What is the profile of an "Ivy caliber" applicant?

siserune · March 19, 2010, 2:48am

Modelling CLASS admission is far easier and more accurate than modelling of individual admission probabilities. The papers and book written by Espenshade and his grad students at Princeton are an example. They constructed a very simple model of individual admissions, ignoring subjective data like transcript, essay, interview, and oboes, and merely assigning fixed numbers of points for SAT score interval, race, athlete or legacy status, US citizenship, and other objective factors available in their data set. They then found that when you run the individual-level model on the whole applicant pool, the predicted class composition closely approximates the true one. This method also reproduced qualitatively, and in some ways quantitatively, other known effects such as affirmative action and legacy preferences.

Those and other quantitative studies — all of which, so far, have used simple point systems and very basic statistical methods — lend credibility to several ideas:

<ol>
<li>Simple models of individual admission probabilities are reasonable and feasible to construct.</li>
<li>Oversimplified models are closer to the truth than admissions offices would like to acknowledge. This is true whether or not admissions uses a point system or explicit algorithm of any kind. It is quite possible that the process itself is complicated, random and subjective but can be accurately caricatured by simple algorithms that tell most of the truth in most cases.</li>
<li>There is a meaningful relationship between data and prediction at the class level, and the individual level. For example, parameters in a model relating year-to-year changes in math SAT distribution to the class composition at engineering schools, may also be useful as parameters in a model of individual admission to those schools, and vice versa.</li>
</ol>

PaperChaserPop · March 19, 2010, 10:04am

Go for it! That’s how I envisioned we can collaborate even with competing models! I’m absorbing the information gained since I posted my last model and haven’t had the time to update it yet. My plan is to put out a more abstract and less variegated list first to get some comments before making it more concrete and variegated again.

Being the third oboist is not going to get you a strike out unless the app is overall weak, or the app fails the minimum bar on some key metrics, or like I said before, the reader is extremely unscrupulous. Assuming the app is a bit weak overall but comparable to the first oboist who continues to move forward in the process, then it goes back to what I said earlier – “such situations are the exceptions, not the norm. I would expect no more than a small percentage of students get in this way.” Small enough not to be a concern to the model which never intends to capture every case.

Excellent example of case 2 of my list of EC’s that “stand out” (post# 804).

epiphany · March 19, 2010, 11:20am

Okay. This weekend.

(Just a postscript to siserune’s last post: I actually don’t doubt there’s a predictability to it; it’s just that there might not be a “science” to the algorithmic construct. FWIW, over the years on the Accepted Threads on CC, from my join date on, I have been surprised exactly three times: 2 for acceptances, 1 for a reject. That’s a pretty good percentage of predictability; it’s just that I didn’t arrive at it ‘algorithmically,’ at least not consciously.;))

EDITED to add: What I’m referring to is when (naturally) I had extensive info about the applicant prior to the results postings. However, in addition, results posted without disclosures but only after the fact were similarly non-surprising.

calmom · March 19, 2010, 8:12pm

That’s where I think you misunderstand the process of highly competitive admissions. I think that at an Ivy, the applicant really needs to strike some sort of “yes” chord with each reader, at every step of the way. Keep in mind that their job is to weed through thousands of applications and pick out the less than 1 of 10 who will be offered a spot in the forthcoming class.

Let’s assume for purposes of argument that the patently unqualified (SAT’s below 500, GPA’s below 3.0, etc.) – have been weeded out at the outset – and the strongly hooked, definite-admit candidates have also been sorted from the pack that goes to admission readers. (At some schools the director of admissions has the authority to make an admit decision from the get go). So let’s hypothesize that there is a pool of remaining candidates, all of whom meet basic qualifications, that are going through the reading process – and that among this pre-culled pile, there is now a 1 out of 4 chance of admission. That is, the readers will read 4 times as many apps as they have spots to fill and they know that. (I think this is a generous assumption – it assumes that 60% of applicants are either automatic rejects or acceptances – but it will serve for purpose of illustration).

When a reader marks a file a “maybe” that’s not going to be good enough. Let’s say that among these remaining applicants, the readers can “grade” the files giving an A, B, or C… (As noted, the A+ apps and the D’s & F’s have already been weeded out). Let’s also assume that the grading is subjective – an “A” simply means that the reader really likes that candidate and would like to see him or her admitted. A “C” is what the reader will give to any app that simply doesn’t stand out from the others.

In most cases a “C” isn’t going to be good enough – unless other readers substantially disagree. So two C’s are the equivalent of a strike-out, not because that applicant has been “rejected” - but because given the level of the competition, admission is going to require at least a B+ average.

