Tuesday, November 5, 2013

Preferences for randomness in German university admissions

Here's a paper that pushed a lot of my buttons: theory, experiments, and institutional detail on university admissions to study medicine (and some other disciplines) in Germany: Flipping a Coin: Theory and Evidence, by Nadja Dwenger, Dorothea Kubler, and Georg Weizsacker

Abstract: We investigate the possibility that a decision-maker prefers to avoid making a decision and instead delegates it to an external device, e.g., a coin flip. In a series of experiments our participants often choose stochastically dominated lottery between outcomes, contradicting most theories of choice such as expected utility. A large data set on university applications in Germany shows a choice pattern that is consistent with a preference for randomization, entailing substantial allocative consequences. The findings are consistent with our theory of responsibility aversion.

Here's some of the institutional description:
"Admissions to German undergraduate university programs in the medical subjects are centrally administered by a clearinghouse. The clearinghouse assigns applicants according to the following three procedures that are implemented in a sequential order:
(1) Procedure A admits students who are top of the class to up to 20% of seats.
(2) Procedure W admits students with long waiting times to up to 20% of seats.
(3) Procedure U represents admission by universities according to their own criteria to the
remaining (at least 60% of) seats.
For each of the three procedures, applicants are asked to submit a preference ranking of universities, which may either be identical or di fferent across procedures. All rank-order lists are submitted at the same moment in time. The central clearinghouse employs the three procedures in a strictly sequential order: all applicants who are matched in procedure A are firmly assigned a seat at their matched university and do not take part in subsequent procedures. All remaining applicants enter procedure W. Likewise, after procedure W, all applicants who are still unmatched enter procedure U. The fact that applicants can submit three (potentially different) rank-order lists of universities, each of which may be relevant, is a unique property of the German mechanism and makes it suitable for our analysis."

Procedure A is apparently an immediate-acceptance ("Boston") algorithm that takes only the preferences of the students as inputs, while procedure U is a university-proposing deferred acceptance algorithm.  While there are strategic reasons for students to submit different preferences in A and U, the authors argue that these can be identified and removed from the data, and that there remain students who order their choices differently in the two procedures, in the manner of experimental subjects who display a preference for introducing some randomness into their assignment.

Georg W. further writes to me as follows:
"Arguably the most important and most interesting strategic motives in the German application system appear because of the multi-stage nature of the mechansim: Applicants need to be careful that they are not matched in the first stage of the mechanism, in cases where they plausibly have a chance to get a better match on subsequent stages."

No comments: