it is theoretically possible for each individual to be confronted with a different income distribution, the most common approach has been to take cross-country data, often repeated cross-section, and include the country-level Gini

coefficient (or something else) on the right-hand side of a satisfaction regression. In this case, the effective number of degrees of freedom in the empirical estimation remains for the most part at the two-digit level.^[862]

Although there are by now many thousands of empirical contributions across the social sciences that relate individual income to some measure of individual well-being, it remains true that only a small fraction of this existing work has considered any role for income inequality.

Perhaps the earliest contribution in economics is Morawetz et al. (1977), which con­trasts two different Israeli communities and shows that the level of happiness is higher in the community with the more equal income distribution. Although interesting, the result essentially relies on two observations and does not control for all of the other factors that might differ between the two communities. A contribution that is more in the regression framework is an innovative article by Tomes (1986). This uses data (from the 1977 Quality of Life Survey) on individuals in approximately 200 Federal Electoral Districts in Canada. Matching in census data on income distribution, it is shown that the share of income received by the bottom 40% of the population is negatively correlated (at the 10% level) with both satisfaction and happiness for men. The same correlations are insignificant for women. Inequality is thus positively correlated with men’s subjective well-being.

Hagerty (2000) is the first of a number of contributions to use U.S. General Social Survey (GSS) data. In his GSS sample from 1989 to 1996, maximum community income and the skew of community income are, respectively, negatively and positively correlated with happiness scores. Hagerty also used aggregate data from eight different countries to show that average happiness is lower in countries with wider income distributions. More recent work using the GSS has, however, come to a variety of results. Whereas Blanch- flower and Oswald (2003) and Oishi et al.

SWB

Continued

SWB

Schwarze and Harpfer (2007) calculated inequality in gross household income at the region and year level in 14 waves of German SOEP data. Life satisfaction is found to be negatively correlated with inequality (although a measure of income redistribution is not significant). Other work establishing a negative correlation between inequality and well-being includes Biancotti and D’Alessio (2008), Brodeur and Fleche (2013), Ebert and Welsch (2009), Oshio and Kobayashi (2010), Verme (2011), Van de Werfhorst and Salverda (2012), and Winkelmann and Winkelmann (2010), using data from a wide variety of different countries.

On the opposite side of the court, a number of contributions have instead concluded for a positive correlation. Along the same lines as the finding in Canadian data in Tomes (1986), Ball (2001) also found that happiness and inequality are positively correlated in raw data from the 1996 World Values Survey (WVS), although the introduction of a number control renders this positive correlation insignificant. The estimated value of #947; in the first 11 waves of the British Household Panel Survey (BHPS) is positive (Clark, 2003), as is that in the first five waves of the WVS (Rozer and Kraaykamp, 2013). Last, in one of the relatively rare contributions entirely outside the OECD, Knight et al. (2009) found that county-level income inequality is positively correlated with hap­piness in the 2002 Chinese national household survey.

One recent intriguing contribution to this empirical debate comes from Grosfeld and Senik (2010). In contrast to a number of the contributions in Table 13.1, their identifi­cation is purely within and not between countries, as they consider data from Poland over its transition period. Using repeated CBOS cross-section data over the 1992-2005 period, they identified a turning point in the estimated relationship between inequality and subjective well-being. This correlation is positive and significant in the first years fol­lowing transition, but then turns negative and significant. The break point that best fits this split in the data is 1996. The interpretation that the authors give is in terms of inequal­ity first being regarded as providing opportunities for future higher incomes, which con­sequently turned into more negative comparative evaluations of disparities as it became clearer that not everyone would be able to benefit from any opportunities that this greater inequality promised.

