This is the condensed version of the deep dive of the same name. Every key number has been checked back against primary sources; for the full argument, the evidence grade on each conclusion, and the citations, read the deep dive. Figures as of July 2026.
It gives no ranking of majors, and it will not tell you "what will be hot over the next decade."
That is not modesty. The single hardest piece of evidence in this issue closes off that road directly.
The US Bureau of Labor Statistics (BLS) publishes ten-year employment projections by occupation every two years, and it is the only producer anywhere in the world that regularly grades itself. On its own published scorecard:
One table says it best of all: of the occupations BLS placed in its "fastest-growing fifth," about 48% really did land in the actual fastest fifth (random would be 20%) — but about 19% fell to the actual slowest end.
The right reading is not "official projections are guesswork" — they are clearly better than guessing, and separate independent research finds that the occupations officials viewed most favorably over past decades did go on to grow much more than the least favored ones. The right reading is: what you get is a coarse signal roughly twice as good as random, not a list you can pick a major from. An occupation on the "fastest-growing" list has about a one-in-five chance of ending up at the bottom ten years later.
And three qualifiers must be added:
One example that stings: after digital cameras arrived, BLS had already lowered its projection for photographic-processing occupations on its own judgment, expecting a decline of a bit over 20% in ten years. The actual decline was 66%. Even in the most favorable case — a technological path already crystal clear and already visible in the data — the officials still understated the fall by roughly a factor of three.
So the conclusion is a gradient: the coarser the level, the more credible; the finer the level, the less credible — and the decision you have to make sits at the very finest end.
This article therefore does something else. It takes apart, one by one, the numbers most often lifted out of popular narratives and into admissions advice; it strictly separates structural changes that have already happened from things that are merely projected; and it ends with a list organized by criterion, plus ten claims you can test yourself.
This is the hardest-evidence section in the article, and the only one you can verify yourself in a single afternoon. It is not a correlation; it is an identity, true by definition.
In the BLS table, the column that people call the "job gap" is headed, word for word, "2024–34 annual average." It is the average annual turnover of people over ten years — not a count of XX0,000 seats actually sitting empty at any moment, and not a ten-year cumulative total either.
Across all occupations, the annual average "gap" is about 18.86 million. Where does that come from?
openings = people leaving the occupation each year + ten-year net job growth ÷ 10
Plugging in: 18.34 million + 0.52 million = 18.86 million.
In other words, 97% of the "gap" comes from people leaving, and only 3% from jobs actually being added.
And this number does not include people who change employers within the same occupation. It is a completely different thing from the "job openings" count on a hiring site — BLS writes that there is "no way" to convert one into the other for comparison.
Within those 18.34 million departures, retirements and exits from the labor force are only 44%; the other 56% are people who moved to a different occupation. Most of the so-called gap is not veterans clearing a seat; it is peers walking off into another line of work. BLS itself writes that for shrinking occupations, "not all workers who separate need to be replaced."
There is an extremely useful criterion here: if an occupation's annual departures exceed its annual openings, that occupation's total employment is necessarily shrinking over the decade. This holds across all 832 detailed occupations, with zero exceptions.
Ranked by annual average openings, in the national top 15, 13 have median annual wages below the all-occupations median, and 6 are actually shrinking over the decade (retail salespersons, cashiers, waiters and waitresses, customer service representatives, general office clerks, and secretaries and administrative assistants).
BLS changed its method once, in 2016. The old table was labeled "ten-year total"; the new one is labeled "annual average." If you simply put the two numbers side by side, you would read it as "the gap was cut by 59%" — when nothing happened at all; one is a ten-year total and the other an annual average. BLS states explicitly that the two editions "should not be compared."
One honest note: the party that judged the old method to have "statistical and conceptual issues" is the very BLS that carried out the change — it is both the executor and the only judge. We found no independent public challenge to the new method, but "not found" is not the same as "does not exist."
Group all occupations by "the education typically needed to enter": of annual average openings, the high school tier accounts for 36% and the no-credential-required tier 33%, while the bachelor's, master's, and doctoral tiers together account for only 21% — even though those three tiers make up 30% of current employment and 58% of the ten-year net job growth.
Why so few openings? Because people at degree levels leave less often. The doctoral tier's annual separation rate is under 5%, while the no-credential tier is close to 16%, in a monotonic gradient.
This cuts both ways, and both must be said: degree-level jobs are more stable, which is good news for people already inside; but the aperture they open outward each year is also narrower, which is not necessarily good news for the new graduate trying to squeeze in.
"Computer science is already saturated" is the most widely circulated narrative of the past three years. Reconciled cell by cell, the result is: the numbers are broadly true, but every one of them is weaker, shakier, and more dependent on definitions than the retellings suggest.
The New York Fed publishes an annual table covering 73 majors, with unemployment rates, "underemployment rates," and median wages. "Underemployment" means: you found a job, but it is a job that does not require a college degree at all.
One discipline for reading the table first: the unemployment rate here covers everyone aged 22–27 holding a bachelor's degree or higher (master's and doctoral holders included), while the median wage covers only full-time workers whose highest degree is a bachelor's — two different populations, which cannot be hung on the same hook.
A few key cells (unemployment rate / share working in jobs that don't require a degree): computer science 7.0% / 19.1%; computer engineering 7.8% / 15.8%; nursing 2.1% / 12.8%; elementary education 1.2% / 16.2%; biology 4.3% / 51.1%; psychology 5.0% / 48.3%; fine arts 7.7% / 58.9%; all majors 4.2% / 39.4%.
A few points worth calling out:
Lining up four editions of this table (2021→2024 data years): CS unemployment rose from 4.3% to 7.0%, up 2.7 percentage points in two years; the starting wage rose from $78,000 to $87,000.
Three things must be said together:
This has to be spelled out, because it determines how far all the numbers above can be used.
The New York Fed has never published, in any public channel, the sample sizes behind this table or any error bounds on the numbers themselves. The four data files have no such columns; full-text searches of the interactive page and its Q&A return zero hits. A harder point: the New York Fed's own methodological work, when analyzing by field, uses 13 broad major categories and three years of pooled data, whereas this web table expands to 73 detailed majors on a single year of data — and that expansion has never been accompanied by any published methodological note.
The only party that has computed error bounds for it is a Washington organization (the Economic Innovation Group). Its results: computer engineering's true unemployment rate could lie anywhere between 4% and 11%; physics between 2% and 12%; public policy between 0% and 13%. In the author's own words, in most cases these intervals are so wide that it "makes no sense" to draw firm conclusions about the returns to particular majors from them, "let alone use them to make policy decisions."
The interest here must be flagged too: this is not a neutral statistical agency but a policy-advocacy organization, founded with funding from tech investors and a long-standing advocate of expanding high-skilled immigration. The conclusion "the CS job market has not collapsed" points the same way as its position. That does not invalidate its statistical work (the method is legitimate and the code is public), but it has to be said when citing it.
There is another way to feel how large the noise is: between 2023 and 2024, the median change in unemployment rate across the 73 majors was 1.25 percentage points, and 20 majors moved by more than 2 percentage points. Early childhood education jumped from 1.3% to 6.6%, nutrition sciences from 0.4% to 4.5%, and foreign languages went the other way, from 4.0% down to 1.6%. CS, by comparison, moved only 0.9 percentage points.
So "consistently high CS unemployment reproduces reliably" must be downgraded one notch: CS's current elevation rests on only two consecutive readings; computer engineering's elevation was created outright by a single-year jump (the prior year read 2.3%), and its two readings sit inside the same error interval, so statistically "reproduction" adds no information at all.
Guard against a double standard here: if you invoke noise to explain art history's one-year blowup, you have to hold CS's two-year elevation to the same yardstick.
The 2025 business-media story that set off the discussion carries a correction notice at the end: the chart data were wrong and were corrected afterward — and what went viral was chiefly that very chart.
The real definitional mishap also has to be described precisely: the story did state the source and year for its by-major data; the problem was another number on the same page, drawn from an entirely different monthly survey, seasonally adjusted, with a single month as its reference period. The mishap is "two surveys placed side by side with only one of them labeled," not "the year wasn't stated." The two really cannot sit side by side: for the same calendar year, the monthly survey's recent-graduate unemployment rate runs 0.74 percentage points higher than the community survey's.
There is also a report widely treated as "a second institution confirming it" (from Georgetown University), giving CS recent-graduate unemployment of 7.2%. It uses the same underlying data as the New York Fed, differing only in age window and pooling method, so it is not an independent measurement. All you can say is "different cuts of the same data source point the same way, which lowers the chance this is single-year sampling noise" — not "two institutions corroborate each other." (That institution's position is explicitly pro-degree, and its funders have raising degree attainment as their mission.)
The conclusion first: over the past three years, hiring for some white-collar entry-level jobs in the US really has cooled markedly, and multiple data sources show it. But on the three questions "was it AI," "how far did it fall," and "will it last until you graduate," no study can currently give a reliable answer. Any statement offering you a specific percentage exceeds what the evidence can support.
One: whether AI can do it in theory. The representative work is the paper by the OpenAI team. It measures how many tasks could, in theory, have their time halved at equal quality using a large model. The authors write in their own abstract: "We do not make predictions about the development or adoption timeline of such LLMs." The body adds that technical feasibility does not guarantee productivity or automation outcomes. And the rating quality itself has hard flaws: human–model agreement runs only 60% to 80%; the model labeled 86 occupations fully exposed where humans labeled only 15; and the raters were company insiders and outsourced annotators — not one was a practitioner in the corresponding occupation.
Two: vendors' product usage logs. The unit of observation is the conversation, not the person; the publishers themselves concede that this does not represent AI use across society, and that they cannot tell whether a user was doing work.
Three: measured employment. Population surveys, payroll records, administrative data, job postings — each with its own blind spots.
There is also a citation-chain problem: many studies that appear to "corroborate each other" actually share the same independent variable (the same machine-scored AI exposure measure), not the data and not the conclusion. All they can corroborate is "after grouping by this measure, employment diverges" — not "this measure really quantifies AI's impact." The most ironic link in the chain: two downstream studies use precisely the machine annotations that the original paper explicitly says are not its primary results ("we present results from human annotators as our primary results").
The interests here run counterintuitively, and must be written out separately: at the high-exposure end stand three OpenAI authors; on the "AI usage is rising" side stands a vendor using its own logs and its own classifier; while the most systematic rebuttal arguing that "AI is not killing entry-level jobs" was written by Google's chief economist and his team's AI-and-the-economy lead, with an acknowledgments list including Alphabet's president and chief investment officer, Alphabet's general counsel, and a Google senior vice president. → The intuition "vendors = exaggerating AI's impact" does not hold; the deepest vendor fingerprints are precisely on the "AI has no effect" side.
It is not that experts are fighting; it is that two lenses are pointed at different things.
The low-power lens (the Yale Budget Lab, using the monthly population survey to look at economy-wide occupational composition) asks: has the distribution of the occupations Americans work in been rearranged? The answer so far is no, the pace of change is still within the range seen when computers and the internet spread, and the change began before AI entered the workplace at scale. The New York Fed, using entirely different data (job postings), independently arrived at the same sentence.
But this lens cannot see a change within an occupation where "the incumbents are still there and newcomers can't get in" — as long as the occupational share is unchanged, it reads flat. That is not an outsider's nitpick; it is their own admission: this survey is underpowered for a small subgroup like "22–27-year-old recent graduates."
The high-power lens (a Danish firm-level study) asks: at the same time, in the same country, did earnings and hours diverge between firms that adopted AI and firms that did not? The answer is no, and it is measured precisely. But by construction this lens can only see differences between firms — if an entire industry cut entry-level hiring in step, it would register as zero. The authors concede this themselves.
That neither lens saw anything does not mean there is nothing. They share one blind spot: the low-power lens cannot see entry flows within an occupation, and the high-power lens cannot see changes that are uniform across an industry. And the shrinking of entry-level opportunity falls exactly inside that shared blind spot.
US Census Bureau researchers measured this using unemployment insurance administrative records: cutting by age, after ChatGPT's release, in high-exposure industry × state cells, new hiring of 22–24-year-olds dropped immediately by about 9%, cumulating over ten quarters into an employment decline of about 12% — and almost entirely through "not hiring" rather than "firing."
The Danish evidence, though, indicates that what happens in that crack is not necessarily AI's doing. The researchers have surveys covering 25,000 workers and 7,000 workplaces, linked to national administrative records, letting them split the trend by whether a firm actually adopted AI. The result: entry-level positions in high-exposure occupations in Denmark are indeed declining, but not at the firms that adopted AI. They computed an upper bound: AI adoption can explain at most about one quarter of the total decline. The remaining three quarters they attribute, speculatively, to the macro environment or to precautionary "freeze hiring first" contraction.
So the correct statement is "AI explains at most a small part of it," not "AI has zero effect on entry-level jobs."
One institutional point has to be told alongside: the three countries where "AI did not cut junior jobs" is corroborated (Denmark, Finland, Norway) are all Nordic, sharing strong unions, high dismissal costs, and centralized wage bargaining; among the countries reporting entry-level declines — the UK, Sweden, the US — US dismissal costs are close to zero. The institutional differences and the data differences overlap completely, and nobody can separate them. So Denmark's "zero" cannot be transplanted to the US, still less to China. All of the studies above are unrefereed.
