316
Journal of Applied Economics
component decreases substantially, suggesting a male wage premium, though not
for the public but for private sector. In the OLS model this premium was
counterbalanced by unobserved heterogeneity, which highlights the importance
of a separate treatment of the selection component. Note that the selection effect
is driven by the sector choice rather than the participation choice. Furthermore, it
becomes apparent that even if observable characteristics and compensation were
identical in the two sectors, male public employees would still earn more due to
differences in unobservable productivity related characteristics.
Tables 4 and 5 report the selectivity corrected wage differentials in the bottom
row. Interestingly, only the male differential is significant and indicates that
correcting for non-random assignments into labour force and sector, the overall
wage gap becomes negative. For females, however, higher productivity
characteristics of public sector workers are fully outweighed by lower compensation
differences.
Hence, not accounting for selection bias may explain a large part of the male
differences compared to other studies for the UK or Scotland. For example, Bender
(2003) controls for sector selection but the unexplained part in the decomposition
does not separate these effects out which may lead to an overestimate of the
public sector premium. Elliott et al. (2004) do not account for selection in the first
place and results may suffer from the same problem.
Figure 1 has shown that there are not only differences in average wages but
also in the distribution of earnings. This is particularly pronounced for women.
The above decomposition, though appealing in its simplicity, fails to capture the
distribution of the unexplained earnings gap. However, employing a technique
developed by Jenkins (1994) shows that the results (with few exceptions) also
hold along the pay distribution.16
Bringing average wages in the public sector more in line with market rates
therefore might disproportionately affect individuals at the lower end of the earnings
distribution. However, the direction of change is very different by gender. While
men will actually benefit from such a policy, women will be affected negatively. As
a result, policy makers face a trade-off between efficiency and equity similar to the
one identified by Elliott et al. (2004).
16 Results are not reported here but can be obtained from the author. In summary, unexplained
earnings differentials are small along the lower part of the male distribution suggesting a minor
public sector premium. Yet, at the top end there is indication of a private sector premium that
seems to dominate the average wage analysis. The reverse holds for females and at the lower
end of the distribution there is indication of a small private sector premium. This is very much
in line with the actual distributions shown in Figure 1.