CHOOSING A YARDSTICK AND ITS COMPONENTS

Multiple perspectives can be used to evaluate the distribution of living standards in a soci­ety. These focal points, as labeled by Sen (1992, p. 20), include monetary indicators such as expenditure, income, and wealth, as well as nonmonetary indicators such as multidi­mensional measures of material standard of living, happiness and life satisfaction, func­tioning, and capabilities.

The distribution of income among persons, or households, has attracted the attention of social scientists at least since Gregory King’s 1688 social tables, “which offer unique quantitative views of social structure and income distribution during a statistical Dark Age” (Lindert and Williamson, 1982). Pareto’s analysis of the revenue curve in 1897 is a more recent formalization of this interest. Income is still the most common indicator of economic resources in rich countries. While consumption expenditure is often used in developing countries, the Hicks-Hansen identity for income (or potential consump­tion), which is equal to actual consumption plus the change in net worth^[360] over a given period, ideally ties income and consumption neatly together. But no one data set con­tains fully comparable measures of all three ingredients in any nation, mainly because change in net worth is difficult to measure (see also Brandolini and Smeeding, 2009; Fisher et al., 2012).

8.2.1 Consumption or Income?

The nearest alternative to disposable income is consumption or consumption expendi­ture, a variable that is often preferred in less developed countries because it is more easily measured in such localities. Consumption can be smoothed over time and therefore is less volatile and less reliant on seasonal variation than is income, especially in agricultural soci­eties (Deaton and Grosh, 2000).

A second argument in favor of consumption is that it is more closely related to per­manent income or lifetime resources than current income. As described by Friedman (1957,p. 209), the distributions of current income.. reflect the influence of differences among individual units both in... the permanent component of income and... the tran­sitory component. Yet these two types of differences do not have the same significance; the one is an indication of deep-seated long-run inequality, the other, of dynamic var­iation and mobility.” If one is interested in “deep-seated long-run inequality,” perma­nent income and, hence, consumption are what matter. However, the simple proportionality between consumption and permanent income in the baseline intertem­poral consumer’s optimization problem does not hold if some of its basic hypotheses are relaxed and simple forms of personal heterogeneity are introduced (effects of accumu­lated or inherited wealth, the degree of intergenerational altruism, the variability of uncertain labor incomes, and capacity to borrow, to name just a few). Therefore, current consumption may not be a very good, and not even the best available, proxy of perma­nent income.

Finally, there is the problem of measuring “true” consumption in rich societies. Con­sumption expenditure data are collected mainly to provide weights and prices for mea­suring the Consumer Price Index, not for measuring consumption. Few surveys actually try to measure actual consumption because purchases of durables such as major appli­ances, automobiles, and especially housing must all be spread out over the useful life of the good, which is bought in one period but consumed in another. Indeed, measures of consumption may differ greatly from consumer expenditures for such persons as older individuals living in an owned but mortgage-free house (Fisher et al., 2012; Johnson et al., 2005; Meyer and Sullivan, 2012a,b).

In brief, there is a priori no cogent or practical reason to prefer consumption to income or permanent income to current income. Indeed Haig (1921) and Simons (1938) recognized that income represents the possibility to consume and therefore estab­lished their famous identity that income equals consumption plus or minus changes in net worth. Most often, the choice is driven by available information, and there is a clear preference among rich nations to rely on income and not consumption.

After settling on income as the focal variable, however, a number of important con­ceptual issues and data concerns remain. In addition to the issues of data availability over time and comparability across countries, the analysis of distributional measures requires decisions and assumptions regarding the income concept, the income-sharing unit, the accounting period, and statistics for measuring poverty, material hardship, or the distribution of income (Johnson and Smeeding, 2013; Smeeding and Weinberg, 2001).

