Brian Nolan, Ive Marx and Wiemer Salverda
over time generally appear more robust in that respect); the use of the modified OECD scale, the square root scale,
and the 1/0.7/0.5 scale also employed by Eurostat would cover the main variants in use in comparative use.
3/ Where possible, it is desirable to check findings across alternative sources - e.g. LIS versus OECD, ECHP
versus EU-SILC. Similarly, there may be a trade-off between the length of time that can be covered in time-series
analysis and the consistency of available data series, and in the number of countries that can be included in com-
parative analyses. In that situation analyses could start with a broader set of observations, but check their findings
by re-estimation for the smaller number of observations that are judged to be most comparable.
4/ Given the myriad measurement and data complexities, it is unwise to rely heavily on small observed differ-
ences across countries or over time in measured inequality. Confidence intervals for summary inequality measures
such as Gini coefficients are now sometimes available, and where they are not it may be possible to use some “rule
of thumb” along the lines suggested in a recent paper by Atkinson, Marlier, Montaigne and Reinstadler (2010) in
the context of trends in poverty, for example that in looking at income poverty rates particular attention should be
paid to changes of 2 percentage points or larger.
We now discuss in turn time-series analysis, pooled time series + cross-section analysis, and purely cross-
sectional analysis. In time-series analysis the inequality measures employed will of necessity be limited to those
available, and variations in the equivalence scale etc. will be similarly constrained. However, where micro-data
are being employed directly, this allows the range of inequality measures and equivalence scales noted above to
be derived and used.
Time-series analysis
a. For comparative analyses of income inequality requiring a long time-period (back to the 1960s) and
wide country coverage, the UN-WIID and UN-SWIID series are the best available option. UN-SWIID
has only the Gini coefficient but may have advantages there, and also has continuous annual series (via
some imputation), while UN-WIID (v2) is not always continuous but also has decile/quintile shares
which could be used directly or from which some other inequality measures could be derived. For earn-
ings dispersion, the OECD database provides the only readily available source.
b. If a narrower set of countries suffices but a long time period is essential, this could be based on a com-
pilation of data from national sources, most often focused in the case of income inequality on the Gini,
and for earnings dispersion on percentile ratios. Data availability/quality differs widely across countries,
especially with respect to a truly (reasonably) consistent time-series, and it would be highly desirable to
include other measures where feasible. A valuable resource for income inequality analysis is the data
Page • 34