For Good Measure Read online

Page 15


  Another instrumental reason for concern with horizontal inequalities is that they are often an outcome of historical and current discrimination against people because of their identity, which is likely to lead to inefficiency when talented people fail to realize their potential. For example, most studies show that affirmative action for African Americans in the United States had a positive impact on economic efficiency (Badgett and Hartmann, 1995).

  Finally, it may be difficult to attain certain targets, such as poverty elimination or universal education, without tackling horizontal inequality and the overall position of a deprived group, because deprived groups often find it particularly difficult to access state services.

  Measuring Horizontal Inequalities

  Given their significance, there is a need for systematic measurement and monitoring of horizontal inequalities. There is a lack of systematic data by group, though economic data by group is increasingly collected by national governments as well as through some global surveys such as the Demographic and Health Surveys (DHS) Program and the Living Standard Measurement Study (LSMS). The collection of data on inequalities in political power or cultural recognition is very rare, undertaken only by some individual scholars (e.g., Gurr, 1993; Langer, 2005; Wimmer, Cederman, and Min, 2009).

  The measurement of horizontal inequalities raises particular issues, beyond those involved in measuring vertical inequalities (Mancini, Stewart, and Brown, 2008). First, there is the question of which group classification to adopt. Second, group size varies, and hence it may be desirable to weight any aggregate measure by the population of each group. Third, it may also be important to take into account intra-group distribution, since the political and policy implications of inequalities between groups can differ according to whether the differences arise at the top of the distribution of each group, or at the bottom, or because of uniform differences throughout the distribution of each group. A common measure of aggregate horizontal inequality in a country is a population-weighted coefficient of variation of average group performance on any indicator. Foster’s general-means approach shows how group differences vary along the distribution (Foster, Lopez-Calva, and Székely, 2003). This involves estimating parametric means for each group at different points in the group distribution. An aggregate measure of horizontal inequalities for a country as a whole is helpful for comparisons across countries and over time, but for domestic policy purposes simple comparisons of each group with the country average are often sufficient.

  What to Do?

  Goal 10 of the UN’s Sustainable Development Goals calls for the reduction of inequalities between and within countries, and makes explicit reference to inequalities based on “age, sex, disability, race, ethnicity, origin, religion or economic or other status.” It is emphasized that no goal should be considered as attained by a country if it is not met for significant groups. This has clear relevance for measurement, monitoring, and policy. This is an issue that applies worldwide. In the European Union, for example, there has been a long process aimed at defining a set of “core social variables,” to be included in all official surveys, that would allow common breakdowns of the population across various well-being dimensions.1 For monitoring, there is a need to develop a common set of group categories and dimensions of opportunities and outcomes across countries, with common standards and definitions, particularly for some horizontal breakdowns such as disability, race, and ethnicity. Given the differences in salient groups and dimensions of deprivation across countries, however, only a minimal or core set of indicators are likely to be applicable worldwide. There is a need for flexibility in monitoring so as to fit the particular context.

  A wide range of policies for tackling horizontal inequalities has been adopted in different countries (Stewart, Brown, and Langer, 2008). The first requirement is to identify which groups are particularly deprived and which dimensions of deprivation are most prevalent. Policies can be universal or targeted. Universal policies provide benefits or impose taxes according to universal categories, applicable equally to everyone in society. Generally, these benefits, such as universal access to health care services, are likely to benefit deprived groups most, and consequently to reduce horizontal inequalities. Targeted policies identify particular groups and grant their members particular favors, such as access to government employment or educational scholarships. Such targeted policies are often known as “affirmative action.” Affirmative action can be effective, but the policies can also have undesirable side-effects, in some circumstances changing behavior, encouraging strong identification with the favored identity (“ethnicization”), and provoking opposition among the nonfavored groups (Hoff and Stiglitz, 1974; Harrison et al., 2006; Brown, Langer, and Stewart, 2012). Yet, in some situations, the visibility and rapidity of affirmative action is desirable to reduce resentment among deprived groups. Anti-discrimination law can be an effective policy when discrimination is at the root of inequalities, but it needs to be enforced and backed up by universal legal access. However, many horizontal inequalities arise from historical reasons, and consequently, anti-discrimination law will only be effective in reducing such inequalities if discrimination is interpreted very broadly, recognizing the historical origin. The most effective approach is to combine universal and targeted policies, as was successfully adopted in Northern Ireland and Malaysia (Faaland, Parkinson, and Saniman, 2003; Todd and Ruane, 2012). But in both cases, while horizontal inequalities were greatly reduced, societal cohesion remained fragile, pointing to the need for complementary policies to promote societal integration.