As well as the sign and significance of the estimated effect, we are also interested in the size. Some of the work cited in Table 13.1 does contain explicit statements about marginal effects. For example, Tomes (1986) wrote that “an increase of 10% in the share of the poor reduces satisfaction by approximately 0.6 of a point. In order to maintain satisfaction unchanged, own income would have to be increased by $4200 for every 1% increase in the share of the poor” (p. 435). This latter figure is larger than the annual income of 3860 Canadian dollars in his data set (although it should be noted that the confidence inter­vals around these estimates are quite large). Alesina et al. (2004) found that a one percentage-point rise in the Gini is compensated by a rise in annual income of 2950 dollars in the United States (8.7% of annual income) and 474 dollars in Europe (4.2% of annual income). The effect size in the SOEP in Schwarze and Harpfer (2007) seems more moderate: “If income inequality would be reduced by a half household income could be reduced by around 10% without changing life satisfaction” (p.244).

Although this kind of compensating differential is attractive in that it is easy to under­stand, it also obviously depends critically on the size of the estimated income coefficient in a subjective well-being equation. It is easy to believe that the coefficient on own income is actually an underestimate here, for standard endogeneity reasons, leading to trade-offs of income against inequality that are too high.

As an alternative, we consider the well-being effect of a one-point rise in the Gini coef­ficient, with the effect size being expressed as a percentage of the range of the subjective well-being measure. For example, the 0—10 life-satisfaction scale used in the SOEP has a range of 10; the corresponding 1—7 scale in the BHPS has a range of 6. It is not possible to calculate a standardized marginal effect using this metric across all of the work in Table 13.1. In the first instance, a number of the contributions here use ordered probit or ordered logit estimations, so that there are as many marginal effects as one minus the number of subjec­tive well-being categories. Restricting ourselves to linear estimation techniques using the Gini, which yield significant estimates, cuts the sample down to five: Hagerty (2000), Schwarze and Harpfer (2007), Knight et al. (2009), Winkelmann and Winkelmann (2010), and Rozer and Kraaykamp (2013). These papers use five different data sets, with subjective well-being measured on a variety of scales.

Expressed as a percentage of the scale range, a 10% point change in the Gini coeffi­cient mostly produces a movement in well-being of between 2% and 8% of the scale range (the exception being Schwarze and Harpfer, 2007, where the figure is smaller). In the SOEP, the standard deviation of life satisfaction is about 18% of scale range (1.79 for a scale of 0-10), with an analogous figure for the BHPS of 21% (1.29 for a 1-7 scale). A broad conclusion is that this very large movement in the Gini has an effect of between 0.1 and 0.4 of a standard deviation in life satisfaction. By way of comparison, the effect of unemployment on life satisfaction in the SOEP and the BHPS is somewhere around 6-10% of the scale range, or 0.3-0.5 of a life-satisfaction standard deviation.^[863]

Some ofthe work on inequality and happiness here has explored the role of mediating variables or subgroup regressions to establish the subjective-groups for which the corre­lation with inequality is the largest to shed some light on the circumstances under which inequality affects subjective well-being. In the perhaps absence of a clear central ten­dency, it is arguably useful for policy purposes to know where and when inequality might be harmful in subjective well-being terms.

One of the best-known findings in this respect comes from Alesina et al. (2004): In Europe, inequality hurts the poor and left-wingers more (in the sense of having a greater negative effect on their well-being scores) than it does richer and right-wingers. This finding has recently been corroborated on more recent (2009—2010) Eurobarometer data by Vandendriessche (2012). Alongthe same lines, in Grosfeld and Senik (2010) the initial positive correlation between well-being and inequality was found only for right-wingers.

Other work has considered the mediating role of individual income. Oishi et al. (2011) found that the effect of inequality on happiness is negative and significant only for those in the bottom two quintiles of the income distribution. Schwarze and Harpfer (2007) found that only those in the first income tercile are negatively affected by post­government income inequality. In Clark (2003), the correlation between regional income inequality and individual well-being is more positive for individuals whose own income has been more mobile over time.

Oshio and Kobayashi (2010) carried out a number of tests of mediating variables and concluded that the correlation between happiness and inequality is more negative for women, the younger, those who have unstable positions on the labor market, and those who are politically in the center (rather than being progressive or conservative).

