The Chartbook aims to provide for each country five indicators covering on an annual basis:
1. Overall income inequality
2. Top income shares
3. Income (or consumption) based poverty measures;
4. Dispersion of individual earnings;
5. Top wealth shares/ wealth inequality measures.
What do the indicators show?
For each of the five indicators, we have a “preferred” definition (or, in one case, a “standard” definition), but we have had to depart from this where no data are available on this basis. To aid the reader, we have in the charts marked by the symbol (*) the series based on the preferred (or standard) definition. In a number of countries, this includes cases where the data are available for the preferred definition only for the later part of the period, and we have had to piece together series with different definitions.
Overall income inequality – In the case of overall income inequality, our preferred definition is the distribution of equivalised (using a scale to allow for differences in household size and composition) household disposable income, defined as income from all sources, including transfer payments, minus direct taxes and social security contributions. The equivalence scale used in most cases is the “modified OECD scale”, which gives a weight of 1 to the first adult, of 0.5 to each additional adult, and of 0.3 to each child. This means that the income of a family of 2 adults and 2 children is divided by 2.1. In some cases, other scales are employed, such as the square root scale, where income is divided by the square root of the household size (2 in the example just given). The distribution is among persons: each individual appears in the distribution with the equivalised income of the household. No allowance is made for within-household inequality. In a number of cases, the definitions in the available statistics depart from this preferred version. For example, income may not be adjusted for household size and composition, or the distribution may relate to gross income, before the deduction of income and social security taxes. Because the income tax is usually progressive, inequality is typically higher for gross income than for disposable income.
The overall distribution is summarised in a single summary statistic, typically the Gini coefficient, which is not our preferred statistic but that most commonly published by statistical agencies. The explanation of the coefficient given by most agencies takes the form of geometry, but we prefer to describe it in terms of the mean difference. A Gini coefficient of G per cent means that, if we take any 2 households from the population at random, the expected difference is 2G per cent of the mean. So that a rise in the Gini coefficient from 30 to 40 per cent implies that the expected difference has gone up from 60 to 80 per cent of the mean. Another useful way of thinking, suggested by Amartya Sen, is in terms of “distributionally adjusted” national income, which with the Gini coefficient is (100-G) per cent of national income. So that a rise in the Gini coefficient from 30 to 40 per cent is equivalent to reducing national income by 14 per cent (1/7).
Top income shares – Much of the evidence about top income shares is derived from tax records, and our standard – although not necessarily preferred – definition is gross income for tax purposes before deduction of allowable outgoings. In some cases, income includes capital gains and losses, although where there is a choice (as for the United States and Sweden), we have omitted capital gains and losses. Transfer income is covered to varying degrees in different countries. Because the tax system is typically progressive, the top shares in disposable income are smaller: for example, in the UK in 2000 the share of the top 1 per cent in before tax income was 12.7 per cent, whereas the share in after tax income was 10.0 per cent. It is also worth noting that the measuring unit is typically not the household but the unit reporting income for tax purposes (the tax unit is typically formed by married couples and unmarried adults or adults only depending on the taxation regime of each country). The evidence about top shares is presented in terms of the shares of, typically, the top 1 per cent. This is readily interpreted: a share of 10 per cent for the top 1 per cent means that they receive 10 times their proportionate share of income.
Income (or consumption) based poverty measures – Our preferred definition of poverty follows that adopted in the European Union (EU) agreed common social indicators: a relative measure set at 60 (or 50) per cent of the median equivalised disposable income in the country in question. In some cases, the figures presented relate to absolute poverty measures based on a poverty line fixed over time in terms of purchasing power. It should be stressed that the relative measure is not simply a measure of inequality. It would be quite possible for the EU measure to be reduced to zero without inequality being eliminated: a situation where no one receives less than 60 per cent of the median is quite consistent with considerable inequality.
Dispersion of individual earnings – Our preferred definition of earnings dispersion refers to the wage and salary received by those in employment and whose employment was not affected by absence. The indicator used in most cases is the ratio of earnings at the top decile (the person 10 per cent from the top) to the median earnings expressed as a percentage. This is a measure of how far the distribution of earnings is spread out at the top: a figure of 180 per cent means that those in the top 10 per cent of earnings receive 80 per cent or more in excess of median earnings.
Top wealth shares/ wealth inequality measures – The indicator of wealth is taken to be the net worth of either individuals (as in estate data) or of households (as in survey data). “Net” means that all liabilities (debts) have been subtracted from the total assets (real and financial); the figure for some households is negative (for example where the mortgage exceeds the value of the property). The summary indicator used in most cases is the share of the top 1 per cent. A figure of 25 per cent means that the top 1 per cent owns 25 times their proportionate share.
