Summer Melt: Research for College Admissions Simulation

summer_melt.md

Summer Melt: Research for College Admissions Simulation

Overview

Summer melt is the phenomenon where students who have been accepted to a college, signaled intent to enroll (often by submitting an enrollment deposit by the May 1 National Candidates Reply Date), and yet fail to matriculate in the fall. The term captures the "melting away" of a college's incoming class during the summer months between high school graduation and the start of college.

Summer melt is distinct from yield (yield measures accepted-to-deposited; melt measures deposited-to-enrolled). A student who melts has already committed — this is post-deposit attrition.

Scale of the Problem

• National estimates: 10-20% of college-intending students do not enroll in the fall (Castleman & Page 2014)
• Harvard SDP estimate: 10-40% of college-intending students succumb to summer melt each year, with the range depending on institution type and student population
• Four-year colleges: Typically lose 2-10% of their deposited class
• Community colleges: Melt rates of 20-40% are common
• Low-income urban districts: Attrition can reach 40%+ (e.g., Philadelphia School District: 36-41% melt for classes of 2022-2024)

Melt Rates by College Selectivity Tier

Summer melt is inversely correlated with institutional selectivity. The more selective the college, the lower the melt, because: (a) admitted students are more committed (high yield), (b) these schools provide generous financial aid that removes financial barriers, (c) the opportunity cost of not attending is perceived as very high, and (d) these schools have sophisticated enrollment management operations.

Tier 1: Highly Selective / HYPSM (2-5% acceptance rate)

• Average melt: ~1-2%
• Range: 0.5-3%
• Mechanism: Near-zero melt because yield rates are 78-85%. Students who deposit at Harvard, Stanford, MIT, etc. almost never fail to show up. The rare cases involve: medical emergencies, family crises, gap year deferrals (which are not true melt — the student retains their spot), or international visa issues.
• COVID exception: In fall 2020, Harvard saw 340 students defer (20% of class, vs. normal 100-130), but these were planned deferrals, not melt. UPenn deferrals jumped ~300% (from ~50 to ~200).
• Financial barriers are minimal: All HYPSM schools meet 100% of demonstrated need; families earning under $75-100K pay $0 tuition.
• Overbooking: These schools overshoot target enrollment by only 1-3% to account for the negligible melt.

Tier 2: Ivy+ / Elite Privates (4-8% acceptance rate)

• Average melt: ~2-4%
• Range: 1-5%
• Mechanism: Similar dynamics to HYPSM but slightly lower yield rates (63-70%) mean more uncertainty. Schools like Columbia, UPenn, Brown, Cornell, Duke, UChicago have strong financial aid but slightly higher student indecision. Cross-admit competition (e.g., a student deposited at Dartmouth who gets off Harvard's waitlist) creates some churn that functionally resembles melt.
• International student melt: A growing concern at this tier (see International section below). Visa denials and policy uncertainty can cause 5-15% melt among international depositors specifically.

Tier 3: Near-Ivy / Selective Privates (7-15% acceptance rate)

• Average melt: ~3-5%
• Range: 2-7%
• Mechanism: Schools like JHU, Northwestern, Vanderbilt, Rice, Georgetown have high brand equity but compete for cross-admits with Tier 1-2 schools. A student deposited at WashU who gets off the Brown waitlist in June will melt from WashU. Waitlist movement at schools above drives melt at this tier. Financial barriers begin to appear for middle-income families ($100-150K) who don't qualify for full need-based aid.

Tier 4: Selective Privates (12-25% acceptance rate)

• Average melt: ~5-8%
• Range: 3-10%
• Mechanism: Schools like Emory, Tufts, Boston College, NYU, USC have yield rates of 25-45%. Melt sources include: waitlist pulls from Tier 1-3, sticker-shock reassessment (families who committed before fully understanding net cost), and competitive offers from state flagships at much lower cost. NYU is particularly vulnerable to melt due to limited institutional aid and high sticker price (~$80K).
• LAC vulnerability: Small liberal arts colleges (Williams, Amherst, Pomona) have low melt (~3-5%) due to committed applicant pools, but their small class sizes (400-500) mean even 15-20 melted students represent a significant fraction.

