In this study we estimate the returns to education on hourly wages in Norway, taking into account the endogeneity of schooling and occupational sector selection. Our results show significant variation in the estimated returns to education across different population groups, with the wage return to an additional year of schooling ranging between 4.8 and 7.8 percent. However, we find strong evidence of non-linearity in the returns across different schooling levels and education fields for all population groups we consider. Interaction effects between education and experience are also present, leading to a rejection of the standard log-linear Mincer wage equation.
Selection effects are found to be significant, though of a moderate size. The magnitude (and direction) of the selection bias varies across different educational groups. In general the OLS estimates are upward biased for the public sector, with the bias being considerably larger for males. These results are consistent with the positive sorting hypothesis suggesting that the more able workers are more likely to participate in the labour force. However, we find the opposite to be the case for private sector workers. These results cast doubt on the existence of positive sorting. Sector-specific differences in the returns to education persist even after correcting for selection effects. We find significantly higher educational returns in the private sector, especially for male workers.
As most previous studies based on IV methods, we find that OLS estimates to be significantly downward biased. In fact, our IV estimates rise even further when we take into account sector choices. These results are puzzling since omitted worker ability, which is likely to correlate positively with schooling such that that more able workers also have higher education, is supposed to give an upward bias in the OLS estimates. Secondly, while our preliminary selection corrections support the positive selection hypothesis, the direction of the selection bias is opposite when we consider endogenous schooling. This is the case for all subgroups we consider. On average, we find that the OLS estimates to have a negative endogeneity bias of about 30 percent and a negative selection bias of about 4 percent relative to the selection corrected IV estimates.