The most widely circulated quantitative study of "AI is compressing youth employment" was rejected in this issue, and none of its headline figures enter the body text. The weak conclusion that can be retained goes only this far:
In a private dataset covering millions of US payroll records (the client-firm composition is not representative of the US economy), since late 2022, in occupations placed in the highest tier by an LLM-scored "AI exposure" measure, the number of employed 22–25-year-olds declined relative to low-exposure occupations at the same firms, while no comparable decline appeared above age 30. This is a descriptive fact, not a causal conclusion.
Moreover, the same authors have publicly acknowledged that once the fullest macro controls are added, this decline becomes significant only after 2024; and when the sample is extended back to 2018, the occupations flagged "most exposed" by this measure were already growing more slowly as of 2020 — before generative AI existed. The current statistical design cannot distinguish "AI substituted for entry-level jobs" from "firms froze hiring in these occupations, and a four-year-wide age band of 22–25-year-olds naturally drains under a hiring freeze." It remains an unrefereed working paper; the data are proprietary and cannot be replicated externally. The data provider is a paying corporate partner of the lab in question, and the paper has no conflict-of-interest statement.
Independent comparisons give a very wide spread of magnitudes: the Dallas Fed, using the monthly population survey, sees only the employment share of 20–24-year-olds in the highest-exposure occupations falling from 16.4% to 15.5% (about one third the magnitude of the former), and states explicitly that the data "do not show AI causing widespread labor market disruption"; the Yale Budget Lab and another institution, using different exposure measures, both find no relationship between exposure and employment; two institutions each used job-posting data to test directly whether, within the same high-exposure occupation, junior roles fell more than senior ones — and both independently found that they did not; one study finds that deterioration in high-exposure occupations began as early as early 2022, months before ChatGPT; and another New York Fed study attributes 64% of the rise in young college graduates' unemployment to remote work.
One unrefereed preprint uses roughly 100,000 consumer chat records to measure "what share of an occupation's work activities show up in chat logs." On that basis, education occupations overlap markedly more than healthcare-support occupations.
But this direction is to a large extent determined by the instrument: explaining, teaching, and drafting materials are the default output form of a chatbot to begin with, while turning a patient, transferring, drawing blood, and monitoring vital signs cannot in principle appear in a text conversation. The paper states explicitly that it does not know the user's own occupation; its pipeline even flags "is this user a student doing homework," yet it never reports that proportion and runs no sensitivity analysis excluding homework conversations — and "postsecondary teacher" is precisely the cell most easily inflated by student homework. So what it measures is topic overlap, not capability, and certainly not substitution risk. The authors warn verbatim that reading a high score as "will be automated, will lose jobs, or will lose wages" — "This would be a mistake." (All five authors are from that company's research institute, the data are its own logs, the classifier is its own model, ethics review was internal, the three human annotators used for validation were the paper's own authors, the raw conversations cannot be released, and not a single cell can be reproduced externally.)
So this article keeps from it only one mechanistic judgment, and no numbers:
The part of teaching work that can be rendered into text (lesson preparation, explanation, item writing, grading, drafting materials, answering questions) overlaps heavily with current chatbot capabilities; the part of nursing work that can be rendered into text is far smaller. So the blanket claim that "nursing and education are both in the automation-resistant camp" does not stand — but that does not mean the employment outlook for education majors will worsen, which is a separate question answerable only with measured employment and wage data.
And a counter-piece must be written down with it: the International Labour Organization's working paper (based on a sample of nearly 30,000 occupational tasks, a survey of 1,640 incumbents, and an expert panel, using no AI product logs at all) places primary school teachers, secondary school teachers, nursing professionals, and general practitioners in the same "unexposed" category. "Education is close to knowledge work" does not hold on that basis either.
Last and most important: no per-occupation "AI exposure ranking" is trustworthy. A third-party audit shows that the rank correlation of occupational orderings across platforms is only 0.43 to 0.79 (and that at the coarse level of just 22 major groups); and within one vendor's own data, 9 of 22 major occupational groups crossed a quintile boundary over 14 months. Switch vendors' logs, or even switch quarters, and the ordering changes.
Where this section lands: once more, because it is the easiest to quote out of context — the cooling in hiring is real; the cause, the magnitude, and the duration are all unknown. This article does not use this section to argue anyone into or out of any particular major.
"Students chase the current employment signal, and four years later graduate together into an oversupply" — this is not a theoretical deduction; it is history you can count year by year.
Per the American Bar Association's mandatory-disclosure data: of the 46,776 law school graduates in the class of 2013, only 57% held full-time, long-term, bar-passage-required positions about nine months after graduation. Excluding positions funded by the law schools themselves, 55%. A completely independent institution (the National Association for Law Placement), using its own survey system, measured that class's figure as the lowest in its records. Two questionnaires, two institutions, two denominators, pointing to the same trough and the same turning-point year.
By the class of 2025, the same measure had returned to 83%. But two-thirds of that recovery arc comes from the denominator, not the numerator: positions rose only from 26,653 to 29,928 (+12%), while graduates fell from 46,776 to 36,206 (−23%). Had the class of 2025 been as large as the class of 2013, the ratio would be 64%, not 83%. The repair came mainly not from the industry taking more people but from the schools sending fewer.
Several qualifiers that must be carried:
The current situation is a divergence: the supply side is re-expanding (42,817 first-year students in fall 2025, the largest class since 2012) while the demand side is softening (large firms are shrinking summer associate recruiting, the median number of offers for 2026 summer 2L programs is the lowest on record, and the total number of entry-level legal jobs is still about 5% below 2007). That is exactly the precondition of the last mismatch. Close this passage with "early warning," not "confirmation."
There is also a popular claim that needs correcting: the widely circulated "law school applicants +33%" in 2026 is a reading from roughly a 15% sample early in the season, and the publisher's own labels are "extremely early data" and "broadly directional at best"; by mid-June 2026 the year-over-year increase was only about +8.5%. Citing 33% without citing 8.5% is selective quotation.
And the publisher's own list of drivers contains no "rising demand for legal jobs" at all; in its survey of test takers, the top three motivations are "to help others," "to advocate for social justice," and "to gain valuable skills."
Where this lands for prospective students: in uncertain years, law school becomes a safe harbor, and a safe-harbor influx is not automatically corrected by wage signals — which makes it more worth watching than the "chase the rally" model, not less.
Per the National Center for Education Statistics' long series (bachelor's degrees conferred in computer and information sciences):
A key causal reading here: the dot-com bubble burst in March 2000, yet the degree peak came in 2003-04, four years later. That is not slow reaction; it is the four-year pipeline delay of a degree — the people graduating in 2003-04 chose their field in 1999–2000, at or before the top of the bubble. The actual behavioral response only shows up in the data from 2004-05, and the 36% decline starts exactly there. This strengthens the "chase the rally" argument rather than weakening it.
