In order to obtain a global impact of individual characteristics on wages, I summarize
the individual characteristics into one variable interpreted as the worker’s skill. 22 To do so, I
estimated a regression of the log wage on education, marital status, sex, nationality, experience
and squared experience, industry and occupation type for the entire original sample of workers.
I used the estimated coefficients related to education, marital status, gender, nationality and
experience to compute the estimated or predicted log wage based on these characteristics. 23
The resulting skill variable has been normalized to 0 and the average of the resulting skill index
by job rank is reported in the last column of the Appendix C Table.
Column 1 of Table 3 presents the results of a regression of wages on rank dummies with con-
trols for occupation and industry, large firm size, public sector and length of the employment
contract. Given that wages are in level, the rank coefficients can be interpreted as additional
dollars value per month from being in a higher rank in the base category for the control vari-
ables. 24 Notice that those coefficients are significant and lower than the raw wage differentials
of the Appendix C Table with no controls for worker and firm characteristics.
Column 2 of Table 3 considers the impact of adding the skill variable on rank effects. 25
On can see that controlling for skills reduces the impact of the rank dummies but that they
remain significant and important.
In order to assess the presence of unmeasured (by the econometrician) individual ability,
the next column of Table 3 presents the results of a fixed-effect estimation. Assuming that
unobserved individual heterogeneity is time invariant and equally valued in the different ranks,
it is possible to eliminate (or control for) this term by using first difference method. If un-
measured ability does not matter in the determination of wages, the fixed-effect estimation
results should be similar to the OLS results. One can see from Column 3 that the fixed-effect
coefficients on ranks significantly lower, and remain significant. This suggests that part of the
rank wage premia is explained by unmeasured ability and part of it still reflects rank effects.
22Given the focus on the role of comparative advantage, this technique, also used in the studies mentioned
earlier applying the quasi-difference and IV method, provides a way to minimize the number of parameters to
be estimated.
23To remain consistent with the Gibbons and Waldman model which focuses on expected productivity equals
to wages in level, the skill variable (estimated with the wage in log) is the exponential of the predicted log wage.
24The base category for occupation and industry is blue collars in the mining and quarrying industry. The
dummy for large firm size is one for firms with more than 500 workers.
25Given that the skill variable is the exponential of the predicted wages, regressing wages on the log of the
skill variable would give a coefficient of 1.
18