Some work has considered a mediating role for individual values, rather than observed demographic characteristics. In Biancotti and D’Alessio (2008), inequality has a more negative effect for individuals who report more inclusive and moderate values. Rozer and Kraaykamp (2013) found that the effect of Gini on well-being is more neg­ative (actually less positive) for Europeans, those with more egalitarian norms (from a question on the relative preference for incomes being made more equal as opposed to needing larger income differences for incentive reasons), and those with greater levels of social and institutional trust. Last, as might be expected if the income distribution reveals information about the individual’s own potential future position, in Ferrer-i- Carbonell and Ramos (2014) the effect of inequality is greater for those with higher (self-reported) measures of risk aversion in 1997—2007 SOEP data. The marginal effect of the Lander-Year Gini coefficient on life satisfaction is twice as negative for those with the highest risk-aversion score (on a 0-10 scale) as compared to the effect for those who report the modal score of 5.

One important individual value in the terms of this chapter, and one to which we shall return later, is the perceived fairness ofthe market system (i.e., the system that transforms individual inputs into individual outputs). In Oishi et al. (2011), the effect of inequality on happiness is moderated by the individual’s perceived fairness of others and whether the individual believes that others can be trusted. Along the same lines, Bj0rnskov et al. (2013) found that the perceived fairness of the income-generation process affects the association between income inequality and subjective well-being.

This burgeoning work on inequality and happiness has then revealed a number of intriguing findings. But perhaps one of the most striking aspects of Table 13.1 is the sheer variety of empirical correlations that have been uncovered. Is there any way of making sense of the variety of different estimated results here, or does sample variability rule the day (with as many positives as negatives as zeros)?

A first point, apparent from the fourth column of Table 13.1, is that there is no empir­ical agreement on the most appropriate measure of inequality. Although the majority of work refers to the Gini coefficient (a point to which we shall return in Section 13.4), it is also true that no consensus has been reached regarding the geographic level at which this coefficient should be evaluated.

Most of the empirical analysis has been carried out using data that contains only coarse-grained information on the distribution of income (i.e., at a very aggregated level, such as the country). Some work on British, Japanese, German, and Russian data has appealed to measures of inequality at the regional level (respectively: Clark, 2003; Oshio and Kobayashi, 2010; Schwarze and Harpfer, 2007; Senik, 2004). One of the few con­tributions to use large-scale data with more local-level inequality measures is Brodeur and Fleche (2013), who appeal to county-level information in the American BRFSS. Another is Winkelmann and Winkelmann (2010), who match in measures of inequality at all of the (in increasing order of size) municipality, region, and canton levels in the 2002 wave of the Swiss Household Panel. The research in Knight et al. (2009) combines more local-level measures of the distribution of income with data from a non-OECD country (China), finding a positive effect of the county-level Gini on respondents’ happiness (see also Jiang et al., 2012).

One of the reasons why the degree of aggregation matters is that the Gini often moves only a little over time, a point made by Graham and Felton (2006), who noted that the Gini coefficient in Chile in the 2000s is not substantially different from that which pertained in the 1960s, despite the considerable social and economic changes that have taken place over the intervening period. Econometrically, it is then difficult to introduce both the Gini and country dummies into a regression, leading to the possibility that the Gini may be proxying for some other fixed country characteristic that is correlated with subjective well-being.

In general, this lack of variation in the measure of inequality does not help us to assuage the doubt that it is strongly correlated with some other variable that is important for happiness. For example, income inequality at the regional or country level could reflect industrial structure or the unemployment rate, both of which may well have inde­pendent effects on subjective well-being. Given a sufficient number of observations, it should be possible to tease out the independent contributions of inequality and other variables. But at the aggregate level it is anything but sure that sufficient observations are available. In general, the list of potentially important aggregate-level variables is often perilously close to the number of degrees of freedom in the analysis. In Di Tella and MacCulloch (2008), for example, income inequality attracts a negative but insignificant coefficient in their analysis of Eurobarometer and GSS data. They noted that this occurs “in part because there is some degree of co-linearity between the included variables. For example, if we do not include unemployment benefits, a variable that is highly correlated with inequality, we find that the coefficient on inequality becomes negative and significant” (p.36). Verme (2011) concurred that the lack of variability in survey measures of the Gini coefficient makes it particularly susceptible to multicollinearity with other aggregate-level variables (a problem he tackled via a number of robustness tests in which the other aggregate explanatory variables are dropped in turn).