In China, the number of institutions offering "data science and big data technology" fell from a peak of about 250 schools in the 2017 approval year to somewhere in the teens in the 2024 approval year — a decay of over 90% in eight years, with the peak arriving in the second year after the program was created. (The year-by-year figures may not be quoted verbatim; the media-compiled series uses inconsistent bases from year to year. But the broad conclusion — peak, final year, and magnitude of decay — is solid.)
For this article's inference rules, this curve is a counterexample, not supporting evidence. Big-data program instances peaked around 2017, yet demand for data jobs went on growing strongly for at least five more years. A leading indicator issuing a false alarm five years early is the norm for administrative counts of this kind, not the exception. And after the decay it stabilized at a floor of about 30 schools a year rather than going to zero, with a cumulative total of about 790 institutions by 2023 — roughly 60% of all undergraduate institutions in the country. The rhetoric of "a 90% collapse" conceals that.
This is something the article must state honestly: the shape "overheating → collapse → contraction → repair" has historical evidence; its parameters (how many years per cycle, how strong the reaction) do not.
The engineering-academy report appendix widely cited as the basis for a "four-year adjustment cycle in the engineer market" is only two pages long, contains a single figure, and has no data table, no regression, no coefficient, and no statistical estimate of any kind; its wording is that the market is "likely" to stay at point b, and the figure's own note says it is adapted from a popular-economics blog.
As for how eagerly students actually chase the rally: one US study finds that degree completions in a field are most strongly related to "wages three years earlier" (that is, the wages students saw as freshmen) — but the paper reports no standard errors, and wages three years earlier explain only about 2% of the variation in degree output. A French study simulating a 10% rise in expected earnings also finds a very small response, which the authors characterize as "very low, which means that the choice of a major is mainly driven by non-pecuniary factors."
Conclusion: students do respond to wage signals, but weakly, slowly, and mainly under the influence of non-pecuniary factors. That weakens both the panic that "everyone will pile in" and the optimism that "the market will correct itself."
The popular criterion is: "this trade requires a license → supply is limited → no glut." That criterion does not hold, and there are three counterexamples:
The correct criterion is:
What determines overheating is not "whether practice requires a license" but "whether training capacity is physically or human-resource constrained," and "how high the marginal cost of opening one more program instance is." A license that binds after graduation (the bar exam, the pharmacist exam) imposes no constraint on admissions at all; only a constraint at the entrance restrains overheating (nursing: clinical placement slots, preceptors, and faculty are all limited at once). And accrediting bodies themselves loosen procyclically during a boom.
At the same time, "overheating is almost certain to be restrained" must be downgraded to: constrained capacity can only lengthen the lag and flatten the amplitude; it cannot guarantee that overheating does not occur.
This criterion is a product of this article's reasoning, not any institution's conclusion — labeled as such.
The supporting data is the "applications turned away" series that the American Association of Colleges of Nursing has published for three consecutive years: 65,766 in 2023 → 80,162 in 2024 → 92,672 in 2025, with the self-stated reasons being insufficient faculty, clinical sites, classroom space, preceptors, and budget.
Three things must travel with that number:
The interests must appear with the numbers: the association is the trade association of baccalaureate-and-higher nursing schools, and its fact sheet openly states that it is seeking federal appropriations for faculty development programs and has long lobbied on related legislation. The larger the number of applications turned away (the numerator), the better for its appropriations case — and the fact that it chooses to publish "applications" rather than "applicants" must itself be disclosed alongside the number.
This is the section with the strongest academic evidence in the article — three Quarterly Journal of Economics papers, all published and peer reviewed. But the limits on cross-national extrapolation are severe.
Institution effects: early research found that students who attended more selective institutions earned no more than students admitted to comparable institutions who chose a less selective one. When the updated version re-estimated using Social Security administrative records, even "higher tuition means higher returns" went to zero along with it.
But the upper tail is the exception: a study published in February 2026 finds that attending the Ivy League plus Stanford, MIT, Duke, and Chicago — those 12 institutions — rather than an "average flagship public university" raises a student's probability of reaching the top 1% of income by 50%, nearly doubles the probability of attending an elite graduate school, and nearly triples the probability of working at a prestigious firm. But the null results in that same paper matter just as much: the effect on the probability of reaching the top 25% is "small and statistically insignificant," and the effect on earnings themselves is modest. The mean figure — $100,000 higher average earnings at age 33 — is introduced in the original by the phrase "as a result of these upper-tail impacts": the mean is pulled up by the tail.
Field effects: a Norwegian study using nearly the complete records of centralized admissions, identified off admission-cutoff discontinuities, writes in its abstract that "different fields have widely different payoffs… for many fields the payoffs rival the college wage premiums."
But the second half, which is usually dropped, must be written with it: "by choosing Science instead of Humanities, individuals almost triple their earnings early in their working career. By comparison, choosing Science instead of Engineering or Business has little payoff." — That is, the conclusion that "the field matters" is carried mainly by the "science vs. humanities" end; among science, engineering, and business there is almost no difference.
Cross-national warning: Norway has centralized admissions, free public tuition, and a markedly compressed wage distribution, so between-field income gaps are compressed to begin with. Transplanting that conclusion directly to China or the US is dangerous.
Within-field gaps are no smaller than between-field gaps: research finds that within a single field, moving from the bottom quarter of earners to the top quarter typically doubles or even triples lifetime cumulative earnings — and that span is wider for lower-earning fields. The same report also says: at the 10th percentile, the difference in lifetime earnings across fields is about $500,000; at the 90th percentile, that difference is over $3.5 million. The two sentences are entirely compatible (different cross-sections of the same distribution), but quoting only the first yields "the major doesn't matter," and quoting only the second yields "the major decides everything."
Decay is a property of occupations, not of majors: a 2020 Quarterly Journal of Economics study finds that computer science and engineering majors earn about 45% more early in their careers but only 33% more by age 50; business goes from 38% to 20%; while the premium for life sciences, physical sciences, and social sciences grows over time. The mechanism the authors give is the most useful sentence in this section: "Declining relative returns is a feature of STEM jobs, not majors." The share of computer science and engineering majors employed in STEM occupations falls from 59% at age 26 to 41% at 50, and those 18 percentage points are almost entirely offset by movement into non-STEM managerial occupations.
→ "Choose this major and you do this for life" — the data itself does not support that premise.
"Working in your field" is three different questions, with different numerators and denominators, and any arithmetic across them is wrong: ① respondents' self-assessment of "is my job related to my highest degree" (among those whose highest degree is a bachelor's, about 25% say "not related at all"); ② analysts' cross-coded determination of "does the job directly match the undergraduate field" (about 27% directly matched; computer science 33%); ③ "is the person employed in a STEM occupation" (engineering 52%, computer/math/statistics 51%, psychology 10%).
And one very common misreading: in that urban wage study, "the job requires a bachelor's degree" corresponds to about +25% in wages, and "the job matches the field" corresponds to a further +5% on top of that — they are two nested thresholds, and the ratio of coefficients is not a ratio of importance. Also, neither coefficient is a causal estimate.
