These sound like guesses. Some examples odds ratios from the Harvard lawsuit are below, assuming full sample and full controls (comparing applicants with similar stats, admission reader ratings, LOR ratings, SCEA/RD status, …). I believe that none of the listed prospective major differences reach statistical significance. They also vary notably between the original analysis and rebuttal analysis, due to the slight difference in controls. I used an average of the two for the concentration coefficients below. Odds ratio of 1.0x means no influence.
If Harvard is significantly penalizing ORMs for being in STEM, then this penalization is not reflected in the analysis. The standard error of all of these coefficients was fairly small, suggesting that there is not a drastic difference between concentration effect for different racial groups. If there is a small difference for URMs, than that small difference for prospective major would be minuscule portion of the impact of URM and other hooks, such that a URM in STEM benefit would be dwarfed by the general URM hook.
Odds Ratio for Harvard Admissions
Athlete: 1800x
Black: 14x
Dean’s List: 10x
Legacy: 6x
Hispanic: 4x
…
Computer Science: 1.2x
Female: 1.15x
Physical Sciences: 1.1x
Engineering: 1.1x
Mathematics: 1.0x
Biology: 1.0x
…
Asian: 0.8x