The regression coefficients showing relatively small effects are from the Plantiff in the Harvard lawsuit. The Plantiff makes a variety of assumptions in the model that magnify the degree of these effects (for example not controlling for parents occupation). One can dispute the merit of these assumptions, but the Plantiff chooses assumptions that magnify the Asian bias, and Harvard’s expert chooses assumptions that minimize the bias. I’d expect the actual effect is somewhere between the Plantiff’s model and Harvard’s model. I think it’s highly unlikely that the Plantiff is dramatically underestimating to the point where the Asian percentage should match Caltech.
Instead the far more likely explanation is that Caltech and Berkeley’s admission model and applicant pools differ from Harvard in a variety of ways beyond just use of race. For example, the lawsuit reports the following racial percentages among the 3 largest non-racial hook groups at Harvard. These 3 groups make up quite a large portion of White admits. Note that these hook groups are all mostly White and rarely Asian, so favoring these groups means White students become more overrepresented and Asian students become more underrepresented. Caltech gives far less preference to legacy and athlete than Harvard, so it follows that a larger Asian percentage is expected. I’d expect that Dean’s special interest list, Z-list, and such play less role in Caltech admissions than in Harvard admissions. Caltech also has a reputation for focusing more on scores in the admission process than Harvard, which again is expected to increase Asian percentage.
Legacy – 76% of admitted legacies are White, 12% are Asian
Athlete – 76% of admitted athletes are White, 15% are Asian
Dean’s Special Interest List – 72% of admits are White, 10% are Asian
One also needs to consider that Caltech and Harvard have very different applicant pools. The “Institute of Technology” fields that Caltech focuses on tend to have a strong Asian male overrepresentation in the applicant pool at selective colleges, particularly among the more qualified applicants. In contrast liberal arts fields for which Harvard is especially renowned are not as dominated by Asian males. Asian males are an underrperesnted group at some selective LACs. Colleges like Caltech almost certainly have a larger portion of Asians in the applicant pool, without even considering the CA location.
In Berkeley’s case, not considering race is a relatively new change; so it’s easy to compare racial percentages before and after the admission policy change. The graph at http://www.dailycal.org/2018/02/02/prop-209-affirmative-action-uc-berkeley-enrollment/ shows the following racial demographic changes shortly after Prop 209 took effect. Nearly half of students were Asian both before and after Prop 209. Berkeley had a far larger percentage of Asian students than White students when they were still considering race. There was not a small change in Asian percentage, not a huge one, and that increase appears to be primarily due to the reduced URM percentage rather than no longer favoring White students.
Asian – Increased from ~40% to ~45%
White – Roughly unchanged at ~30%
If Harvard changed their admission policy to just consider GPA and SAT scores, Asian students would become a much larger percentage of the class. However, that is not Harvard’s admission model. They consider other factors besides just stats and “personal” rating. The Plantiff’s analysis estimated the demographic percentages would change as follows, without consideration of race and various other factors. The Plantiff claims the Asian penalty only decreases Asian percentage by 1-2%. The far more influential factors are the boosts given to URMs, legacies, and athletes. Removing Dean/Director’s special interest list would also more significantly increase Asian percentage.
Including “Personal” Rating
Actual Class – 22% Asian, 45% White
No Asian Penalty – 23% Asian, 45% White
No Asian Penalty or URM Boost – 28% Asian, 52% White
No Asian Penalty or URM/Legacy/Athlete Boost – 32% Asian, 48% White
Excluding “Personal” Rating
Actual Class – 22% Asian, 45% White
No Asian Penalty – 24% Asian, 44% White
No Asian Penalty or URM Boost – 29% Asian, 51% White
No Asian Penalty or URM/Legacy/Athlete Boost – 34% Asian, 47% White