Most of these coefficients for planned area of concentration appear to be below statistical significance. As such many have very different degrees of magnitude or event direction, depending on nuances of the modeling used. For example, female + CS was listed as one of the highest coefficients in the table above. However, if you instead look at the original model (first document, not rebuttal) with unhooked baseline + full controls (model 6), then the coefficient changes to -0.019 = 0.98 odds ratio… the opposite direction and far below any measure of statistical significance.
I noted this effect as well. I believe it relates to the combined strength of two hooks is generally weaker than the effect predicted by the strength of each hook individually. Rather than creating a superhook for kids with multiple strong hooks, the effects tend to be blunted. For example, recruited athletes who are also legacies may not see a huge boost in chance of admission over recruited athletes who are not legacies. If the legacy benefit is indeed weaker for recruited athletes than otherwise unhooked legacies, then it follows that an interaction variable for recruited athlete + legacy, it would be a strong negative. Similarly I’d expect interaction variables between legacy + other hooks to be negative.
The models indirectly consider essays, as reflected by how they impact category ratings given to the applicants. With any model, two important factors to consider are how much variance the model explains, and the statistical significance of the conclusion. Both sides of the lawsuit created models that explained the majority of variance in Harvard’s admissions decisions. While the models are by no means perfect, they generally work well. Some of the regression coefficients reached statistical significance, and some did not. For example, earlier in this post, I mentioned that preference for planned area of concentration generally did not reach significance. However, preference for recruited athlete was extremely significant. We can say with near certainty that Harvard has an extreme preference for recruited athletes, even though the model is not perfect.
The preference for SCEA was far weaker than the preference for recruited athlete, but also reaches statistical significance. All 12 models (6 models in original publication + 6 models in regression) show a reasonably similar and noteworthy preference for SCEA. Effects of factors not included in the model, such as your essay resonating in a way that is not reflected in AO ratings example, are reflected in a higher standard error and decreased confidence level. If my math is right, the confidence level of a preference for SCEA is significant at the 99.999999999999999999999999999999999999999999999999% level . This is a higher confidence level that the preference for legacy, faculty/staff kid, dean’s special interest list, and numerous other powerful hooks. While these LDC hooks appear to be stronger than the preference for SCEA, they also have a higher standard error, with an apparent more variable degree of benefit for the hook. SCEA is weaker, but appears to have a more consistent degree of benefit for the hook among different individual applicants, so the confidence level is higher.
C’mon, guys. If you dont give them what they look for, these tables and more massaging wont get a kid in.
Find a chart that shows: poor grades and scores, low rigor, blew his essay, lukewarm LoRs, bad Why Us, incomplete course expectations, weak ECs, immature or antisocial showing, etc.
Early doesn’t benefit those kids. Not in your dreams. Legacy won’t. Etc.
Any “benefit” goes to fully compelling applicants, but filtered by institutional needs, incl geo diversity and gender balance.
It depends how luke warm stats, LORs, ECs, etc. Obviously being a legacy or having other hook does not mean you automatically get admitted, no matter how bad of a student you and no matter how poor your application. Instead it means a certain degree of preference. That preference is enough for some students to be admitted, who would be rejected without the special preference. The model gives some hints about how powerful that preference is and what type of applicants might be admitted with the hook preference, who would otherwise be rejected. For example, specific regression coefficients from the model with full controls are below, as well as the reader guidelines for academic rating.
Applying early appears to offer a similar benefit to increasing academic (or most other) rating from 3 to 2. A double legacy who also applies early may get a similar benefit or greater benefit as increasing academic rating from 4 to 2, although I expect most legacies do well in the academic categories and instead the legacy preference more typically helps legacy applicants with less amazing EC or personal ratings get admitted, who would otherwise be rejected without the legacy preference.
Regression Coefficients: Full Sample, Full Controls
Legacy + Double Legacy: +3.06
Legacy: +2.33 (0.16)
Applies Early: +1.53 (0.10)
Academic Rating Drops from 2 to 3: -1.51 (0.10)
Academic Rating Drops from 2 to 4: -3.84
**Academic Rating = 2 **
Magna potential. Excellent student with top grades and,
a. SAT and SAT Subject tests: mid 700 scores and up
b. 33+ ACT
c. Possible local, regional or national level recognition in academic competitions
**Academic Rating = 3 **
Solid academic potential; Cum laude potential: Very good student with excellent
grades and
a. SAT and SAT Subject tests: mid-600 through low-700 scores
b. 29 to 32 ACT
**Academic Rating = 4 **
Adequate preparation. Respectable grades and low-to mid-600 scores on SAT and subject tests or 26 to 29 ACT.
This OP has made no reply in 2+ weeks. Frankly, with no background info for him/her, just legacy, needing time to improve grades, there’s nothing we can add.
The point here is that all statistical models are wrong, some are just better than others. They help to quantify an otherwise completely subjective process. Holistic admissions is a black box. And it really shouldn’t be.
These models are not 100% accurate. But they sure help to understand what were the key factors that drives admission decisions. Apart from Athletes, being on the Dean’s interest list, FACTOR1, legacy or child of a faculty really boosts your chances. Not by a little, but by a lot.