Tuesday, October 20, 2009

Affirmative action: the numbers

Princeton sociologist Thomas Espenshade has a new book on elite college admissions. He is an author of an earlier study which quantified the advantages and disadvantages conferred by ethnic group at virtually all universities (with only a few notable exceptions such as Caltech). Already, some results from the analysis in the book have leaked out, and the numbers are not pretty -- see below.

US News: ... Translating the advantages into SAT scores, study author Thomas Espenshade, a Princeton sociologist, calculated that African-Americans who achieved 1150 scores on the two original SAT tests had the same chances of getting accepted to top private colleges in 1997 as whites who scored 1460s and Asians who scored perfect 1600s.

Espenshade found that when comparing applicants with similar grades, scores, athletic qualifications, and family history for seven elite private colleges and universities:

Whites were three times as likely to get fat envelopes as Asians. Hispanics were twice as likely to win admission as whites. African-Americans were at least five times as likely to be accepted as whites.

To clarify, 1150 is similar to the average at a typical state university, 1460 similar to the average at a top Ivy, and 1600 is obtained by only a few thousand high school seniors each year (note, this is post-1995 re-centering of the SAT). Cognitively, these are very different groups of kids.

Related posts: Asian-Americans hurt by affirmative action, the ugly truth, the price of admission, Asians at Berkeley, the new white flight.

See these slides for more results. The bottom slide below suggests that if race-blind admissions were enacted the percentage of Asian students at elite universities would jump from 24 percent to 39 percent. Note that 39 percent is similar to the current Asian populations at Caltech and Berkeley, two elite institutions with (roughly) race-blind admissions; the former due to meritocratic idealism, the latter thanks to Proposition 209.

(Click for larger versions.)


Luke Lea said...

Interesting numbers. Do they break it down by geographical region? I get the distinct impression that the Scots-Irish are woefully under-represented at Ivy League colleges.

In my view we should either have affirmative action for all groups or else affirmative action for none. For none makes sense at technical schools like Caltec and MIT, where brains are all that matter. So what if Asians and Jews are vastly over-represented? There's no social harm. To the contrary, advances in science benefit us all.

But by the same token "affirmative action for all" makes sense at elite liberal arts colleges like Harvard and Yale because -- and let's be honest here -- this is the pool from which tomorrow's political class will be chosen.

Right now that class is unrepresentative of the ethnic and geographical diversity of America, and it shows. Not a healthy situation.

Unknown said...

Well, elite Ivy's like Harvard are basically investment management companies, so why wouldn't they approach admissions as an asset allocation problem? You don't want your portfolio of students to be too concentrated in one asset class, no matter how attractive that class might be. Asians are like value stocks, good fundamentals, a little boring, allocate 30%. Blacks are like a speculative venture fund, risky but occasionally a big payoff (like BHO), so allocate 5 or 10%.

Ian Smith said...

This is a 10 on the BS meter. It confirms Steve's world-view, so he believes it.

1600 was extremely rare on the old SAT. Do I misremember? Didn't the (at most) two who scored 1600 have television interviews?

Some sort of extrapolation has been used, but any East Asian or South Asian who scored 1600 on the old SAT would have been admitted everywhere he applied. In other words, the extrapolation is BS.

Me: northwestern European, old SAT = 1460, father Harvard, grandfather Princeton, small envelopes from everywhere I applied.

AA's real intent is to obscure discrimination against white male gentiles and discrimination in favor of Jews. These are the only two groups with any power in the US.

Steve conforms to the Mongoloid stereotype.

Steve Hsu said...

1. From the perspective of the individual elite university, there is of course every reason to admit each class to maximize things like future endowment contributions, political influence, popular cachet, etc. There's no reason to be an academic or intellectual meritocracy -- as Golden makes clear in The Price of Admission, it is a disadvantage for Caltech. MIT has moved away from this strategy over the years, so as to emulate the top Ivy's and Stanford.

