Read on Twitter
College education in the OECD is characterized by high rates of dropout and of major switching. Is the system good at letting students learn about match-quality through experimentation, or bad at matching students to the programs they are most likely to graduate from? 1/
This question is incredibly relevant for policy, but it is difficult to answer. @tlarroucau, a @Penn JMC, came up with an ingenious way to answer it, and to find if admission systems can be re-designed to reduce inefficiencies (hint: they can, and the econ job market did it!) 2/
What is evidence of a mismatch? That not being assigned to your most preferred program causes you to drop out/switch. But assignments are not random = hard to estimate their causal effect. And application preferences!= true preferences when applicants are strategic. Solution: 3/
Tomas combines admin data from the DA centralized admission system in Chile with novel surveys. Using an RD design that exploits admission cutoffs and elicited preferences, he finds clear evidence of mismatch, an inefficiency that calls for policy intervention. 4/
How much of the dropout is inefficient? We need to know how much dropout there would be in a world in which students are always assigned to their truly most preferred programs. This is not seen in the data, so Tomas simulates this counterfactual with a structural model. 5/
He finds that 1/3 of switches are inefficient. Redesigning assignment mechanisms in ways that increase truth-telling can increase efficiency. For example, a device like the signalling used in the econ job market can reduce inefficient dropouts by up to 20%! 6/
There's a lot more in the paper, such as the effects of dynamic incentives like changes in future switching costs, distributional effects on college graduation etc.
Top-notch research combining different tools to tackle a policy relevant question!
Paper: https://tlarroucau.github.io/JMP_Larroucau.pdf
Top-notch research combining different tools to tackle a policy relevant question!
Paper: https://tlarroucau.github.io/JMP_Larroucau.pdf