This is the only honest landing point for the cross-national comparison in this article.
The accurate statement is: China has no publicly released, nationwide, comparable, continuous numerical employment indicator by field of study — no unemployment rate by field, no starting wage by field, no education-mismatch rate by field.
But "China has no official employment data by field of study" is too strong a statement. Employment data by field do exist inside the system, are collected, and are being used for administrative action (red and yellow cards, orders to suspend admissions, program abolition). The authorities have also publicly released ranked outputs — what gets published is name lists, not numbers:
At the provincial and institutional levels, destination rates by field really are published: Jiangxi issues a yellow card when the destination rate falls below 50% and a red card, with an order to suspend admissions, when it falls below 50% for two consecutive years; Sichuan sets yellow/red cards at two/three consecutive years below 50%.
The hardest piece of positive evidence sits in the Ministry of Education's own document: on page 29 of the Statistical Indicator System for Monitoring and Evaluating Chinese Education (2025 Edition), the indicator "graduate destination rate" is broken down verbatim as "by level; by form of employment; by type of provider" — there is no "by field of study." Across all 108 pages, the finest disciplinary dimension of any indicator is the broad-category level. On the same page the officials state that this indicator "cannot fully reflect the actual person–post match of graduates" — the best first-hand basis for the fact that the authorities acknowledge having no "working in a job that doesn't need the degree" indicator.
→ So the correct landing point is: China has data by field but does not publish the values; the US by-field data come from the Census Bureau's community survey, which anyone can download and recompute, and the New York Fed's table is a derived product. The asymmetry is in reproducibility, not existence.
And this article never puts the Chinese and American numbers in the same table, on the same axis, or in the same sentence. The reason is that the constructs are entirely different. China's "graduate destination rate" (毕业生毕业去向落实率) has total graduates as its denominator, and its numerator includes those going on to further study, those in flexible employment, freelancers, and the self-employed. The US unemployment rate has the labor force as its denominator, and "working in a job that doesn't require a degree" has employed graduates as its denominator. Those are three different things. Setting them side by side is the usage in this whole article most likely to cause real harm — a prospective student will read it directly as "this Chinese major has better job prospects than that American one."
Per the 2021 edition of the Definitions and Standards for Graduate Destinations (the operational document that decides what counts as employment is a disclosed-upon-request document, not in the Ministry of Education's proactively disclosed catalogue):
Here is the load-bearing point: in the broadest tier of the official destination rate, the supporting materials are provided by the graduate themselves and adjudicated by their own university's employment office — the auditor and the audited belong to the same organization.
The same document also states: "Before publicly releasing their province's graduate destination rate, provincial employment authorities must verify the data with the Ministry of Education's Department of College Student Affairs, and may not publish without such verification." — That data-release control clause was written by the party being measured.
And the document's "three prohibitions" (do not set unrealistic targets, do not pile on quotas level by level, do not tie the destination rate alone to counselors' performance pay) are themselves official evidence that quota-pushing exists.
To be clear: "loose recognition" is an implication of the text, not an official admission — the document's tone throughout is one of strengthening statistical discipline. The two must be stated separately. Also, a new classification system took effect in 2024 and the thresholds have been tightened at the operational level (employment through other forms of hiring now requires a term of at least 6 months, and e-commerce entrepreneurship requires transaction records).
The most widely circulated calculation divides the official flexible-employment rate by a figure from a commercial survey and yields a "three- to fourfold discrepancy." That calculation is wrong, and this article does not use it. The official "flexible employment" and the commercial survey's "flexible employment" are two sets that barely overlap (the authorities count entrepreneurship in the entrepreneurship rate and not in flexible employment, while the commercial survey folds it in; the authorities count people with no contract but with wage records as flexible employment, while the commercial survey classes them as "employed full-time"). The only intersection is "freelance work," and the authorities never publish a standalone figure for that category.
The substitute conclusion that has an independent measurement and does bear weight is this: for the same graduating class (2021) and the same all-levels basis, official administrative data report flexible employment at about 16%, while an independent national sample survey by the research group at Peking University's Institute of Economics of Education (34 institutions, 20,269 responses) measured 11.4% — the official figure is about 4.9 percentage points higher, roughly 40% higher in relative terms.
That direction is consistent with the Ministry of Education's own 2023 statement that it would focus on "verifying flexible employment and related data." Conclusion: flexible employment really is the cell in the administrative basis most prone to inflation, but the magnitude of that inflation is around 40%, not three- to fourfold.
On falsification, both the official statements and the official silences are checkable: an August 2023 Ministry of Education release states that "fake signed contracts and fake certificates" will be investigated with "zero tolerance." But exhaustive searches for "notices of disciplinary outcomes" and "numbers of confirmed cases" return nothing but reprints of that same release — no named notice, no confirmed case count, no disciplinary outcome announcement of any kind. → What can be written is "falsification is acknowledged; no public report of how much was found."
One more thing: an official independent measurement capable of checking what universities report does exist — each September the Ministry of Education commissions the National Bureau of Statistics to run a sample survey of graduate employment status. And the system design of that survey hardwires in that results are "fed back by official letter" to provincial education authorities. → The only official independent measurement capable of checking the administrative data is never publicly released.
The old series ends with the June 2023 urban surveyed unemployment rate for ages 16–24 of 21.3% (the highest since the statistic began; publication was suspended a month after release); the new series begins with December 2023's 14.9% for the 16–24 labor force excluding students. The difference between these two numbers must not be read as improvement — they are not comparable.
The key blank is a documented fact: the official explanation of the change contains not one unemployment rate figure anywhere in its full text, and offers no side-by-side old/new figures, no dual computation over an overlap period, and no conversion factor. The only quantitative material it provides is the denominator stock: of the urban population aged 16–24, about 62 million are students and about 34 million are not. → It explained only "who was removed and how many," and never gave the resulting unemployment rate for the same period.
And the year-over-year figure must be given at the same time: June 2026 is 14.9%, while June 2025 on the same basis was 14.5% — 0.4 percentage points higher than a year earlier. "Three consecutive monthly declines" is the seasonal shape of every March-to-June; the year-over-year direction is upward. Giving only month-over-month without year-over-year is materially misleading.
Interests at stake: the statistics bureau is both the setter of the definition and its sole publisher; the definitional change occurred after the old series hit a record high and publication was suspended, and that sequence itself must be disclosed to readers.
One: the supply side in computing was substantially amplified within five years. 955 institutions now offer computer science and technology and 661 offer software engineering (per the Ministry of Education's Sunshine Gaokao platform, as of November 2025), and the two together graduate more than 100,000 students a year; from 2018 to 2022, 1,047 new program instances were added in the computing category, the most of any program category. (A "program instance" — 专业点 — is one major offered at one institution, a count of school–program combinations, not of enrollment.) These come from the Ministry of Education's filing and approval records — administrative statistics, not a sample survey. This is the one part of this section that needs no discounting.
