Preferences for redistribution and social structure

Corneo and Fong (2008) estimate on data from the 1998 Gallup Social Audit that the monetary value of justice for US households amounts on average to about one fifth of their disposable income.

There is a growing amount of evidence that people have wrong perceptions about inequality and its causes and that these perceptions strongly influence their ideas about whether more redistribution is needed. Recent examples include Gimpelson and Treisman (2015) and Page and Goldstein (2016).

We understand the “reference group” of an individual as the group of people that he observes and from which he derives information about the overall income distribution. Note that in a different context, a reference group is commonly understood as the group of people to which an individual compares himself in order to evaluate his relative welfare position. Corneo and Grüner (2000, 2002) analyse the determinants of redistributive taxation in such a setting in which individuals are concerned about their relative consumption.

The assumption of a linear tax scheme is very common in the literature and it has also been shown that such a scheme may be a good empirical approximation of the more complex non-linear tax schemes that are found in the real world (see, e.g., Roemer et al. 2003). Nevertheless, exploring the consequences of relaxing this restriction is an interesting topic for future research. The assumption of a quadratic income tax scheme (as in De Donder and Hindriks 2003) may be an interesting starting point for this extension.

In political economy models of the determination of tax rates, each individual voter has a zero impact on outcome. We focus on individual preferences: the optimal \(\tau \) is then determined as if the individual is a dictator.

In the welfare economic literature, the treatment of luck is far from obvious—see, e.g., Lefranc et al. (2009). We follow the terminology in which “luck” stands for circumstances.

As one reviewer notes, the empirical literature has investigated the acceptability of income differences due to effort, not those due to taste for effort. However, in most models the former is monotonically related to the latter. At a more basic level, the distinction relates to the fundamental debate in the literature on responsibility-sensitive egalitarianism on the question for which factors individuals should be held responsible: for factors under their control, or for their preferences. Within the economic model, effort choices are determined by preferences and by constraints and there is no really free choice. See, e.g., Fleurbaey (2008) for a deeper discussion of these issues.

The index i is typically used for the consumer assessing his/her preferences for redistribution, and j concerns typically consumers observed by him/her.

To save on notation, we do not use subscripts for these cultural traits, but we will derive comparative statics results with respect to them.

In an earlier version of this paper (Schokkaert and Truyts 2014), we did not impose the assumption of an homogeneous taste for effort. This complicates the derivation and the interpretation of the results, without adding useful insights. The results derived in this paper still hold in the more general model.

Note that this means that we confine ourselves to the voters’ ex ante mental exercise of determining their preferred tax rate, based on local information. As one reviewer correctly noted, in our simplified model observing the true demogrant \(\bar{m}\) would allow sophisticated voters to infer the society’s true average ability. We keep to our simple setting for the present purposes, mainly because we want to stay close to what empirical research has revealed about the informational limitations of citizens in forming their opinion about redistributive policies. Similar results as in our model can be obtained with fully Bayesian voters observing the true demogrant in combination with less lenient other information assumptions, but we leave such a significantly more complicated analysis for a later paper.

Since i has zero mass, this immediately follows from \(\left| \mu \left( i\right) \right| >0.\)

Equation (5) might suggest that individuals take a parochial attitude and are only interested in justice within their reference group. This is not our interpretation, however. We could have started from a more general society-wide measure of injustice. However, as will become clear, only the means and the variances of the different variables will enter the expressions for the preferred tax. Since we assume that these are estimated by the individuals on the basis of their own reference group, choosing a more general formulation would not change any of our results. Moreover, in our theoretical analysis, reference groups can be interpreted very broadly, e.g., they can be seen as a (probably biased) sample of the overall population.

This is similar to the egalitarian equivalent approach in the theory of responsibility-sensitive egalitarianism—see, e.g., Fleurbaey (2008).

