Statistical Correlations for Chances at MIT

Colleges and Universities A-Z Massachusetts Institute of Technology
1

<p>I compiled all of the statistics from the MIT early decision results thread, and drew together some correlations between various stats and acceptance rates. Here are the results: </p>

<p>SATs: </p>

<p>.32</p>

<p>SAT Subject Test Math Level 2: </p>

<p>.86</p>

<p>SAT Subject Test Highest Science: </p>

<p>.64</p>

<p>Unweighted GPA: </p>

<p>.97</p>

<p>Number of APs passed with 4 or 5: </p>

<p>.48</p>

<p>While this is all very interesting, I should point out a couple of things: </p>

<p>-I ignored deferred applicants while doing the calculations</p>

<p>-Data set was extremely limited, only 17 students</p>

<p>-Not a random sample, only based on those who post on CC</p>

<p>-Only a very limited range of GPAs and SAT scores were seen (very high, of course), so conclusions are difficult to draw. </p>

<p>But with just this data, using a very rough around the edges method, I created a formula for figuring out your chance at MIT! </p>

<p>Step 1: Take your SAT score (out of 2400), and multiply it by .000846, then subtract .9888127 from that number. Then multiply that number again by .0978593</p>

<p>Store this number</p>

<p>Step 2: Take your Math Level 2 score (out of 800), and multiply it by .0052432, then subtract 3.1642369 from that number. Multiply that number by .2629969</p>

<p>Store this number</p>

<p>Step 3: Take your highest Science Subject test score (out of 800), and multiply it by .00343525, and subtract 1.6776978 from that number. Multiply , that number by .1957186544</p>

<p>Store this number</p>

<p>Step 4: </p>

<p>Take your unweighted GPA (on 4.0 scale), then multiply it by 1.619754, and subtract 5.4597 from that number. Multiply this number by .296636</p>

<p>Store this number</p>

<p>Step 5: </p>

<p>Take the number of AP Tests you've gotten a 4 or a 5 on, and multiply this number by .0467625899, then add .726618 to that number. Multiply this number by .1467889908</p>

<p>Store this number</p>

<p>Step 6: </p>

<p>Take the numbers from all the previous steps and add them together, then multiply by 100. This is the percent chance of you being admitted to MIT. </p>

<p>Since most of you reading this have probably applied to MIT, I bet you have some nifty suggestions on improving this method to get better results. The biggest problem is see with it today is that it fails to consider how SATs, GPA, and AP tests all overlap each other, and how GPA might be the really important factor, but GPA only seems important since it correlates to SATs. </p>

<p>I'm currently thinking about some ways of getting around this (like reducing the weight of one factor if it correlates really highly to another factor, but less to the admissions decision).</p>

2

<p>Using this method, I myself have a 66% chance of getting in, which seems way too high. Oh well.</p>

3

<p>sweet according to your formula i have a 90% chance of being accepted to MIT...</p>

4

<p>The problem is that this method doesn't consider context. Most schools organize applicant information by school/district/region. For example, if 35 students from Palo Alto's Gunn High School applied to MIT, all of them might be considered together. Although they might all merit admission based solely on scores and GPA, it's very possible that the admissions staff might want to take only 2 or 3. Top schools like MIT could fill their entire freshman classes with students from only 100 high schools in New York and California, if they wanted. But no school wants such a narrow slice of the population. </p>

<p>Admittedly, I'm an MIT parent, not on the MIT admissions staff. However, as a professor at another institution, I've participated in admissions decisions occasionally, and this experience informs my perspective.</p>

5

<p>What do you mean by "correlation" between a stat and whether or not you are accepted? Correlation requires some sort of linear relationship, no? And "getting accepted or not getting accepted" is categorical. Can you describe your methods?</p>

6

<p>why do i get a 104%? maybe its a calculation error. oh well, im too lazy to go back and check.</p>

7

<p>I think my method so far over-estimates student's chances a bit, so I'm going to integrate last years RD results with the rest of the data to make it fit a little better. </p>

<p>CalAlum: I didn't know MIT looked at regions so much. But nevertheless, I think the correlation between admission percentage and SATs/other factors is sufficient to draw good conclusions. </p>

<p>And nonetheless, I really made this formula as a rough predictor, not an attempt to take all possible factors into consideration. </p>

<p>@Lalaloo6: </p>

<p>My method was to draw a line of best fit correlation between admissions factors and whether the student was admitted or not. The formula plots those students factors on each of the lines of best fit, and uses a weighted average based on correlation to mesh those different results together. So in other words, if SAT Subject Tests have twice the correlation of regular SATs, they get twice the weight. </p>

<p>Getting accepted or not accepted isn't really categorical, it's either a 1 or a 0. It follows the logic that others who had a 2320 on their SATs had say a 10% chance of admission, so you should have a similar chance. </p>

<p>@everyone: </p>

<p>With that said, I'm going to take in the RD stats from last year, since my formula REALLY over predicts chances (maybe by 50% or so).</p>

8

<p>@tongchen, </p>

<p>My data set was very limited so it tended to over-predict results, but there's no error in getting 104%. It just means that overall you're more qualified than the average student from my data set who got accepted to MIT. </p>

<p>But even though it overpredicts, 104% is still pretty darn good.</p>

9

<p>So yes admitted = 1, not admitted = 0? Note that then you're assigning values; it's still categorical - there are two categories, namely, not admitted and admitted. Values that you ASSIGN can make it "numerical" in some sense.... but that's like saying SAT scores "correlate" to gender, and making male = 0, female = 1. Gender is still categorical, no matter what values you assign.</p>

<p>But that's pretty fun :-)</p>

10

<p>haha, i think im one of your 17 "data" to begin with. i was just curious as to how amazing your formula was.</p>