  As noted earlier, horizontal inequalities affecting people’s well-being go well beyond the strictly economic realm and include cultural discrimination, official and nonofficial behavior (e.g., by the police or the media), and political discrimination, all of which can affect economic opportunities as well as well-being. Consequently, the policy arena needs to be correspondingly extensive.

  Intra-household Inequality and the Measurement of Money-Metric Inequality

  Why Intra-household Inequality Matters

  Consider any indicator of economic or social well-being, such as consumption, education, or health. Our normative frameworks are typically built on realizations of such indicators for each individual. When the value of an indicator falls below a normatively determined critical value, that individual is identified as being in deprivation. This critical value can be the poverty line for consumption, or other similar lines such as an adequate level of nutrition. The variation in the indicator across individuals in the population under consideration is the basis of inequality measurement. An important strand of the literature then begins with accounting for this variation along different dimensions. For example, how much of this variation is due to variations by caste, race, or ethnicity is often the starting point for a deeper investigation of the role of these factors in inequality. Similarly, variation accounted for by gender is a key element of discussion of gender inequality in a society. Indeed, as discussed in the previous section, inequality across groups with shared characteristics is the basis for analysis of horizontal inequality.

  Gender inequality raises a troubling question: Could it be that boys and girls, and men and women, are treated so differently within the household that their well-being differs from each other? In other words, is there intra-household inequality? Intra-household inequality would lead us to question many normative frameworks where the household is meant to be an institution for cooperation and equity. If intra-household inequality exists, it contributes to overall inequality, and its patterns can in turn shed light on inequality across genders, and across age groups, in the population as a whole.

  Measuring Intra-household Inequality

  The standard instrument for measuring individual well-being is the household survey, which collects a mixture of individual- and household-level information. A key piece of information collected at the household level is data on household income and consumption (or, more accurately, on consumption
expenditure). This is the central data source for generating headline poverty and inequality measures in a large number of countries. In the case of consumption data, leaving to one side a number of well-discussed issues such as the length of the recall period for expenditures, allowing for home-produced consumption, housing services, and price variations, the question arises as to how to go from household-level consumption to information on individual-level consumption, which is needed to generate inequality and poverty measures.

  The answer for official figures for most countries is straightforward and somewhat disconcerting. Total household expenditure is typically divided by the number of members of the household, and each individual is allocated the per capita consumption of the household. In other words, it is assumed that there is no intra-household inequality. This is also the implicit assumption when adult equivalent scales are used to allow for different consumption needs by demographic characteristics. There is assumed to be no inequality across equivalized individuals. Put another way, our standard method of generating headline inequality and poverty measures systematically suppresses intra-household inequality. It therefore understates overall inequality, focusing only on inequality in household per capita consumption.

  Before turning to empirical studies that try to establish the magnitude of intra-household inequality, it is as well to take up the argument that an understatement of inequality levels is not necessarily important when the focus is on changes in inequality over time, as a constant understatement will not affect the trend as such. This is of course true, but the following points should also be considered. First, if we are interested in overall inequality, surely the level matters as well—at the very least, a constant understatement may matter very differently at different levels of inequality. Second, how do we know that the understatement is constant? We will not know this unless we explore the matter empirically, and allow at least for the possibility of understatement.

  How much understatement of inequality is there as the result of the neglect of intra-household inequality? The question is not easy to answer given the nature of standard data sources. If we had true individual-level consumption, which we do not measure in standard household surveys, the question would be irrelevant since we could observe the true overall inequality. There are two possible strategies we can follow.

  The first is to use structural econometrics. In this approach, you start with a model of intra-household allocation, with a free parameter from which intra-household inequality can be inferred; then you estimate this parameter from observed patterns of household level consumption. This is the approach followed by Lise and Seitz (2011), who conclude that “previous work underestimates the level of individual consumption inequality by between 25% and 50%” (p. 352).

  The second approach is to use indicators for which we do have individual-level data, either in the standard household surveys or in especially collected data sets. Since in these cases we do indeed have the “true” distribution of the indicator across all individuals, we can construct the hypothetical distribution where each individual in a household is allocated the household’s per capita value of that indicator. The difference between inequality in the true distribution and the synthetic distribution gives us an estimate of how far wrong we would have gone had we not had individual level data on the indicator.

