Reconstruction of MIT 2009 domestic (USA) admit rates by SAT score.

siserune · March 17, 2010, 4:45am

MIT posted at its website admission rate data for all 2008-9 applicants together, US and international:

[MIT</a> Admissions: Admissions Statistics](<a href=“http://www.mitadmissions.org/topics/apply/admissions_statistics/index.shtml]MIT”>http://www.mitadmissions.org/topics/apply/admissions_statistics/index.shtml)

Because admissions are disparate for Americans and internationals, I made some estimates of the US-only numbers based on the overall admission rate in each math SAT tranche. The method could be refined or applied to some of the other tables MIT provides, but the basic message is clear: combining US and International pools in the reported tables downplays the apparent sensitivity of admission to (math) SAT scores.

MIT’s data for 2008-9:

Nobody was accepted with math SAT below 600 (576 applicants).

These numbers indicate that SAT strongly correlates with admission, but the effect is masked by the presence of internationals. US applicants are admitted at four times the international rate, but the international pool has higher math scores, and the combination will depress the relative admission rates in the upper ranges of the table compared to the lower ones.

The acceptance rate for US applicants was 12.9% (1552 of 12026), 3.8 times higher than the internationals rate of 3.38% (123 of 3636). We recalculate the above table for Americans under the following assumptions:

<ol>
<li>All internationals submit SAT. (Errors in this assumption are suppressed in the end result by factors in the range of 10-20 as long as the percentage submitting SAT is relatively high, so this is not a major problem). </li>
<li>All internationals have math SAT score 700 or higher, and between 85 and 100 percent have score in the 750+ range. That is, the international pool is heavily stacked toward the upper SAT tranche. (As we will see shortly, separating internationals affects the rates in MIT’s table mainly in the 750-800 range so the only assumption with any force is what percentage of internationals are in that stratum. If a substantial fraction of the below-750 internationals are really below-700 it would not change the conclusions.)</li>
<li>US-to-international acceptance ratio is, in each range, the same as the overall ratio for MIT. </li>
</ol>

Assuming this, I get the following numbers for the US applicants for each possible percentage of internationals in the highest math SAT range of the table. R code for the calculations will be posted on request (no private messages, please, just ask in the thread).

%intl rate750-800 rate700-750 
85 22.3 14.0 
86 22.4 13.8
87 22.5 13.7
88 22.6 13.5
89 22.8 13.4
90 22.9 13.3
91 23.0 13.2
92 23.1 13.0
93 23.3 12.9
94 23.4 12.8
95 23.6 12.6
96 23.7 12.5
97 23.8 12.4
98 24.0 12.3
99 24.1 12.2
100 24.3 12.0

We see from this that the acceptance rate of US applicants in the high math SAT range is around 22-24 percent (not the overall 15 percent when including internationals) but the other figures don’t move as much. According to this calculations, the acceptance rate for the top tranche is about 2 and 3 times better than the rates in the second (700-740) and third (650-690), not ratios of approximately 1 and 2 as in MIT’s aggregated figures..

This suggests that SAT scores above 750 (particularly, scores of 800) are more correlated with acceptance, and more causal of it, than MIT admissions officers have said in these boards. This would, one assumes, be even more the case for non-minority, male US applicants, because MIT’s gender-balancing and affirmative action policies lead it to accept at higher rates from SAT math score distributions that, for women and minorities, are located a bit lower than that of white and Asian males. The statistical effect of this is, like the inclusion of internationals, to squeeze the acceptance rates of the different SAT tranches closer to each other.

jessiehl · March 17, 2010, 9:41am

Am I not fully awake yet this morning? I’m missing where this data indicates anything whatsoever about causality when it comes to high SAT math scores.

Do you have a cite for that, or are you making an assumption?

(Note that even if you had a cite that more women and URMs were accepted with lower SAT scores, it still doesn’t follow that this was because they were women or URMs. But as things are now, you haven’t even provided that cite.)

siserune · March 17, 2010, 10:45am

As it happens, yes, I do have citations.

People who sat on Caltech admissions committees have confirmed in other discussions (here) that their female applicants are, as one might expect from the national data, weaker than the male applicants by the usual objective measures such as test scores and math/science competitions, but stronger on some subjective factors such as self-selection, perceived maturity, focus, or clarity of goals.

Similar reasoning is used in MITChris’ discussion of minority applicants to MIT and their admission rates, in his recent postings (these applicants are really, really, really focused on being nuclear physicists, etc).

