2.1 The setup
Let a distribution of individual incomes,
y, be represented by its cumulative distribution function (cdf) denoted by
F(
y) and let
\(\mathcal {F}\) be the set of all such cumulative distributions. We are interested in judging
F from a normative perspective. We assume that a social planner is endowed with preferences over such income distribution denoted by
W(
F). To represent such preferences, we resort to the rank-dependent model proposed by Yaari (
1987).
1 It assumes that the welfare derived from a risky distribution can be written as a weighted average of all possible realizations, where the weights are a function of the rank of the realization in the distribution.
2 Thus, let
\(p\in\) \(\left[ 0,1\right]\) denote the rank in the income distribution
F(
y), so that
\(F^{-1}(p)=\inf \left\{ y: F(y) \ge p \right\}\) represents the inverse of the cdf
F, that is, the level of income benefited by the individual ranked
p in
F. The social preference over such income distribution can be defined as follows:
$$\begin{aligned} W(F)=\int _0^1\omega (p) F^{-1}(p)dp \end{aligned}$$
(1)
where
\(\omega (p) \ge 0 \ \forall p \in [0,1]\) is a differentiable function, such that
\(\omega (p)=f'(p) \ \forall p \in [0,1]\) with
f continuous and increasing so that
\(f(0)=0\) and
\(f(1)=1\). The function
\(\omega (p)\) can be interpreted as a preference function of a social planner that assigns weights to the incomes of each agent in the distribution according with her relative position in that distribution. By analysing the functional form of
\(\omega (p)\), it is possible to infer the distributional attitudes of a social planner adopting
W(
F) to compare states of the world.
Then, the evaluation of a given society will be the result of a weighted aggregation across individuals, where the weighting scheme is constructed upon the relative position of individuals in the income distribution. Restrictions on weights will define different classes of social welfare functions, characterized by different normative implications. Hence concerns for compassion and envy, separately, will be expressed throughout such restrictions.
Now, if the functional form of
\(\omega (p)\) was known, we could directly check welfare dominance between the different income distributions compared. In practice, this is not possible. Hence, we need to reformulate the dominance expressed in terms of
W in a dominance expressed in terms of a restriction that involves only the observables, i.e. the income distributions under alternative states of the world. This can only be realized by imposing restrictions on the class of social welfare functions denoted by
Z, defined as follows:
$$\begin{aligned} \begin{aligned} Z&=\{ W: \omega (p)=f'(p) \ \text {continuous and differentiable on} \ [0,1], \\&\quad f'(p) >0 \ \forall p \in [0,1], f(0)=0 \ \text {and} \ f(1)=1 \}. \end{aligned} \end{aligned}$$
Hence, our departing point is a class of rank-dependent social evaluation functions, which are explicitly sensitive to information about the relative position of individuals. This is consistent with the increasing interest given in the economic literature on the role of individual economic positions as indicator of social status, and complements the growing discussion about the formal integration of other-regarding preferences into a normative measurement model (see Decerf & Van der Linden,
2016; Treibich,
2019). Other-regarding preferences, in fact, acting as externalities could lead to substantive welfare gains or losses. It is, therefore, essential to understand how they should be incorporated into normative evaluations. Indeed, the rank of each agent in the distributions of income (or of other general economic resources) in her society, represents a standard proxy of her social status and plays an important role in her appraisal of her own welfare and the welfare of the whole society. The relevance of the rank suggests that attitudes of envy and compassion are important components of individual judgements and cannot be overlooked when assessing alternative states of the world (Weiss & Fershtman,
1998).
Before illustrating the results, it is useful to clarify the notion of compassion and envy endorsed in this model. Compassion considers altruistic individuals that evaluate their own welfare on the basis of two components: their income and the income of the disadvantaged individuals, that is, those individuals whose relative position in the income distribution falls below a given threshold. Both components act by increasing the utility of altruistic individuals who, thus, feel compassionate with respect to the disadvantaged agents. The sentiment of compassion that arises overall in the society constitutes the basis of social judgement. Because both components positively affect the utility of altruistic individuals, compassion can be quantified on the basis of the cumulative effect of the income of the disadvantaged individuals on the income of the advantaged altruistic individual.
Also envious individuals evaluate their welfare on the basis of two components. The first is identical to the case of altruistic individuals, that is, their own income. The second corresponds to the income of the advantaged individuals, that is, those individuals whose relative position in the income distribution is found above a given threshold. The utility of an envious individual is increasing in the first component but decreasing in the second. Envy can then be quantified on the basis of the distance between the income of each worse-off individual and the average income benefited by the advantaged segment of the society.
Hence, envy has to do with the frustration an agent experiences because some other individuals are in a better position than she is, while compassion arises from becoming aware of the existence of individuals in worse situations.