Tier 5: Top Publics / Regional Selectivity (15-35% acceptance rate)

• Average melt: ~5-10%
• Range: 3-12%
• Mechanism: State flagships (UVA, Michigan, Berkeley, UCLA, UNC, Georgia Tech) face a bifurcated melt pattern:
• In-state students: Low melt (2-4%) — committed, affordable, often only option
• Out-of-state students: Higher melt (6-10%) — may choose in-state flagship or private school instead
• International students: Highest melt (10-15%) — visa issues, currency fluctuations, last-minute decisions to attend home country universities
• Georgia Tech data: Reports ~2% in-state melt, ~8% out-of-state, ~15% international

Tier 6: Selective Public / Broad Access (25-50% acceptance rate)

• Average melt: ~8-15%
• Range: 5-20%
• Mechanism: Schools like UIUC, UW-Madison, Purdue, Virginia Tech have large classes (5,000-8,000+) and lower yield rates. Melt sources include: cost sensitivity (students choosing cheaper community college), academic undermatch (students who applied broadly and commit late), and life circumstance changes. These schools overshoot enrollment targets by 5-10% to compensate.

Open Access / Community Colleges (not in our simulation)

• Average melt: 20-40%
• Range: 15-50%
• Mechanism: No deposit requirement at most community colleges means "intent to enroll" is an extremely weak signal. Castleman & Page found 37% melt at community colleges vs. 19% at four-year institutions. Financial barriers, work obligations, family responsibilities, and lack of residential commitment all contribute.

Melt Rates by Student Demographics

Summer melt disproportionately affects vulnerable student populations. The disparities are stark:

First-Generation College Students

• Melt multiplier: 2.0-3.0x the rate of continuing-generation students
• Source: Castleman & Page 2014; NACAC data
• Mechanism: First-gen students lack family knowledge of college processes. Parents may not understand FAFSA verification, housing deposits, orientation registration, immunization requirements, or other pre-enrollment tasks. There is no family "institutional memory" of college enrollment. 28% of first-gen students report not feeling mentally prepared for college (EAB survey).
• Simulation mapping: Students with parentalEducation <= 2 should have elevated melt risk.

Low-Income Students

• Melt multiplier: 2.0-2.5x
• Source: Castleman & Page 2014; Harvard SDP
• Mechanism: Financial barriers are the primary driver. Even after committing, students may: discover unexpected costs (housing deposits, orientation fees, textbook costs, travel), fail to complete FAFSA verification (requires tax documents many low-income families struggle to produce), lose financial aid due to paperwork errors, or face family pressure to earn money immediately rather than attend college. In Philadelphia, low-income students had melt rates of 56% vs. 19% for higher-income peers.
• FAFSA disruption (2024): The troubled rollout of the new FAFSA form caused submission rates to lag 20-30% behind prior years, disproportionately harming low-income students who depend on aid. This drove elevated melt in summer 2024.

Underrepresented Minorities (URM)

• Melt multiplier: 1.5-2.0x (highly correlated with income and first-gen status)
• Source: Castleman & Page 2014; Hechinger Report
• Specific data: Latino graduates had a 59% melt rate in one urban district study; Black graduates had similarly elevated rates
• Mechanism: URM melt is substantially explained by the intersection of low-income + first-gen status. After controlling for income and parental education, the racial gap narrows but does not disappear — suggesting additional barriers (cultural expectations, family obligations, lack of representation at the institution).
• Lancaster, PA data: District is 60% Hispanic, 16% Black; pre-pandemic melt was 26%, rising to 43% during COVID

International Students

• Melt multiplier: 2.5-4.0x (highly variable by origin country and political climate)
• Mechanism: International student melt has a distinctive profile:
• Visa denials: F-1 visa approval is not guaranteed. Consular officers can deny visas for insufficient financial documentation, perceived immigration intent, or administrative processing delays. Denial rates vary by country (5-40%).
• Travel restrictions: Post-COVID and under varying political administrations, travel bans and visa processing freezes have caused significant disruption. In 2025, at least 4,736 international students had visa records terminated.
• Currency fluctuations: Families in countries with volatile currencies may commit when costs seem manageable, then face a 20-30% increase in local-currency cost between deposit and enrollment.
• Last-minute alternatives: Students may receive late offers from universities in the UK, Canada, or home country that eliminate the complexity/risk of US enrollment.
• Family reunification concerns: Some families change plans based on evolving immigration climate.
• Post-2024 trend: International student melt has increased substantially due to visa policy uncertainty. EAB identified international students as one of two primary populations experiencing "move-in melt" (failing to show up at orientation/move-in).