An additional drawback to the empirical analysis of the relationship between individ­ual well-being and aggregate income inequality is that it does not adequately distinguish between the comparative and normative aspects of the reference group. Even though some of the empirical analyses in Table 13.1 (although far from all) do introduce some measure of the mean of the income distribution into the analysis, they are unable almost by construction to calculate measures of relative deprivation and relative satisfaction from the survey data used. As such, any partial correlation between aggregate income inequal­ity and individual subjective well-being very likely mixes together aspects of the com­parative and normative reference groups, which perhaps explains the variety of estimated coefficients in Table 13.1.

Given the perhaps natural limits on the analysis of the relationship between aggregate inequality and individual subjective well-being, any evidence from this type of analysis will probably have to remain suggestive. This is arguably not the case for experimental work, where the reference group and the degree of inequality can be exactly manipu­lated, and it is to this that we now turn. Experimental work is of course not free of prob­lems, in that what people say in a controlled setting may well differ from the way in which they would actually behave in reality, and their perceptions of inequality will likely be influenced by many factors. For a thorough discussion of these aspects and problems with experiments regarding social preferences, see Levitt and List (2007).

13.3.2 Experimental Economics

The experimental economics contributions to inequality aversion from the more aggre­gate perspective have appealed to two different approaches: (1) inequality and risk aver­sion with a parametric social welfare function; and (2) general social welfare functions. In the first of these, two types of experiments have been run. The first is similar to that adopted in the experiments on status or relative income discussed earlier in Section 13.2.2, that is the choice between alternative societies with different income dis­tributions behind the veil of ignorance. The second type is based on the leaky-bucket experiment, which we introduce later.

Johansson-Stenman et al. (2002) carried out hypothetical-choice experiments. An individual’s relative risk aversion is interpreted as the social inequality aversion from a utilitarian social welfare function’s perspective. Inequality aversion is evaluated via indi­viduals’ choices between two types of society, from behind a veil of ignorance. Individ­uals are asked to choose the society that would be the best in terms of the well-being of their imaginary grandchild (in order for choices to be abstracted from the respondent’s own circumstances and environment). The income distributions in the two societies, A and B, are uniform, and the respondent is told that their grandchild has an equal prob­ability of receiving any income level within the range.

For example, Society A has a uniform income range of 10,000 to 50,000 Swedish kroner, whereas Society B has a uniform income range of 19,400 to 38,800 Swedish kro­ner. The student subjects in the experiment are told that prices are the same in the two societies, that there is no welfare state, and that there are no growth effects of the different income distributions.

An individual who is risk neutral will prefer Society A, in which expected income is higher. Someone who is indifferent between the two societies will have a relative risk­aversion parameter, #951;, that can be calculated by assuming a CRRA utility function^[864] (see their Equation 5). In the example given earlier, indifference between societies A and B implies a value of #951; of 0.5; equally, an individual who prefers A (B) over B (A) will have a value of #951; of lt; (gt;) 0.5. There are eight different conditions in their experiment. Society A always remains as described earlier, whereas there are eight society Bs, ordered such that indifference between A and B implies increasing risk aversion (see their Table 1). The higher is the value of #951;, the more income society is willing to give up to bring about a more egalitarian distribution of income, corresponding to a more concave social­welfare function.

