The notion that workers have a comparative advantage in some job ranks is equivalent to
saying that skills are differently rewarded along the successive rungs of the job ladder, and
that workers sort into a given rank according to their level of ability or skills. Column 4 of
Table 3 considers the possibility that comparative advantage and non random selection operate
on measured skills. To take this into account, I added interactions of the skill index and the
worker’s job rank to the baseline regression of column 1. One can see that the coefficients
on the interactions are significant. A test of equality of these coefficients shows a value of
2.09) for the χ2 (3) statistic. This shows evidence of the existence of distinct evaluations of
measured skills in each rank. Finally, column (5) shows the results of a fixed-effect estimation
of the specification with comparative advantage in column (4). The rank coefficients decreases
substantially compared to column (4) and resemble more those of column (3) obtained with a
fixed-effect estimation. This suggests that unmeasured ability matters, even in the presence of
comparative advantage based on measurable skills.
The second panel of Table 3 show the results of a similar analysis applied to the sub-
sample of workers in the private sector only. Results are very similar to those based on the
entire sample. Given that a larger majority of workers in the private sector experience a change
in rank, 26 the similarity in the results is not surprising.
This section has shown evidence that workers self-select into the different levels having a
comparative advantage in a given level based on their level of measured skills. In addition, the
results of the first difference estimation lead us to suspect that unmeasured ability may also
matter in the explanation of the inter-rank wage differentials and thus, in the wage dynamics
within firms.
These results are consistent with the Gibbons and Waldman framework of analysis of wage
and mobility dynamics inside firms. Given that unmeasured ability is likely to be correlated
with measured ability, it is reasonable to expect to find evidence on the fact that workers also
have a comparative advantage based on unmeasured ability. The next section presents the
results based on the Gibbons and Waldman model specification presented in Section 2.
4 Results
This section proceeds in three parts. The first part presents the estimation of the comparative
advantage and learning effects on wage dynamics for the sample of workers staying with their
26See Appendix Table B2.1
19