A REEXAMINATION OF THE NRMP MATCHING ALGORITHM

Most graduating medical students in the United States find their first professional appointments through the National Resident Matching Program (NRMP). This service receives rank-order lists of preferences from students and from hospitals, and then generates final assignments of students to hospitals through the use of a specific computerized matching algorithm. The author uses recent findings from the mathematics and economics literatures to demonstrate three difficulties with the NRMP's matching algorithm and the official descriptions thereof. First, the algorithm favors hospitals over students, a feature known to the NRMP since at least 1976, but, in the author's opinion, not made clear in NRMP literature for students. Second, the author argues that the NRMP's justification that its algorithm mimics orderly, noncentralized admission processes is not correct. Institutions operating under non-centralized procedures must typically make more initial offers than there are positions, in the realization that some fraction of their offers will be declined. This arrangement enlarges the choices available to many applicants, and thereby benefits them, whereas the NRMP's algorithm unrealistically assumes that no institution would ever send out any extra offers. Third, the NRMP's algorithm contains incentives for students to misrepresent their true preferences when constructing their rank-order lists. This feature is a substantial disadvantage of the current algorithm and is incorrectly described in literature distributed to students and in published articles from the NRMP. The author discusses the implications of these issues, specifically considering workable alternatives to the NRMP's algorithm; a rough estimate of how many students' assignments might be improved with-a change of algorithm; information for students about the difficulties in accommodating the incentives to misrepresent their preferences under the current algorithm; and ramifications for hospitals. The author recommends that research be done with authentic rank-order lists from students and hospitals to compare alternative algorithms and alternative strategies for participants to use when constructing rank-order lists; that the NRMP revise its literature for students to portray the current algorithm more accurately and in more detail; and that the choice of matching algorithm be reopened to debate because it is possible that the current algorithm is not the fairest one overall.