8.2.2 The Definition of Income and Other Essentials

The most basic income concepts collected by national statistical agencies and used by researchers are market (factor) or pretax and transfer income and disposable income. On the basis of the recommendations of the reports of the Expert Group on Household Income Statistics—The Canberra Group (2001, 2011), market income should include all types of earnings gross of employees’ social insurance contributions; net self-employment income^{[361] [362]; all types of capital income, including interest, rent, or dividends received (but not accrued); and subtracting interest paid and adding private pensions.}

Disposable income takes market income and subtracts direct taxes (including an employee’s contributions to social insurance) but ignores other “indirect” taxes (prop­erty, wealth, and value-added taxes); then it adds back in regular interhousehold cash transfers received net of those made, as well as all forms of cash and near-cash public income transfers including social insurance benefits (for social retirement, disability, and unemployment); universal social assistance benefits; and targeted income transfer programs such as social maintenance.

However broad these definitions might be, they exclude imputed rents, capital gains and losses, and other unrealized types of capital income, home production, and in-kind transfer benefits such as education and health insurance. Because these items may account for an important share of the economic resources at the household’s disposal, their inclu­sion in the income definition may affect measured inequality. Indeed, research on the United States suggests that uncounted realized and unrealized income from capital increases measured incomes by over 40% at the mean and more than 20% at the median (Smeeding and Thompson, 2011).

Imputed rent for owner-occupied dwellings tends to benefit a wide range of low- to high-income units, especially the elderly, but their overall effect may vary across coun­tries, depending on the level of housing prices and the diffusion of home ownership (Frick and Grabka, 2003). Unrealized appreciation and untaxed income from capital, as well as capital gains, mainly benefit higher-income units. Indirect taxes have a relatively larger impact on the budget of lower-income units (Newman and O’Brien, 2011), but the opposite happens with the imputation of in-kind public benefits for health care, hous­ing, and education valued at their cost of provision. Because the value of these benefits is spread more or less evenly among beneficiaries (“potential” beneficiaries in the case of health insurance), the typical approach is to augment income by a fixed amount, which accounts for a larger fraction of income at lower-income levels (Burkhauseret al., 2012b). In general, elder households and households with children are net gainers from the impu­tation through health insurance and education benefits, respectively, whereas middle­aged childless units are net losers (Garfinkel et al., 2006, 2010).

As stressed in the first Canberra Group Report (2001, pp. 62-67), the undercoverage of property and self-employment income, own account production, imputed rent for owner-occupied dwellings, in-kind social transfers, capital gains, and other unrealized income from wealth are major issues to be addressed in expanding internationally com­parable income measures. But the analysis of these augmented notions of income is also scarce at the national level. Despite these omissions and shortcomings, market income and disposable household income (DHI) remain the standard concepts measured and published by national statistical agencies and research institutions.

8.2.2.1 Reference Period, Income Units, and Resource Sharing

Income is a flow of resources received by people over a given period. To have a coherent concept of income that can be used to compare distributions across countries and analyze trends over time requires common units to describe the period over which income is received and the groups of people who are sharing the income.

The statistics and trends analyzed in this section are all based on annual data, in part because of convention and data availability. The choice of the reference period does, however, have implications for the degree of inequality in the distribution measured at any given point in time. In the presence of fluctuations in income, where some house­holds experience positive or negative shocks or lumpy income streams, the distribution of income will seem more unequal the shorter the reference period (Atkinson, 1983; Atkinson et al., 1995). At intra-annual frequencies, income may fluctuate because of sea­sonal factors (e.g., in agriculture), movement of workers into or out ofjobs, or the timing of payments (e.g., interest on financial assets or liabilities, dividends on stocks). Aggre­gating over the year implies averaging out these differences, although the overall impact on measured inequality may be small (Boheim and Jenkins, 2006). By the same token, lengthening the reference period beyond the year reduces measured inequality by smoothing the variability due to the business cycle or the life cycle (e.g., Bjorklund, 1993; Bjorklund and Palme, 2002). Longer periods of time may come closer to approximating the “lifetime income” concept preferred by some economists, but in practice these data are quite rare. Using Swedish data, Bjorklund (1993) found that the dispersion of four decades’ worth of cumulative income data for individuals was up to 40% lower than dispersion measured from a standard cross section.

