Forms of Economic Inequality
The various arguments for reducing economic inequality turn out to reflect concerns about distinct kinds of inequality that can differ in several dimensions.
The first dimension is the economic variable or variables whose unequal distribution is at issue. The potential variables of interest include (a) income, (b) consumption, (c) wealth, and (d) access to goods and services provided by governmental or non-governmental organizations(hereafter “public services”)—in particular, those that provide capability building services such as health care, sanitation, and education. In most cases it is the amount of the variable accruing to an individual that is most critical, but in some cases the endowment of the individual’s household or family is more important.
The second dimension involves the nature of the distributional entity whose unequal possession of an economic variable is the source of concern. Most often this is the individual person or household. In this case differences among them in possession of the relevant economic variable can be described as differences of economic class, and the population may be divided into a hierarchy of classes, each of which is defined by a pre-specified range of values for the relevant economic variable. (The range can be defined in absolute or in relative terms, i.e., as fractiles) Alternatively, the distributional entity of interest may be a group of people who share a pre-defined characteristic, independently of their economic status. Such groups may be defined ethnically (e.g., by race, caste, tribe, religion, native language) or geographically (by politico-administrative or topographical region). In this paper I will consider only ethnically defined groups, because concern about inter-ethnic inequality is generally much weightier than concern about inter-regional inequality—if only because ethnic identity is difficult or impossible to alter, while regional identity can be altered through migration. Inter-group inequalities—as well as inequalities across separate hierarchical classes of individuals—are most readily measured by assigning to each group the group median for the variable at issue, which sets up a frequency distribution with a number of observations equal to the number of groups.
The third dimension addresses the part (or parts) of an unequal distribution on which concern is focused. I think one can usefully distinguish four distinct configurations of inequality, as follows:
- (a) Accentuated inequality at the lower end of the distribution. This configuration involves a predominant concern with the extent of poverty conceived of in relative terms—i.e., in relation not to a pre-specified poverty line, but to the societal median. It is motivated by Sen’s distinction between income and capability: “Relative deprivation in the space of incomes can lead to absolute deprivation in the space of capabilities” (Sen 1992, p. 115). In the case of class distributions for any given economic variable x, it can be measured analogously to a head count measure of poverty by the number (or proportion) of people falling below y*xm, where xm is the societal median value of x, and y is a pre-specified fraction no higher than—say—50%. Or it can be measured analogously to a gap measure of poverty as the total deficit in x under y*xm of those in the head count, taken as a share of societal total x. The distributive share of the bottom 5 or 10% of the population provides a very rough indicator of the latter measure. In the case of group distributions one would want to focus on the number of ethnic groups whose median falls below the overall median, as well as the proportionate extent to which each group median falls short.
- (b) Accentuated inequality at the upper end of the distribution. This configuration involves a predominant concern with what one might best characterize as “privilege,” in opposition to poverty, also conceived of in relative terms. Just as absolute deprivation with respect to capabilities is linked to relative deprivation with respect to economic resources, so absolute advantage with respect to power, influence and autonomy in a society is linked to relative advantage in terms of economic resources. For class distributions a head count of people with more than some very high level of x would not be very informative, since it is their aggregate economic power that is the major source of concern. Thus it would be best to use a gap-like measure—i.e., the total surplus in x above z*xm of those who have at least that amount of x, taken as a share of societal total x, where z is a pre-specified multiple of at least—say—10. Indicators such as the distributive share of the top 1% of the population provide a very rough approximation of this measure. For group distributions accentuated inequality can be measured by the proportionate extent to which the medians of the highest-placed ethnic groups exceed the overall median.
- (c) Inequality in the form of a weak middle of the distribution. This configuration reflects concern about what recent literature has labelled “polarization,” or more specifically “bipolarization,” which means that the size of the “middle class” is small in comparison with the sizes of the upper and lower classes in the distribution. In the case of class distributions one would need to pre-specify a middle range of values of the variable x at issue, from (1 - v)*xm to (1 + v)*xm, where v takes on a value—say—between 25 and 50 %. The extent of bipolarization could then be measured by the ratio of the head count of those outside that middle range to the head count of those within it, or— probably less informatively, because it would be dominated by the upper class —the share of total x accruing to those outside the middle range. The distributive share of the middle quintile or the share of the middle four deciles of the population provides rough approximations of the latter measure. In the case of group distributions, one would compare the number of group medians relatively distant from the overall median to the number of group medians relatively close to it.
- (b) Inequality spread over the full distribution. This configuration of “entirety” represents distributions that do not show, to any significant extent, the particular attributes encompassed by the three configurations listed above. It can be measured—if imperfectly—by any of the traditional measures of overall inequality, such as the Gini coefficient. Such measures can be applied either to economic class distributions or to ethnic group distributions, even though the number of different pre-defined ethnic groups is bound to be far, far smaller than the number of individuals (or families) in the relevant population.
In the following four sections I will discuss in turn the major moral, political, economic, and social arguments that have been made for limiting economic inequality, characterizing each argument in terms of the societal objective to which the reduction of inequality is expected to contribute. In each case my aim is to determine what form(s) of inequality are at issue.
-  I am using the term “class” in this paper simply as a short-hand for the alternative to “group” as adistributional entity. In sociological and/or Marxist analyses, of course, classes are themselvesdefined as groups of people that share certain important characteristics.
-  To paraphrase Sen (1992, p. 117, fn.), we are interested in inequality between different groups notso much because of intrinsic interest in group differences but because of what such differences cantell us about inequality as between individuals placed in different groups. Sen (1992, pp. 121-22)goes on to note that “The way a person is viewed in a society with racial disparity may be deeplyinfluenced by his or her visible racial characteristics, and that can act as a barrier to functioningpossibilities in many circumstances. Distinctions of caste similarly have influences of theirown... .
-  For some purposes it may be preferable to work with the mean rather than the median. In the restof the paper I will mention only the median, but it should be understood that one might wish to usethe mean instead.
-  For some purposes it is useful to measure also the degree of inequality within each group—on thepremise that, ceteris paribus, greater within-group inequality (which implies greater likelihood ofoverlap of individuals in different groups) reduces the salience of differences in group means. Thisis the logic of a “multidimensional polarization index” proposed by Zhang and Kanbur (2001),which is defined as the ratio of a measure of between-group inequality to a weighted average ofmeasures of within-group inequality. I believe, however, that it is mainly differences in groupmedians that drive concern about group inequalities.
-  Sen (1992, pp. 115-16) elaborates on this point as follows: “In a country that is generally rich,more income may be needed to buy enough commodities to achieve the same social functioning,such as ‘appearing in public without shame’. The same applies to the capability of ‘taking part inthe life of the community’. These general social functionings impose commodity requirements thatvary with what others in the community standardly have.”
-  See Motiram and Sarma (2011) and the references therein. Note that bipolarization is closelyrelated to the notion of bimodality in a distribution.
-  One major economic argument for reducing inequality that I do not consider in this paper is thatgrowing inequality generates macroeconomic instability, by stimulating the growth of anincreasingly fragile financial services industry. This argument has been advanced rather persuasively to explain the global financial crisis that began in 2008; see Kumhof and Ranciere (2011)and Galbraith (2012). But an essential ingredient of the explanation is inadequate regulation of thefinancial sector, and it is not clear that this was itself a consequence of growing inequality, or thatgrowing inequality necessarily entails an out-of-control financial sector.