But I think people should actually understand how elite admissions works. In my own experience I've found that the individual admissions officers at elite schools don't have a fully articulated understanding of what it is they are doing. (What function are they maximizing?)

2. anon, Espenshade's data is from 1997. The SAT was "recentered" (or whatever) in the mid-90's, and the new 1600 became much easier to obtain than it used to be . As I mentioned, something like a few hundred kids each year get a 1600, so it's more like the old 1500. (On the old SAT scores there was an interesting correspondence to IQ: the SD was about 150, so you could, very roughly, just divide by 10 to convert to IQ. The history of the SAT and its links to psychometricians who worked on IQ is quite interesting.)

3. What Espenshade shows is that if you quantify each component of an application -- grades, scores, athletics, leadership, family background / social status -- and look at candidates of roughly equal quality, race plays a *huge* role.

4. If elite schools were doing to Jews what they are currently doing to Asians, there would be a huge public outcry.

AlchemX said...

Race based admissions are a tricky idea. We don't want people of certain races to become permanently locked in low opportunity situations, providing a homogeneous elite with cheap labor. Then success would just correlate to where a person is born. On the other hand we want the best and smartest people to succeed and put our money on the best minds. Is there an alternative way to make sure that peoples opportunities are not simply related to race. Though income is a big deal breaker, race obviously plays a major role. Asians are obviously being discriminated against, but they are not filling the prisons any time soon like black and hispanic people.

LondonYoung said...

You know Steve, I so wish I understood what the university staff are thinking sitting at those admissions committees. I imagine (correctly or not) black and hispanic applications going into a preferred pile and committee members telling themselves that these groups are discriminated against so they are righting social wrongs.

But what do these people tell themselves when they identify asians to be discriminated against? Do they do it solely by using correlated factors so they can kid themselves, or do they really just use the race box? Do profs ever discuss this among themselves? Do college presidents control this, or the boards of trustees?
(no reason to answer, I am just venting)

Ian Smith said...
This comment has been removed by the author.
Ian Smith said...

Then what is meant by "the two original SAT tests"?

Steve Hsu said...

LY: I don't think any admissions people consciously set out to discriminate against Asians.

1. There are categories for preferred admission (like legacies and athletes) that tend to disfavor Asians. Esp.'s startling 1460 vs 1600 result might include these categories.

2. There is a tendency to stereotype the Asian applicants, so that even if the Asian kid is, e.g., on the track team the feeling might still be that she/he is still a nerd whereas a white kid with the same athletic accomplishments gets a higher "extracurriculars" or "leadership" rating. Golden explores some of this in his book -- he accumulates some good crypto-racist comments from actual admissions people "these Asian students are all alike -- good grades, scores; want to study science; play the piano ... How can we tell them apart?"

The real apples to apples comparison is among "unhooked" kids -- those with no legacy connection, no significant wealth, no chance of making a varsity team. This is the majority of applicants, and the question what is the admit rate of Asian/white kids of similar strength. The answer is that Asian kids face a pretty big penalty, but it takes pretty serious analysis to show this. (There was an internal Stanford study done some time ago that came to this conclusion.) Espenshade is essentially reverse-engineering the process and building a model of admissions that (statistically) "explains" the outcomes.

In my experience at, e.g., Yale (where I was a professor), talking to admissions people, they don't have the equivalent of this model in their heads. They just do their jobs (following guidelines from above -- the dean of admissions, etc.) and it works how it works. Interested professors can figure out what is going on, but aren't really in a position to force any changes on the admissions office.

anon: there are now 3 sections (math-verbal-writing; many claim the third section is useless and was added to make life harder for Asians). When they refer to the two original tests I believe they mean math and verbal, but post (1995?) recentering.

Steve Hsu said...