Again, it doesn’t have to be the first chair, first in the door oboist. But it has to be the application that in some way seems better than the previous 3 that the reader has read that day (on average). Partly that stems from the quality of the applicant, partly that stems from the quality of the paperwork that has been submitted in support the application, and partly it depends on luck. If it is the end of the day and the reader is getting tired, or the reader has stuck too many apps in the “recommend admit” (A) pile and not enough in the “let’s pass” (C) pile – it may be tough going.

It’s not a “crap shoot” but it is a process where the applicant has no way of knowing the metrics that will be in play at the time of evaluation – because it is not about what the applicant presents, but what every other applicant remaining in the pool presents.

OldbatesieDoc · March 19, 2010, 8:17pm

The point is, that even with great stats, and “heart” and even a few unique properties, there are some factors of randomness in play, ergo the term: “crap shoot”

Kelowna · March 20, 2010, 1:56am

How do you think a true renaissance kid would fit a profile of an “Ivy applicant”?
By renaissance kid I mean someone who is truly good in almost anything they touch.
Great writer (how do you prove that) great in math/sciences(again, what is the prove), a fine artist. A kid that succeeds in anything that comes his way.
I know that there are a lot of kids out there, kids that CAN be great in almost anything they
try.
How are they going to be viewed? What kind of proves do they need to show?

OldbatesieDoc · March 20, 2010, 5:00pm

I have such a child. He got wait-listed everywhere he applied.
He won awards for his writing and editing of our student newspaper-best editorial in NYS for a paper at a school in his division, best editorial page, and best paper.
He was student government president, won a national award for a computer sym contest and several thousand dollars for the United Way in our town.
He also has perfect pitch, plays the piano, and composes.
He has a side business fixing computers.
He got 800 verbal 760 math and 690 writing , which we thought was odd. He had all 740 plus on his SAT2s
He ran varsity XC and track and went to sectionals.
He had a 4.0 in IB at our unweighted non-class rank public HS in NY, with 3 6s(History, English, and Theory of Knowledge) 2 5s(Math and French) and a 4 in Bio
He is white, not poor, not a legacy.
It’s a jungle out there.
The happy ending is that he got off the Midd wait list and is really thriving there.
I know I tell this story over and over, but I mean it to be instructive.
We didn’t groom him, create a fabulous packet, or send him to special camps or summer programs.He spent his summers playing, going to scout camp, and watching his sister.
We thought he could get in anywhere all by himself.
It all worked out, but if I had to do it again, I would do everything I could to “package” my child, because even with those stats, he just wasn’t “special” enough in today’s top college market.

Pea · March 20, 2010, 8:35pm

Great post OldbatesieDoc. We too ended up with only one really good choice. I know all you need is one but it was cutting it a little too fine. If I had it to do all over again we would look harder for schools that were a good fit but not with the really prestigious name because I think it is just too hard to get into the creme de la creme anymore.

With your son’s stats he should have gotten in everywhere he applied and 30 years ago that is what would have happened. Not today.

siserune · March 22, 2010, 1:13am

That’s interesting. Of course, one interesting thing is that rising from 88% percentile to 100% Harvard’s admit rate only goes from 10% to 20%. Still only a 1/5 chance even with perfect SATs.

[/quote]

No, the chances at Harvard are in the 40-50 percent range for applicants with perfect SAT. The 1-in-5 admission rate seen in the Revealed Preferences data is for scores in the top percentile, which on the current SAT is the 2290-2400 range. This implies quite a strong sensitivity to SAT (or things that correlate so well with SAT that they are functionally the same as SAT) at the high end.

bovertine · March 22, 2010, 8:20am

You’re going to have to esplain this to me. The chart I’m looking at shows a 20% level right about the 100 percentile (akthough it could be 99 I guess). THe accompanying text says this (for Harvard) -

“THe probability of admission rises from close to zero at the 88th percentile to about 10 percent at the 93rd percentile. It then rises very gradually to the 98th percentile and finally rises steeply to 20 percent.”

Are you saying that this 20% actually refers to the 99th percentile?

Where do you find this big discontinuity from 20% at the 99th percentile to 50% at the 100 percentile? Is that information/data actually in here somewhere? I’ll admit I just glanced at this.

siserune · March 23, 2010, 9:17am

There is no 100th percentile on the SAT, because a substantial number of individuals attain the maximum score. People with 1600 would have outranked more than 99 but less than 100 percent of the test takers, which is why College Board tables list the upper scores as “99+ percentile”. The Revealed Preferences data only included about 200 applications to MIT, and to plot the graph with that few data points they had to aggregate ranges of scores into buckets called 96, 97, 98th, 99 and 100th percentiles. The last interval is anyone above the 99th percentile in the College Board tables.

