Flip Klijn writes:
"Hey Al,
Recently, a letter in the Spanish newspaper "La Vanguardia" questions the new mechanism to assign students to universities in Spain.
Link to the letter in Catalan:http://www.lavanguardia.es/lv24h/20100623/53951521785.html
The writer of the letter (a high school student) wonders whether the introduction of *multiple* access grades (the novelty) is desirable. He discusses a hypothetical scenario in which multiples access grades may in fact lead to an undesirable assignment.
I was happily surprised with his analysis. Basically, he implicitly describes the deferred acceptance mechanism through a simple but very nice example that has a priority structure with a cycle (as in Ergin, Econometrica 2002). Then, he makes the point that the assignment is fair (or stable) but not very desirable (since it is inefficient).
It was impossible for me to not react to the letter. My response (in Spanish) was published in the Blog section of the same newspaper:http://www.lavanguardia.es/lv24h/20100624/53951867284.html
In my response, I first mention that with a single access grade there is indeed no incompatibility between stability and efficiency. Next, I argue that in certain situations it might be convenient, however, to have multiple access grades in which case the deferred accepted mechanism is a natural candidate mechanism (referring to the original work of Gale and Shapley, 1962, and the application in the National Resident Matching Program). I also point out that the incompatibility between stability and efficiency in the situation with multiple access grades cannot be solved by using some other mechanism. Finally, I mention that the deferred acceptance mechanism is "hassle-free" (i.e., strategy-proof) and that the experience in Boston and New York (high schools) has been very positive. Therefore, there are reasons to believe that the deferred acceptance mechanism with multiple access grades will work satisfactorily in the assignment of students to Spanish universities."
In essence, you're likely to get high enough correlations among a student's grades, and high enough correlations between those grades and the student's preferences, that the theoretical worst-cases aren't illustrative of what's likely to happen in practice. (I imagine the fact that nothing solves the worst-cases is likely to be important to assuage concerns as well, though.)
ReplyDelete