Two: establishment-post channels for replenishing teachers are tightening in some provinces (direction only, no magnitude). ("Establishment posts" — 编制 — are budgeted, permanent public-sector headcount slots.) Hubei's provincially unified recruitment plan has fallen for four consecutive years; Jiangxi's has likewise fallen year by year; and the national rural compulsory-education "Special Post Plan" allocation is also declining year by year. But this speaks to one channel in some provinces, and is not the same as the total number of teaching posts. It excludes channels outside the Special Post Plan, off-establishment contract hires, and private schools; and the category composition of the posts changed between years (the 2026 tables no longer include kindergarten teacher posts or posts reserved for demobilized military personnel, whereas 2025 did) — the bases differ, which is why this article gives no percentage decline. Recruitment plan numbers are also a product of establishment quotas and fiscal budgets: establishment freezes and fiscal pressure could depress them even with school-age population unchanged.
A tense error that must be corrected: you cannot juxtapose the 2026 recruitment cuts with the 2024 decline in primary school enrollment and claim that together they prove "posts are already contracting." The first-graders who enrolled in 2024 will not finish primary school until around 2030; the 2026 cut cannot mechanically have been "already" caused by it — at most it is a forward-looking contraction of establishment quotas.
There is also a transmission-chain problem worth naming: the whole narrative of "recruitment halved in many places" and "hundreds-to-one competition for some subjects" is carried by admissions-consulting social media accounts whose monetization directly benefits from manufacturing anxiety about choosing a major. This article substantiates only the direction, and gives no magnitude.
Three: the IT wage premium is narrowing, but is far from gone. Per the National Bureau of Statistics' 2025 average wages for employees in urban units: information transmission, software and IT services ranks first on the non-private basis, running 92% above the overall average; on the private basis, 79%. → "IT doesn't pay anymore" does not hold on the statistics bureau's basis. But the premium really is narrowing: IT's annual growth (+4.0% private, +4.1% non-private) is below that of both manufacturing and education. One more cell worth knowing: within the IT industry, the non-private basis is 1.94× the private basis, a wider gap than the 1.81× for the economy overall — showing that IT's "high-pay impression" depends heavily on the non-private sample.
Four definitional notes: these are pre-tax total wages (including individually paid taxes and social insurance and housing fund contributions); coverage is limited to legal-person units with 5 or more employees, so self-employed businesses and freelancers are not in the sample — and those two are precisely an important absorber of youth employment; these are means, not medians; and they are averages across all incumbent employees, not new graduates' starting salaries.
One thing readers must know: almost every domestic ranking of employment rates by major uses the "graduate destination rate," and that indicator's numerator counts enrollment in graduate school, at home or abroad, as employed. Which means programs with high further-study rates automatically look like they have "good employment."
And in the most-cited firm's own data, computing is the program category within engineering with the lowest graduate-school rate (13.6% for the class of 2022, against 32.4% for materials and 24.1% for electronics and information in the same year). So subtracting the destination rate of computing from that of history or foreign languages, whose further-study rates are far higher, compares not difficulty of employment but differences in further-study structure. The line circulating widely of late — that computing graduates now have a harder time finding work than humanities graduates — is essentially a product of this definitional mismatch, and this article does not use it.
The firm has to be named, because the institution is part of the story: MyCOS is a commercial survey company formerly listed on the NEEQ ("New Third Board"; delisted in 2020), with data-monitoring services accounting for 85% of its revenue, and its paying customers are precisely the universities it surveys and evaluates; its public materials contain no disclosure of interests. Its technical report gives only a sample size of 135,000 and has never published the total number of invitations issued or the response rate — a numerator with no denominator. On the most damaging bias of all — that people who found jobs are more willing to respond — the report offers only the assertion that "testing found no self-selection sample bias," without specifying the comparison variables, the test method, or the significance level. Not once in the entire volume does an error bound appear, so whether the few percentage points separating a program category's figure from the national average are even distinguishable from the unknown sampling error cannot be judged. The corresponding rankings are likewise not reproducible.
Interestingly, in the same report, computing's monthly income six months after graduation, its employment satisfaction, and its job–field relatedness are all above the undergraduate average, and six of its constituent programs have repeatedly made the "green card" list over the past five years. One indicator says cold, three say hot — which itself shows that no single employment-rate ranking can support a ten-year decision about a field of study.
On its most famous product — the red/yellow/green card lists — the methodological conclusions reached in this issue are more useful than any single year's list:
In fairness, one sentence must also be written: you cannot say "the report was bearish on computing back then and has been proven wrong today." In the same release that assigned the red card, the firm wrote that computing programs "have a large number of unemployed graduates, but demand for talent in the relevant industries is very strong" — the judgment at the time was "a mismatch between supply quality and demand," not "the industry is failing."
First, the nature of this list, stated flatly: this is how this article organizes the evidence, not any institution's conclusion. It is not ordered by hot and cold, it assigns no score to any field, and it does not predict which field will do well in ten years. What it does is apply the conclusions above to three transferable criteria — the criteria matter more than the list, because in four years the list will be stale and the criteria will not.
The dividing line is not "technical vs. non-technical," nor "education vs. nursing."
| Field / direction | Position | Basis |
|---|---|---|
| Nursing, rehabilitation therapy, medical imaging technology, and other hands-on medical care | Low-codability end (robust) | Two vendor log measurements and one independent measurement using no AI product logs at all agree in placing nursing in the low-exposure group; turning a patient, transferring, drawing blood, and monitoring vital signs cannot in principle appear in a text conversation |
| K–12 education | Low end, but the gap from the row above is far smaller than popular claims suggest | The only non-vendor measurement places primary school teachers, secondary school teachers, nursing professionals, and general practitioners in the same "unexposed" category |
| Postsecondary teaching, writing, translation, junior legal drafting, general copywriting | High-codability end | Explaining and drafting materials are the default output form of a chatbot to begin with; three datasets consistently show postsecondary and graduate teaching markedly more exposed than K–12 teaching. No rankable number is given in this column |
| Software engineering / computer science | Internally split; cannot be filed as a whole | In the official ten-year projections, the growth rate for software developers alone is higher than for the group combining quality assurance and testing — the most easily automated testing portion grows more slowly, direct evidence that the split occurs within the field; on the job-postings side, the net increase is concentrated in senior roles and AI-related titles |
The strongest qualifier for this column: no per-occupation "AI exposure" score is load-bearing. Cross-source rank correlations of occupational orderings run only 0.43–0.79, and that at the coarse level of just 22 major groups; within one vendor's own ordering, 9 of 22 major groups crossed a quintile boundary in 14 months.
