College Admissions as a Matching Market

Source: college_matching_market.md


College Admissions as a Matching Market

Research on how US college admissions fits (or fails to fit) the stable matching framework from market design theory, drawing on the work of Gale-Shapley, Roth, and subsequent researchers.


US College Admissions as a Matching Market

The Two-Sided Matching Framework

College admissions is a canonical example of a two-sided matching market: students have preferences over colleges, and colleges have preferences over students. Prices alone do not clear the market -- you cannot simply purchase a spot at Harvard. Both sides must agree to a match, making this fundamentally different from commodity markets where price equilibrates supply and demand.

The formal theory begins with Gale and Shapley's 1962 paper "College Admissions and the Stability of Marriage," which introduced the deferred acceptance (DA) algorithm and the concept of stability in matching. A matching is stable if there is no student-college pair who would both prefer to be matched with each other over their current assignments. Gale and Shapley proved that a stable matching always exists and can be found by their algorithm.

How It Differs from the Textbook Model

While the Gale-Shapley framework was literally named after the college admissions problem, the actual US admissions system departs from the theoretical model in several critical ways:

  1. Decentralized process: There is no centralized clearinghouse. Students apply directly to colleges, and colleges make independent admission decisions. This contrasts with the NRMP (National Resident Matching Program) for medical residencies, which uses a centralized DA algorithm.

  2. Incomplete and asymmetric information: Colleges cannot perfectly observe student quality, and students cannot perfectly observe their probability of admission. The Gale-Shapley model assumes complete preference orderings on both sides.

  3. Multiple rounds with commitment devices: The US system operates through sequential rounds (ED, EA/REA, EDII, RD, waitlist) rather than a single simultaneous match. Each round has different strategic implications.

  4. Capacity uncertainty and yield management: Colleges do not know how many admitted students will enroll, so they admit more students than they have seats for. This "yield management" problem has no analog in the standard matching model.

  5. Financial aid as a contract term: Matching is not simply student-to-college; it includes the financial terms of the match. Hatfield and Milgrom (2005) extended matching theory to "matching with contracts," which can accommodate financial aid as a dimension of the match, but this adds substantial complexity.

  6. Non-strict preferences and holistic review: Colleges use "holistic" review with subjective components (essays, recommendations, institutional priorities), producing preferences that are not strictly ordinal and may be intransitive or context-dependent.

Comparison with NRMP (Medical Residency Matching)

The National Resident Matching Program is the paradigmatic example of a centralized matching clearinghouse that successfully resolved market failures:

Feature NRMP (Medical Residency) US College Admissions
Structure Centralized clearinghouse Decentralized, sequential rounds
Algorithm Student-proposing DA (Roth-Peranson) No algorithm; colleges decide independently
Strategy-proofness Truthful reporting is dominant strategy for students Strategic behavior is pervasive and rewarded
Stability Produces stable matching by design No guarantee of stability; blocking pairs likely exist
Information Rank order lists submitted after interviews Holistic review with subjective criteria
Timing Single match date ED (Nov) -> EA (Nov) -> EDII (Jan) -> RD (Mar) -> Decisions (Apr) -> Waitlist (May+)
Binding Match is binding on both sides Only ED is binding on students; colleges can rescind

The NRMP was established in the 1950s precisely because the medical residency market had suffered from severe unraveling -- offers were being made progressively earlier, eventually two years before graduation. Roth (1984) showed that the NRMP algorithm produces a stable matching, and Roth (1991) demonstrated that markets with stable clearinghouses tend to succeed while those without them tend to fail.

The key question is: why has US college admissions not adopted a similar centralized mechanism? The answer involves institutional resistance, the value colleges place on "demonstrated interest" as a selection tool, the role of financial aid in the match, and political economy considerations (see "Proposed Centralized Mechanisms" below).


The Unraveling Problem

Roth and Xing (1994): Jumping the Gun

Roth and Xing's landmark 1994 paper "Jumping the Gun: Imperfections and Institutions Related to the Timing of Market Transactions" documented a pervasive phenomenon across dozens of markets: unraveling, in which transactions occur progressively earlier over time, to the point where matches are made with severely incomplete information.

The paper examined markets ranging from medical internships (where offers were made two years before graduation) to postseason college football bowls, and identified common patterns:

  1. The race to go early: If some participants make early offers/applications, others have incentives to go even earlier, creating a cascading dynamic.
  2. Exploding offers: Offers with short deadlines (sometimes hours) that force decisions before alternatives can be explored.
  3. Thin markets: As transactions spread over time, any given moment has few participants, reducing match quality.
  4. Loss of information: Early transactions occur before participants have full information about their options or the quality of potential matches.