11

<p>Also... if answers can be over 100% it can't really be called "chances" of getting in, as a probability obviously can not be greater than 1. Maybe then you can think of it more as "how qualified you are compared to the average accepted CC poster." Lol.</p>

12

<p>Okay, I redid my calculations, with over 60 students from last year's regular decision thread. </p>

<p>Let's just say that there's a large negative correlation between taking AP Tests and getting in. Ughhhhhh... There's no correlation between SAT scores and getting in. </p>

<p>The only factors that really seem to matter are </p>

<p>Then again, I should put bounds on my answers, since I haven't seen results for people who have really low SATs, so I'd say: </p>

<p>For SAT scores from 2100 to 2400, since I don't have good data on students outside that range. </p>

<p>But this is really intriguing. A 2400 is no better than a 2200, assuming people on CC are honest and vindictive of the general trend. I'll have a new formula up within the hour.</p>

13

<p>Yup 102.5% chance. Got deferred.</p>

14

<p><strong><em>ATTENTION: UPDATED CHANCES FORMULA</em></strong></p>

<p>Here's my updated data, taken from this years early decision thread and last year's regular decision thread. The stats from over 73 students were sampled (I really have no life). </p>

<p>Correlations: </p>

<p>SATs (all 3 sections, super scored): </p>

<p>.002997</p>

<p>Math Level 2 Subject Test: </p>

<p>.2475</p>

<p>Highest Science Subject Test: </p>

<p>.1191</p>

<p>Unweighted GPA: </p>

<p>.302</p>

<p>Number of AP Tests passed with 4/5: </p>

<p>-.0531 (Doing well on AP Tests actually hurts your chances)</p>

<p>Ranking of most important factors: </p>

<ol>
<li> GPA/Class Rank</li>
<li> Math Subject Test</li>
<li> Science Subject Test</li>
<li> SATs</li>
<li> AP Tests</li>
</ol>

<p>As for SATs not correlating to chances, I very honestly believe that I did my calculation correctly, however nearly every student who posted had SATs in the 2200+ range, and very little data existed for students with SATs below 2100. This is interesting, since it seems to defeat the myth that there's a big difference in college's mind between a 2300 and a 2400. </p>

<p>I have no idea why the AP relationship exists. Its possible that the kids who took no APs went to unconventional schools, and therefore stood out. But this is a very interesting result. </p>

<p>Anyway, onto the formula (I provide a whole lot of precision with the decimal places, it's somewhat unnecessary, depending on how precise you want your final chance to be): </p>

<p>Step 1: </p>

<p>Take your SAT score (out of 2400), and multiply it by .0000098545 and add .69614, take this number and multiply it by .0048456</p>

<p>Store this number</p>

<p>Step 2: </p>

<p>Take your Math Level 2 SAT Subject test score (out of 800) and multiply it by .002628 and then subtract 1.325722243, then multiply this number by .4</p>

<p>Store this number</p>

<p>Step 3: </p>

<p>Take your highest Science SAT Subject Test score (out of 800) and multiply it by .001054, then subtract .076445 from that number, then multiply this number by .19256</p>

<p>Store this number</p>

<p>Step 4: </p>

<p>Take your unweighted GPA (out of 4.0), and multiply it by 1.265733423, then subtract 4.270284 from this number, then multiply by .4883224064</p>

<p>Store this number</p>

<p>Step 5: </p>

<p>Take the number of AP Tests you've gotton a 4 or a 5 on, and multiply by -.0091718924, then add .7635234132 to this number. Then multiply by -.0857756388</p>

<p>Store this number
Step 6: </p>

<p>Add the results from steps 1-5, and multiply by 100. This is your percent chance of being admitted into MIT.</p>

15

<p>Use the updated equation, it takes into consideration the results from over 73 students. Its much better, I get a 51% chance of admission. Still too high. I'm going to continue looking through it to find errors and stuff. But I think its a pretty decent formula, once I get that AP thing figured out (why higher AP scores hurts your chances).</p>

16

<p>Lol 127.4% now. And I've gotten 8 AP exams with scores 4 or 5.</p>

17

<p>al6200, could you please post / upload the data you are using? I'd like to experiment with this myself;)</p>

<p>Thanks :)</p>

18

<p>Did you use the updated formula, balladechina? </p>

<p>I think part of the problem is that accepted people are much more likely to post in results threads than the rejected people, and only perfectionists even bother to go on CC without tearing their hearts out in shame and humiliation (a little over the top, I'll admit), so my results aren't a true random sample. </p>

<p>Are SATs meaningless? Of course not, but for the people who use CC and post in decisions threads, they have no bearing on getting into college. That really shocked me. </p>

<p>Also, do people lie in decisions threads? Like, positing "2400, 4.0, 1st in class, rejected, GOD I HATE MIT" just for the fun of it. Cause if they do, that would explain a lot.</p>

19

<p>At Sgreben, I've been doing the regressions on my calculator, and the data comes from last year's regular decision results thread and this year's early decision results thread. </p>

<p>I'll put the data into an excel file pretty soon, which I can upload, but its a lot of data (73 students). But like I said, I have no life. If you're looking to play around, please share your results, since I'd be interested to see more viewpoints on this. </p>

<p>It feels like a lot of people assume 2300 is much worse than 2400, but my data suggests that the advantage of 2400 is relativly small. There is some data to suggest than a 2300 is only slightly better than 2200, but there is little data showing the performance of 2100 or lower, since very few people apply in those brackets.</p>

20

<p>Whoops, miscalculation on the GPA part. 78.6%</p>

<p>But still...I think that's too high.</p>