Every intervention that reduced summer melt could answer a student back

An admitted student in Atlanta checks her portal in July and finds a hold she does not understand. She was admitted in the spring. She filed Georgia State University's commitment-to-enroll form, the school's equivalent of a deposit. The hold traces to her financial aid file, which was selected for verification: the federal government wants documents proving the income figures on her Free Application for Federal Student Aid, the FAFSA. Her summer list also carries a final high school transcript, an immunization record, orientation registration, and a loan counseling module. In the summer of 2016, Georgia State's admissions office reported that the share of students who filed that commitment form and never arrived ran as high as 18 percent.

The melt rate runs from 8 to 40 percent, and the population sets the number

Enrollment offices call the loss summer melt: admitted students who state an intention to enroll and never start. The number depends on where the intention gets measured. Researchers have anchored it to an acceptance plus a filed aid application, to a college named on a scholarship form, to a commitment-to-enroll form, and to a paid deposit. Each anchor produces a different rate.

In the national sample of the high school class of 2004, where college-intending meant applied, accepted, and filed for aid, 10 to 15 percent never enrolled. Among Boston scholarship applicants who named the college they planned to attend, 18 to 22 percent melted. Georgia State put its commitment-form figure as high as 18 percent. In Providence, researchers described a summer in which up to one-third of graduates who had paid deposits reconsidered where, and even whether, to enroll. Within the Boston sample, nearly 40 percent of students who intended to start at a community college failed to matriculate, about double the nearly 20 percent among their four-year-intending classmates.

There is no single national rate underneath these figures. Each one names its population and its anchor, and the span runs from 8 to 40 percent.

The losses climb as family income falls

The Boston sample splits by the expected family contribution, the EFC, the dollar amount the federal aid formula concludes a family can pay toward college. Applicants whose EFC was zero, the formula's marker for the lowest incomes, melted at 22 percent. Everyone else in the sample melted at roughly 18 percent.

The national data shows the same slope at equal test performance: students of lower socioeconomic status (a composite of income, parental education and occupation) at the sample's mean performance melted at 15 percent, and higher-SES students at similar performance melted at 10 percent.

The successful trials worked one student at a time

The intervention record comes from randomized trials run between 2008 and 2016. The trials that raised enrollment share a design. A sender the student could recognize took on specific summer tasks and could answer whatever the student asked back.

Counselor outreach is the most studied intervention. In the summer of 2011, trials in Boston and in Fulton County, Georgia, offered college-intending, low-income graduates two to three hours of proactive counselor support. Enrollment rose 3 percentage points overall and 8 to 12 points among low-income students, at a cost of $100 to $200 per student. The Boston gain held: offered students were nearly 9 percentage points more likely to be enrolled continuously through the fall of sophomore year. Fulton County's version cost about $115 per student and raised on-time enrollment of low-income students by 8 points. The earliest pilot, in Providence in 2008, cost under $200 per student, and the study authors later reported a 14 percentage-point enrollment gain.

Texting cost $7 per participant. In 2012, personalized messages from local programs reminded graduates in Dallas, Boston, Lawrence, and Springfield of required summer tasks. Where other college-going support was thin, enrollment rose 4 to 7 percentage points, and the largest subgroups gained 11.3 to 14.6 points. In Boston, the estimated effect was null. Peer mentors, at $80 per student, raised four-year enrollment 4.5 points.

Georgia State automated the reply channel in 2016. A chatbot texted 3,745 admitted students and answered individual questions about individual holds. Among committed students, on-time enrollment rose 3.3 percentage points, a 21 percent reduction in melt that works out to roughly 116 additional students enrolled at Georgia State in a cohort of about 3,500. The gains showed up task by task: verification holds fell 3 points, immunization holds fell 5, final transcript submission rose 4, and loan counseling completion rose 6.

