firms. The second part describes the results in terms of the validity and predictive power of the
instruments. The last part compares wage and mobility dynamics within and between firms
estimating the effects of comparative advantage and learning on the sample consisting of firm
stayers and firm changers.
4.1 Comparative Advantage and Learning Within Firms
The estimation results, shown in Table 4, are presented in two parts. First, equation (8) is
estimated under the assumption of perfect information to emphasize the impact of comparative
advantage on θi , observed by the market but unmeasured by the econometrician. Second,
the estimation is performed for the model under imperfect information about θi , where both
comparative advantage and learning effects are possible.
Results from the first part of Table 4 confirm the importance of the non random selection
of workers based on unmeasured ability. The cj coefficients which evaluate the impact of
unmeasured ability in each rank j are all significant. More importantly for the comparative
advantage hypothesis, they are significantly different from one another. The joint test for
equality of slopes shows a value of 9.36 for the χ2 (3) statistics which is significant at the 5%
level. The coefficients related to measured skills by rank (the βj ) are still significantly different
from one another (χ2 (3) of 9.71 for the joint test) implying that comparative advantage based
on measured ability is still important. Compared to column (4) of Table 3 however, the impact
is smaller when comparative advantage based on unmeasured ability is controlled for.
For rank to rank differences in the coefficients on measured and unmeasured ability, the
effect of unmeasured ability is significantly different between the middle and upper rank (χ2
of 5.11 significant at the 5% level) implying that it is at that level of the job hierarchy that
comparative advantage based on unmeasured ability plays the most significant role. From the
results on measured skills, rank to rank differences in coefficients are significant between the
lower and middle rank of the hierarchy. Together these results suggest that measured and
unmeasured ability play significant roles in determining the assignment of workers into ranks,
but have different effects at different levels of the job hierarchy with unmeasured ability being
important when moving to the upper part of the hierarchy while measured skills are important
when moving from the lower to middle part. Note that the pure rank effects, dj ’s, all remain
significant implying that measured and unmeasured skills are not the only determinants of
wage increases.
Given the different patterns of transition between ranks for blue and white collared workers
20