<p>al6200: …just for kicks I put in your latest methodology on a
spreadsheet and played around </p>
<p>IMHO, some questions for you to address in case you write about it
…</p>
<p>a) … the negative multipliers
given to the AP scores implies that doing no APs or getting scores less
than 4 is better than doing APs and getting 4’s and 5’s. …</p>
<p>b) …the implication that GPA is 1/3 as important as the SAT II
Math IIC score and less important than the science SAT II score…</p>
<p>the scientist in you can rise to the occasion and hypothesize to prove
or disprove a and b and accept the veracity of the results…you could write
an essay about how you got all excited only to find the human element was missing and that your forumala though fun was flawed…which is why you
want to really attend so you can learn to come up with formulae that uses
fuzzy logic and op amps instead of …</p>
<p>Yeah, the AP thing is a really puzzler. But, like I said, I’m going to make an update to the whole formula, which will hopefully be more accurate. </p>
<p>I don’t think I said that GPA was less important than Subject Tests. Did you mean SATs? That part of it can be misleading, since my data only refers to SAT scores between 2200 and 2400. I can’t really make predictions for SATs outside of that range.</p>
<p>Not enough people do USAMO for me to draw good statistics. Also, the new method for doing the weights is considerably more complicated then the last. </p>
<p>Any Mathy people have some statistical criticisms? Errors? </p>
<p>I think I’ll probably send this to MIT as a supplement, and I’d imagine that the admissions officers (being from MIT and all), will be quick to spot errors.</p>
<p>well, I’m pretty sure MIT (and also Caltech?) are the only schools that specifically ask for AMC scores. I don’t know about whether they ask for AIME and USAMO scores, but even if they don’t, I suspect they expect applicants to volunteer them if they did well.</p>
<p>Yale has an “other tests” section on the supplement where they ask for AP scores or AMC or AIME or whatever you want to give. There is no specific box that says “Put AMC here” like on MIT’s. :)</p>
But of course this isn’t a method for how admissions officers make their decisions, it’s a correlation between different statistical aspects of the application and the probability of admission.</p>
<p>In the same way, the admissions officers say (and I believe them) that an SAT of 2250 is equal in their eyes to an SAT score of 2400, but I’d be willing to bet a pretty large sum of money that 2400s are admitted at a higher rate than 2250s. It’s not that having a 2400 by itself increases a student’s odds of admission, it’s that other academic factors correlate positively with having a very high SAT score.</p>
<p>Of course not. Schools will understand that most students have never even heard of these competitions, and that it doesn’t reflect poorly on their character. </p>
<p>
</p>
<p>No one from my school ever does Siemens, USAMO, AMC, or AIME, since the students are never told about what is required to get into college. But still, a few of our school’s kids have gotten into Harvard and Ivys, and a significant number have gotten into Hopkins/CMU/Georgetown.</p>
<p>You hit the nail on the head right there. The most recent formula is not perfect, but I still believe that it has some predictive validity. </p>
<p>
</p>
<p>My data shows neither trend. My data in fact suggests that SATs actually hurt a student’s chances for admission, since SATs don’t affect the probability of getting in, yet they correlate to GPA pretty well. </p>
<p>
</p>
<p>That’s exactly what I’m trying to consider in the new method. But its pretty difficult, and requires quite a bit of advanced statistics. The challenge is trying to isolate the real affect that SATs have on chances from their correlation to other factors.</p>
<p>There are good reasons why MIT does not give too great a weight to very high overall SAT scores (beyond 2250). They are simply not great indicators of sucess at MIT. The top SAT scorers are historicaly not necessarily the best performers once admitted. First, MIT SAT scores tend to be more lopsided than other elite colleges: great SAT math, good SAT verbal, average SAT writing. MIT will not penalize a student who is not the greatest writer as much as HYP may, but a low or average SAT math score is not helpful. </p>
<p>On the other hand, MIT does place a lot more weight of other indicators it believes are better predictors such as SAT II scores in math and science and AP scores such as Calc BC or Physics C which are the only APs that are considered direct equivalents to intro calc and physics at MIT. MIT’s education is intense and the school wants to be assured you can handle the academics. </p>
