It’s obvious that Harvard and other elite colleges favor hook groups that of which Asians are underr-erpresented among almost all – recruited athletes, legacy, URMs, dean and director’s special interest, Z-list, etc. However, this was not the focus of the lawsuit. I don’t know enough about the relevant law to guess at the outcome of a theoretical SCOTUS decision on the lawsuit, but looking at the lawsuit docs, it doesn’t seem obvious that SCOTUS would theoretically “strike this down.”
I talked about my opinion on the personal ratings earlier. My opinions on the admission decisions models and how they reflect Asian vs White preferences is below, which key for determining merit. Like the personal rating, I think it looks suspicious, but not conclusive. And anti-Asian bias beyond favoring hooks (including URM) also has less effect than is often suggested. As discussed below, an increase from 22% to 23% assuming full controls and expanded sample. It does not appear to occur for Asian females – only Asian males, particularly ones outside of California.
The lawsuit mentions many different admissions models that come to varying conclusions. The most simple and basic one is model is the Harvard OIR one, which was referenced in an old internal memo. This is the one with the graph that shows that the Asian percentage drops from 26% to 18% when considering “demographics” (both URM and Asian). The model controlled for academic index, adcom ratings in the 4 main categories (academic, athletic, ECs, and personal), athlete, legacy, gender, and race. With these controls, it found the following regression coefficients.
Athlete: +6.3
Legacy: +2.4
Black: +2.4
Hispanic: +1.3
Asian: -0.4
The Asian coefficient was negative and statistically significant. However, it is important to note this is a very basic model that does not consider things like LOR ratings, interview ratings, intended major (lawsuit notes large differences between races) , whether the student applied SCEA, whether the student was disadvantaged, whether the student was on the dean/director’s list (16% of white admits), etc. Both the plantiff and Harvard’s expert agree that the one from the old Harvard OIR memo is not adequate and have their own models that incorporate the additional categories mentioned above, as well as many others. In spite of the hundreds of pages of complaints largely relating to these models, the plantiff and Havard expert’s models are quite similar.
The plantiff’s model found that the percentage of Asian students would change as follows under different considerations, assuming full sample and full controls (including personal rating). The “Asian penalty” only increases the percent Asian by 1% from 22% to 23%. Favoring URMs, legacies, and athletes have larger influences, but these categories are not the main focus of the lawsuit.
Default: 22% Asian
No Asian Penalty: 23% Asian
No Racial Preferences (no URM or Asian): 28% Asian
No Race, Legacy, or Athlete: 33% Asian
The regression coefficients with full controls and full sample are below. The specific numbers all differ from the Harvard OIR model with the additional controls, but a similar pattern occurs, with a small, but statistically significant negative coefficient for Asian.
Athlete: +~7.849 ((0.153)
Black: +3.674 (0.103)
Legacy: +2.329 (0.164)
Hispanic: +1.959 (0.086)
Asian: -0.257 (0.070)
The Asian coefficient varies from year to year, varying from a minimum of approximately -0.6 in 2018 to a maximum of +0.1 in 2019. It also appears to vary by gender. There was no statistical difference between female Asian and female White. Harvard’s expert found a small advantage for both Asian female and Asian from California. The degree of various suggests there is significant noise in the measure, likely depending on additional variables that were not included in the model.
Harvard’s expert found an overall Asian coefficient that was not statistically different from 0 using the same data source that the plantiff did. The reason for the different conclusion, primarily relates to the model differences listed below:
Personal Rating Worth +0.31 – As noted above the previous numbers include the personal rating. If you exclude it, the Asian coefficient decreases significantly. I previously stated my opinion.
Controlling for Parents’ Occupation: Worth +0.19 – Parents’ occupation has a notable overlap with race, so if you compare applicants with similar parents’ occupation, then it decreases the appearance of a bias against Asian applicants. Harvard says it’s valid to do so because adcoms have access to parents’ occupation and are instructed to consider it. The plantiff says parents’ occupation is primarily used to determine whether the applicant is disadvantaged and is not a key component of the admission decision. I don’t know that level of detail in how applicants are reviewed, but the plantiff’s view seems more reasonable to me.
Add Intended Career and Staff Interview: Worth +0.11 - Intended career should absolutely be included. Only a small minority receive staff interviews, so I can see the merit of either including or excluding them. The effect is fairly minor.
Remove Disadvantaged & Race Interaction: Worth +0.10 – It seems reasonable to me that disadvantaged and race would have significant interaction, so I see no good reason to remove it.