Unraveling in College Admissions

US college admissions exhibits classic unraveling dynamics:

Causes of Unraveling in This Market

Roth and Xing identified multiple causes that apply to college admissions:

  1. Intertemporal instability: If the "regular" (later) market produces matches that are unstable relative to what could have been achieved earlier with less information, participants have incentives to contract early.
  2. Strategic complementarity: When some colleges move to early rounds, remaining colleges face a depleted applicant pool in regular decision, incentivizing them to also move early.
  3. Market power from heterogeneity: The most desirable colleges (HYPSM, Ivy+) have market power that lets them set the terms of early engagement.
  4. Exploding offers as the binding mechanism: ED is essentially an exploding offer -- students must commit before seeing other options. Roth's research shows that exploding offers are a hallmark of unraveled markets.

Why Unraveling Persists Without a Clearinghouse

Markets that have successfully resolved unraveling (NRMP, some law clerkship markets) did so by establishing centralized clearinghouses. College admissions has not done this because:


Early Decision as a Commitment Mechanism

The Signaling Interpretation

Avery, Fairbanks, and Zeckhauser's 2003 book The Early Admissions Game: Joining the Elite (Harvard University Press) provided the definitive empirical analysis of early admissions as strategic behavior. Their study of over 500,000 applications to 14 elite colleges found that:

Matching Theory Interpretation

From a matching theory perspective, Early Decision serves as a decentralized commitment mechanism that partially substitutes for a centralized clearinghouse:

  1. Reducing instability: By having students commit early, ED reduces the number of "blocking pairs" (student-college pairs who would prefer each other to their actual match) that would otherwise exist.

  2. Yield certainty as a substitute for stability: In a DA algorithm, colleges know exactly how many students will enroll because the algorithm produces a final matching. ED provides partial yield certainty by guaranteeing enrollment from ED admits.

  3. Sequential mechanism as approximate DA: The round structure (ED -> EA -> RD) can be viewed as an approximation of sequential proposals in the DA algorithm, but with the crucial difference that ED makes proposals binding on one side, which the DA algorithm does not require during the proposal phase.

  4. Preference revelation: ED applications reveal cardinal preference intensity (not just ordinal rankings), which the standard Gale-Shapley framework does not capture. This is information-theoretically valuable but creates strategic asymmetries.

Avery and Levin (2010): Formal Model

Avery and Levin published "Early Admissions at Selective Colleges" in the American Economic Review (2010), providing a game-theoretic model that rationalizes the observed patterns:

ED as Exploding Offer

From Roth's perspective, ED is an exploding offer in reverse: instead of the college making an offer the student must accept immediately, the student makes a binding commitment before receiving the offer. This is even more extreme than a traditional exploding offer because:

This creates a strategic asymmetry where colleges capture most of the surplus from the early matching.

Equity Concerns

The matching theory lens reveals distributional consequences:


Proposed Centralized Mechanisms

Yenmez: Centralized Clearinghouse with Commitment

M. Bumin Yenmez proposed a centralized clearinghouse for college admissions that preserves the desirable features of ED (signaling, yield management) while eliminating undesirable aspects (unfairness, unraveling). Key features:

This approach extends Hatfield and Milgrom's (2005) "matching with contracts" framework to accommodate the specific institutional features of college admissions.

Why Centralization Has Not Happened

Several factors explain the persistence of the decentralized system:

  1. Elite college resistance: Top colleges benefit from the current system's strategic asymmetries. A centralized mechanism would be strategy-proof for students (truthful reporting as dominant strategy), which reduces colleges' ability to exploit demonstrated interest and yield management tactics.

  2. Financial aid complexity: Unlike medical residencies (where salary is roughly standardized), college financial aid packages vary enormously and are a key dimension of the match. Incorporating financial aid into a centralized mechanism is technically feasible (via matching with contracts) but institutionally complex.

  3. Holistic review: Colleges value the discretion of holistic admissions (essays, interviews, institutional priorities). A centralized mechanism requires colleges to submit explicit rank orderings, which conflicts with the narrative of "holistic" review. In practice, colleges do have implicit rankings, but formalizing them raises political and legal issues.

  4. Common Application as partial centralization: The Common Application already centralizes the application process but not the matching process. Moving to centralized matching would require colleges to cede decision-making authority to an algorithm, which no US institution has been willing to do.

  5. Political economy: No entity has the authority or incentive to impose centralization. The Common Application organization, NACAC, and individual colleges all have different interests.

International Examples

Several countries use centralized college matching:

These international examples demonstrate that centralized college matching is technically feasible and can improve match quality, but they also reveal trade-offs: increased stratification, potential reduction in diversity, and the need for standardized assessment metrics.


Relevant Equilibrium Concepts

Stability (Gale-Shapley)

The foundational equilibrium concept for matching markets. A matching is stable if:

Applicability to US admissions: The decentralized system almost certainly produces unstable matchings. Students frequently end up at schools they prefer less than schools that would have preferred them (because they did not apply, applied in the wrong round, or were waitlisted). The sequential round structure creates artificial constraints that prevent stable outcomes.

