This paper quantifies the change in the causal effect of education
on labor market earnings in the United States between 1979 and 2002. In
absence of valid instrumental variables for schooling, I develop a causal
model for earnings and schooling that incorporates heterogeneity in absolute
and comparative advantage across individuals. The model is used to impose
some structure on the observed relationship between schooling and earnings.
A simple intuition arises from the model: if individuals with higher returns to
schooling acquire more schooling, the relationship between log earnings and
schooling will be convex. Likewise, for a fixed cohort of individuals, the
degree of convexity will rise over time if the causal return to education rises.
Differences across cohorts in the mapping between schooling and ability will
lead to permanent differences in the profiles of the earnings-schooling
relationship. Changes in the observed relationship between schooling and
earnings can therefore be decomposed into year-specific and cohort-specific
effects corresponding to causal and confounding components. Using CPS
data for cohorts of men born between 1930 and 1970, I find that the causal
return to education increased by 20-40% between 1979 and 2002, after
controlling for the confounding effects of time-varying ability and
comparative advantage biases across cohorts. This increases explains most
of the observed change in the educational wage structure in the U.S. over
that time period.