Programs die small: which programs colleges are quietly closing, and which fields churn fastest

The median program that vanished from the IPEDS completions file awarded 2 credentials in its final year. The median program that survived awarded 16. That gap is the finding, and it is worth more to a program review committee than any list of closures, because it turns "is this program viable" into a question with a national base rate attached.

The scale first. 3,804 of the 109,790 institution x CIP-4 programs active in the 2022-23 award year reported no completions in either 2023-24 or 2024-25, across 5,635 institutions that stayed open the whole time. That is 3.46 percent. Those 3,804 programs carried 18,950 completions between them, 0.36 percent of the 5,261,032 completions those institutions awarded.

76.7 percent of them awarded 3 or fewer credentials in their final year. 49.2 percent awarded exactly one. Only 3.5 percent were awarding 25 or more. Programs rarely go down fighting at the faculty senate. They go out at one graduate a year, and nobody outside the department notices until the teach-out memo circulates.

That is why the analysis is worth running at all. Any single discontinuation is invisible from outside until it has already been decided, and by then the interesting question is years behind you. In aggregate the pattern shows up in the completions file long before the memo does, so when a dean asks whether four graduates a year is unusual you can answer. It is not especially: 3.25 percent of programs that size were gone within two years. One graduate a year is a different conversation, at 23.33 percent.

A caveat that has to travel with every number here. There is no federal field that says "this program was discontinued." The completions file records awards reported, not administrative status. A program that stops reporting may have closed, gone dormant, been renamed, or been recoded into a different CIP. The next section puts numbers on how much of the 3,804 is noise, and the noise is not small.

The unit of analysis is one institution reporting one CIP-4 field in one award year: a UnitID paired with a four-digit CIP code, first majors only, at least one award. A college with three bachelor's tracks inside a single CIP-4 counts once here, not three times. That is coarser than your catalog and finer than a department.

A program counts as discontinued in a base year if it reported completions that year and reported nothing in either of the next two, at an institution that itself kept reporting in both of those years. The institution condition does more work than it looks like. Drop it and college-level exits get scored as program decisions: 279 institutions that awarded credentials in 2022-23 failed to report completions in one or both of the next two years, taking 1,186 programs and 38,171 completions out of the base. Those are excluded from every discontinuation figure below and reported here once, separately. A college going dark is a different event from a curriculum decision.

Size throughout is credentials awarded per year. No federal dataset publishes enrollment by program, so nothing here is a headcount and nothing here should be read as students enrolled. For orientation on the file itself, the 2024-25 award year contains 112,189 live institution x CIP-4 programs at 5,758 institutions awarding 5,701,913 first-major credentials.

On a fixed base of 102,942 programs active in 2014-15, the two-year rule flags 3,649, the three-year rule 3,001, and the five-year rule 2,507. The two-year rule overstates durable discontinuation by about 46 percent, and 1,142 of its 3,649 flags (31.3 percent) reappear within the following three years.

We use the two-year rule anyway, for one reason: it is the only rule that can see the most recent award year. A five-year rule would have nothing to say about anything after 2019-20, which makes it useless for the decision you are actually facing. So read every figure below as "went dark," not "was closed," and discount accordingly when the number is going in front of a board.

Sort every program active in 2022-23 by how many credentials it awarded that year, then ask what share of each band had gone dark two years later. This is the benchmark that program review committees keep trying to construct by anecdote.

The gradient is 172.8x from the smallest band to the largest. Programs awarding a single credential in 2022-23 disappeared at 23.33 percent. Programs awarding more than 100 disappeared at 0.13 percent, 16 out of 11,853. Somewhere in the middle the risk goes quiet: once a program clears 10 graduates a year it drops below 1 percent and stays there.

Read the two columns against each other before anyone uses this to defend or condemn a small program. Small is not the same as doomed. 39.4 percent of the programs that survived awarded 10 or fewer credentials, and a quarter of them sat at 6 or below. Plenty of small programs keep going for years.

What separates the two groups is how hard the dying ones cluster at the floor. The 90th percentile of the discontinued distribution is 9 credentials; the median survivor is at 16. Nine tenths of the programs that disappeared were smaller than a typical program that stayed.

The other thing the totals tell you is that closing programs is not a capacity lever. All 3,804 of them together carried 18,950 completions, 0.36 percent of the 5,261,032 at these institutions. Whatever the case for pruning a portfolio, it is not a volume case. Forecasting enrollment and program viability covers the modelling side of the viability question; this is the base rate that modelling needs underneath it.