The median value of inequality aversion in these experiments is in the interval between two and three. The respondents were fairly evenly distributed between the cat­egories, with 43% of the respondents having inequality aversion of between one and five. Furthermore, a considerable number of respondents (17%) exhibited zero or negative inequality aversion. In addition, 19% of respondents exhibited extreme aversion compat­ible with the Rawlsian maxi-min strategy, which is the case of maximum aversion in the experiment. In a similar experimental setting, Carlsson et al. (2005) confirmed a median value of relative risk aversion of between two and three, and found a larger fraction of respondents (63%) with a value of relative risk aversion between one and five. In their experiment, 8% of respondents were found to be risk-lovers.^[865]

Some work in this area has tried to distinguish further between two types of inequality aversion: the first is the individual’s level of risk aversion, as explained earlier, whereas the second is the individual willingness to pay to live in a more equal society. The estimation of individual inequality aversion only via risk aversion disregards any preferences that individuals may have regarding inequality per se.

To separate out these two attitudes, two types of experiments are carried out, one for each type of aversion. To this end, Carlsson et al. (2005) extended the analysis of Johansson-Stenman et al. (2002). The first experiment concerns the traditional imaginary grandchild, as described earlier, where the respondents do not know the position of their grandchildren, but only the income distribution and hence also the probability distribu­tion in each society. In the second experiment, subjects choose between pairs of hypo­thetical societies with different income distributions, where the grandchild’s income is known and is set equal to the mean income in the society. In other words, “In the first experiment individuals choose between hypothetical lotteries, where the outcomes determine their grandchildren’s incomes in a given society. This experiment allows for the estimation of the individual’s risk aversion in a setting where the level of social inequality is fixed. In the second experiment individuals choose between hypothetical societies with different income distributions, where the grandchildren’s incomes are known and are always equal to the mean income in each society. This experiment enables us to estimate parameters of individual inequality aversion in a risk-free setting” (Carlsson et al., 2005, p.376).

In the second experiment, with a value of inequality aversion of zero, the individual is indifferent to income inequality; with a value of one, a 1% increase in own income yields as much utility as does a 1% fall in inequality. The median value of inequality aversion is found to be in the interval between 0.09 and 0.22, and most responses reflect positive inequality aversion. Only 7% of respondents appear to be inequality-lovers, in the sense that they are willing to sacrifice their own income to make society more unequal, whereas 6% are found to be extremely inequality-averse. Kroll and Davidovitz (2003) also found that subjects prefer more equal income distributions. However, when they had to give up part of their reward to shift to a more equal distribution, they chose not to do so.

Amiel et al. (1999) belongs to the second type of experiment in method (1), in which social inequality aversion is estimated via the leaky-bucket experiment. A sample of stu­dents were asked to indicate the amount of “lost money” that they were willing to accept for a transfer of money from a richer to a poorer individual, where this loss came about for example due to administrative costs. The median value of inequality aversion was esti­mated to be between 0.1 and 0.22, which is much lower than the existing estimates from the alternative approach, such as in Johansson-Stenman et al. (2002). However, the cir­cumstances of the two experiments are very different, making a clear comparison of the results rather difficult.

That these large differences in the value of inequality aversion result from the different measurement techniques is confirmed by Pirttila and Uusitalo (2010). The authors esti­mated inequality aversion using a questionnaire approach in a representative survey of Finns. The advantage of this questionnaire is that the same individual was asked questions based on two different measurement techniques: the leaky bucket and the preferred wage distribution under the veil of ignorance. The median value of the inequality aversion parameter from the leaky-bucket questions lay below 0.5. However, the results from the preferred distribution question gave a much higher value for inequality aversion, with the parameter being over 3. There are thus a considerable number of respondents who are willing to sacrifice the mean wage to bring about a more equal distribution of wages, but who at the same time are not willing to carry out costly transfers from richer to poorer.

Pirttila and Uusitalo proposed a number of explanations for this rather radical differ­ence in the results. One possibility is that people simply have different attitudes toward the implied efficiency-equity trade-off in different situations. The leaky-bucket question is specifically focused on redistribution, whereas the change in the wage distribution is a bargaining result. The two questions may also be measuring the same phenomenon but at a different scale. In addition, the leakage, that is the efficiency loss, is explicitly visible in the leaky-bucket question, whereas the respondent would have to calculate it in the wage-distribution question. Respondents may have had efficiency concerns in mind in the leaky-bucket question, and their preferences over efficiency could explain part of their unwillingness to support the transfer.