Income is typically shared across family or household units. Analysis of the distribu­tion of income across countries and over time requires both adjustments for the econo­mies of scale associated with income sharing and the use of comparable income units. The typical income-receiving unit is the household, but some data sources report income for individuals, families, or tax-paying units, which potentially include individuals, families, and subfamily units. The broader the definition of household, the more measured inequality tends to decrease, because the dispersion of individual incomes is abated by their aggregation and supposedly egalitarian distribution among all members of the unit (Redmond, 1998). The poverty trends discussed below in Section 8.3 and the dis­tributional measures for the entire population discussed in Section 8.4.1 are based on household income surveys and use the household as the income unit. The trends in high-income shares discussed in Section 8.4.2 are typically based on tax-paying units; they are commonly based on national income tax statistics, and multiple tax units may be included in one household.

It is widely accepted that there are greater costs associated with larger households and economies of scale in consumption that are generated by cohabitation. A family with two children faces greater costs than a family with one child, with greater expenses for food, clothing, education, transportation, and housing. As a result, the same level of after-tax income implies a lower material standard of living for the larger family. With economies of scale in a household, though, providing for the second child will not be as costly as providing for the first. Similarly, a couple living together will spend more on housing, utilities, food, and transportation than a single person, but the couple does not need to spend twice as much to obtain the same standard of living, all else being the same.

To account for costs associated with household size and the related economies of scale, researchers have developed different “equivalence scales” to create comparable incomes of different household sizes and compositions. The most commonly used equiv­alence scale, taken from Buhmann et al. (1988), further described by Atkinson et al. (1995), and recommended by the Canberra Group, divides household income by the square root of the household size. Using the square root scale, costs increase with the household size, but at a declining rate. The square root scale, though, does not explicitly acknowledge differences in the cost of living between adults and children. The LIS pro­ject uses the square root scale, and the OECD has used it in its publications since 1995. The EU uses an alternative scale when calculating distributional statistics with the Statistics on Income and Living Conditions (SILC) data (Atkinson et al., 2010a,b). The scale used by the EU divides household income by the weighted number of household members; different weights are applied to adults and children. The household head is given a weight of 1, each additional adult household member a weight of 0.5, and each child a weight of 0.3.^[363] The U.S. Census Bureau adopted a three-parameter equivalence scale that further differentiates between children in different household types. The census scale, discussed by Short (2001), reflects the idea that children in single-parent families represent a greater increase in costs than do children in two-parent families.

The choice of the equivalence scale affects inequality comparisons. It also affects pov­erty comparisons, especially between those who typically live in small units (elderly) or larger units (families with children or multigenerational units) (Buhmann et al., 1988; Coulter et al., 1992).

Finally, the welfare weighting of the single observations may vary. Each observation may receive a weight of 1 (household weight) or may be weighted according to its size (person weight) or its size and composition (equivalent adult weight), again bringing differences in poverty and inequality outcomes (Danziger and Taussig, 1979; Ebert, 1997).

8.2.3 Data Source Comparability: Surveys, Tax Records, and the Rich

The last cause of limited comparability may be attributable to differences in the source of data. Income data are available both from national household surveys and from adminis­trative archives. Of the latter, the most important are income tax records, which have historically provided long runs of continuous data and have been exploited in the literature on top incomes (Atkinson and Piketty, 2007). Income tax records suffer from potentially serious problems, including the incomplete coverage of those with incomes below the tax threshold, inability to adjust for household size, and the tendency to underreport certain types of income. These and other methodological issues related to tax records and calcu­lation of top income shares are discussed in greater detail in Section 8.4.2.