Our setting allows for more extreme positions. At one extreme, we have the laissez-faire or libertarian conviction that considers all income differences to be justified, such that fairness warrants no redistribution at all. This can be modelled as \(\hat{c}_{j}^{\zeta }=m_{j}\) for all j. In this case, social injustice \(\Omega _{j}\) equals the average income change due to taxation and it is minimized by setting \(\tau =0.\) Note that this fairness ideal conflicts with self-interest for consumers with an income below average. At the other end of the spectrum, the pure egalitarian position corresponds to \(\hat{c}_{j}^{\zeta }=\bar{m}_{\left( i\right) }\) for all j. We will not analyse this position as such (it is rarely defended explicitly).

We assume that when deciding about the tax rate, the consumer does not yet know the realization of luck \(\varepsilon _{i}\) and therefore estimates it as the average \(\overline{\varepsilon }_{(i)}=0\).

The condition for strict concavity of the optimization problem is that \(a_{i}<2\overline{a}_{(i)}.\) We only consider individuals for which this assumption holds.

An alternative approach would take as the reference a no-tax situation, with effort equal to \(\alpha _{j}\beta \) and fair pre-tax income \(\beta \left( \zeta a_{j}+\left( 1-\zeta \right) \bar{a}_{(i)}\right) \). This would mean, however, that our “naive” consumer uses a sophisticated model to go from the observed situation to the counterfactual no-tax world. As a matter of fact, our comparative statics results remain valid under this assumption (simply set \(\tau ^{a}=0\) in all expressions).

Note that the naive idealist in our model does not take into account incentive effects in determining his/her own preferred tax rate. However, if \(\tau ^{a}\) is large, (s)he will observe a lower contribution of effort in the actual process of income creation and this has an effect on what (s)he considers to be the fair tax rate.

This mechanism is analogous to the one that is described by Alesina and Angeletos (2005) to explain the differences between the European and the US welfare states.

Alesina et al. (2012) mention that it is debatable whether or not \((1-\tau )\) should enter the definition of the “fair” wealth. An alternative is to take as reference the no-tax situation \(\tau =0.\) This alternative assumption does not change the fundamental results.

This result stands in sharp contradiction to the optimality result for linear redistributive taxation in the traditional approach with a welfarist (e.g., utilitarian) social welfare function. In fact, our sophisticated consumer is only concerned about fairness (and self-interest) and does not consider the society-wide trade-off between the global level and the distribution of income.

The (reasonably weak) conditions for strict concavity can be obtained from the authors on request.

Consumers may use other information sources, such as what they see and hear in mass media. Of course, this information is also biased. The (interesting) question of how consumers may combine different biased information sources is an open question for further research.

A discussion of reference group formation for generic distribution functions in the case where only natural talent matters is presented in Appendix.

Hence, we restrict our attention to cases where the support of \(\mu \left( i\right) \) is in the interior of \(\left[ \theta _{L},\theta _{R}\right] \times \left[ \theta _{D},\theta _{U}\right] .\)

The General Social Survey was set up by the National Opinion Research Center at the University of Chicago in 1972, and collected its 30th round in 2014. See: http://​www.​gss.​norc.​org.

The previous literature has often made use of redist1, one of the four variables that make up our factor (see, e.g., Alesina and Giuliano 2011). The results with that variable are very similar to the ones obtained with FactRedis, and can be obtained from the authors on request.

“Explanatory” is meant to refer to statistical features and does not imply causality. Indeed, one does not have to be particularly cynical to note that regressing “attitudes” on other “attitudes” is bound to lead to strong associations.

As shown in Table 13, the range of the education variable is from 0 to 20.

As can be seen from Table 12, income is just household income measured in 1986 dollars. However, the combination of rlincome with prospect can arguably be interpreted as a proxy measure of permanent income, which may better fit the income concept in our theoretical model.

Note that these data requirements w.r.t. the reference groups go well beyond the data used in the previous section. The 1987 round of the GSS contains detailed information about different memberships of organisations (sports clubs, science clubs...), but this does not allow for a characterization of the sample of society that individuals meet in such an organization. Even if individuals from all parts of society are members of sports clubs, this does not mean that they meet each other in these clubs, because the individual sports clubs can be very homogenous in terms of their members’ socio-economic and other characteristics.