In addition to those postings, which are repeating obvious information but are URL-locatable if you want to pursue the issue, I have read and heard any number of comments from MIT personnel, including some on CC, that are essentially the same as the positive half of the Caltech remarks about the supposedly decisive intangible or subjective strengths of the female applicants at MIT, with this advanced as a full or partial explanation of the gender disparity in admission rates. I can’t say that my memory of any of the last five times I’ve heard that one is as specific as for the first two items or that I could locate such an instance on request, but it is, again, pretty much the party line from admissions when pressed on the issue. Caltech is just more forthcoming, so far.

At any rate, the upshot of those sociological observations, if you take admissions people at their word, is that women at any given SAT range will, on average, have higher admission rate than men with the same SAT’s (since in comparison to those men they tend to also have other subjective attributes desired by Admissions and that Admissions says are influential — attributes that may or may not include being female). This is all just the standard statistical assertion that if your probability of selection is positively influenced by two factors A and B (e.g., SAT and maturity, or SAT and female gender), then having A and having B will be negatively correlated among those with any given probability of selection. The application of this principle to the MIT admissions table is as I said: if admission rate, for both men and women, rises with SAT, then mixing the two pools together will tend to mask the effect of SAT on admission rate within each gender separately.

jessiehl · March 17, 2010, 11:08am

Who cares (for purposes of this topic, anyway) what people from Caltech admissions said about their applicant pool? The Caltech applicant pool and the MIT applicant pool are not the same thing. By a cite, I mean some actual numbers from a reliable source indicating that women and URMs accepted to MIT have lower SATs than others accepted to MIT. Or even an MIT admissions officer saying that this is the case, without providing numbers. This may indeed be the case - I don’t actually know - but you’ve not provided evidence of that.

MITChris did discuss the self-selection factor. The self-selection factor doesn’t necessarily mean that people’s stats were weaker. Sometimes it means that they are stronger. MITChris and others have also brought up the “strong intangibles” thing. Again, that doesn’t actually mean that those people’s stats are weaker.

And I’m still not seeing the causality thing that you claim in your original post. I’m not questioning your point that SAT correlates with likelihood of admission, or your point that international data partially masks this. But where’s the causality that you’re purporting?

MITChris · March 17, 2010, 11:13am

@siserune -

Some of your assumptions are incorrect.

As for our treatment of SAT scores, see our website: [MIT</a> Admissions: Standardized Test Requirements](<a href=“http://mitadmissions.org/topics/apply/standardized_test_requirements/index.shtml]MIT”>http://mitadmissions.org/topics/apply/standardized_test_requirements/index.shtml)

As I have posted several times in the last several days, we look at SAT scores as predictors of success or causes for concern, not as qualifiers for MIT in and of themselves.

Since you’re a fan of statistical analysis, you’ll be happy to know that’s what drives our treatment of scores. Statistically speaking, there are SAT score ranges (on each individual test) that indicate a great likelihood of success at MIT and those that indicate a great likelihood of trouble at MIT.

The break point exists somewhere between the high 600s and low 700s depending on the test. Below that, and the record shows the student would likely struggle here; above that, and students succeed with statistically insignificant differences as scores increase.

These scores are then considered with grades and coursework as additional predictors of success or causes for concern. For extreme examples, As in multivariable calc is a
predictor of success; Cs in precalc (as a senior, with no calculus taken) is a cause for concern.

Once you clear the success/concern bar, we don’t really care about scores anymore, and we consider grades as part of your context (how much initiative have you shown? is your coursework especially outstanding? etc).

As for your citations:

Nowhere have I said that our women or URM applicants have lower scores than other applicants. I have not said this for two important reasons.

<h1>2 - Even if it were true, it would be immaterial. A student who scores a 2300 on the SATs is not “less qualified” than someone who scores a 2400. They are equally qualified. They have both cleared the break point, and thus, when making decisions about “qualifications”, we look elsewhere for dispositive data.</h1>

I hope this helps clear up any confusion!

MITChris · March 17, 2010, 11:14am

@jessiehl

You are correct - “self-selection” does NOT mean that applicants are weaker. In fact, the converse is usually true: self-selection means that applicants are STRONGER than the average.

Kei-o-lei · March 17, 2010, 11:26am

Do I have this right:

overall, higher SATs are accepted at a higher rate than lower SATs?

Anything else?

What am I missing here?

Thanks.

