6 things we learned
We joined 62,406College Scorecard programs to the occupations graduates enter, the AI exposure of those jobs, and metro rents. Here's what the data actually says about whether a degree is worth it.
More debt doesn't buy more pay.
Across 226 Bachelor's majors, the correlation between how much debt a field carries and what its graduates earn is just r = 0.056 — essentially zero. The lever that decides your earning power is the field you choose, not the price tag. Computer Engineering graduates out-earn Dance graduates by $74,865 five years out, on roughly the same debt.
The best-paid fields are often the most exposed to AI.
Plot every major by pay and by how much of its work generative AI could already do, and the uncomfortable cluster is the top-right "danger zone" — well paid and highly exposed. Computer Engineering, Computer Science, Computer Programming all land there. (This is task overlap from GPT-4-era measures — what could be assisted, not a prediction that the job disappears.)
The 'premium' is often about who gets in, not the major.
A major's raw earnings edge mixes the field's value with the fact that selective schools admit higher earners. Adjust for who enrolls and Computer Science keeps almost all of its edge (+$48,460 raw → +$45,982 adjusted), while Neurobiology and Neurosciences's +$5,511 edge nearly vanishes to +$-5,141. It's an observational adjustment, not causal proof — but it separates signal from selection.
Where you live decides whether the pay is enough.
The same degree's paycheck is comfortable in one metro and a stretch in another. We pit graduate-weighted pay against market rent across ~370 metros: a tech salary clears the 30%-of-income rule almost everywhere, but coastal and Hawaii metros flip it into a stretch. Geography is the variable the headline salary hides.
The gender pay gap persists inside the same field.
Even comparing men and women who studied the same thing, the median field shows an 8% gap in 5-year earnings, widening to 30.6% in fields like Religion/Religious Studies. Same major, same credential — different outcome.
Two degrees, opposite job-market futures.
Pay is only half the story — the other half is whether the jobs will still be there. Using BLS 2024–2034 projections weighted over each major's occupations, the jobs that Mathematics and Statistics graduates enter are projected to grow +21.9% over ten years, while those for Graphic Communications are projected to shrink -15.2%. Same diploma timeline, very different headroom.
How we built it
The spine is the U.S. Dept. of Education's College Scorecard (program-level earnings and debt). We map each major to its occupations via the NCES CIP→SOC crosswalk, weighted by employment; attach wages and AI-exposure measures (O*NET, Eloundou, AIOE); and join metro wages (BLS OEWS) to rents (Zillow), and add each occupation's 10-year BLS growth outlook. Every number is a real, published figure — privacy-suppressed cells are left out, never imputed — and the analysis is reproducible from the committed data.
We show the modeled crosswalk against reality. For 39 fields we could match by name, we add an independent empirical view from Census ACS microdata (2023 1-Year PUMS) — the occupations real graduates of that field actually report, by share. The two are built from entirely separate data (ACS never feeds the model), so they sometimes diverge: the model weights occupations by their total size, while ACS reflects where graduates actually land. Showing both — rather than hiding the disagreement behind a single number — is the honest way to present a lossy CIP→SOC map.