It also induces downward
biases in the late career earnings to the extent of
8.3-11.5% across education groups. The performance of
semilog cubic function is marginally better than the semilog
quadratic. The quadratic earnings function makes a
significant improvement in reducing the biases and thus
provides a much better approximation to the actual earnings.
Measures of out-of-sample prediction errors also
show the superiority of quadratic function over others.