Public-PRIVATE Pay Differentials
307
Employees who regard trade unions as important for the protection of working
conditions and wages select themselves into public sector jobs compared to
individuals who do not adhere to this perception.
Unsurprisingly, individuals employed in small or medium sized firms have a
higher probability of working in the private sector.11 Education and job tenure on
the other hand do not impact significantly on either decision with some exceptions
in the univariate probit case.12 Finally, occupation matters for the sectoral decision.
Individuals are significantly more likely to be employed in the public sector if they
work in managerial or non-manual occupations.
The results for women are very similar, with some exceptions. First and somewhat
surprisingly, females with children have a lower participation probability compared
to women with very young children. Furthermore, obtaining a higher degree
significantly increases the likelihood of labour market participation. Similarly,
professional occupation increases the probability of public sector employment, as
does job tenure.
In the second step, wage equations for public and private employment have
been estimated using the probit results to construct selection correction terms.
Tables 1 and 2 report the results for men and women respectively. Alongside OLS
results, selection corrected wage equations are estimated for both the univariate
and bivariate case. Standard errors for models 3 to 6 are based on a simple re-
sampling bootstrap method (see Efron and Tibshirani 1993 for details) as the
calculation of the corrected variance-covariance matrix is cumbersome. Thus, 1000
samples of size N are drawn from the original sample (parent sample) with
replacement. For each sample all coefficients are re-estimated and then used to
derive standard errors and confidence intervals.13
11 The majority of public sector workers working in small establishments is employed with local
governments or work in town halls.
12 Yet, once one does not control for occupation, education exhibits a significant impact on
the sector decision. Hence, the main effect of education is on occupation and occupation then
affects the sector choice.
13 Three different types of intervals have been calculated, the normal (N), the percentile (P)
and the bias correct (BC). If the bootstrap statistics are roughly normally distributed, the
normal and percentile intervals will be fairly similar. However, if there are significant differences,
percentile intervals are usually preferred. Furthermore, the point estimate of the original
sample and the average statistic of the bootstrap do not necessarily agree and their difference
is referred to as bias. Then, the bias corrected confidence interval takes these possible
discrepancies into account. If the bias is small, percentile and bias corrected confidence intervals
are roughly identical. Hence, all three intervals will be very similar for an approximately
normally distributed bootstrap statistic and a small bias.