This is common orthodox wisdom on these forums, but The College Board says that SAT scores are more likely to be a strong point versus HS GPA for students from income backgrounds (possibly because of increased test prep resources and embedded test prep at higher SES schools): https://files.eric.ed.gov/fulltext/ED562878.pdf . Hence, a greater emphasis on SAT will generally tend to favor students from higher income backgrounds.
3 Likes
Iâm going to proactively remind posters to stay on topic. Weâve had countless threads about TO and the value of standardized tests.
If you are referring to the Plaintiff assessment, the Plaintiff regression models included 6 different combinations of controls. The full controls model included controls personal rating. The full controls model came to a similar conclusion as the model that excluded controls for personal rating.
The counterfactuals they discuss implicitly assume away the personal factor, and therefore do not credit higher personal scores as at least part of the reason why LDC applicants might get admitted at higher rates. .
As noted the regression model with full controls included controls for personal rating. The specific odds ratio for LDC and personal ratings are below. This is the estimated average change in odd of admission if all other criteria remains the same â same test scores, same academic rating, same personal rating unless otherwise noted, same race (numbers below assume White), âŠ
- On Special List â 36x (high variation)
- Double Legacy â 27x
- Personal Rating increases from 3 to 1 â 26x (an average of only ~6 applicants per year get personal rating of 1, making sample size too small to be meaningful)
- Faculty/Staff Kid â 14x (high variation)
- Legacy â 10x
- Personal Rating increases from 3 to 2 â 8x
- Personal Rating decreases from 3 to 4 â 1/28x
LDC applicants had higher average ratings in all categories than non-LDCs⊠When controlling for race, LDC applicants had the greatest degree of difference in athletics and personal. And the smallest degree of difference in GPA and LORs. The regression analysis above controls for all of these ratings. Among applicants who had similar ratings (and other criteria), what is the LDC boost?
The degree of boost from personal rating was smaller for LDCs than non-LDCs on average, which may relate to avoiding admitting persons who lack basic academic qualifications. A similar effect occurs when combining other hooks. For example, legacies who are URMs get a smaller average legacy boost than legacies who are ORMs. When strong hooks are stacked, the combined effect gets blunted.
1 Like
Statistically? Sure. But I meant IRL.
Societally, in the real world, in ways that actually matter and are going to meaningfully impact people. Not IMO. Even if you point to the optics of it (which is what frankly makes it a clickbaity stat) do we really believe that this particular outcome, of something like a few hundred-thousandths of a percent of the US population, is especially problematic? How many F500 CEOs should the Ivies produce? Zero? Eight? How many colleges should F500 CEOs come from? 50? 100? 500?
The focus this gets is way out of proportion to the actual underlying issues, which center around access and opportunity that begins looooong before college admissions.
I believe it was @NiceUnparticularMan who wrote about this chain of opportunity and access at length up thread.
Again, youâre focusing on normative values.
Again, to be quite clear, I was not.
I am not debating you. I am stating my view, and not in relation to yours.
To be clear, posts regarding the value of standardized tests will be deleted on this thread.
And, reminder that DD is not a debate society.
1 Like
Thanks for presenting your perspective. I disagree, but I understand the argument you are making now.
Indeed. Harvard parents disproportionately raising kids that Harvard likes? Kinda plausible. It is easy to be cynical, but there are some deep questions here about what parents value in a person themselves, how those values might be informed or at least reflected in their own college choice, and then how those values would inform or be reflected in their parenting choices.
I note the models in question controlled for normal personal demographics (entity, gender, etc.), various family socioeconomic indicators, neighborhood indicators, high school type, academics, intended majors, ECs, and teacher recs.
But would all that REALLY be enough to predict someoneâs personality as evidenced in their application?
As another poster has rightly pointed out, this is really a conceptually impossible task, at least with this sort of data. You could maybe do something like a controlled experiment where reviewers got identical applications with or without the legacy tag, and see if that made a difference. But as the other poster pointed out, since that didnât happen . . . I am not sure this is answerable.
That being said, my completely non-expert back of the envelope estimate is personal factors (and maybe some undescribed overall factors) that are merely correlated with being legacy might account for something like half of what these expertsâ models attributed entirely to direct legacy preference.
So I personally have no doubt there would still be a direct legacy preference effect even if you put the personal factor back into the model. I just am not sure the direct legacy effect was as powerful as these particular experts argued.