The question is "what extra inputs does a school need to open a new program instance," not "does this line of work require a license."
| Field / direction | Position | Basis |
|---|---|---|
| Nursing, medicine, and engineering fields requiring labs or accredited beds | Expansion constrained physically and by human resources | Three consecutive years of the trade association's turned-away-applications series, with the stated reasons being faculty, clinical sites, classrooms, preceptors, and budget. But: the publisher is a trade association, and the larger the number turned away the better for its appropriations case; it publishes "applications," not "applicants"; master's and doctoral levels are about a fifth of the total, so using the total to describe the bachelor's-level bottleneck overstates it by about 22% |
| Same as above | But "the bottleneck is worsening" does not hold | The same page states that the 2025 faculty vacancy rate of 7.2% is below the ten-year average of 7.64%. These two numbers must always appear together |
| Law, management, most computing fields, most humanities | Low marginal cost of expansion | Three counterexamples: US law requires an accredited school plus a license and is still the textbook oversupply case; pharmacy had all three thresholds and still nearly doubled and produced a glut; nurse practitioner programs grew several-fold within a decade. On the China side: computing added the most new program instances of any category in 2018–2022 |
| The criterion itself | — | A license that binds after graduation (the bar exam, the pharmacist exam) imposes no constraint on admissions; only a constraint at the entrance restrains overheating. And accrediting bodies loosen procyclically during a boom. This is a product of this article's reasoning, not any institution's conclusion |
The strongest qualifier: constrained capacity can only lengthen the lag and flatten the amplitude; it cannot guarantee that overheating does not occur.
The question is "over the four years between your enrolling and your graduating, how many people will end up with the same label as you."
| Field / direction | Position | Basis |
|---|---|---|
| China: computer science and technology, software engineering | Penetration already extremely high | 955 and 661 institutions offering them; more than 100,000 graduates a year between them; the most new program instances of any category in 2018–2022. Administrative records, not subject to sampling problems — the strongest cell in this section |
| China: artificial intelligence | The label is no longer scarce | As of the 2024 approval cycle, about 626 undergraduate institutions had filed this program, roughly half of the national total. The only robust implication is that the label itself is no longer scarce. It cannot tell you anything about supply and demand for jobs in 2030 |
| US: computer and information sciences | Two complete cycles already run; currently in the third at a high level | Forty-year series: two peak-to-trough declines of −42% and −36%; the third ascent is +186% from the trough; the degree pipeline has a four-year delay |
| US law | The only case to complete all four steps; currently a divergence of "supply re-expanding + demand softening" | The trough was computed independently by two institutions and two questionnaires; two-thirds of the repair came from the denominator; the fall 2025 first-year class is the largest since 2012 |
| China: teacher-training fields (the establishment-post channel) | One channel is tightening, but the magnitude is not credible | Three independent bases point the same way, and a dedicated search found no contrary provincial announcement. No magnitude is given; one province cannot be extrapolated to the country |
| Resource extraction and heavy industry | A sample of historical reversal | The nine programs given green cards and recommended for expanded admissions by a commercial firm in 2010–2011 are today prime "avoid" territory |
| Weaknesses of the criterion itself | — | The shape "overheating → collapse" has historical evidence; its parameters do not: the widely cited "four-year cycle" comes from a two-page appendix containing no statistical estimate of any kind; and students' response to wage signals is weak and slow |
One, where the returns come from. Institution effects fall essentially to zero once self-selection is controlled (the upper tail excepted — elite institutions mainly affect the probability of reaching the top 1% of income, an elite graduate school, or a prestigious firm, while the effect on the probability of reaching the top 25% is "small and statistically insignificant"); between-field gaps are real but carried mainly by the "science vs. humanities" end, with almost no difference among science, engineering, and business; and within a single field, moving from the bottom 25% to the top 25% typically doubles or even triples lifetime earnings, with that span wider for lower-earning fields.
Two, return decay is a property of occupations, not of majors. The share of computer science and engineering majors employed in STEM occupations falls from 59% at age 26 to 41% at 50, and those 18 percentage points are almost entirely offset by movement into non-STEM managerial occupations.
One last emphasis: what hangs in every cell above is evidence about a criterion, not a promise about employment. The correct use of this list is to test the recommendations you read elsewhere — not to take it as a recommendation.
Ordered by strength of evidence. Each item comes with an action you can take yourself.
(1) "How many schools the Ministry of Education newly approved to offer an artificial intelligence program this year."
Many people will tell you to watch this number and see whether it falls. The advice sounds testable but is unusable:
For contrast: program instances of "data science and big data technology" peaked in 2016–2017, yet demand for data jobs went on growing for more than five years afterward. A leading indicator issuing a false alarm five years early is the norm for administrative counts of this kind, not the exception.
So this article does not list it as a test signal. If you see someone elsewhere using this number to tell you that AI programs "have already peaked," that is an argument you can now dismantle yourself.
(An institutional fact, worth knowing but not a signal: from the 2025 batch onward, the Ministry of Education changed its release granularity — the lists were sent down to provincial departments and institutions, and the public must now piece things together from provincial press releases and third parties' incomplete compilations. The narrowing of release granularity is worth knowing in itself, but it cannot substitute for a signal.)
(2) Any single year's employment-rate ranking of majors.
The main indicator in domestic rankings has further study in its numerator, so programs with high further-study rates automatically look like they have "good employment"; and the first criterion for the red and yellow cards is the absolute number of unemployed graduates, which is driven mainly by how many people are in the program. Structurally, rankings of this kind cannot distinguish "this program is bad" from "this program is big."
(3) "Occupation X will have a gap of XX0,000 over ten years."
That number is a ten-year annual-average flow, of which 97% comes from people leaving, and 56% of those departures are moves into a different occupation. A large gap almost never means abundant opportunity; it mainly means people leave fast.
A test you can run yourself: open the BLS employment projections' occupation.xlsx and find Table 1.10. Take any occupation you have seen described in admissions advice as having a "huge gap," and divide that row's annual openings by its annual separations. If separations ≥ openings, that occupation's total employment is shrinking over the next decade — the entire "gap" comes from people walking out. Then divide openings by that occupation's 2024 employment and you get its annual turnover rate. The higher the turnover, the larger the "gap," and it has nothing to do with the industry expanding.
The corresponding move on the China side: when reading any employment quality report, first ask "who is in the denominator" — all graduates? "graduates with job-seeking plans"? or "those already employed full-time"? The three give completely different pictures.
How to test: go to bls.gov/emp/evaluations and open any round's evaluation. Look at the major-group scorecard first, then the detailed-occupation scorecard, and compare the official model's win rate against the "predict nothing" lazy model. You will see with your own eyes the gap the same evaluation gives at two levels of aggregation.
A quicker move: take the list of "20 fastest-growing occupations" from the projection edition of ten years ago and look up each one's actual employment today. Mind the baseline: drawing 20 at random from 338 occupations, you would expect about 1.2 hits — the standard is not "is the hit rate high" but "is the hit rate clearly above 1.2 out of 20."