2.2 Ranking criteria
The first ranking criterion we propose is formalized in the following Proposition
1 and provides a robust method to compare alternative states of the world when the feeling of compassion is accounted for in the normative assessment.
3
According to Proposition
1, in order to judge one distribution as superior to another distribution two conditions need to be satisfied. The first is a second order inverse stochastic dominance condition that must hold for every percentile of the distribution up to
\(\bar{p}\). This dominance means that the cumulated income in the dominant distribution must be nowhere lower than the cumulated income in the dominated distribution up to the threshold level
\(\bar{p}\). The second condition requires checking that the average income of the dominant distribution is not lower than the average income of the dominated distribution. This proposition is based on the ideal that the higher the income of the more disadvantaged part of the society the lower will be the feeling of compassion suffered by individuals the more socially desirable will be a distribution, by safeguarding at the same time efficiency through condition (ii). The distinction between the two parts of the distribution depends on the value of the threshold
\(\bar{p}\) that we assume it is defined exogenously.
4
This Proposition formalizes the ranking criterion that a social planner with particular preferences should adopt. These preferences encompass a common monotonicity (or efficiency) property represented by \(\omega (p)=f'(p) \ge 0\) for all \(p \in [0,1]\), according to which an increase in individual income never reduces social welfare, independently on where the individual sits in the distribution, and a property that we define as aversion to the feeling of compassion and that is obtained by imposing \(f^{''}(p) \le 0\) \(\forall \; p \in [0,\bar{p}]\), \(f^{''}(s) = 0\) \(\forall \; s \in [\bar{p}, 1]\), which implies that \(f'(p) \ge f'(s) \ge 0\) \(\forall \; p \in [0, \bar{p}]\) and \(\forall \; s \in [\bar{p},1]\) such that \(0\le p \le \bar{p} \le s \le 1\). This property, focusing on the part of the distribution encompassed between 0 and \(\bar{p}\), introduces a concern for the disutility generated by the compassion suffered by those individuals ranked above \(\bar{p}\). Technically this property says that the marginal effect of a reduction of income on social welfare is higher the poorer is the individual and hence the higher the feeling of compassion that is generated among those richer than him. Whereas, the marginal effect of a reduction in income is constant for those that are not considered as disadvantaged and hence that do not represent a source of compassion. Hence, a progressive transfer between individuals ranked above \(\bar{p}\) does not have any effect on social welfare, whereas a progressive transfer from an individuals ranked above \(\bar{p}\) to an individual ranked below, as well as a progressive transfer between individuals both ranked below \(\bar{p}\), reduces the feeling of compassion and may increase social welfare. This is the case of designing a policy intervention that must account for the potential creation of negative externalities in the society, through acting especially on alleviating the economic hardship of the most disadvantaged individuals.
Special cases of Proposition
1 are obtained by selecting two particular values for the threshold
\(\bar{p}\). That is, for
\(\bar{p}=1\) Proposition
1 boils down to a standard second order inverse stochastic dominance condition, corresponding to a situation of pervasive compassion in the distribution. Whereas, for
\(\bar{p}=0\) Proposition
1 boils down to a dominance test applied on the average income of the two distributions, corresponding to a situation of absence of compassion. A third special case worth emphasizing is obtained by assuming
\(\bar{p} \rightarrow 0\) and
\(\omega (s) = f(s)'=0 \ \forall \; s \in [\bar{p}, 1]\) and is formalized in Remark
1.
The class of social welfare functions considered in Remark
1 is consistent with the preferences of a social planner valuing an income distribution socially desirable if and only if the most disadvantaged individuals experiences an increase in income, which echoes back the maximin criterion
à la Rawls.
It is interesting to note that the SWF referred to in Proposition
1 is consistent with individual utility functions that are separable on individual income and the income of all other individuals with lower income, as expressed below:
$$\begin{aligned} U(q)=\alpha (q) y(q)+\int _{0}^{q}\beta (p)y(p)dp \; \forall \; q \in [0,1] \end{aligned}$$
(2)
with
\(\alpha (q), \beta (p) \ge 0\),
\(\frac{\delta \alpha (q)}{\delta p}, \frac{\delta \beta (p)}{\delta p} \le 0\), and
\(\omega (p)=\alpha (p)+ (1-p) \beta (p)dp\). This complements well-known results in Lambert (2003) and Schmidt and Wichardt (
2019) outlining the relationship between relative concerns and social welfare in the case of utilitarian social welfare functions.
5
The second ranking criterion we propose is formalized in the following Proposition
2 and provides a robust method to compare alternative states of the world when the feeling of envy is accounted for in the normative assessment.
