Public-PRIVATE Pay Differentials
311
Table 2. (Continued) Estimation results for females and three different model
specifications
OLS |
Univariate probit |
Bivariate probit | ||||
(no correction) | ||||||
(1) Public |
(2) Private |
(3) |
(4) Private |
(5) Public |
(6) Private | |
λs (Sector) |
0.192 |
0.249 |
0.180 |
0.227 | ||
(0.188) |
(0.202) |
(0.167) |
(0.191) | |||
ρju |
0.143 |
0.223 |
0.495 |
0.317 | ||
(0.253) |
(0.308) |
(0.409) |
(0.397) | |||
ρjv |
0.499 |
0.594 |
0.158 |
0.643 | ||
(0.406) |
(0.396) |
(0.306) |
(0.564) | |||
σ |
0.384 |
0.419 |
0.364 |
0.353 | ||
(0.072) |
(0.076) |
(0.052) ** |
(0.060) ** | |||
Observations |
214 |
302 |
214 |
302 |
214 |
302 |
R-squared |
0.47 |
0.38 |
0.48 |
0.39 |
0.48 |
0.39 |
Notes: OLS, sample selection terms based on separate probits, and sample correction terms
based on bivariate probit. Dependent variable is wage in public and private sector. Cross-section
weights applied. *significant at 10%, **significant at 5 %, ***significant at 1%. Standard errors in
parentheses, where OLS standard errors are robust and the s.e. for the remaining models are
bootstrapped. (N) refers to normal.
effect on male public sector wages. The picture for females is more diverse, e.g.,
having a higher degree increases wages in the public sector; in contrast, A- and O-
level holders perform worse in all three models in the public sector.
For males, sector selection has a significant impact on wages. Given conditional
expected wages (equations (9) and (10)), employees working in the public or private
sector perform better than a random individual would have done as the positive
and significant coefficients on the correlation terms show. However, there is no
indication for a participation bias in the univariate model and only weak evidence
in the bivariate specification. For women neither selection coefficient is significant
regardless of the model specification. This is surprising since the labour force
participation for females in the sample is much lower compared to men.14
14 The reported correlation terms and the standard deviations are constructed following Tunali
(1986).