312
Journal of Applied Economics
C. Wage gap and switching regression
Given the above results, predicted log wages in the public and private sector
can be estimated consistently. Tables 3 to 5 report predicted log wages, differences
in predicted log wages and the decomposition of predicted log wages into various
parts according to equation (11) by gender and for three different model specifications.
Furthermore, results are shown for two different weighting specifications with
different base categories, β* = β2 = βpri and β* = β1 = βpub, as results are
usually different. Confidence intervals for significance tests have been bootstrapped
and refer to 1300 replications.
Regardless of the specification, the unadjusted wage gap between public and
private sector is statistically significant and around 10 % for males and 24 % for
females, which lies well in line with Rees and Shah (1995) but is slightly higher than
Bender’s (2003) findings for the whole of the UK. For men in all three models and
for either weighting scheme, the gap is more than accounted for by differences in
characteristics. In contrast, differences in returns reduce the unadjusted gap,
though the term is statistically insignificant. Hence, while returns to productive
and job-related characteristics are lower for those in the public sector, differences
in these characteristics of public sector employees more than counterbalance this
effect leading to higher wages in the public sector.15
Results are more sensitive against the weighting scheme for women. Using the
public wage structure the absolute figures are fairly similar to the male ones.
However, if the private sector wage structure is used as weight, only between 46
and 67 % of the wage difference is explained by differences in characteristics
suggesting a substantial public sector premium.
The model specifications with sample selection correction have an additional
term besides the explained and unexplained component. In the presence of sample
selection, unobserved productivity related characteristics will be captured in the
unexplained component in the simple OLS specification. The existence (or non-
existence) of wage premiums may, therefore, be simply due to these unobserved
characteristics rather than discrimination.
As the results show, the selection term is positive for both males and females
and statistically different from zero for the former. At the same time the unexplained
15 Note that Rees and Shah (1995) find the same for males but not for females while Bender
(2003) reports mainly positive contributions from both the explained and unexplained
components. Yet, since both the data and methodology are very different, comparisons have
to be drawn with caution.