How to test: for each field you are considering, ask three questions —
Data you can watch: in the US, the nursing colleges association's annual "applications turned away" and "faculty vacancy rate" — remembering that applications ≠ applicants, and that these must be compared to the ten-year average, not to last year. In China, the urgently-needed-program list and program-warning list your provincial education department publishes each year, plus the number of institutions in your province offering a given program — once that count approaches the province's total number of undergraduate institutions, the label's scarcity is gone.
How to test: wherever you are given a "major income ranking," ask whether it gives percentiles. Rankings that give only medians or means systematically conceal the information that matters most to you.
A concrete move: in the New York Fed's "Outcomes by Major" table, look at the early-career and mid-career columns together, and then look at that field's share of graduates working in jobs that don't require a degree. Fields with a high share there but a low unemployment rate (biology 51%, psychology 48%, fine arts 59%) tell you their graduates can find work but cannot find work that requires the degree — which predicts your actual situation better than the unemployment rate does.
How to test: ask yourself "which part of the distribution is the outcome I want in?" If the goal is the top 1% of income, an elite graduate school, or a prestigious firm, the choice of institution has a substantive effect; if the goal is only the top 25%, the best study available measures an effect that is "small and statistically insignificant."
An observable signal: when reading a target institution's graduate destinations report, do not look at average salary; look at the absolute number of upper-tail destinations (how many went to a handful of target employers or programs), then divide by that institution's total graduates in that field.
How to test: for any claim that "field X is cooling / recovering," demand a series of at least five consecutive years. A single year's direction is not a signal — the median year-over-year change in unemployment rates by field is about 1.25 percentage points, and 20 of 73 fields change by more than 2 points in a year.
Checkable on the US side: the National Center for Education Statistics' long series on computer and information sciences degrees conferred (1964-65 to the present), where you will see two complete declines of −42% and −36%.
Checkable on the China side: the annual admissions plan books by program published by each province's admissions examination authority (countable school by school, corresponding directly to headcount supply), and the year-by-year change in admission ranks and cutoff scores for your target program (each province's score-distribution table plus institution-program-group cutoff lines). These two are primary data you can look up yourself during application season, and they have no "school count ≠ headcount" unit mismatch.
How to test: do not use the national flexible-employment rate as a baseline for subtraction — that mistakes definitional differences for falsification.
The correct move: open your target institution's Graduate Employment Quality Report and make a within-institution cross-program comparison — look at the sum of "flexible employment + awaiting employment + not currently seeking employment" for a given program, and whether its flexible-employment share is markedly higher than that of other programs at the same institution. A within-institution comparison is unaffected by definitional differences, and that is the usable criterion.
Institutional background worth remembering: freelance work is verified "on the basis of supporting materials signed and confirmed by the graduate themselves, adjudicated by the responsible officers of the employment offices at the university and school levels"; e-commerce entrepreneurship counts toward the destination rate and need not meet the minimum wage.
Abolition is harder behavioral evidence than addition, because it carries a real cost. The Ministry of Education's annual batch includes abolition and suspension figures (in the 2024 cycle, 1,428 program instances abolished and 2,220 suspended nationwide).
But before citing any "top 5 abolished programs" list you must ask three questions: ① is there a denominator for the abolition rate? (Nowhere has one ever been given.) ② Do suspensions enter the ranking? (No — and suspension is often the precursor to abolition.) ③ Where does the underlying table come from? (The Ministry of Education's annexes for 2023 and 2024 contain no abolition list at all, and the provenance of the circulating tables is unknown.)
And remember: whether a program's name contains the words "information / computing / network," whether it belongs to the computing category, and whether it is being phased out are three separate things. The programs leading abolition counts are management and mathematics programs carrying "information" in their names (information management and information systems is a management program; information and computing science is a science program), along with programs with no IT flavor at all such as public administration, marketing, and product design; no core computing program appears in any of these top-5 lists.
A move closer to you: check your own province's program-warning list and urgently-needed-program list (already covering 473 program types), and whether your target institution's program has ever received a provincial yellow or red card. A provincial list is closer to the decision you have to make than the national count of newly added schools.
There is also a policy chain here: an employment rate that is too low is an officially stated trigger for "suspension of admissions," and the stated condition for abolition is "no admissions for five consecutive years" — the two are linked by the same policy chain. So "many abolitions/suspensions = bad employment" is not folk speculation; it is the intended outcome of the policy design. What you really need warning about is the lag: it can take five years or more from a too-low employment rate to eventual abolition.
This is the weakest evidence in the article. It is listed here because it is the most frequently asked.
For any claim that "software jobs are recovering / collapsing," ask three things first:
Interests at stake: this is a hiring platform's data, and the platform's commercial interest lies in "job-posting counts are a meaningful labor market signal"; the index covers only its own postings, and platform share drift contaminates it. It also revises and restates history continuously, so the same sentence "up Y% since date X" gives different numbers depending on when it was scraped — any citation must lock the retrieval date.
How to test: every time you see a "field match rate" or "field relatedness," first identify which of the three it is:
The three have different numerators and denominators, and any arithmetic or comparison across them is wrong. On the China side, "job–field relatedness" has those already employed full-time as its denominator — not all graduates, and not the administrative "field-matched employment rate."
One very common misreading: "the job requires a bachelor's degree" corresponds to about +25% in wages, and "the job matches the field" corresponds to a further +5% on top of that — they are two nested thresholds, not competitors, and the ratio of coefficients is not a ratio of importance.
| Signal | Why it is better | Where to look |
|---|---|---|
| Enrollment plan headcounts by program | Corresponds directly to headcount supply, with no "school count ≠ headcount" mismatch | Each province's admissions examination authority's annual admissions plan book, countable school by school |
| Year-by-year change in admission ranks and cutoff scores for your target program | Directly reflects candidates' collective judgment, and you can look it up yourself during application season | Each province's score-distribution table plus institution-program-group cutoff lines |
| Appearance of relevant program codes in the annual batch's abolition/suspension lists | Abolition is harder behavioral evidence than addition, because it carries a real cost | The Ministry of Education's annual Results of Program Filing and Approval |
| Each province's "urgently needed program list" and "program warning list" | Already covering 473 program types; which program your own province puts on the warning list is closer to the decision you have to make than the national count of newly added schools | Provincial education departments, published by July 31 each year |
| The total planned headcount in your province's Public Recruitment Announcement for Primary and Secondary School Teachers | Checkable year by year, and a primary administrative document | Provincial education / human resources departments. You must also check whether the post categories match the previous year's — when the bases differ, do not compare percentages |
One last piece of discipline, for you and for me: in judging any recommendation about fields of study, you can apply the same standard — was its decision rule set before it saw the data, or after?