PS Somewhere at the upper levels of the admissions office of each of these elite schools there must be people who are 1. familiar enough with the system and 2. smart enough that they have the intuitive equivalent of Espenshade's models in their heads. These people, who set the policies, know full well what the effects are. And I bet they're the ones who feel the heat if the Asian population gets too large. In the old days it would have been the Jewish population. Then it just takes a little leaning in the right direction to push the percentages back in the desired direction. I doubt you or I or Golden will ever get any of those people on the record. And I bet there is some unspoken tension about this among the admissions officers. But as we know, groupthink and political correctness are quite strong, and it is shockingly easy for people to avoid looking behind the curtain.

Anonymous said...

I find myself wondering how all this works out given the real constraints under which admissions officers operate.

Suppose you're admitting a class to Harvard, and, let's say, you want to get one quarter of them to be the brainiest available, one quarter to be great leaders who also have fine minds, one quarter to be top quality athletes who also have fine minds, and one quarter to be geographically representative students who also have fine minds.

Which of these categories will be heavily Asian? Only the first, I should think. Asians aren't better athletes than others, so they won't have a greatly disproportionate presence among the best athletes with brains. Asians aren't better leaders than others, so they won't have a greatly disproportionate presence among great leaders with brains. Asians aren't present in significant numbers across most of the US, so they won't have a greatly disproportionate presence in the geographical representation segment.

Statistically, it may look like a conspiracy against Asians, but unless the goals themselves I suggested above are misguided for a college like Harvard, the outcome may not be unreasonable.

LondonYoung said...

Thanks Steve, I think this is good insight.

To LiberalBioRealism, I think you have a sketch of a possible explanation as well, but I suspect this book (which I haven't read) considers and rejects this point. There's no way to say without a statistical analysis. I am also kinda uncomfortable saying "asians aren't better leaders" because absent the cold power of mathematical inequality we (as liberals in the 19th century sense) would want to say they aren't better at the SAT either - but it turns out they are. What does Harvard do to measure leadership and how do asians fare on those measures? I would like to know, and maybe I will even read this book!

Unknown said...

Too bad Obama supports racial preferences. Oh yeah, he has been a beneficiary his whole life, and he has reluctantly admitted as such, be it Columbia, Harvard, or the Harvard Law Review. Most recently, our nation was embarrassed by the affirmative action Nobel Peach Prize. How humiliating.

Unknown said...

They once of course limited by a White/Middle Eastern ethnicity--the Jews.

Now, as Steve Sailer pointed out, Hispanics are the only ethnicity that has a quota? Why is this? Well, if all ethnicities had a quota, Jews would be hit and hit hard.

My solution: race-blind admissions.

And, if you can, my advice to prospective students is to check the Hispanic box even if you are not ethnically Hispanic. See, Hispanic can also mean someone who comes from a Hispanic culture. Well, ever live in SoCal? There you go! Have some linkage to S. Africa or other African country. You are now African-American. All ethnicity and race is a matter of "self-identification," so there you go. Now, some elite schools, but not all, require interviews. This is the only way you could be questioned, but remember you would be technically be telling the truth.

My point as long as the Frankenstein monster of racial preferences exist, engage in legal civil disobedience to fight it!

Unknown said...

"4. If elite schools were doing to Jews what they are currently doing to Asians, there would be a huge public outcry."

Of course you are right. But Asians will not speak up, and are not feisty like many Jews. And the media ownership thing...

I would also point out that Catholic & Protestant Whites are underrepresented at Harvard--vastly.

Anonymous said...


Mostly I'm just assuming that, above a certain minimum threshold of cognitive ability -- certainly less than the threshold of ability required to get into Harvard in any case -- then leadership ability is mostly randomly distributed across the population.

If Asians represent, say, 4% of the larger population, they would represent only perhaps (here I'm guessing) 8% of the population beyond the threshold of cognitive ability to get into Harvard in any case. 8% would of course still be relatively high -- double their representation in the larger population -- but well less than their representation in the "pure brains" segment Harvard might select for.

Similar arguments apply for the other categories I mention, athletic ability and geographical representation (of course, there are other things as well Harvard might seek out).

