<p>The admissions offices themselves don’t dismiss the linear models. The core of the Early Admissions Game study was extremely simple statistical modelling of admissions probabilities – chances go up by X percent per SAT point, Y percent for each unit of GPA, etc. Admissions offices did not ridicule the methodology as naive; they conceded the correctness of the study conclusions, and either defended early admission on other grounds, cancelled it (Harvard, Princeton, UVA), advertised it as an admissions boost (Duke), or kept quiet and hoped for the best. Nobody has ever seriously claimed that the study is misleading for not taking account of intangibles. </p>
<p>Espenshade’s many papers and recent book are based on nothing more than logistic regression, something taught in intro statistics classes. In his case the methodology does have serious problems, but even there it is valuable as a snapshot of the data, and matches up with a lot of published information on how the individual applications are considered. His model takes “context” into account only by adding or subtracting a fixed number of points based on one’s race category, income category, immigrant generation, gender, high school type, feeder school attendance, and many others.</p>
<p>Espenshade had access to internal admission data, but Avery and his colloborators used external data, tracking thousands of applicants through the process. Admissions prediction web sites are doing the same thing with regressions based on self-reported data by anonymous users, same as CC. I see no reason why such models are faulty in principle, and neither did the Harvard or Princeton admissions offices.</p>
<p>Pizzagirl is right. My S’s go to a top private school in Minnesota. I went back on Naviance over the last 2 years. Of the 200 grads per year (of which 100% went on to college), only 20 left the 3 state area and only 4 went to Ivies or Stanford. There is simply not an Ivy league mindset here.</p>
<p>Re: post #772 and not related to the topic.</p>
<p>bovertine - Thanks for the link. UC-Berkeley is on my S’s list and I just learned from that document that they require 3 SAT IIs. Son was planning to take 4, but SAT II Composition (one of the required tests for Berkeley) was not one of them.</p>
<p>But the purpose of the model is to indicate a very high probability to get acceptance from one such school which is still not refutable even after the MIT result.
MITChris does indicate that it doesn’t matter whether an applicant has 700 on SATI math or 800 but the median SAT1 math at MIT contradicts that as it exceed 770.</p>
<p>keylime pie, that’s an old document (2005) from Berkeley. Go to the University of California website for more accurate information about admission requirements. They are in the process of dropping the SAT II requirement completely. I think that takes effect for HS class of 2012.</p>
No. Imagine that 90% of the applicants have SAT math scores over 770. Even if 700 is deemed good enough, they are still going to end up with SAT math scores that average over 770.</p>
<p>mathmom: ^^^ But that is what the model predicts that with a SAT1 math > 770 there is more probability to get accepted at MIT than with a SAT1 math of 700.</p>
<p>That means that 50% of all admitted students have math SAT’s somewhere below 770. We don’t know how much – if you had a group of 99 students, and 49 had scores of 800, and 49 had scores of 500, and 1 student had a 770, the “median” would still be 770. It doesn’t matter how low the tail end of scores go, the “median” is still whatever the student in dead center has. Obviously that is an implausible scenario, but I find it slightly amusing that there are people who are talking about admission standards to MIT don’t seem to understand what a “median” is.</p>
<p>FWIW, MIT publishes its common data set, and we know that 85% of students have SAT math scores between 700-800; and 14% have scores between 600-690; with 1% having a math scores below 600. I’d view that as confirming what MITChris says – obviously he has seen many students admitted with scores in the 600 range.</p>
<p>in re:the 6,000 students nationally with 2300s or above at a single sitting</p>
<p>The implication is that there there are many more “Ivy caliber” students than one would predict using the model, because many more people get in without the stats that the model measures. The model does not include many factors that go into admissions decisions. Therefore the model is not accurate in predicting chances of admissions.</p>
<p>To the extent the model is used for a more limited purpose - increasing the chances of admissions to the HYPed school - all it is doing is assigning numbers in an arbitrary fashion to factors THAT WE ALREADY KNOW increase one’s chances (e.g., if you are a valedictorian, or an Intel winner, or scored 2400 on your SATs, etc.) = faux-scientific verification of the obvious. </p>
<p>The combination - the model is redundant and ignores important admissions factors - renders it close to useless.</p>
<p>Kei</p>
<p>P.S. But , hey, go for it . . . I suppose there are folks out there that really want to know that getting 30 points on the model makes them “Ivy caliber”,or that by having more of the quanitfied accomplishments their chances for admissions go up. </p>
<p>Will that change behaviors? Will one now apply to more Ivies with a 63 . . . or fewer if they have a 33? Or neglect safety schools with a 67?</p>
<p>POIH said: “mathmom: ^^^ But that is what the model predicts that with a SAT1 math > 770 there is more probability to get accepted at MIT than with a SAT1 math of 700.”</p>
<p>No, we already knew that, before the model existed . . .what value does the model add?</p>
<p>Yes. I was just using it as an example of somebody modeling admissions. It’s always best to check up-to-date admissions info. Sorry if I confused/panicked anybody.</p>
<p>Just thought I’d point out that this, in my opinion, is the most important and useful comment here. </p>
<p>Once one establishes how an Ivy Caliber applicant generally is as a person (the person that gets the prestigious awards, the 2400, the 4.0, etc.), then one can have a complete model. This is so because, when combined with the other high achievements, the recs that the student receives will reflect what it is that makes this student as great as s/he is and, when placed with the other high achievements of said students, can allow one to understand (generally) who is admitted. Sorry if that made no sense (I didn’t expand upon some of my points, so yea…) but I wrote it rather quickly and didn’t think into it much.</p>
<p>And this becomes a tautology – higher achieving students are more likely to be admitted to Ivies (etc). In other news, the sun rises in the east and sets in the west.</p>
<p>These are what I call horizontals in my business – and they are useless unless you know the profile of applicants. </p>
<p>If 20% of MIT applicants have SAT math scores of 700-800, then the 85% of accepted or matriculating students having SAT math scores between 700-800 is meaningful, because they are overrepresented dramatically.</p>
<p>If, however, 80% of MIT applicants have SAT math scores of 700-800, then the fact that 85% of accepted or matriculating students having SAT math scores is a meaningless number.</p>
<p>It’s rather like saying that 55% of all crime happens between Jan and June. Well, 50% of the year happens between Jan and June, so it’s not a meaningful statistic. You need to know your base before you can draw any conclusions.</p>
<p>As was explained already, several schools publish this data, including MIT. You really should have a look. Performing the obvious web search and sorting out the US rates from those of internationals (e.g., by estimating that 85+ percent of internationals submit SAT math scores 750 or higher), one finds that the admission rates for US applications are, approximately:</p>
<p>The differences by SAT tranche would become more pronounced if the data were further separated into US white/Asian males, US females, US black applicants, etc. </p>
<p>Also note that these figures are not particularly consistent with the idea of imposing a cutoff and then selecting on other factors. One does see a cutoff pattern with grades; valedictorians were admitted at about the same rate as ranked applicants in the top 5% of class, according to numbers from MITChris. For SAT there is no such levelling off, indeed the admission rate appears to increase more rapidly as the scores escalate. </p>
<p>See the graphs of MIT (and Princeton, Harvard, etc) admission rate by SAT for a sample of domestic applicants from 10 years ago in:</p>
<p>This was also seen in Espenshade’s measurements of the influence of SAT scores (mostly pre-1995) in his regression studies and his book. There isn’t a flat cutoff, the SAT effects increase at the high end.</p>
What would you expect the math SAT scores to be of persons who have won math competitions and have taken highly advanced and accelerated math courses? I think it’s not clear what is cause and effect in this analysis.</p>