Causes of Summer Melt (Detailed)

1. Financial Barriers (Primary Cause)

The #1 driver of summer melt across all demographics. Specific mechanisms:

• FAFSA verification failures: ~30% of FAFSA filers are selected for verification, requiring additional tax documents. Low-income families often cannot produce required documentation, causing aid packages to be rescinded or delayed.
• Award letter confusion: Students misunderstand the difference between grants (free money), loans (debt), and work-study (requires work). A package that looks generous may include $15K+ in loans.
• Unexpected costs: Housing deposits ($200-500), orientation fees ($100-300), meal plans ($4,000-6,000/year), textbooks ($1,000-1,500/year), and transportation costs are often not included in the "sticker price" families see during the commitment process.
• Family financial shocks: Job loss, medical emergency, or other financial crisis between May and August can make college unaffordable.
• FAFSA 2024 disruption: The new FAFSA form launched late and with errors, causing 20-30% lower submission rates nationally and delaying aid packages well past the May 1 deposit deadline. This was described as potentially "the worst season of summer melt in memory" (Chronicle of Higher Education).

2. Administrative/Bureaucratic Barriers

• Missing transcripts: Students must request final transcripts be sent to the college; many fail to do so.
• Immunization records: Most colleges require proof of vaccinations (MMR, meningitis); obtaining records can be difficult.
• Placement testing: Some schools require math/English placement tests before registration.
• Housing and orientation deadlines: Missing these deadlines can mean losing guaranteed housing or preferred course registration.
• Technology barriers: Online portals, student accounts, and multi-factor authentication can be barriers for students without reliable internet access.

3. Social and Psychological Barriers

• Isolation: During the school year, students have counselors, teachers, and peers providing support. Over the summer, this support network disappears.
• Imposter syndrome: First-gen and low-income students may question whether they "belong" at a selective college.
• Family pressure: Some families (particularly in immigrant communities) may prefer the student stay close to home, earn money, or help with family responsibilities.
• Decision regret: Students may second-guess their college choice after committing, especially if they learn more about the school over the summer.
• Mental health: 17% increase in anxiety disorder diagnoses among young people over the past decade contributes to enrollment anxiety.

4. Gap Year Decisions

• Pre-COVID baseline: 40,000-60,000 students per year took gap years nationally
• COVID peak: 130,000 students took gap years in 2020-21 (Gap Year Association)
• Elite college deferrals:
• Harvard: 340 deferrals (20% of class) in 2020, vs. normal 100-130
• Williams: 90 deferrals vs. normal 25
• Bates: 10% deferral rate vs. normal 4%
• UPenn: ~200 deferrals vs. normal ~50 (~300% increase)
• Post-COVID: Deferral rates have normalized but remain elevated above pre-2020 levels, as gap years have become more culturally accepted.
• Note: Gap year deferrals are NOT true melt if the student retains their enrollment spot for the following year. However, some students who defer never return (~3% of deferred students do not ultimately enroll).

5. Visa Issues (International Students)

• See International Students section above
• 2025 data: At least 1,220 students at 187 colleges had visas revoked or status terminated since late March 2025
• F-1 visa processing times can exceed 3-4 months, meaning students who receive admission in March may not have visa appointments until July-August

6. Competing Opportunities

• Employment: In tight labor markets, immediate employment at $15-20/hr can seem more attractive than taking on college debt
• Military enlistment: Some students commit to college as a backup but ultimately choose military service
• Waitlist upgrades: Students deposited at a safety school may melt when pulled off a waitlist at a higher-ranked school (this is functionally a "lateral melt" — one school's gain is another's loss)

Castleman & Page (2014): Key Findings

Benjamin Castleman and Lindsay Page's seminal research on summer melt, published as both a paper ("A Trickle or a Torrent?", Social Science Quarterly 2014) and a book (Summer Melt, Harvard Education Press 2014), established the field. Key findings:

Melt Rates by Institution Type

District-Level Data

Castleman & Page studied multiple school districts including Boston Public Schools, Fulton County Schools (GA), Fort Worth ISD, and Chicago Public Schools:

