The returns to observable and unobservable skills over time : evidence from a panel of the population of Danish twins

This paper provides estimates of the private financial return to education based on large
samples of monozygotic (MZ) and dizygotic (DZ) twins which we obtain from Danish
population registers. Our estimation exploits the fact that our data is a long panel. We show
that the rising inequality, which we observe in the raw data, is due to rising returns to
observable skills. Indeed, our results suggest that the inequality associated with unobservable
skills appears to have fallen since the late 1980's. The fact that we have both MZs and DZs
allows us to separate the rising residual variance into changes in returns to unobservables
and changes in the variance in unobservables across successive cohorts.
Measurement error has been a concern in the twins literature since the usual methodology is
based on within-twin differences. We exploit two instruments that provide additional measures
of the within twin schooling difference: differences in when the twins first join the labour force on a full-time basis, which comes from a register that is independent of the education
registers; and the strong assortative mating in the data which allows us to use twins spouse's
education as an instrument. We also address a further concern in the literature: that
differencing between twins fails to remove individual fixed effects as opposed to family fixed
effects resulting in schooling differences being correlated with the residual. This would induce
the within twin schooling difference coefficient to be biased. Here we exploit the Danish
equivalent of Maimonides' rule which generates potential variation in education within twin
pairs associated with being placed in different classes if they attended a small school in a
larger than average cohort. This different experience across twin pairs is shown to generate
differences in within twin schooling. Our baseline estimates suggests that correcting for selfselection
in schooling, and measurement error, gives returns that are about two fifths higher
than OLS for men and about one fifth higher for women.