In general, I very much doubt that the statistics would show anything else.

In fact, the statistics mentioned in the article Steve quotes don't mention anything about leadership ability. Rather it only mentions "comparing applicants with similar grades, scores, athletic qualifications, and family history".

Now I don't think that either "family history" or "athletic qualifications" add anything here for Asians in particular in terms Harvard might care about. I don't see what "family history" is supposed to add in any case. Harvard really doesn't care a lot about socioeconomic status, given that a vast proportion of its class is selected from the middle and upper middle class. And unless an Asian student is good enough an athlete to be selected to be part of a team, athletic ability is mostly irrelevant.

So, really, the things controlled for in this comparison across ethnic groups seems mostly to reduce to making comparisons across SAT scores and grades: i.e., purely signs of cognitive ability.

In other words, as stated anyway, what is described as being a "race blind" comparison seems actually to be a "race blind" comparison taking into account only markers of cognitive ability.

But why should Harvard and other Ivies restrict their selection criteria only to cognitive ability? Why isn't selection across other dimensions a good thing for the Ivies to pursue? Is it really sensible to think that everything -- everything -- important is reducible to IQ or its like?

Steve Hsu said...

If you look at the slides you can find one addressing the effect of social class (family background). When all other factors are held equal the social class strongly affects admission probability. In the case of Black-Hispanic-Asian applicants the least affluent applicants have the highest probability (slide 19, fig 3.9). For some reason this isn't the case with white applicants. That's a bit mysterious.

I don't know how you measure "leadership" except through high school activities. I'd be surprised if Espenshade doesn't have access to that data and therefore can (try to) control for it when he, e.g., holds all factors fixed and examines race as an independent variable. My suspicion is that if a white applicant is President of the student body the admissions people tag her as a future go-getter, whereas if an Asian kid is President they tag him as a nerd padding his resume. But we can't know without looking deeper into the data. It's also possible that recommenders (teachers, coaches) are more often rating the white applicants as future leaders and less often doing so with the Asians. In that case you can't blame the admissions people -- the letters say what they say.

BTW, you can see that in Espenshade's simulation model race blind admissions would lead to every other group losing slots and only Asians gaining. According to that model it is *only* Asians that are facing discrimination relative to their merits on the various factors (grades, scores, athletics, family background) -- everyone else is getting additional slots relative to strength on those factors. So whites are getting affirmative action under the current system -- they should actually have -6.5% fewer slots. (Again, this is just relative to Espenshade's model, but consistent with what happens at Caltech and Berkeley.)

Anonymous said...


Just to amplify on my point, let's take the very marker for leadership you mention: being elected student body president.

I would expect that Asians are so elected across the nation in close proportion to their underlying numbers: say 4%. If one restricts oneself to those who pass the minimum threshold for getting accepted to Harvard (say, the equivalent to a 1350 on the 1997 SAT unless the candidate is affirmative action) then perhaps this increases to 8% of the pool.

In this subgroup, then, Asians start losing representation at Harvard compared to what they would be at a strictly cognitive ability based institution such as Caltech.

Again, return to the (purely speculative) numbers I stipulated in my original post. Namely, 25% of Harvard's class is selected based purely on ability, and the other 75% based on other attributes, with a high minimal threshold required for cognitive ability.

Then even assuming that 45% of the brainiest are indeed Asians (which seems to be true at Caltech), we get .25 x .45 +.75 x .08 = about 17%.

Now the real mix is of course likely quite different from my speculation. But I think the calculation starts to get us in the right ballpark in terms of Asian representation -- and there is no presumption at all of any prejudice against Asians.

What one would certainly NOT expect to see is that Harvard would have numbers nearly exactly matching those of a purely cognitive ability based institution like Caltech. The very fact that Espenshade's model seems to be making that prediction for what he apparently calls "race blind" admissions in the Ivies strongly suggests that he really IS only taking into account markers for cognitive ability.

