Comparable Indicators of Inequality Across Countries
6. Some Guidance and Recommendations
The nature of the analysis clearly must determine the choice of indicators and sources; analyses relying on
trends over time requiring robust measures of year-to-year variation will need a different approach to ones relying
on cross-country variation. In addition, availability is the key consideration, and studies wishing to encompass a
long time-span or many countries will be constrained in terms of options. None the less, awareness of the many
and various sources of potential non-comparability or limitations in comparability across data sources, as well as
the nature of the inequality measurement process, point towards some general guidelines that we propose the con-
tributions to GINI seek to follow insofar as possible. We set these out at a high level of generality first, and then
discuss their application in different settings.
1/ Since summary inequality measures differ in their sensitivity to changes at different parts of the income
distribution, it is essential to use a range of inequality measures rather than relying on just one, so that results which
are indicator-dependent can be seen. The ubiquity of the Gini coefficient in the literature suggests that it should
be used, to facilitate comparison with existing studies. In complementing this with other measures, the Atkinson
measure is often available and has attractive properties. However, it is also important to be able to capture and
distinguish what is going on towards the top versus the bottom of the distribution: the ratio of P90 and P10 to the
median are easy to derive and interpret, and are also unaffected by imprecision in the estimates of incomes towards
the very top and bottom. After that, measures such as the Theil index, the mean log deviation and the squared coef-
ficient of variation can be used as available. Where there is a specific interest in poverty, standard relative income
poverty measures can also be used to capture the distribution at the lower end. In analyzing earnings dispersion,
percentile ratios provide for ready comparison with existing studies and allow movements towards the top versus
bottom to be distinguished.
2/ In deriving income distributions and inequality measures, equivalence scales must be used, and we propose
as the default setting that where possible the modified OECD (1, 05, 03) scale widely used in European research
and by Eurostat be employed. Where children cannot be distinguished from adults, the square root scale could be
used but the possible bias from not distinguishing children from adults must be kept in mind. It is particularly im-
portant in cross-sectional analysis to check the sensitivity of results to the equivalence scale used (whereas changes
Page • 33