Kei

nemom · March 17, 2010, 11:45am

Perhaps it would be helpful to remember (correct me if I am wrong, MITChris) that , basically, applicants are either qualified or not in terms of SAT scores and (probably) GPAs. So, although the very highest SATs and GPAs may also count in the admissions process, they don’t indicate ‘more qualified’ because there is no such critter.

metapod · March 17, 2010, 12:52pm

You are right. But that doesn’t mean high SAT scores cause an increased chance of admission. It could mean that people with very high SAT scores are more likely to have a stronger application overall; the kind of person who is really serious about going to an elite school might learn the test inside and out, so that their scores don’t disqualify them. That doesn’t mean the scores themselves are important past a certain point. They aren’t.

collegealum314 · March 17, 2010, 1:38pm

concerning Caltech:

they don’t practice AA, gender or otherwise
even if girl admits have lower stats on average than guy admits, this doesn’t imply preference. It could just mean that the superstars admitted were overwhelmingly guys and girls tended to be middle-of-the-pack (of the admit pool, not the applicant pool) in terms of stats. I’m not saying this is actually the case. I have no idea. I am just saying this is one scenario in which girl admits could have lower stats than guy admits without any preference.
people have explained the phenomena of self-selection, so I won’t go over it again.

People need to think more about statistics and what they really mean (and don’t mean.) Before they go off on a diatribe, they should make sure there isn’t an alternative scenario that is supported by statistics that would invalidate their conclusions.

silverturtle · March 17, 2010, 1:44pm

Can you specifically share some of this data? It is extremely difficult to grasp that students at MIT who scored 2400 on the SAT do not do better on average than those who scored around 2100.

PaperChaserPop · March 17, 2010, 1:56pm

MITChris, thanks again for sharing admissions information. The information on the “break point” of SAT scores is especially valuable.

The above example on grades is as you said, extreme; it does not narrow down your break point like the excellent example you provided on SAT scores. Would you be able to give examples on grades that are less extreme and more indicative of your actual break points on grades?

MITChris · March 17, 2010, 2:02pm

@silverturtle -

I don’t have the numbers in front of me, but the differences are so small as to not be dispositive. A student with a 2400 is a bit less likely to struggle (broadly defined - low grades, dropout rate, etc) than a student with a 2100, but we’re talking a few percentage point difference here, not a matter of life and death.

Scores are just part of our prediction analysis. That’s what standardized tests were designed to do.

MITChris · March 17, 2010, 2:05pm

@PaperChaserPop -

Unfortunately, no. Grades are really complex, not in the least because we deal with a half-dozen different grading systems. While we feel comfortable evaluating an applicant in their context, it’s impossible to give guidelines that are generally helpful.

What we say is this: try to get mostly As in mostly the most difficult classes you can take. More specific than that and we need a whole lot more contextual information to evaluate the break point.

silverturtle · March 17, 2010, 2:09pm

Are there any significant differences in the predictive values of the ACT and SAT? Also, what is the point at which higher SAT Subject Test scores no longer indicate a higher chance for success?

Thanks.

PaperChaserPop · March 17, 2010, 2:15pm

MITChris, I’m a bit disappointed, but thanks for the quick response.

Does MIT have more or less established “break point” on grades for high schools that regularly send in more than a couple kids a year?

PaperChaserPop · March 17, 2010, 2:27pm

Does your data further filter on their GPA’s? For example, is difference more pronounced in the likelihood to struggle between a 2100 with 3.7 GPA versus a 2400 with a 4.0 GPA (assuming the two are from high schools of similar rigor)?

sorghum · March 17, 2010, 2:30pm

There is no logical way that they are equally qualified. If you knew nothing else about 2 applicants, and took only one, of course you would take the 2400.

Kei-o-lei · March 17, 2010, 2:53pm

sorghum said: “If you knew nothing else about 2 applicants, and took only one, of course you would take the 2400.”

Exactly!!! In other words, if SATs were THE ONLY admission criterion you would be right.

But you know that is not the case; the are both equally qualified because they are both over the threshold to do the work AND to go further in the admissions process.

And as a practical matter, any institution saying that the ~300 students who get 2400s are qualified and the other ~5,700 or so who score 2300 or above would e limitd to an extremely small entering class :-)

Kei

silverturtle · March 17, 2010, 3:20pm

Yes, being qualified should be thought of on a continuum, not as an absolute threshold as MITChris seems to be suggesting is MIT’s policy.

Reconstruction of MIT 2009 domestic (USA) admit rates by SAT score.

CONNECT WITH US