Nash Equilibrium in Application Games

When students must choose where to apply (with limits on number of applications and strategic round selection), the relevant concept is Nash equilibrium in the application game:

Research shows that Nash equilibria in truncated preference lists (where students don't apply everywhere) produce at most one matching, and any stable matching in equilibrium must be unique. However, multiple Nash equilibria may exist, and they need not be stable in the Gale-Shapley sense.

Competitive Equilibrium

Kelso and Crawford (1982) showed that under a substitutability condition (adding a student to a college's choice set never causes the college to reject a previously accepted student), stable matchings correspond to competitive equilibria with personalized prices. This connection is powerful because:

Core and the Large Core Problem

The core of the college admissions market is the set of all stable matchings. In the standard Gale-Shapley model with substitutable preferences, the core is always non-empty and has a lattice structure (with student-optimal and college-optimal extremes).

However, Yenmez and others have shown that when financial aid is included (the "need versus merit" problem), the core can be very large -- many different stable matchings exist with very different distributional properties. This "large core" result means that the choice of mechanism (student-proposing DA vs. college-proposing DA vs. some other stable mechanism) has significant distributional consequences, even though all mechanisms produce stable outcomes.

Matching with Contracts (Hatfield-Milgrom)

Hatfield and Milgrom (2005) generalized matching theory to include contract terms (not just binary match/no-match). Their framework encompasses:

Under a substitutes condition (and a "law of aggregate demand"), the DA algorithm generalizes to matching with contracts and retains its desirable properties (stability, strategy-proofness for the proposing side). This is the most promising theoretical framework for a centralized college matching mechanism that incorporates financial aid.

Protocol-Free Equilibrium

Recent work on "protocol-free equilibrium" shows that an outcome is a subgame perfect equilibrium robust to the choice of bargaining protocol if and only if it corresponds to a stable matching. This result applies to college admissions problems, matching with contracts, and assignment games. It provides a game-theoretic foundation for stability that does not depend on the specific mechanism used.


Implications for Simulation

What the Theory Tells Us About Modeling

  1. The current system is not a stable matching mechanism. The simulation should not model US admissions as a DA algorithm. Instead, it should model the decentralized, sequential process with its strategic complexities.

  2. Round structure matters. The ED -> EA/REA -> EDII -> RD -> waitlist sequence is not just administrative; it is a mechanism that creates strategic incentives. The simulation should model each round separately with appropriate behavioral rules:

  3. ED: Binding commitment with admissions boost (the "signaling premium"). Model as ~1.5-2x admission probability multiplier.

  4. EA/REA: Non-binding signal of interest. Smaller boost than ED.

  5. RD: No signaling premium. Largest applicant pool, most competitive.

  6. Waitlist: Yield management tool. Colleges admit from waitlist to hit enrollment targets.

  7. Student application portfolios are strategic. Students do not simply apply to their top-N choices. They construct portfolios balancing reach/match/safety schools and strategically allocate their ED application (if any). The simulation should model this portfolio construction.

  8. Yield management is central. Colleges over-admit because they cannot predict yield. The simulation should model yield rates by round and adjust admission thresholds accordingly.

  9. Hook multipliers as preference heterogeneity. Legacy, athletic, donor, and first-gen hooks represent heterogeneous college preferences that violate the simple "rank by quality" assumption. These are exactly the kind of complementarities that the Hatfield-Milgrom framework accommodates but that simple DA does not.

  10. The simulation can measure instability. After producing a matching, the simulation can count the number of blocking pairs -- cases where a student ended up at school A but would have been admitted to (preferred) school B, which would have preferred them to a marginal admit. This is a direct measure of how far the decentralized outcome is from a stable matching.

  11. Equity analysis through the matching lens. The simulation can compare outcomes for students with and without access to early rounds (a proxy for wealth/information access) to quantify the distributional effects documented by Avery et al.

Specific Modeling Recommendations

Key Parameters Informed by Theory

Parameter Theoretical Basis Suggested Range
ED admission multiplier Signaling premium (Avery & Levin 2010) 1.5x - 2.0x
EA admission multiplier Weaker signal than ED 1.1x - 1.3x
EDII admission multiplier Late commitment signal 1.3x - 1.5x
Yield rate by round ED: ~100%, EA: 30-50%, RD: 20-40% Calibrate to real data
Hook multipliers Preference heterogeneity / complementarities Legacy 2-3x, Athlete 3-4x, Donor 3-5x
Student portfolio size Strategic portfolio construction 8-15 applications
Information asymmetry Well-counseled vs. poorly-counseled students Varies by high school type

Key References