Brian Nolan, Ive Marx and Wiemer Salverda
This array of possible measures leaves the researcher with a surfeit of choice. The properties of different in-
equality measures have been intensively investigated, bringing out for example which are more sensitive to chang-
es at the top versus the middle or bottom of the distribution.8 (Note also that some measures have to exclude or
amend incomes of zero - notably those based on log or exponentials such as the Theil and Atkinson measures and
the mean log deviation.) Percentile-based measures are distinctive in ignoring information about the distribution
between the percentiles being compared, but do provide a direct means of distinguishing trends towards the top
versus the bottom, and also are insensitive to outliers at the very top or bottom which may be particularly poorly
measured. Recent theoretical developments in the inequality literature have also clarified the relationship of ‘tra-
ditional’ inequality indices such as the Gini coefficient to concepts such as relative deprivation and ‘complaints’/
concerns about income distribution that are also relevant for the shaping of policy (Cowell 2008). Measures of
polarisation of the income distribution, which is related to but may be distinct from `traditionar inequality, have
also been developed (Wolfson 1997); again, relatively straightforward measures such as interquartile ranges/ratios
may be helpful in this context.
As well as summary measures, it may also be relevant to focus on the income shares accruing to specific parts
of the distribution: the bottom from a poverty perspective, the middle from a “squeezed middle”/polarisation per-
spective, and the very top given the dramatic increases in top income shares in many countries documented and
explored by Atkinson and Piketty (2007, 2010). Tax data provide the source for tracking top income shares, rather
than the household surveys on which one generally relies for the bulk of the distribution, and one must be aware
of the different features and limitations of these sources in capturing incomes at different points in the distribution.
8 The Gini is less sensitive to changes at the extreme of the distribution than e.g. the mean log deviation (at the bottom) and the squared
coefficient of variation (at the top).
Page • 20