Jeff K. said...

"My suspicion is that if a white applicant is President of the student body the admissions people tag her as a future go-getter, whereas if an Asian kid is President they tag him as a nerd padding his resume."


This is not one of your least-biased statements.


Steve Hsu said...

Jeff: I think you should read the Golden book before you judge whether my suspicion is reasonable or not. I could be wrong about this subconscious bias, but I wouldn't be surprised to be right.

LBR: I think I understand what you are proposing. It involves both a model of admission, with different categories (brainiac, future leader, etc.) and also a model of how abilities and personalities are distributed in various ethnic groups (the usual HBD stuff).

1. Do you think your model explains the (even larger) overrepresentation of Jews at elite schools?

2. Technical comment: you need to look at the normal distribution so that you can do better than your 4% vs 8% estimates. I think you might be surprised about how overrepresentation grows as you go out on the tail.

3. Regarding your admissions model, it may actually describe Harvard, but probably not Yale or a number of other schools. When I was at Yale the admissions people liked to describe the Yale kids as "balls" whereas Harvard kids were "spears". That is, a Yale kid had to be a balanced all-arounder (so, *evenly* weighted sum of various factors, scoring well on almost all factors) whereas Harvard would take very unbalanced kids but who had extraordinary special talents (math whiz, Yo Yo Ma, novelist, etc.). I think Espenshade's model is more of a ball model and would fit a lot of the non-Harvard elites.

4. Don't you think that very determined (high conscientiousness and g) kids can pad their resumes so as to appear to be great future leaders, even if they lack the "innate abilities" that make for leaders? I think it would be much harder for the admissions people to judge this factor than the cognitive or athletic ones, for which we have pretty good measures that extend far out into the tail. I don't know what a +4 SD "future leader" looks like at age 17. Do you? (Bill Clinton was an exceptional case.)

Anonymous said...

"4. Don't you think that very determined (high conscientiousness and g) kids can pad their resumes so as to appear to be great future leaders, even if they lack the "innate abilities" that make for leaders? I think it would be much harder for the admissions people to judge this factor than the cognitive or athletic ones, for which we have pretty good measures that extend far out into the tail. I don't know what a +4 SD "future leader" looks like at age 17. Do you? (Bill Clinton was an exceptional case.)"

Steve has it right, here. The right tail of leadership credentials is not nearly as dominated by non-Asians as the right tail of SAT/Grades is by Asians, in proportion to their numbers.

But I don't think the solution is to have HYPS 40-50% Asian. Justifying this ethically is of course difficult or impossible, but I don't think the Asian minority making a fuss would be helpful. I think drawing attention to the fact of the tremendous over-representation of Asians at top schools would result instead in some backlash, as most people aren't now aware that Cal Tech, Cal, Bronx HS, etc are 60% Asian. Ethnocentrism still holds sway in America at the top level of decision making (which elite school admissions certainly is), just as in all countries.

Ian Smith said...
This comment has been removed by the author.
Ian Smith said...

Argument will continue until the US does what almost every other country in the world does: admit for intelligence alone as measured by standardized tests.

Because the US does not do this its elite is not as smart as that of other countries, and is much pushier, much more "sales-oriented", much more cliquish, and much more homogeneous in world-view.

Unknown said...

This brought to mind a story from my branch chief at NASA's Ames Research Center. He was at UC Berkeley discussing NASA's educational expectations with University officials. They said that the freshman class at UC Berkeley would be 100% asian if admission was based on grades and test scores.
This discussion took place in the mid 1970s. Old Story

Unknown said...

The growing Asian student population also poses a "dilemma" at Berkeley and UCLA. The holistic approach can only do so much to manage the population without using race as an explicit variable. See Tim Groseclose's paper here.
The proposed elimination of SAT subject tests is an obvious attempt to control the Asian student population. It's ironic that just few years ago UC ran an anlysis that showed SAT II to be a stronger factor for collge success than SAT.

Blog Archive