Next the quadratic specification was experimented on.
The problem of hetroscadesticity was detected only for
the `low’ and `medium’ education groups. For these
groups, the weighted least squares regressions were run
whereas for the higher education group the OLS estimates
are retained. All the estimated coe cients are statistically
significant and have the expected signs. The values of raw
R2 are also higher than those for the earlier specifications
of the earnings function (Table 5).