Discontinuation rates run from 1.09 percent to 7.59 percent across the two-digit CIP families, a 7x spread. The fast end is Science Technologies and Technicians at 7.59 percent (17 of 224), Area, Ethnic, Cultural and Gender Studies at 7.39 percent (67 of 907), Philosophy and Religious Studies at 6.85 percent (95 of 1,386), Theology and Religious Vocations at 6.41 percent (74 of 1,155), and Engineering Technologies and Technicians at 6.14 percent (227 of 3,697).

The quiet end is Liberal Arts and General Studies at 1.09 percent (22 of 2,011), Personal and Culinary Services at 1.59 percent (34 of 2,143), Psychology at 1.69 percent (44 of 2,609), Biological and Biomedical Sciences at 1.94 percent (86 of 4,436), and Precision Production at 1.96 percent (20 of 1,022).

The ended-rate column on its own will get the story wrong, so the table carries a new-program rate next to it.

Two definitions before you read that table, because the rate columns do not share a denominator. "Ended" is the rule used everywhere else in this post: active in 2022-23, silent in both 2023-24 and 2024-25, over the 2022-23 active count the table shows. "New" is that rule run backwards: active in 2024-25, silent in both 2022-23 and 2021-22, over the field's 2024-25 active count, which the table does not show. The two percentages therefore sit on different populations and should not be subtracted from each other, though the Net column does net the underlying counts.

The new-program column is what keeps the ended column honest. Multi/Interdisciplinary Studies has the sixth-highest discontinuation rate in the country at 5.88 percent and the highest new-program rate at 11.54 percent, 479 of the 4,152 programs the field was running by 2024-25. Net gain of 263 programs, more than any other field. It is the busiest field in the file in both directions, and what looks like heavy churn there is turnover inside a growing portfolio. Science Technologies is the only other field above 10 percent new, at 10.59 percent, and it also ends the window up.

Program counts and graduate counts do not move together either, which is the same lesson from the other direction: nursing added 113 institutions offering the BSN and 40 graduates over six years, while computing added 15 institutions and 50,966 graduates. The field-level completions breakout has that side.

Eight of the 36 fields shown lost programs on net: Engineering Technologies at -27, Family and Consumer Sciences at -26, Philosophy and Religious Studies at -21, Legal Professions at -20, Theology at -17, Area and Ethnic Studies at -15, History at -7, and English at -3. That is the actual contraction list, and it is a short one. The growth side is much larger in magnitude: Multi/Interdisciplinary Studies +263, Health Professions +172, Computer and Information Sciences +119, Engineering +109, Education +96, Biological and Biomedical Sciences +76.

Coverage note for anyone quoting the table: these 36 fields cover 109,595 of 109,790 programs (99.8 percent) and 3,793 of the 3,804 discontinuations. Two of the 38 two-digit families hold under 200 programs each and are suppressed rather than shown at rates a handful of programs could swing.

Drop to the four-digit level and the rates get sharper. Among fields hosted at 100 or more institutions, Religious Music and Worship leads at 17.00 percent, followed by Energy Systems Technologies at 14.06 percent and Germanic Languages and Literatures at 13.99 percent. By raw count the most-dropped fields are different again: Health and Medical Administrative Services lost 68 host institutions, Business Operations Support 59, Entrepreneurial and Small Business Operations 53, Multi/Interdisciplinary Studies Other 48, Computer Software and Media Applications 45, and Religion/Religious Studies 44. Count and rate answer different questions and neither one substitutes for the other.

Switching the grain to institution x CIP-4 x award level changes the base rate, and the two rates must never appear in the same sentence as if they were comparable. At this finer grain the overall discontinuation rate is 4.85 percent (8,035 of 165,786), higher than the 3.46 percent CIP-4 rate, because a college that drops its certificate in a field while keeping the bachelor's counts here and does not count there.

Inside that grain the credential ordering is stark. Post-master's certificates churn at 14.46 percent, 4.2 times the bachelor's rate of 3.43 percent. Postbaccalaureate certificates run 12.25 percent (727 of 5,935). The most durable credential in the file is the professional practice doctorate at 1.81 percent, 29 of 1,599.

The last two columns explain most of the ordering. Graduate certificates are structurally tiny: the median postbaccalaureate certificate that disappeared awarded exactly 1 credential, and the median one that survived awarded 5. The size-at-death rule from earlier does not stop applying at the credential boundary, it just meets a category where almost everything is small to begin with. Associate's and bachelor's degrees show the same pattern in miniature, a median of 1 at death against 9 and 13 for survivors.

