Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Standard view
Full view
of .
Look up keyword
Like this
0 of .
Results for:
No results containing your search query
P. 1
Stable Matching Theory, Evidence, And Practical Design

Stable Matching Theory, Evidence, And Practical Design

Ratings: (0)|Views: 118|Likes:
Published by Milton Recht

More info:

Published by: Milton Recht on Oct 15, 2012
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less





Stable matching: Theory, evidence, and practical design
This year’s Prize to
Lloyd Shapley 
 Alvin Roth
extends rom abstract theory developed in the 1960s,over empirical work in the 1980s, to ongoing eorts to fnd practical solutions to real-world prob-lems. Examples include the assignment o new doctors to hospitals, students to schools, and humanorgans or transplant to recipients. Lloyd Shapley made the early theoretical contributions, which wereunexpectedly adopted two decades later when Alvin Roth investigated the market or U.S. doctors. Hisfndings generated urther analytical developments, as well as practical design o market institutions.
Traditional economic analysis studies markets where prices adjust so that supply equals demand. Boththeory and practice show that markets unction well in many cases. But in some situations, the standardmarket mechanism encounters problems, and there are cases where prices cannot be used at all toallocate resources. For example, many schools and universities are prevented rom charging tuitionees and, in the case o human organs or transplants, monetary payments are ruled out on ethicalgrounds. Yet, in these – and many other – cases, an allocation has to be made. How do such processesactually work, and when is the outcome ecient?
Matching theory
The Gale-Shapley algorithm
Analysis o allocation mechanisms relies on a rather abstract idea. I rational people – who know theirbest interests and behave accordingly – simply engage in unrestricted mutual trade, then the outcomeshould be ecient. I it is not, some individuals would devise new trades that made them better o.An allocation where no individuals perceive any gains rom urther trade is called
The notion o stability is a central concept in cooperative game theory, an abstract area o mathematical economicswhich seeks to determine how any constellation o rational individuals might cooperatively choose anallocation. The primary architect o this branch o game theory was Lloyd Shapley, who developed itsmain concepts in the 1950s and 1960s.Unrestricted trading is a key presumption underlying the concept o stability. Although it allows clearanalysis, it is dicult to imagine in many real-world situations. In 1962, Shapley applied the idea o stability to a special case. In a short paper, joint with David Gale, he examined the case o 
: how individuals can be paired up when they all have dierent views regarding who wouldbe the best match.
Matching partners
Gale and Shapley analyzed matching at an abstract, general level. They used marriage as one o theirillustrative examples. How should ten women and ten men be matched, while respecting their indi-vidual preerences? The main challenge involved designing a simple mechanism that would lead to astable matching, where no couples would break up and orm new matches which would make thembetter o. The solution – the Gale-Shapley “deerred acceptance” algorithm – was a set o simple rulesthat always led straight to a stable matching.The Gale-Shapley algorithm can be set up in two alternative ways: either men propose to women,or women propose to men. In the latter case, the process begins with each woman proposing to theman she likes the best. Each man then looks at the dierent proposals he has received (i any), retains
what he regards as the most attractive proposal (but deers rom accepting it) and rejects the others.The women who were rejected in the rst round then propose to their second-best choices, while themen again keep their best oer and reject the rest. This continues until no women want to make anyurther proposals. As each o the men then accepts the proposal he holds, the process comes to an end.Gale and Shapley proved mathematically that this algorithm always leads to a stable matching.The specic setup o the algorithm turned out to have important distributional consequences; it mat-ters a great deal whether the right to propose is given to the women – as in our example – or to themen. I the women propose, the outcome is better or them than i the men propose, because somewomen wind up with men they like better, and no woman is worse o than i the men had been giventhe right to propose. Indeed, the resulting matching is better or the women than any other stablematching. Conversely, the reverse algorithm – where the men propose – leads to the worst outcomerom the women’s perspective.The clarity and elegance o the Gale-Shapley paper placed it on academic reading lists or economicsstudents worldwide. But its real-world relevance was not recognized until much later. In the early 1980s,Alvin Roth set out to study a very practical allocation problem: the market or newly examined doctors.
Markets or new doctors
In the U.S., students who graduate rom medical school are typically employed as residents (interns) athospitals, where they comprise a signicant part o the labor orce. In the early 1900s, this market waslargely decentralized. During the 1940s, competition or scarce medical students orced hospitals tooer residencies (internships) increasingly early, sometimes several years beore graduation. Matcheswere made beore the students could produce evidence o their qualications, and even beore theyknew which branch o medicine they would like to practice. When an oer was rejected, it was otentoo late to make oers to other candidates. A market ridden with such problems does not producestable matches, because not enough oers can be made in time to ensure mutually benecial trades.In order to make more oers quickly, hospitals imposed strict deadlines or responding to oers. This,in turn, orced students to make early decisions without knowing what other opportunities wouldbecome available later on.In response to these problems, a centralized “clearinghouse”, called the National Resident MatchingProgram (NRMP), was introduced in the early 1950s. In a paper rom 1984, Alvin Roth studied thealgorithm used by this clearinghouse and discovered that it was closely related to the Gale-Shapleyalgorithm. He then hypothesized that the undamental reason or the success o the NRMP was thatit produced stable matches. In the early 1990s, Roth went on to study similar medical markets in theU.K. There, he ound that dierent regions had adopted dierent algorithms, some o which producedstable matches and others not. Those which resulted in stable matches had turned out to be successul,whereas the other algorithms had broken down in various ways.
Practical design
Matching doctors and hospitals
Despite its success, the NRMP still encountered problems. The number o emale medical studentshad grown, and it became increasingly common that dual-doctor couples looked or internships in thesame region. The NRMP could not accommodate these requests, so that many applicants chose not touse the mechanism: a sign that it was not stable. The NRMP – where the hospitals oered positionsto students – was also criticized or systematically avoring hospitals over students. Indeed, as Gale
and Shapley had shown theoretically, the proposing side o the market (in this case, the hospitals) issystematically avored. In 1995, Roth was asked to help design an improved algorithm that wouldeliminate these problems. Along with Elliott Peranson, he ormulated an algorithm, built on appli-cant proposals and designed to accommodate couples. The new algorithm, adopted by the NRMP in1997, has worked well and over 20,000 positions per year have since been matched with applicants.The research underlying the revised design prompted the development o new theory. It seemed thatapplicants could manipulate the original algorithm – by turning down oers which they actually pre-erred and keeping those which were worse – in order to achieve a better outcome. In several theoreticalpapers, Roth showed how misrepresentation o one’s true preerences might be in the interest o thereceiving side (students in the original NRMP) in some algorithms. Drawing on this insight, the revisedNRMP algorithm was designed to be immune to student misrepresentation. Furthermore, computersimulations veried that, in practice, it was not sensitive to strategic manipulation by the hospitals.
Doctor’s first choiceDoctor’s second choiceHospital’s first choiceHospital’s second choice
Matching doctors and hospitals.
When the doctors make oers, they all frst choose hospital a, which accepts doctor 1 (the hospital’s frstchoice). In a second stage, doctor 2 makes an oer to hospital b, and doctor 3 to hospital c, which gives a stable matching. When the hospi-tals have the right to make oers, the result is instead that doctor 2 is matched with hospital c and 3 with b.
Outcome if the doctors make offersOutcome if the hospitals make offers
1+a 2+b 3+c a+1b+3 c+2

You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->