Which is a bit of a bummer since my S24 might be applying to my college, but I am skeptical merely listing me as his parent will really give his personal score a big boost in any direct sense. I do think it is possible I am one of those parents who had some relevant values when parenting. But I suspect that at this point, whether that gets him a good personal score is up to him, not me being on his application.
It sure did. But Model 6, as I know it, was not used to test the effect of LDC factors on the personal rating. See Table B.6.7 in the original expert report. It couldnât be used that way, because then you would be using it to explain itself.
OK, then in this paper, they say this when defining what they call the estimated admission model:
We incorporate a broad set of applicant controls in the model, including numerous measures of socioeconomic status, neighborhood and high school attributes, region, intended major, and academic aptitude, among others. We incorporate many of Harvardâs internal ratings, including the academic, extracurricular, athletic, the school support measures, and the alumni interviewer ratings. For each rating we create separate indicator variables for rating levels from 1 to 5. We do not include either the overall rating or the personal rating. The overall rating is not included because the rating is specifically designed to incorporate admissions preferences; the personal rating is not included because there is empirical evidence that is influenced by preferences as well.79
And Note 79 says:
For example, Document 415-9 shows that, in contrast to the other ratings (with the exception of the overall rating), an ordered logit model of the personal rating shows that legacies receive a bonus for this rating (see Table B.6.7R).
But again, they couldnât use Model 6 in B.6.7R, and did not.
OK, then again they select a particular model to generate their counterfactuals. They reiterate:
To put the size of the LDC preferences in context, we examine how the probability of admission would change for non-ALDC applicants of different races and different baseline probabilities of admission using the results from our preferred admissions model.84 Consider, for example, how the probability of a white non-ALDC applicant with a baseline probability of admission of X would change if they happened to be a legacy, but if all other characteristics remained the same.
Note 84 reads:
Our preferred model excludes the personal rating since there is evidence that the personal rating incorporates preferences for ALDC applicants. Using a model that includes the personal rating does not alter the basic findings.
I believe this, if âbasic findingsâ means that alternative model would still indicate a substantial direct legacy preference. I also believe the magnitude might be lower. Unfortunately, to my knowledge they do not further explain this claim.
OK, and then finally, Table 10, which provides the counterfactuals we have been discussing, uses this preferred model:
We use the estimated admission model to calculate an admission probability for each applicant in the absence of legacy preferences by setting the coefficients related to legacy status to zero and keeping all other coefficients the same.
And I am personally confident the same analysis would apply if they instead included the personal factor (and did not zero it out)âthe same âbasic findingsâ would still be true, but the magnitude might be lower.
Where exactly are you getting this from? Is this in the NBER paper we have been discussing?
While there are many paths to becoming a corporate or financial CEO, the most over represented narrative begins with a stint of management consulting at one of the big three or former I banker.
The big three and I banks continue to recruit heavily from elite schools (although they are expanding). The past, current, and continued likelihood is that Ivy plus schools will continue to be over represented based on this trend and well traveled career path.
Should we punish these wealthy students because they were born into wealth? Is it their fault?
CEOâs of Fortune 500 companies wonât last very long if they are not competent and not increasing value for the companies. Ditto for investment bankers, hedge fund managers etc.
1 Like
Part of that personal rating is based on alumni interview, in which a legacy applicant probably has a leg up.
At least for my part, Iâm not suggesting/asking that anything should/needs to change with respect to company CEOs. Rather, the overwhelming sourcing from the Ivys is evidence that admittance should broadly reflect the countryâs diversity, economically and otherwise.
Is it âpunishingâ a student from wealth for a college admission reader to notice that a student from non-wealth with similar achievements may have actually achieved more due to having to climb over barriers and work for opportunities that were purchased by the parents of the student from wealth?
Although even CEOs who fail are often compensated quite well, including with golden parachutes.
1 Like
F500 CEOs are always going to be more highly educated, more intelligent, and more hard-working than the vast majority of people, so they will not reflect average Americans.
I agree. But the average American doesnât have a 4 year college degree either (just pointing out the difference is much greater than that between a highly rejective school and one that accepts most applicants (like most 4 year colleges do).
Letâs not forget that these schools also produce politicians in the US (and even around the world). With the current state of affairs, Iâm not sure they demonstrated any competence.
1 Like