Pirttila and Uusitalo also confirm the results in Beckman et al. (2004): the actual posi­tion of the respondent in the income distribution affects the answer given in the leaky- bucket experiment. As expected, support for this transfer is higher among the individuals who would benefit from it.

In the income-distribution literature the indices that are deemed appropriate to mea­sure inequality are those that conform to the Lorenz dominance criterion. These indices fulfill four basic axioms: scale invariance, symmetry, the population principle, and the Pigou-Dalton transfer principle. For a recent survey of these properties and the domi­nance criteria see, among others, the excellent chapter in Chakravarty (2009). The first three properties are commonly assumed in the majority ofindices of well-being; only the transfer principle, as we mentioned in the introduction, is at the heart of inequality measurement.

Attitudes toward inequality have been interpreted by some authors as being revealed by the reaction of (some relatively informed part of) the general public to these four basic properties. This is the contribution of the authors in group (2), where some general social welfare function is assumed but without any a priori functional form. The main question that is addressed in this part of the literature is what inequality seems to represent for the general public, and in particular whether these four basic axioms are reflected in individ­uals’ views. The seminal book is this area is Amiel and Cowell (1999). Given that the defining concept for inequality measurement is the Pigou-Dalton transfer principle, we will discuss only those experimental results that cover this aspect of inequality.

In Amiel and Cowell (1992), the transfer principle is presented to respondents both as a numerical problem and verbally. In the former, they are asked to say which of two dis­tributions of income are more unequal: A = (l, 4, 7, 10, 13) versus B = (l, 5, 6, 10, 13).

Verbally, they are asked to say what happens to inequality in the following scenario: “Suppose we transfer income from a person who has more income to a person who has less, without changing anyone else’s income. After the transfer the person who formerly has more still has more.”

Nearly two-thirds of the student sample in Amiel and Cowell (1992) did not think that inequality was lower in B than in A, whereas 40% did not agree that inequality would fall following the verbal scenario. The difference in these figures likely comes from indi­viduals thinking of some kind of Robin Hood redistribution in the verbal case, whereas the actual numerical problem involves redistribution from the fairly poor to the even poorer. Amiel et al. (2012) examined many “flavors” or interpretations of the transfer problem. Only 21.6% of the sample are found to be in line with the researcher’s standard view. A critique of the way in which some of these kinds of questions are asked is pro­vided by Jancewicz (2012).^[866]

Similar to Kroll and Davidovitz (2003) and Carlsson et al. (2005), Amiel and Cowell (2002), Gaertner and Namazie (2003), and Cowell and Cruces (2004), using method (2), examined the degree to which the principle of transfers is followed by people who eval­uate inequality and risk. About 60% of respondents in the latter contribution viewed an equalizing transfer as inequality/risk reducing, and consistency in the risk version of the questionnaire was higher than consistency with the principle of transfers in the inequality version. This finding is confirmed by Gaertner and Namazie and Amiel and Cowell (2002), where the proportions of acceptance in the sample are 23% in the risk question­naire and 17% for inequality.

Overall, individuals do have normative preferences over the distribution of income. It is, however, hard to argue that these are isolated in happiness regressions, as the latter are not able to separate out the comparative and normative components of attitudes to inequality. The experimental literature has been more successful in this respect, but even there the variety of different methods have produced quite a large range for the estimated value of inequality-aversion. Part of the problem here seems to be that the different methods make salient different preferences (such as risk aversion or pref­erences over efficiency). Another is that there are almost an infinite number of ways in which we can change the inequality of the income distribution, and preferences over taking money from the rich to give to the poor, and taking money from the middle or lower-middle class to give the poor may reasonably differ, even if the final impact on the Gini coefficient is the same.

15

13.4.