Proposition
2 formalizes a ranking criterion that the social planner should apply in case of aversion to the feeling of envy generated by an income distribution. This criterion is composed of two conditions. As in Proposition
1, the first is a second order inverse stochastic dominance condition, but differently from Proposition
1 it must hold for every percentile of the distribution between
\(\bar{p}\) and 1. The second condition imposes that the average income in the lower part of the distribution, specifically from 0 to
\(\bar{p}\), cannot be lower in the dominant distribution. The principle underlying this proposition is that the higher the income of the richest individuals the higher will be the feeling of envy in a society. The condition
\(\omega (p) = f'(p) \ge 0 \forall \; p \in [0,1]\) ensures that an increase in income of one individual, ceteris paribus, does not decrease social welfare. At the same time, the envy aversion endorsed by the social planner is reflected in the structure of the weighting scheme:
\(f''(s) \le 0\) \(\forall \; s \in [\bar{p}, 1]\),
\(f''(p) = 0\) \(\forall \; p \in [0,\bar{p}]\). Such weights, by focusing on the part of the distribution encompassed between
\(\bar{p}\) and 1 introduce a concern for the feeling of envy that is felt by those individuals ranked below
\(\bar{p}\). According to these conditions, the marginal effect of an increase of income on social welfare is lower the richer is the individual and hence the lower the feeling of envy that is generated among those poorer than him. Whereas, the marginal effect of a reduction in income is constant for those that are not considered as advantaged and hence that do not represent a source of envy, but it safeguards efficiency in the lower part of the distribution through condition (ii). Hence a progressive transfer between individuals ranked below
\(\bar{p}\) does not have any effect on social welfare, whereas a progressive transfer between individuals both ranked above
\(\bar{p}\) reduces the feeling of envy and may increase social welfare. Clearly, for
\(\bar{p}=1\) and
\(\bar{p}=0\), Propositions
1 and
2 will coincide. In its role of defining the ideal policies, the social planner must be aware that the economic gains generated in favour of the more advantaged individuals also act as negative externalities imposed on the society. They must be taken into account in elaborating a policy intervention, for instance in case of potential utilitarian gain when that gain is due to a negative welfare externality imposed by envy. Similarly to the argument made above, the SWF referred to in Proposition
2 can be obtained as an aggregation of individual utility functions that are separable in individual income and the income of all other individuals with higher income:
$$\begin{aligned} U(p)=\gamma (p)y(p)-\int _{p}^1\sigma (q)y(q)dq \; \forall \; p \in [0,1] \; \end{aligned}$$
(3)
with
\(\gamma (p), \sigma (q) \ge 0\),
\(\frac{\delta \gamma (p)}{\delta p}, \frac{\delta \sigma (q)}{\delta q} \le 0\) and
\(\omega (p)=\gamma (p)-p \sigma (p)\).
Note that the social preferences considered in Propositions
1 and
2 are reminiscent of the approach to inequality measurement based on the notion of complaints about income distribution, whose philosophical foundations can be traced back to Temkin (
1986,
1993). This approach is based on the ideal that each individual may express a complaint when comparing her welfare to that of a reference person or a group in a given society. This is exactly the situation taking place in our framework, where both envy and compassion are oriented towards a group (those below
\(\bar{p}\) for compassion and those above
\(\bar{p}\) for envy). Therefore, compassion and envy can be regarded as the weighted sum of these complaints, with weights increasing with the size of the complaint (see also Cowell & Ebert,
2004). Although it has the features of an individualistic approach, the aggregations that are generated are also normatively relevant since—as also argued by Temkin (
1986)—the overall complaint in the community represents a form of social bad.
In case of intersection in condition (i) of either Propositions
1 or
2, or if condition (i) and (ii) of either Propositions
1 or
2 result in a conflicting dominance, the two distributions cannot be ranked. Hence, we need to identify the minimal refinement on the set of admissible preferences that allows unambiguous assessments of distributions. This is done in the following two propositions. In particular, one can resort to the test proposed in Proposition
3 if Proposition
1 results in a clash of the distributions compared. Similarly, the test proposed in Proposition
4 can be implemented if it is Proposition
2 that generates incomplete rankings.
Proposition
3 characterizes a ranking criterion based on two conditions. Condition (i) is a (upward) third order inverse stochastic dominance condition up to
\(\bar{p}\). Condition (ii) is a sequential test composed of two steps. The first step consists in checking the dominance between the average income of the bottom part of the distribution up to
\(\bar{p}\); it coincides with the last step of condition (i) in Proposition
1. In the second step, the average income of the upper part of the distribution is added, coinciding with condition (ii) of Proposition
1. Differently from Proposition
1, we are restricting the set of preferences for which the dominance must hold to those satisfying
\(\omega ^{''}(p)=f'''(p) \ge 0\) \(\forall \; p \in [0, \bar{p}]\) asking that one should prefer progressive transfers taking place at the very bottom part of the distribution to progressive transfer taking place somewhere above. If the transfer takes place from one individual ranked above
\(\bar{p}\) to one individual ranked below
\(\bar{p}\), then one should prefer transfers that benefit individuals that are the poorest among those ranked below
\(\bar{p}\). Both kinds of transfer that one should prefer are those that focus on individuals whose economic conditions generate the highest feeling of compassion.