  In a large number of surveys in the Luxembourg Income Study, Malghan and Swaminathan (2016) find that, for two-earner households, within-household inequality accounts for 30% or more of total inequality. Ponthieux (2015) uses a question in the EU-SILC 2010 thematic module (“What proportion of your personal income do you keep separate from the common household budget?”) to construct a “modified equivalised income” measure. The author finds that “departing from the assumption of full income pooling within couples results in increased levels of various indicators of inequality.” For calorie intake, in one of the first studies to quantify intra-household inequality, Haddad and Kanbur (1990) use a specially designed survey of a small number of households in the Philippines that collected information on nutritional intake of each individual. Using calorie adequacy as the well-being indicator, they find that possible errors in inequality could be of the order of 30%.

  These are all, of necessity, indirect approaches to estimating the understatement of inequality when intra-household inequality is suppressed as in our standard headline measures. But they all indicate a significant scaling up of standard measures of overall inequality that neglect intra-household inequality.

  Intra-household Inequality and the Growth Elasticity of Poverty Reduction

  Clearly, the estimated level of overall inequality is significantly affected by the neglect of intra-household inequality. This understatement must surely affect the assessment of well-being in a society for any given level of per capita income. Empirical work is not sufficiently advanced to test if the understatement is constant or not but, in terms of changes over time, a constant understatement will obviously not affect trends. But are there other aspects of the development discourse, and indeed the discourse in developed countries, that are affected by the understatement of true inequality?

  A key concept introduced in development economics in the last quarter century is that of the “growth elasticity of poverty reduction.” The basic idea behind this notion stems from the argument that the reduction of absolute poverty between two periods can be broken down into a “growth component” and an “inequality change component.” To derive the first component, analysts construct a distribution where all incomes grow at the growth rate of per capita income between the two periods. Then, by construction, inequality is unchanged since each income has grown in the same proportion. You can compute the poverty in the synthetic distribution, and label the change in poverty the “growth component” of poverty change, since it is the result of this “distribution neutral” growth. The remaining part of the actual poverty change can then be attributed to inequality change.

  The percentage change in the “growth component” of poverty divided by the growth rate of the economy (which is, of course, the percentage change in per capita income) is designated the “growth elasticity of poverty reduction,” measuring the responsiveness of poverty to distribution-neutral economic growth. However, the “growth elasticity of poverty reduction” is itself a function of the level of inequality. While the general case is technically ambiguous, Bourguignon (2003) has shown that, for specified functional forms and empirical simulations, the growth elasticity is lower the higher is the level of income inequality. This finding has been interpreted as implying that reducing inequality could not only have a direct level effect on poverty, for a given per capita income, but also have an indirect effect by increasing the responsiveness of poverty reduction to economic growth. For his specific parametrizations, Bourguignon (2003) finds that when the Gini coefficient rises by a third, the elasticity falls by a third.

  One implication of the above discussion is that the true level of inequality is understated because standard methods suppress intra-household inequality. This must mean, by the Bourguignon (2003) argument, that the true growth elasticity of poverty reduction is overstated in standard calculations, since they rely on measures that understate true inequality. And the quantitative magnitudes are significant.

  Estimating the “True” Levels of Inequality

  Quantifying intra-household inequality is a first step toward getting a more accurate measure of the level of inequality, and of the responsiveness of poverty reduction to economic growth. It can also provide a platform for investigating inequality across gender and age groups, both of which are aspects of horizontal inequality. But, as we have seen, so far as the headline money-metric measures of inequality are concerned, standard national household surveys collect consumption information only at the household level, so that understatement of inequality is endemic to official statistics.

  It is unlikely that official national household surveys can be turned to collecting individual
-level consumption information, especially in developing countries. But there are alternatives, following the small empirical literature that exists. First, structural econometric methods can be used to estimate intra-household inequality parameters. Second, systematic investigation of other indicators available at the individual level in standard household surveys can be analyzed to develop a sense of the understatement in these cases if individual information is not available. Thus information on personal income streams and questions on the extent of income pooling can be used creatively by researchers to explore and estimate intra-household inequality. Third, small specialized surveys, like the one in Haddad and Kanbur (1990), can be mounted. As more data is collected we will get a sharper sense of the understatement of inequality as the result of suppressing intra-household inequality.

  The Gender Wealth Gap

  Why the Gender Wealth Gap Matters

  As seen in the previous section, a growing literature has demonstrated that household and individual welfare are not necessarily the same, and that intra-household inequality may condition economic outcomes. What has been of much interest, specifically, is how a woman’s fallback position (those resources she controls should the household dissolve) conditions her bargaining power within the household (Deere and Doss, 2006). To test this proposition, much of the literature on bargaining power has focused either on nonlabor income (data on which is readily available in household income surveys and can be derived from either asset ownership or public or private transfers) or on the ownership of particular assets, such as land or financial assets.