• Boston ACCESS program: 22% of college-intending graduates melted during the summer
• Urban district average: One in five college-intending graduates failed to enroll
• Low-income Latino graduates: 59% melt rate in one urban district
• Low-income overall: 56% melt rate in one district vs. 19% for non-low-income peers

Root Causes Identified

Castleman & Page framed summer melt as "an information deficit rather than a motivational one" — students want to go to college but lack the knowledge, support, and resources to complete the transition. They identified:

1. Absence of school-based support during summer months
2. Confusion over paperwork and administrative requirements
3. Lack of parental guidance (especially for first-gen families)
4. Teenage tendency to procrastinate on complex, unfamiliar tasks

Intervention Results

The text messaging intervention was particularly notable: among 4,882 students across five cities, 68.0% of texted students enrolled vs. 64.9% of control group (p < 0.05). For students with $0 EFC, the effect was 72.1% vs. 66.4% (p < 0.01).

Interaction with Waitlist Movement

Summer melt and waitlist activity are mechanically linked: melt creates the seats that waitlist admits fill. This creates a cascading chain:

The Melt-Waitlist Cascade

1. May 1: Deposit deadline. Colleges count deposits and compare to target class size.
2. May-June: Initial waitlist movement. If deposits are below target, schools pull from waitlist immediately. This is "yield miss" driven, not melt-driven.
3. June-July: Summer melt begins. Deposited students start withdrawing — financial aid falls through, gap year decisions finalize, visa denials arrive, competing offers materialize.
4. July-August: Second wave of waitlist pulls. Colleges that were on-target in May but lost students to melt may go back to waitlist. Some schools keep waitlists active through August.
5. Late August: "Move-in melt" — students who maintained enrollment but fail to arrive on campus. At this point, it's too late for most waitlist activity.

Implications for Simulation

In the current simulation, waitlist processing is a single round after student decisions. Summer melt could be implemented as a post-enrollment step:

1. After all enrollment decisions are finalized, apply a melt probability to each enrolled student
2. Melt probability should be a function of: student parentalEducation (proxy for first-gen), student income, college tier, and international status
3. Melted students are removed from their enrolled college
4. Colleges below target can then pull from waitlist (second waitlist round)
5. This creates realistic waitlist timing: some waitlist activity happens weeks/months after initial decisions

Cascade Effects

A student melting from a Tier 3 school doesn't just affect that school — it may be caused by a waitlist pull from a Tier 1 school, which itself was triggered by a Tier 1 melt. The chain can propagate:

• Harvard has 2 unexpected melts → pulls 2 from waitlist
• Those 2 students were deposited at Northwestern and Duke → they melt from those schools
• Northwestern and Duke each pull from their waitlists → those students melt from Emory and NYU
• And so on, cascading down the tiers

This cascade means that the total system-wide melt from a single HYPSM melt event can be 3-5x the initial number.

Anti-Melt Interventions

Text Nudging Campaigns

The most cost-effective intervention. Key implementations:

• Castleman & Page (2014): Personalized SMS reminders about pre-enrollment tasks. 10 messages over summer. +3.1pp enrollment at $7/student.
• Georgia State "Pounce" chatbot (2016): AI chatbot delivered 200,000+ answers to incoming freshmen. Reduced melt by 22%. Treatment group had 21.4% lower melt and 3.9% higher enrollment. 63% of 3,114 treatment students engaged on 3+ days, exchanging 60 messages each on average. Less than 1% of student messages required human staff attention.

Counselor Outreach

• Providing 2-3 hours of summer counseling support per student
• Most effective for low-income students: +8pp enrollment
• Focus areas: FAFSA verification, housing applications, orientation registration, final transcript requests

Peer Mentoring

• Connecting incoming students with current students from similar backgrounds
• Particularly effective for first-gen males
• Reduces isolation and imposter syndrome during summer months
• Cost: ~$80/student

Financial Aid Interventions

• Proactive outreach about FAFSA verification requirements
• Emergency microgrants ($200-500) for housing deposits and orientation fees
• Bridge loans for students awaiting delayed aid packages
• Clear, jargon-free communication about actual out-of-pocket costs

Campus Engagement

• Summer orientation programs (virtual and in-person)
• Social media groups for incoming class
• Pre-arrival course registration
• Housing roommate matching and communication