Proposition
4 characterizes a ranking criterion based on: (i) downward third order inverse stochastic dominance condition between
\(\bar{p}\) and 1; (ii) a test on the difference in the magnitude of the mean income of the bottom part of the distribution as for condition (ii) of Proposition
3; (iii) a test on the difference in the magnitude of the mean income of the upper part of the distribution. Differently from Proposition
2, we are restricting the set of preferences for which the dominance must hold to those satisfying
\(f'''(s) \le 0 \; \forall \; s \in [\bar{p}, 1]\), asking that among all possible progressive transfers that could be implemented in the group of advantaged individuals, one should prefer progressive transfers taking place between individuals ranked very close to 1 to progressive transfer taking place between individuals ranked very close to
\(\bar{p}\). Such kinds of transfer are those that focus on individuals whose economic conditions generate the highest feeling of envy.
From a technical point of view, it is interesting to notice that while condition (i) of all previous Propositions is based on an upward dominance test, condition (i) of Proposition
4 introduces a dominance test to be implemented through a ‘downward’ procedure, which means that one has to start from the highest percentile, add sequentially the cumulated income corresponding to lower percentiles, and check that the dominance holds at every step of the downward cumulative process.
6
Last, we propose a generalization of the dominance conditions characterized so far, which considers the advantaged and disadvantage groups as pushed far apart from each other. This implies two thresholds need to be introduced: a lower threshold for identifying the disadvantaged group, denoted by
\(\underline{p}\), and an upper threshold for identifying the advantaged group, denoted by
\(\bar{p}\).
7
Proposition
5 derives two conditions that must be satisfied in order to identify the dominant distribution and establish a rank across distributions. The first condition is a second order inverse stochastic dominance condition that must hold for every percentile of the distribution up to the lower threshold
\(\underline{p}\). It corresponds to condition (i) of Proposition
1 except for the selection of the threshold percentile. The second condition is a sequential test composed of two steps: the first step requires checking that the average income of the individuals ranked up to
\(\bar{p}\) of the dominant distribution is not lower than the average income of individuals ranked in the same position in the dominated distribution; the second step requires checking the dominance of the overall average income between the two distributions compared.
The first condition of Proposition
6, as above, is equivalent to condition (i) of Proposition
2, except for the threshold percentile: it is a dominance of the cumulated sum of incomes ranked between
\(\bar{p}\) and 1. The second condition requires checking that the average income of the individuals ranked up to
\(\underline{p}\) of the dominant distribution is not lower than the average income of the dominated distribution. The third condition, instead, concerns the dominance between the average income of individuals ranked in the middle, more precisely between
\(\underline{p}\) and
\(\bar{p}\).
As discussed in the introduction, the theoretical results developed in this section contribute to the literature on game theory, happiness economics, and normative economics. However, they are also related to the literature in the filed of applied mathematics and insurance that devotes particular attention to the properties of the Yaari preference functionals and their implications for inverse stochastic dominance.
8 Most prominent contributions include, among others, Muliere and Scarsini (
1989), Wang and Young (
1998), Dentcheva and Ruszczyński (
2006). Muliere and Scarsini (
1989) are concerned with the implications of resorting to second-degree Lorenz dominance as an additional criterion to be implemented for the construction of Lorenz curves orderings. They also argue that there may be a closer relationship between Lorenz dominance of first or higher degrees and rank-dependent measures of inequality than between Lorenz dominance and utilitarian measures of inequality, as the former are explicitly defined in terms of the Lorenz curve, whereas the utilitarian measures are not.
9 Similarly, Wang and Young (
1998) characterize partial orders to be applied for intersecting distribution functions. Dentcheva and Ruszczyński (
2006) introduce sequences of orders that are based on iterated integrals. They also present optimality and duality conditions for convex optimization problems that satisfy the stochastic dominance constraints. With respect to this literature, this paper contributes by proposing dominance conditions that are consistent with a Yaari functional that is expression at the same time of different attitudes towards the income benefited by different parts of the distribution. This is realized by imposing different constraints on social preferences that depend on the introduction of a threshold identifying the different partition of the distributions toward which social preferences do not coincide. In some specific cases, our dominance results can be interpreted as generalization of existing inverse stochastic dominance orderings. Because of the different attitudes incorporated in the same social preferences, our results also consider situations of downward dominance that were not contemplated in the above mentioned literature. In fact, the existing frameworks impose an arbitrary focus on compassion by focusing on the bottom of the distribution. This paper contributes to the existing literature by providing new normative criteria, within a self-contained and unifying framework, that will not suffer from these restrictions.