to other industries) as a share of total employment is not known. But the incomplete
coverage of OES surveys suggests that low-skilled workers are likely to be undercounted,
which could bias the estimates of wage inequality.
Another problem is the discontinuity in the dataset with the change from Standard
Industry Codes to the North American Industrial Classification System in 2002. In this
transition, a number of production activities allied to the motion picture industry were
assigned to other industries, while non-production activities such as movie exhibition and
videotape rental were added to the motion picture industry. As a result, the post-2001
classification is less representative than before. Moreover, these redefinitions prevent
comparisons between 1998-2001 and 2002-2004. The analysis therefore examines 1998-
2001 and finds strong evidence of increased wage inequality in this period. After a
discussion of measurement issues, the section presents the results of these statistical tests.
Measuring Inequality
Measuring inequality is not a simple task. As Cowell (2000, 89) explains, “inequality…
is not self-defining.” Inequality is a property of a distribution of a resource– payment for
work in the form of wages, in this case– and inequality measurement is designed to allow
comparisons of wage distributions. Probably the most popular inequality measure is the
Gini coefficient. However, other measures such as the Theil index are preferable on
theoretical and empirical grounds, particularly when the units of observation have been
aggregated into groups, as generally occurs with compensation and wage data due to
confidentiality requirements.
The analysis in this section therefore employs the Theil
18
An empirical appeal is that bootstrap estimates of standard errors tend to be nearer to their asymptotic
values for the Theil index than for the Gini coefficient. Mills and Zandvakili 1997, 148. Theoretically,
Cowell (2000, 150) notes that the Gini coefficient, unlike the Theil index, is not additively decomposable
(that is, changes in inequality cannot be separated into “between group” and “within group” effects).
However, both the Gini coefficient and the Theil index fulfill two other desirable characteristics of
inequality measures: they are scale invariant, so proportionate changes in wages (due to inflation, for
18
Convention