Linking of series over time
Discontinuities in statistical series on inequality are frequent. The US Census Bureau “selected measure of household income dispersion” covers the period from 1967 to the present, but there are no fewer than 19 footnotes indicating changes in the processing method. This is more than one every third year. Dealing with these is a matter for judgment. In constructing the series in the Chartbook, the rules we have followed are (a) to accept in general continuous published series; (b) to link overlapping series given within a single source by assuming they share a proportional relationship (i.e. if an overlap begins in 1970, the series are linked by multiplying the pre-1970 series by the ratio of the new to the old observation for 1970); (c) to link in the same way overlapping series from different sources where there appears to be a sufficiently close definition (we recognise that this is a matter for judgment); and (d) in some cases, where there is no overlapping year between two series, to join them by linking adjacent years (i.e. implicitly making the additional assumption that there was no change over the intervening period). In a few instances, where a discontinuity is present in very recent years, we have applied the proportional linking, as described above, forward rather than backward. This avoids recent methodological changes affecting observations for the distant past in long-run series.
The proportionate linking means that the reader can rely on the year-to-year percentage changes, but means that the figures graphed here may differ from those in the original sources.
Where the conditions stated above are not satisfied, then we show multiple series without links.
Sources
The sources are described for each country on the page following the chart. We have tried in all cases to check the figures against the original sources. The importance of such checking may be illustrated by reference to South Africa. In seeking data on the overall distribution, we had identified a series for the Gini coefficient covering the years from 1960 to 1987 in the World Income Inequality Database (WIID). Given the problems of securing long-term distributional data for that country, this appeared too good to be true. This proved to be the case. Investigation of the original source (Lachmann and Bercuson, 1992, Table 2) revealed that the title was “Gini coefficients assuming income equality within racial groups”. The data showed the differences between races, which is an important part, but only part, of the story. These data do not measure overall inequality and are not used here.
In this exercise, we have made use of valuable building blocks. In particular the studies of top incomes, largely resulting from the project organised by Atkinson and Piketty (2007 and 2010), provide an anchor for the empirical analysis of top shares. This project gave rise to the World Top Incomes Database subsequently subsumed into the World Wealth and Income Database (referred to below as ‘WTID’ and ‘WID.world’ respectively). But we wish also to cover, as far as possible, the distribution as a whole, and to follow what happens to poverty as well as riches. The series that we present therefore show not only top income shares but also measures of overall inequality and measures of low incomes. Here we are able to draw on the collection of historical data assembled over the years by Atkinson and Brandolini (see for example, Brandolini, 2002).
The general sources on which we have drawn are:
- (a) Atkinson, A B, 2008, The changing distribution of earnings in OECD countries, Oxford University Press, Oxford.
- (b) Atkinson, A B and Piketty, T, editors, 2007, Top incomes over the twentieth century, Oxford University Press, Oxford.
- (c) Atkinson, A B and Piketty, T, editors, 2010, Top incomes: a global perspective, Oxford University Press, Oxford.
- (d) Brandolini, A, 2002, “A bird’s eye view of long-run changes in income inequality”, Bank of Italy Research Department, Rome.
- (e) Luxembourg Income Study (LIS) Key Figures, downloaded from LIS website. In June 2016, the Key Figures covered 47 countries, including 19 of those covered by this Chartbook: http://www.lisdatacenter.org/data-access/key-figures/inequality-and-poverty/
- (f) World Top Income Database (WTID), by F Alvaredo, A B. Atkinson, T Piketty, and E Saez. Online between January 2011 and November 2015.
- (g) World Wealth and Income Database (WID.world), created by F Alvaredo, A B Atkinson, T Piketty, E Saez and G Zucman, http://www.wid.world . The database and the project (managed also with the contribution of Lucas Chancel) is the expansion of a previous version publicly known as World Top Income Database.
- (h) OECD iLibrary, Employment and Labour Market Statistics, Gross earnings decile ratios http://www.oecd-ilibrary.org/employment/data/earnings_lfs-ear-data-en
- (i) Eurostat data based on EU-SILC (Statistics on Income and Living Conditions) http://ec.europa.eu/eurostat/web/products-datasets/-/ilc_di12
In the case of the last of these, it should be noted that the results are published on the basis of the survey year, whatever the underlying income year. The income reference period in EU-SILC is a fixed 12-month period prior to the survey year (such as the previous calendar or tax year). This holds for all countries except the UK, for which the income reference period is the current year and Ireland (not included in the Chartbook) for which the survey is continuous and income is collected for the last twelve months. (This may be seen by consulting the Metadata on the website.) The income year has therefore been taken here, for all countries apart from the UK, as the year preceding the survey year.
As for the WID.world data on Top income shares, we mostly refer to data downloaded in December 2016. At the same time, it is worth stressing that not all data on top income shares is taken from the WID.world. This is the case, for instance, of Brazil and Iceland, where estimates are taken from existing literature.
We owe a considerable debt to the many researchers who have contributed to these sources.