A campaign that reached 800,000 students left enrollment unchanged

Researchers then ran the message logic at national scale. Campaigns through the Common Application and one statewide sample reached more than 800,000 lower-income and first-generation students with reminders about FAFSA completion, varying the framing, the delivery, the timing, and access to one-on-one advising. The study found no impact on aid receipt or on college enrollment, overall or for any subgroup, and no variant of framing, timing, or delivery performed differently. The zero was precise. In the Common Application sample, where 82 percent of the control group enrolled anyway, the authors could rule out enrollment effects larger than half a percentage point. In the statewide sample, where 49 percent enrolled, they could rule out effects above 1.3 points.

In the authors' own accounting, the earlier trials had a local partner with closer connections to and knowledge of the students, the content at national scale was more generic and less personalized, and current students may already arrive with better information about the FAFSA. The engagement numbers agree. In the 2012 local trials, 40 to 60 percent of targeted students interacted with a counselor or mentor. In the national campaign's Common Application arm, the share of students who ever answered a text was 11.6 percent. The statewide arm drew responses from more than 40 percent of students, and enrollment still did not move.

Financial aid verification runs a second melt season

Verification is the federal process that selects some FAFSA filers and requires their families to document the application's income figures before aid is disbursed. For the 2016-17 award year, it selected 5.4 million of the 10.6 million students eligible for a Pell Grant. That is a 51 percent selection rate.

Among Pell-eligible students selected for verification, 56 percent went on to receive a Pell Grant. Among Pell-eligible students never selected, 81 percent did. The 25-point gap separates two groups the aid formula had already judged eligible. The National College Access Network, the advocacy group whose analysis of federal Pell end-of-year data produced these figures, estimated verification melt at 22 to 28 percent across the 2015-16 through 2017-18 award years and wrote that "the verification melt rate is even higher than the summer melt rate." Federal Student Aid's own estimate, stated at its December 2019 training conference, was 11 percent.

72.2 percent of applicants who completed verification in the 2019-20 award year saw no change to their Pell award. Among the lowest-income filers, those with an automatic zero expected family contribution, the share seeing no change was 93.4 percent.

The newest evidence points the same way

The evidence on summer melt now spans more than fifteen years. The first counselor pilot ran in Providence in the summer of 2008, and the national campaign results were published in 2021. The most recent large study is observational. Researchers followed roughly 54,000 New York City seniors from the class of June 2020 and matched students who received remote near-peer support, help from a mentor close to the students' own age, to similar students who received none; the supported students matriculated at a higher rate, and the strongest gains went to Black and Hispanic students and to students from low-income neighborhoods. Matching is a weaker design than random assignment. Within that limit, the measured gain was 7 percentage points.

Methodology

Every figure in this piece was verified at a primary-source URL on August 21, 2026. The piece covers traditional-age, deposit-taking or enrollment-commitment admissions. The community-college-intending figure from the Boston sample appears only as a within-sample population comparison from the same primary study; it describes 13 percent of one scholarship sample and is never a national rate. Every melt figure carries its population and its measurement anchor, because the published rates move with the anchor: acceptance plus aid filing, declared intent, commitment form, or paid deposit. No published single national melt rate exists, and no published sector-wide deposit-to-enrollment baseline exists.

The counselor-outreach result (Castleman, Page and Schooley 2014) meets What Works Clearinghouse standards without reservations. The Providence pilot's figures are verified only as reported in the authors' own later working paper; the original journal article is paywalled, and the chain is labeled in the sources. The verification analysis comes from the National College Access Network, an advocacy nonprofit working from federal Pell Grant End-of-Year Report data; Federal Student Aid's lower estimate (11 percent) is presented alongside NCAN's wherever verification melt is quoted. The Georgia State chatbot is a commercial product; the result cited is the peer-reviewed randomized trial, and the paper itself discloses platform pricing. The New York City near-peer study is observational, built on propensity-score matching of a COVID-era cohort, and the piece labels it as carrying less certainty than the randomized trials.