Incident 42: Inefficiencies in the United States Resident Matching Program

Description: Alvin Roth, a Ph.D at the University of Pittsburgh, describes the National Resident Matching Program (NRMP) and suggests future changes that are needed in the algorithm used to match recently graduated medical students to their residency programs.
Alleged: National Resident Matching Program developed and deployed an AI system, which harmed Medical Residents.

Suggested citation format

Yampolskiy, Roman. (1996-04-03) Incident Number 42. in McGregor, S. (ed.) Artificial Intelligence Incident Database. Responsible AI Collaborative.

Incident Stats

Incident ID
42
Report Count
1
Incident Date
1996-04-03
Editors
Sean McGregor

Tools

New ReportNew ReportNew ResponseNew ResponseDiscoverDiscover

CSET Taxonomy Classifications

Taxonomy Details

Full Description

Alvin Roth, a Ph.D at the University of Pittsburgh, describes the National Resident Matching Program (NRMP) and suggests future changes that are needed in the algorithm used to match recently graduated medical students to their residency programs.

Short Description

Alvin Roth, a Ph.D at the University of Pittsburgh, describes the National Resident Matching Program (NRMP) and suggests future changes that are needed in the algorithm used to match recently graduated medical students to their residency programs.

Severity

Negligible

Harm Distribution Basis

Other:Medical doctors matching to residency

System Developer

National Resident Matching Program

Sector of Deployment

Human health and social work activities

Location

United States

Named Entities

National Residential Matching Program, University of Pittsburgh, Alvin Roth

Technology Purveyor

National Residential Matching Program

Beginning Date

1996-04-03

Ending Date

1996-04-03

Near Miss

Unclear/unknown

Intent

Unclear

Lives Lost

No

Infrastructure Sectors

Healthcare and public health

Incident Reports

The National Residency Matching Program as a Labor Market

[Pulse: Communication]

Roth, Alvin E. PhD

Mellon Professor of Economics, Department of Economics, University of Pittsburgh.

Graphics Table 1

Although medical students are unaccustomed to thinking of the National Resident Matching Program (NRMP) as a labor market, as an economist I am struck by its similarities to other entry-level professional labor markets. What makes the NRMP unusual, although far from unique, is that matching residents to hospitals is organized by a computer algorithm. What makes it familiar are the timing problems this market has experienced, which led to the NRMP's current organization.

An examination of the residency market and other labor markets with timing problems that have employed matching algorithms shows that students and hospitals have a great deal of common interest in preserving an orderly market. [1] It also shows there is room for disagreement between them about what kind of matching algorithm should be employed. The study of the match that I have been asked by the NRMP to direct is intended to shed some light on this latter issue.

Development of the Residency Market ^

Internships, the predecessors of today's residency positions, were introduced in the United States around 1900. Hospitals competed for interns by setting their hire dates earlier than their competitors. Consequently, the date by which most internships were finalized began to creep forward from the end of the senior year of medical school. Dates of appointment unraveled from one year to the next, first slowly then faster, so that by 1944 medical students were arranging their postgraduate employment as interns 2 years in advance of graduation. This meant students had to apply for positions long before they were far enough along in their education to know either their tastes or talents. Hospitals had to hire future staff with little information about how they would develop in their remaining 2 years of medical school. As a result, there were reasons for dissatisfaction on all sides.

In 1945 US medical schools embargoed letters of reference and moved the date of appointment to 1 year before employment was to begin. In subsequent years the dates at which letters were released and appointments made were moved back to the senior year. But the problems in this market did not end when the appointment date was controlled. Students were called on to make increasingly prompt decisions whether to accept offers. In 1945 offers were supposed to remain open for 10 days. Each subsequent year that interval shortened, until by 1949 a grace period of 12 hours was considered too long.

Hospitals found that if an offer was rejected near the deadline, it was often too late for them to reach their next most preferred candidate before that student had accepted other offers. Even when there was a long deadline, much of the action was compressed into the last moments because a student who had been offered a position at, say, his or her third-choice hospital would be inclined to wait as long as possible before accepting, in the hope of eventually being offered a preferable position.

The period before the deadline was frenzied, with students seeking to improve their positions by contacting the hospitals they preferred, and with hospitals sometimes pressuring students into early decisions in order to avoid having to contact students on their waiting lists after the deadline had expired. This congestion in the market, with its collateral missed opportunities and hasty agreements that occasionally were not honored, led in 1952 to the use of a matching algorithm, the National Intern Matching Program, the predecessor of the NRMP.

Other Professional Markets ^

Many markets and submarkets have experienced the unraveling of transaction dates that characterized the medical market before 1945. [2] The (Table 1) concentrates primarily on professional labor markets, but timing problems are not restricted to them; note that the list includes the market for postseason college football bowls. Even fraternities and sororities have experienced the unraveling of selection dates, leading to the term "rush." [3]

Table 1. Examples of Markets With Timing Problems*

In these markets considerable effort has been spent to halt and reverse the unraveling of transaction dates; the (Table 1) lists the organizations entrusted with this task for many of the markets. Some of these organizations, like the NRMP, were created solely to stabilize transaction dates. Many have considerable compulsory power, but often a solution to the timing problem has proved elusive. The difficulties encountered by these other markets may illuminate some pitfalls we need to keep in mind when thi

The National Residency Matching Program as a Labor Market

Similar Incidents

By textual similarity

Did our AI mess up? Flag the unrelated incidents