Does your major
matter more than
your college?

Run 546 fit selectivity against earnings across 988 colleges and named its own biggest blind spot on the way out: no field-of-study variable at all. An engineering-heavy selective school's earnings premium was, in that data, indistinguishable from a liberal-arts-heavy selective school's. This run supplies the missing variable, using the Scorecard's own field-of-study file — one row per institution × specific bachelor's program — instead of the institution-level file. 27,617 bachelor's-level rows report a non-suppressed median earnings figure 4 years after completion; 5,767 of them sit in a conservative “hard STEM” bucket (Computer & Information Sciences, Engineering, Engineering Technologies, Biological Sciences, Mathematics & Statistics, Physical Sciences, Science Technologies) and 2,369 in an equally conservative Humanities bucket (English, Foreign Languages, Liberal Arts & General Studies, Philosophy & Religious Studies, History, Area/Ethnic/Gender Studies) — both defined before any fit ran, deliberately excluding ambiguous adjacent series like Agriculture, Psychology and Visual & Performing Arts that only partly belong to either camp.

The naive gap is not close, and it survives clustering by major. STEM programs' graduates report a mean earnings of $81,502 4 years out (median $81,574); Humanities programs' graduates report $52,281 (median $51,141). Regressing earnings on a STEM/Humanities indicator, with standard errors clustered by the 115 distinct majors involved (since the same major recurs at dozens of colleges and those rows are not independent draws): $29,220 more a year for STEM, 95% CI [$20,506, $37,935], p=5.0e-11, R²=0.29 — excludes zero. On the log scale the same comparison reads as +53% more earnings, CI [+36%, +72%].

Weight each program cell by how many graduates actually stand behind its earnings figure, and the gap widens rather than shrinks: $32,301 a year, CI [$31,078, $33,524] — though this spec does not cluster by major, so its interval should be read as the more optimistic of the two, not a tighter estimate of the same thing.

The sharpest check available in this data: compare STEM to Humanities at the very same institution. 734 colleges report a non-suppressed earnings figure for at least one STEM-bucket program and one Humanities-bucket program. Take each college's own mean STEM earnings minus its own mean Humanities earnings — a design that holds school quality, region, cost of living and admissions selectivity fixed by construction, since both numbers come from the same building — and the gap is $25,196, 95% CI [$24,182, $26,210], p=2.5e-232. 96% of the 734 colleges show STEM ahead of Humanities at that same school; the gap is not carried by a handful of outlier universities.

Named, and the full 32-field ranking undercuts easy cherry-picking in either direction. The single highest-earning row in the whole sample is Duke University's Mathematics program ($297,029, n=17 graduates) — a STEM row, as expected. But the field that pays worst of all 32, lower than either bucket's average, is Visual & Performing Arts ($44,402 mean, n=1,672) — a field in neither this run's STEM nor Humanities bucket. Transportation & Materials Moving, also outside both buckets, out-earns three of the seven STEM fields. The lowest single row in the sample is Caribbean University-Bayamon's Allied Health and Medical Assisting Services program ($16,087, n=26). The six Humanities fields cluster tightly between $49,298 and $54,977 — solidly in the bottom third of all 32 fields, but not literally the six lowest-paying fields on the page.

Read plainly, and with the limitation this design cannot remove. The STEM-Humanities earnings gap is real, well-estimated, survives major-clustering, survives graduate-count weighting, and survives holding the college itself fixed. What none of these three specs can do is separate a major's own causal return from who chooses it: students who select Engineering over English may differ systematically — in quantitative preparation, risk tolerance, or family background — from students who select the reverse, the same selection problem run 546 could not fully rule out for choice of college. Nor is this a full cost-benefit ROI calculation: the Scorecard's field-of-study file has no tuition-by-major figure, so this run compares earnings only, not earnings net of what each degree cost to obtain. And it is a single snapshot, 4 years after completion — not a lifetime-earnings trajectory, where some contested claims argue humanities majors narrow the gap later in their careers. This run cannot speak to that; it can only say what the numbers say at year 4, which is that the gap is large, it is not close, and it holds up inside the same school.