COVID and Post-COVID Trends

2020: The COVID Melt Surge

• Summer melt rates increased 20-30% above baseline at many institutions
• Gap year deferrals surged (130,000 vs. normal 40-60,000)
• Two-year college enrollment fell nearly 50%
• Overall undergraduate enrollment dropped ~10%
• Philadelphia melt: 36.2% (2020) vs. 31.5% pre-pandemic (2019)
• Lancaster, PA melt: 43% (2020) vs. 26% pre-pandemic

2021-2022: Elevated Baseline

• Melt rates remained above pre-COVID levels
• Students who delayed during COVID were unlikely to return: only ~3% of deferred students enrolled the following year
• Community colleges continued to see depressed enrollment
• The percentage of 18-24 year olds enrolled in college hit 38.1% in 2021, the lowest since 2006

2023-2024: Partial Recovery with New Disruptions

• Undergraduate enrollment rose 1.2% in fall 2023 — first increase since the pandemic
• However, the 2024 FAFSA rollout caused a new crisis:
• Submission rates lagged 20-30% behind prior years
• Many students did not receive aid packages until well after the May 1 deadline
• Described as potentially "the worst season of summer melt in memory" (Chronicle of Higher Education)
• Philadelphia melt: 40.5% for class of 2024 (highest in three years)

2025: Visa Crisis

• Trump administration visa crackdowns created new international student melt
• At least 4,736 international student visa records terminated
• International students at elite schools canceled summer travel plans, fearing re-entry denial
• Broader chilling effect on international enrollment expected to reduce the 1.1 million international student population

Post-COVID Structural Changes

• "Move-in melt" has emerged as a distinct phenomenon: students who maintained enrollment all summer but fail to arrive on move-in day. EAB reports this is increasing.
• Mental health concerns: 28% of first-gen students report not feeling mentally prepared for college
• Value questioning: ~60% of surveyed students question the value of a college education (EAB)
• Gap years normalized: Post-COVID, gap years are more culturally accepted, contributing to ongoing elevated deferral rates

Melt at Our Simulation's College Tiers

Mapping summer melt to the simulation's 6-tier structure:

Tier 1: HYPSM (5 schools)

Harvard, Yale, Princeton, Stanford, MIT

• Estimated melt: 1-2%
• Melt source: Almost exclusively gap year deferrals (not true melt) and rare international visa issues
• Yield context: 78-85% yield means these schools barely need to overshoot enrollment targets
• Waitlist impact: Even 1-2% melt from a 1,650-student class = 16-33 seats → drives waitlist pulls
• Simulation note: Apply near-zero melt probability. Any melt should trigger waitlist pulls.

Tier 2: Ivy+ (8 schools)

Columbia, UPenn, Brown, Dartmouth, Cornell, Duke, UChicago, Caltech

• Estimated melt: 2-4%
• Melt source: Cross-admit competition with HYPSM, international visa issues, occasional financial reassessment
• Yield context: 63-70% yield; schools overshoot by 3-5%
• Waitlist impact: Cornell's large class (3,600) at 3% melt = ~108 melted students → significant waitlist activity
• Caltech exception: Small class (235) means even 2-3 melted students matter

Tier 3: Near-Ivy (8 schools)

JHU, Northwestern, Vanderbilt, Rice, Notre Dame, Georgetown, CMU, WashU

• Estimated melt: 3-6%
• Melt source: Waitlist pulls from Tier 1-2 are the dominant cause. Financial reassessment for middle-income families. International student visa issues.
• Yield context: 40-55% yield; significant overbooking required
• Georgetown note: Non-binding EA creates yield uncertainty that amplifies melt effects

Tier 4: Selective (16 schools)

Emory, Tufts, BC, Williams, Amherst, Middlebury, Pomona, Swarthmore, Bowdoin, Wellesley, CMC, USC, NYU, Wake Forest, Tulane, Northeastern

• Estimated melt: 4-8%
• Melt source: Cost sensitivity (especially NYU, USC, Tulane, Northeastern which have high sticker prices and/or limited institutional aid). Waitlist pulls from Tier 1-3. State flagship competition on price.
• LAC sub-tier: Williams/Amherst/Pomona/Swarthmore/Bowdoin have committed applicant pools → lower melt (~3-5%)
• NYU: Likely highest melt in this tier due to cost ($80K+) and urban setting challenges

