Returns to Education: New Evidence From a Discontinuity in School Entry Tests

Research project funded by the Swiss National Science Foundation (CHF/USD 215,000), the Jacobs Foundation (CHF/USD 150,000), and a seed grant from the Basic Research Fund (CHF/USD 87,000). Co-PI (with Stefan Boes, Lukas Schmid, and Marco Steenbergen).


Many theories developed in the social sciences argue that education is of major causal importance for a variety of outcomes. The empirical quantification of education effects, however, is a challenging task. First, education is a choice variable driven by numerous factors, only some of which are observed, and which often have a direct effect on the outcome, too. Second, education is commonly measured by the years of schooling which largely disregards the complexity of a diverse education system. Third, the effects of education are likely heterogenous over different educational levels and across individuals. As a consequence, if potential confounders and heterogeneity are neglected in a regression-type model, then standard estimators will generally be biased for the true causal effect.

This project aims at the estimation of causal education effects using a unique feature of the Swiss education system. Pupils around the age of 12 had to pass a centrally organized exam (the "Sekundar- or Gymnasialpr├╝fung"). The result of this exam determined the level of secondary school that pupils could attend, and their educational track. Thus, the exam had a large impact on the highest education achieved. For estimation purposes, we can take advantage of a discontinuity in the classification scheme. Pupils below a certain threshold were classified in lower level secondary school, pupils above the treshold in higher level secondary school. Strategic sorting near the treshold can be plausibly ruled out because grading was delegated to external experts. This allows us to analyze the education outcomes of pupils near the discontinuity as if they would come from a randomized experiment.

The project is structured in two parts. The first part is concerned with a large survey of former pupils. In 9-months preparatory work we have identified several schools that conducted the test and archived the old exam results. We have collected this data in extensive field work and have updated the contact details of about 3000 former pupils. Now we plan to conduct a large scale survey about demographic aspects (e.g., civil status, fertility, and health related behavior), economic aspects (e.g., employment prospects, earnings, risk preferences, time discounting), political interest (e.g., turnout, party identification, political attitudes), and sociological and psychological aspects (e.g., altruism, discrimination, social capital). This will provide us a unique database.

The second part tackles the empirical quantification. Using fuzzy regression discontinuity methods, we estimate causal education effects by comparing pupils just below and above the threshold and how they differ in the outcome of interest. These pupils are assumed to be similar, and thus comparable. Given the nature of our data, we expect to obtain causal education effects for a large range of topics. Moreover, due to the lack of credible exogenous variation in almost all previous studies we expect to be among the first that provide such evidence. The results of a pre-study are promising and confirm our expectations. From a policy perspective, our results are, among other things, of great relevance in the on-going debate about harmonizing education systems.

Project period 2012-2014