Professional practice doctorates sit at the other extreme, a median of 49 completions among survivors, and they almost never disappear. Certificates are the flexible layer of the portfolio and behave like it. That is not a failure mode; it is the point of a certificate. It does mean that any institutional dashboard counting "programs discontinued" without splitting credentials will report mostly certificate turnover.

A program at a private nonprofit four-year college is 1.85 times as likely to disappear as one at a public four-year: 4.39 percent (1,587 of 36,117) against 2.37 percent (949 of 40,085). That is the single comparison worth carrying out of this section, and it runs against the usual framing of who is under pressure, since public four-years are simultaneously the biggest net adders in the file at +652 programs.

Rolled up to control of institution: public 2.95 percent (1,986 of 67,386), private for-profit 3.74 percent (219 of 5,851), private nonprofit 4.37 percent (1,599 of 36,553). If you need to see which institutions sit in each of those buckets before you build a comparison, the institutions directory filters by control and by two-year against four-year.

The sector table below uses the same two rules and the same split denominators as the field table: ended rates over the 2022-23 active count in the third column, new rates over each sector's 2024-25 active count, which the table does not show.

Geography adds another factor of three. Among the 46 states and territories with at least 500 active programs, rates run from 1.72 percent in Utah (19 of 1,104) to 5.60 percent in Montana (32 of 571). Small-state rates move on a handful of programs, so the large states are the more useful comparison: California 2.71 percent (307 of 11,310), New York 2.78 percent (181 of 6,515), Texas 2.99 percent (186 of 6,214), Florida 3.06 percent (127 of 4,149), Illinois 3.43 percent (146 of 4,254), Ohio 4.63 percent (213 of 4,603).

Pennsylvania is the outlier worth a second look: 291 of 5,898 programs ended, 4.93 percent, the highest rate of any state with more than 5,000 programs. A private-nonprofit-heavy state landing near the private nonprofit four-year sector rate is roughly what the table above would lead you to expect, which is the useful part: read any state figure with that state's sector mix sitting next to it. If you are constructing the comparison group properly, how to build a peer group walks through the selection logic.

Nine base years, one number each. The eight non-2019 base years run between 3.46 and 3.71 percent, with a mean of 3.58 percent. Program discontinuation looks like a constant background rate rather than a recent shock. The demographic cliff does not show up as a break in this series. Neither does the pandemic, nor the run of consolidation announcements.

The one exception is base year 2018-19 at 4.48 percent, with 51,413 completions lost against roughly 22,000 in neighbouring years. That is the CIP 2020 taxonomy revision, not a year of mass closures, and anyone who presents this series without flagging it has invented a trend out of a coding change.

This is the question that decides whether the rest of the post is worth anything, so it gets its own test rather than a hedge. Across adjacent year pairs from 2016 to 2025, every code-level boundary moves fewer than 400 completions except one. Between data year 2019 and 2020 the file retires 5 CIP-4 codes carrying 11,389 completions and introduces 33 carrying 22,945. The count of distinct CIP-4 codes reporting completions goes 367 in 2016, still 367 in 2019, then 395 in 2020 and 401 by 2025.

One boundary moves real volume, and it is not the one this post is built on. The 2022-23 base year sits inside a clean stretch of the taxonomy, which is why the headline analysis uses it.

Up close, the 2019 to 2020 boundary shows entire fields vanishing without a single program closing.

CIP 51.24 Veterinary Medicine ends at 3,297 completions in 2018-19 and has no rows afterwards. CIP 01.80 begins at 3,367 in 2019-20. CIP 51.25 ends at 782 and CIP 01.81 begins at 776. CIP 51.19 Osteopathic Medicine ends at 6,700 and has no rows after, while CIP 51.12 Medicine jumps from 19,720 to 27,564, an increase of 7,844. CIP 51.08 drops from 99,679 completions to 77,649 while the brand-new CIP 01.83 Veterinary Technology appears at 8,295. Nobody closed veterinary medicine in 2019.

Every row above would be scored as a discontinuation by a rule that only checks whether a code kept reporting. That is why the base year matters, and why the within-institution check from the caveats section matters too: 24.9 percent of discontinued programs are at an institution that added a new CIP-4 in the same two-digit family within two years, against a 21.6 percent baseline for survivors. The excess is 3.3 points, roughly 125 of the 3,804.

There is a broader version of the same point. 74.9 percent of discontinued programs (2,850 of 3,804) are at an institution that still awards credentials somewhere in the same two-digit family afterwards. Three quarters of the time the college did not exit the field, it exited one code inside it. Whether that reads as a closure or a reorganisation depends on a catalog you have and the federal file does not.