The admission rate for perfect SAT scorers is periodically disclosed for Harvard and its peer schools, and is historically in the 40-60 percent range. The merit of the Revealed Preferences data set (not the work on ranking!) is that it also measured the admission rate at other SAT ranges and one can see the huge effect even of moving from 1500-1590 to 1600, or 700-790 to 800 on the SAT-2 subject tests.

Here’s what the Revealed Preferences data set disclosed for applicants to Ivy League, MIT and Stanford. In a sample of about 1400 applicants (with top 10-20% class rank from good high schools) sending 3300 applications to those schools, rates of admission were:

34% 1400-1490 SAT, 700-790 SAT-2 average
50% 1500-1590 SAT, 700-790 SAT-2 average
71% 1500-1590 SAT, 800 SAT-2
74% 1600 SAT, 700-790 SAT-2
84% 1600 SAT, 800 SAT-2

The admission statistics on school web sites, and the SAT/ACT regression coefficients in Espenshade’s admission studies, also show the steepening and nonlinear effect of testing percentile on admission.

One consequence of this is that the admissions office mantra that test scores don’t matter all that much, is a load of polite nonsense.

bovertine · March 23, 2010, 11:19am

Yes, obviously I understand this. As I understand percentile ranking, it refers to the percent scoring under a certain value, so even if only one person attained a perfect score it could never technically be exactly 100%.

What I’m interested in is where in this “Revealed Preference” document you get this info-

If you could refer me to a section(s) or page(s) it would help, because I’m not really interested in reading through this entire thing carefully, since most of it is about where students prefer to attend rather than where they are admitted. And in the free version I downloaded I can’t easily identify any data about the SAT2 test at all. Of course, maybe I’m looking at a different document than you are.

epiphany · March 23, 2010, 11:36am

I don’t think that’s their mantra, siserune. I think their mantra is that test scores as stand-alone elements are not decisive for admission, but are one – albeit important – piece of information among many other elements of predictable college performance.

proud_mom_2 · March 23, 2010, 5:28pm

OldbatesieDoc: I have a great daughter who needs to meet your son. :)

phineas · March 23, 2010, 6:16pm

This thread is out of control.

Forced recitation of all 834 entries may now be used as a form of CIA approved torture of terrorist suspects.

By number 379, anyone would crack.

siserune · March 23, 2010, 6:53pm

The Revealed Preferences numbers I posted are not from the published paper, they are from a separate 20-page report on the data set. Apart from the SAT pattern in admission rates, the numbers from the report are further evidence that the graph in the published paper aggregates quite a few different scores into the “100th percentile” with its approximately 20 percent rate of admission. The raw data in the survey was that out of about 500 applicants to Harvard, 26 percent were admitted, whatever their SAT scores. The rate at the upper scores is higher than that, as we also know from the published 40-60 percent rates for those with perfect SAT. The 20 percent rate in the graphs probably excludes early admission applicants, as the published article was about students choosing between multiple offers of admission.

bovertine · March 23, 2010, 7:57pm

Okay. Thought I was missing something. Assuming these numbers are true, and I have no reason to doubt them, I’d agree that’s pretty strong support for your theory about the predictive value of these standardized exams. Along with the regression coefficients.

siserune · March 24, 2010, 2:45am

Let me also mention that the book The Early Admissions Game contains essentially the same graphs of admission rate in the Revealed Preferences sample, for most of the top schools, in 50-point SAT intervals.

Figure 5.13b, page 153: Admission rates at Harvard (1999-2000 cycle, survey data)

SAT 0 - 1400 2% regular 6% early
SAT 1410-1450 8% regular 21% early
SAT 1460-1500 7% regular 22% early
SAT 1510-1550 12% regular 55% early
SAT 1560-1600 33% regular 57% early

To summarize this: under 1400 only special cases are admitted ; 1400-1500 acceptable but no sensitivity to SAT in this range; high and steepening sensitivity to SAT or things that are functionally equivalent to SAT at the upper range. This is assuming that the story for all applicants today is similar to that for regular applicants ten years ago.

TheGFG · March 24, 2010, 8:47am

I’d say my S was very good at a lot of things but not great at any one thing–the kind of kid who could stand out as accomplished on a high school level but not on a larger stage. For example, he could make varsity and captain but not be superstar; be first chair and soloist in school band, but not make the all-region stuff; achieve high SAT 1 and 11 scores and high GPA, but not perfect/near perfect scores and wasn’t val or sal; could win school level subject matter awards but no national prizes (didn’t try really though), etc. He got into two lower Ivies, a few top LAC’s (one with a full scholarship), but was waitlisted or rejected at the other Ivies.

E2012CY · April 5, 2010, 9:33pm

so what is the average GPA range of the applicants attending ivies? average ranking?

What is the profile of an "Ivy caliber" applicant?

CONNECT WITH US