Tuesday, November 16, 2010

Marginal returns of education policies

It is well-known that the returns to education are high, higher than financial returns in fact. Especially for primary education, estimation of Mincer equations has yielded returns over 12% a year, returns that decline somewhat with additional years of education. These are personal returns, that is, how much one's wage increases with an additional year of education. This indicates that one should choose more education than less. From a policy point of view, it is, however, not clear that one should try to stretch as much as possible education. Indeed, higher education is more costly and its returns may differ by individual.

Pedro Carneiro, James Heckman and Edward Vytlacil address this heterogeneity by estimating returns from the National Longitudinal Survey of Youth of 1979. They are certainly not the first ones to do so with this dataset, but the innovation is in the use of instrumental variables. Indeed, they identify a serious shortcoming in interpreting the latent (Corr: local) average treatment effect because the people induced to go to school by a change in an instrument may not be the same that are induced to go to school by a given policy change. As a consequence, the returns for the two types of people can be quite different, and they are in this case. They improve the estimation technique by identifying what sections of an economically interpretable mean marginal benefit surface are identified by different instruments.

Carneiro, Heckman and Vytlacil conclude from their analysis that returns of higher education differ indeed from individual to individual, and in a way that is highly predictable by both the econometrician and the individual. In other words, people who sort themselves into higher education are those who have already experienced high returns and are likely to experience high ones in the future. This indicates that with current policies the right people go to higher education, and that encouraging more to go to college would not yield returns as high as for those who already go there.

1 comment:

Anonymous said...

Hi there - just a quick correction - the L in LATE stands for local, not latent.