Estimation of Educational Choice and Returns Parameters using Subjective Expectations Data: Evidence from a Sample of Swedish High School Students
Speaker(s)Mikael Lindahl (University of Gothenburg, Sweden)
Date and time
November 03, 2020
16:00 - 17:00
Per Johansson, Mikael Lindahl, Ariel Pihl
For a sample of public high school students in Stockholm, we elicited beliefs and expectations about future earnings, and other outcomes, associated with university fields of study degrees, as well as for the option of not continuing to university. We apply this data using the framework in Arcidiacono, Hotz, Maurel and Romano (2020) who uses subjective expectations data to obtain ex-ante treatment effect estimates of earnings returns to college majors and occupational choice for students at one US university. In addition to looking at fields of university study, we extend their approach by also focus on the university vs high school dimension, a classical returns to education topic, as well as in addressing more diverse aspects of ex-ante returns to education (i.e., estimating ex ante MTE parameters), of educational sorting patterns and its consequences for bias in estimated returns. We have also access to relatively large sample of students, for which we have added administrative data on background characteristics and follow up data on actual application and enrolment at university.
We find that the individual ex ante returns to university education and university fields of choice varies a lot, are correlated with some observable characteristics, but also that a lot of the variation remains unexplained. Individual’s returns to education differ by educational choice because prospective students sort into educational states, at least to some degree, based on unobservable gains from treatment. Hence, ex ante treatment effect parameters differ, and heterogeneity in returns needs to be taken into account in constructing and estimating models of educational choice and of returns to education. Our estimated ex-ante returns to fields of study are lower than what has been found for data using data from for some selective US colleges, but both university premium and field of choice returns line up fairly well with studies using observational data for Sweden. Ex ante OLS estimates are lower than ex ante ATE and TT estimates, although they converge when using application or enrollment data on education. We find consistent evidence of TT estimates being larger than TUT estimates, which implies sorting based on comparative advantage, both for university education, and especially, for fields of study choice (in line with results in Kirkeboen, Leuven and Mogstad, 2016, using observational data for Norway).