Household surveys are also subject to problems, including sampling errors, which depend on the size and structure of the sample, and nonsampling errors caused by non­response and underreporting (see Chapter 2 of Atkinson et al., 1995). For these reasons, the upper tail of the income distribution tends to be unsatisfactorily covered in sample surveys, unless the rich are oversampled and reporting errors are minimized. The survey-based evidence discussed later in this chapter may be seen as being about the incomes of the bottom 95-99% of the population, and it is thus complementary but not always fully comparable to the results of high incomes based on tax records reported in the final section of this chapter.^[364] The specific statistics used in the calculation of poverty and income inequality using household survey data are discussed below in Sections 8.3 and 8.4.1.

All these factors need to be kept in mind in the analysis of the national trends in income inequality or in cross-national comparisons. While the data include a great deal of “noise” or possibly unknown errors, the important assumption is that the signal derived from the analysis exceeds the noise for most careful analyses, which also include sensitivity tests of assumptions (Atkinson et al., 1995; Gottschalk and Smeeding, 2000). In examining trends, we are aided by the fact that errors may be more consistent across multiple rounds of the same survey, and therefore trends may be more cross-nationally reliable and comparable than levels of inequality (Gottschalk and Smeeding, 1997). But even then, almost all surveys undergo often substantial changes over multiple decades, producing artificial changes in results due to changes in sampling, survey mode, or other changes in procedures.

Finally, full comparability is an impossible goal. Surveys within countries as well as across countries are subject to changes in methods and are characterized by differences in sampling and nonsampling errors. Comparability is vastly increased when the researcher can access the individual observations on household incomes available in a national archive or in international databases, where the original databases are harmo­nized, such as the LIS and the EU-SILC. Here, both levels and trends are more compa­rable than using other methods. Ex ante instructions to compute a series of harmonized data also are available from the OECD (2008, 2011, 2013).

Since 1983, the LIS Cross-National Data Center has been creating “harmonized” income data sets for a growing number of countries. LIS works with the existing income surveys of different countries and converts them to a format with consistent definitions and concepts that make cross-national comparisons possible. By way of nondisclosure agreements and secure remote access servers, LIS also makes possible for research access to income surveys from a number of countries that traditionally do not share their under­lying data. By 2012, LIS included eight different waves of harmonized data covering roughly equivalent points in time between 1967 and 2010 across countries. The initial LIS wave included 7 countries, but the number has grown steadily, reaching nearly 40 countries in the most recent waves.

The EU's statistical agency, Eurostat, provides comparable income survey statistics for the EU member countries. Eurostat initially used a common survey instrument across the European counties but has since switched to an “ex-ante harmonized” framework (Atkinson et al., 2010a,b). The European Community Household Panel Survey covered 15 different countries from 1994 to 2001 and was replaced by the SILC. SILC works through the statistical agencies of the different EU member countries and achieves cross­national comparability through adoption of common definitions and concepts key to cross-national comparability of income and other policy-relevant matters in the EU.^[365] In 1995 there were 13 countries initially represented, but the number of countries expanded to 22 by 2000 and 30 by 2005. The income distribution measures produced by Eurostat now cover 32 different countries. In contrast to LIS, the Eurostat distribution statistics are produced annually; covering the years between 1995 and 2011 there are 380 year-country observations for the different distributional measures.

The OECD also regularly releases income distribution and poverty measures for its member countries. These releases have been highlighted in major publications, includ­ing Growing Unequal? (2008), Divided We Stand (2011), and Crisis Squeezes Income and Puts Pressure on Inequality and Poverty (2013), and are also available in the organization’s Household Income Distribution and Poverty online databases (www.oecd.org/social/ inequality.htm). The OECD figures are based on the national statistical agency house­hold surveys and are created by a network of country specialists using common mea- sures.^[366] Since the figures are calculated from country-specific surveys in different years, the data are not always based on the same years. In several of these publications, the OECD data compare fairly well with the LIS data observed in the same year, but then the OECD methods add more up-to-date data than those available from the LIS. In the mid-1970s, 8 countries were represented in the distributional statistics, but by the late 2000s the number of countries had grown to 34. Because these data tend to be more immediate and can be updated with less ex-post harmonization than, say,

LIS, we use OECD (2013) poverty data to capture the effects of the Great Recession (GR) on poverty below.

8.3.