Programs that would go on to disappear were already distinguishable two years out, and the sharpest signal has nothing to do with trend. Only 50.1 percent of them had reported completions in both of the two prior years, against 88.9 percent of programs that survived. Half of the doomed group had already skipped a reporting year. What that looks like in a record is a flicker: two credentials, then none, then one.

The slope is there too. Among dying programs that reported in 2020-21, 42.1 percent had at least halved by 2022-23, against 12.8 percent of survivors. And the level gap is already open: a median of 3 completions in 2020-21 for the programs that would disappear, against 18 for the ones that would not.

If you are building a watch list, the intermittency test is the cheaper one and it separates the groups more sharply than the halving test does.

The OBBBA earnings test asks whether the median earnings of a program's graduates, measured four years after they left, clear a benchmark built from what people with a high school diploma or a same-field bachelor's degree earn. The statutory math is in the OBBBA workforce Pell earnings premium explainer, and that is as much background as this section needs.

The full field and credential breakdown of who fails that test is in 1,220 programs fail the OBBBA earnings test, and the field rates in the table below come from the same query set.

The question here is empirical: do the fields colleges are actually retiring line up with the fields the federal test flags? At the field level they do not line up at all. Take the 28 CIP-2 families that carry at least 200 active programs and at least 200 evaluated programs, so that both rates rest on a workable base. The Pearson correlation between discontinuation rate and earnings-test failure rate across those 28 is r = 0.087, and the Spearman rank correlation is rho = -0.002. Neither the levels nor the orderings have anything to say about each other.

Two cases make it concrete. Visual and Performing Arts has the worst earnings-test failure rate in the country, 16.96 percent of evaluated programs, and a middling discontinuation rate of 3.50 percent. Engineering Technologies has the fifth-highest discontinuation rate at 6.14 percent and a 0.27 percent failure rate.

Three boundaries before anyone quotes the null result. First, the denominator: the OBBBA earnings test resolves for 44,052 of 209,321 PPD programs, 21.0 percent, and 1,220 of those fail, a 2.77 percent failure rate among evaluated programs. A null earnings flag means not evaluated, not passed, and every field rate above is computed on the evaluated subset only. Among fields with enough evaluated programs to rank, the lowest failure rates are Engineering at 0.00 percent (0 of 2,163), Physical Sciences at 0.18 percent, Mathematics and Statistics at 0.23 percent, Engineering Technologies at 0.27 percent, and Architecture at 0.34 percent. Several smaller families also read 0.00 percent, on evaluated counts small enough that the zero is a sample size rather than a result.

Second, vintage. Completions here run through the 2024-25 award year, while PPD earnings pool the 2017-18 and 2018-19 completer cohorts measured four years after exit. The comparison sets a recent closure decision against an older earnings signal. Which is why a correlation of 0.087 is not evidence that colleges ignore earnings. It says only that two field-level rates, built on different populations and different vintages, do not move together, and no field-level rate can reach down to an individual college deciding about an individual program.

Third, and this is the practical one: PPD cannot identify a discontinued program at all. Its exists_in_2024_or_2025 flag is true for all 209,321 rows, 100.00 percent, because PPD:2026 only publishes programs that existed in 2024 or 2025. The flag is a universe filter, not a discriminator. If you want to know which programs went dark, the IPEDS completions file is the only source that can tell you.

Every completions figure comes from IPEDS Completions, part A, first majors only, data years 2012 through 2025, queried through Clema's public data layer. Data year 2025 is file C2025_A with 313,566 rows; 2024 is C2024_A with 307,707; the base year 2023 is C2023_A_RV with 303,460. The IPEDS data dictionary defines CTOTALT as awards and degrees conferred between July 1 and June 30, so data year 2025 is the 2024-25 award year and the base year is 2022-23. Institution characteristics come from the IPEDS HD directory, data year 2023. Earnings figures come from PPD:2026, release year 2026, published January 2026. The full query log and column definitions are in the evidence files linked under Sources.

Six limits, and each one changes what the number is allowed to say.

Analysis of IPEDS Completions, part A, data years 2012 through 2025, and the IPEDS HD institutional directory, data year 2023, via the IPEDS Data Center. Variable definitions from the IPEDS data dictionaries. Taxonomy background from the Classification of Instructional Programs, including the CIP 2020 revision that produces the 2019 base-year spike. Earnings-test figures from the U.S. Department of Education's Program Performance Data (PPD:2026) release, program outcome file, release year 2026. Full reproducible query log: docs/data-evidence/which-programs-colleges-are-quietly-closing/queries.md, with table and column definitions in schema.md alongside it.