<p>While it is true that most college students may never have heard of the AMC, AIME or USAMO, MIT applicants are highly self-selecting and have definitely heard of them. I would venture to guess that the vast majority of admitted students have taken at last the AMC. It is not by accident that MIT is one of the only schools that list the test on its application. It is because it is COMMON among its applicants. That does not mean that failure to take the AMC will automatically exclude a candidate, but if a student is from a high school where students take the test and they themselves did not, it will at least raise a question. Same thing with AP classes and tests. If most top students at your HS take AP Calc BC and you did not, you have some explaining to do. </p>
<p>As far as how much of an edge a high AMC or AIME score provides, the only thing I know from meetings with adcoms as an EC (educational counselor) , is that an AIME score greater than 7 makes the candidate a “super candidate”. USAMO, Siemens or Intel finalists get even a greater boost and MIT tends to attract and aggressively recruit a disproportionate number of them. A USAMO finalist or Intel finalist is much more attractive to MIT than a 2400 SAT scorer. MIT participates in a lot of collegiate math competitions such as the Putnam as well as many science competitions and wants to perform well at them. MIT may not recruit olympic caliber athletes but it does recruit math and science olympians. </p>
<p>Because of all of these factors, a multiple regression model trying to predict admission chances at MIT is simply not the right approach. Other models that deal with non-continous variables such as Bayesian or discriminant analysis models may be better suited. Under a Bayesian approach, You would essentially first compute a pre-test admission probability based on some academic record (SAT + SATII scores + GPA). The maximum pre-test probablity would be reached at 750/test and an UW GPA somewhere around 3.75. This provides the baseline admission probablity. According to MIT adcoms they could fill their class 3 to 4 times over with academically qualified candidates, so we can estimate the maximum baseline probability will be around 25%/33%. A top SAT scorer with nothing else major to show would therefore have no better than 1 in 3 or 1 in 4 admission chance which is consistent with admission practices at MIT. You would then set up successive Bayesian tests for factors you know MIT values: AP scores, AMC and AIME scores, science competition results,published research etc… While valedictorian status is big factor at some elite schools it appears less so at MIT, but since the MIT President brags about hundreds of valedictorians each year during commencement it must have some incremental value. You are assuming independence between the tests which is mostly true. After each successive test, your admission probablity has either increased or decreased. You probably will never reach 1 except in extreme cases (USAMO Gold medalist or Intel or Siemens winner).</p>
<p>Wow, very informed post. This gives me a lot to think about. </p>
<p>
</p>
<p>I don’t have any statistics on the success of admitted students. But it sounds interesting. I’m honestly surprised by that, I would’ve expected top-scorers to have dramatically higher success rates. </p>
<p>
</p>
<p>Writing scores are still very high (600+), but that seems to agree with the data I have. </p>
<p>
</p>
<p>My data shows that Subject Tests are one of the best predictors of admissions chances. </p>
<p>
</p>
<p>I’m not sure about that, but it seems reasonable. However, many of the brightest students at my school have never heard of it, and there’s certainly no teachers/parents/guidance counselors informing kids about these tests. </p>
<p>I self studied for some AP tests at my school, and some people considered that insane (although I know of a handful of people who have done it). </p>
<p>
</p>
<p>
</p>
<p>That seems sensible. If I was running college admissions, I would look at math contests as a key factor. </p>
<p>
</p>
<p>
</p>
<p>Wow, I’ve never studied Bayesian networks, but a lot of the stuff in my new model seems to be a crude version of the network. </p>
<p>I don’t know what you mean by non-continuous. Do you mean non-independent? </p>
<p>Also, when you talk about a computing a pre-test admissions probability, are you suggesting that this would be done by a multiple regression model?</p>
<p>And finally, I’m applying to MIT this year, and I’m thinking about sending a finished model in as a supplement. Any opinions on that?</p>
<p>Can you send me/us exact conclusions you developed with your data? Why don’t you send your new model to accepted/deferred students (over mail) and see what happens (You can even send it to people you know who got into MIT)? I mean for double-checking. Im sure you wouldn’t want your formula to spit out a 13.37% chance for a guy who got accepted and a 97% for a guy who got deffered. If that’s the case, you might re-do you formula before you send it to MIT.</p>
<p>Noncontinuous is like not a continuum - for example, SAT scores (you can’t have a score of 20pi, for example), AP scores (can only score integers from 0 - 5). It’s like the difference between milk (continuous) and cookies (discrete) (talking about each individual cookie, and assuming half a cookie or a cookie crumb does not count as “one cookie”.)</p>
so you can learn to come up with formulae that uses