Tier 5: Top Public/LAC (13 schools)

UVA, UCLA, Michigan, Colby, Wesleyan, Hamilton, Davidson, Colgate, UC Berkeley, Georgia Tech, UNC, UT Austin, UF

• Estimated melt: 5-10%
• Bifurcated pattern: In-state students melt at 2-4%, out-of-state at 6-10%, international at 10-15%
• Georgia Tech data: Confirms ~2% in-state, ~8% OOS, ~15% international
• UC system: Large classes (6,000+) mean even 5% melt = 300+ seats
• LAC sub-tier: Colby/Wesleyan/Hamilton/Davidson/Colgate ~4-6%

Tier 6: Selective Public (5 schools)

UIUC, UW-Madison, UW Seattle, Purdue, Virginia Tech

• Estimated melt: 8-12%
• Melt source: Cost sensitivity for out-of-state students, competing offers from other publics, community college alternative for in-state students, international student visa issues (particularly relevant for UIUC and Purdue which have large international populations)
• Large classes: These schools enroll 5,000-8,000+ freshmen; at 10% melt that's 500-800 students → very active late-summer waitlist and over-enrollment management

Simulation Implementation Considerations

Proposed Melt Step

Add a post-enrollment, pre-final-stats step:

For each enrolled student:
  base_melt_prob = tier_melt_rate[college.tier]

  // Demographic multipliers
  if student.parentalEducation <= 2:  // first-gen proxy
    melt_prob *= 2.0
  if student.incomeBracket <= 2:      // low-income
    melt_prob *= 1.8
  if student.international:
    melt_prob *= 2.5

  // College-specific adjustments
  if college has generous financial aid (meets 100% need):
    melt_prob *= 0.5

  // Cap at reasonable maximum
  melt_prob = min(melt_prob, 0.30)

  // Bernoulli trial
  if Math.random() < melt_prob:
    student.melted = true
    college.enrolledCount -= 1

Waitlist Second Round

After melt, colleges below target can pull from remaining waitlist:

For each college with enrolledCount < targetClassSize:
  deficit = targetClassSize - enrolledCount
  pull from waitlist up to deficit (using existing waitlist priority)

Base Melt Rates by Tier

Key Parameters

• MELT_FIRST_GEN_MULTIPLIER = 2.0 — first-gen students melt at 2x base rate
• MELT_LOW_INCOME_MULTIPLIER = 1.8 — low-income students melt at 1.8x base rate
• MELT_INTERNATIONAL_MULTIPLIER = 2.5 — international students melt at 2.5x base rate
• MELT_FULL_NEED_MET_DISCOUNT = 0.5 — colleges meeting 100% need halve the melt probability
• MELT_MAX_PROB = 0.30 — cap individual melt probability at 30%

Data Sources

• Castleman, B.L. & Page, L.C. (2014). "A Trickle or a Torrent? Understanding the Extent of Summer Melt Among College-Intending High School Graduates." Social Science Quarterly, 95(1), 202-220.
• Castleman, B.L. & Page, L.C. (2014). Summer Melt: Supporting Low-Income Students Through the Transition to College. Harvard Education Press.
• Castleman, B.L. & Page, L.C. (2015). "Summer Nudging: Can Personalized Text Messages and Peer Mentor Outreach Increase College Going Among Low-Income High School Graduates?" Journal of Economic Behavior & Organization, 115, 144-160.
• Harvard Strategic Data Project. Summer Melt Tools and Data.
• Education Northwest. "What the Research Says About Summer Melt."
• NACAC. "Avoiding Summer Melt."
• EAB. "Move-in melt is on the rise" (2024); "Summer melt: Why 2024 is different" (2024).
• Georgia State University / AdmitHub. "Pounce" chatbot intervention results (2016-2017).
• School District of Philadelphia. Summer Melt Reports (Classes of 2022-2024).
• Hechinger Report. "Counselors and colleges struggle through the summer to make sure students show up" (2022).
• National Student Clearinghouse Research Center. Enrollment data (2020-2024).
• Chronicle of Higher Education. "This May Be the Worst Season of Summer Melt in Memory" (2024).
• Gap Year Association. Deferral statistics (2020-2021).