On a recent check, I did a "head to head" comparison between UPenn and a bunch of its peer schools. According to the website, 53% of students pick Penn over Yale, Penn gets about a 50/50 split with Stanford, Penn takes 71% of cross-admits from Brown, Penn wins 62% of cross admits from Columbia, and Penn takes 72% of cross admits from Dartmouth. Moreover, Penn takes 88% of cross admits from Cornell, and 72% of cross-admits choose Penn over Duke. Penn also takes 60% of cross-admits from MIT.
11% choose Penn over Yale (not surprising)
15% choose Penn over Stanford (not surprising)
35% choose Penn over Brown (I was surprised Brown did so well)
41% choose Penn over Columbia
46% choose Penn over Dartmouth (surprised Dart beat Penn)
63% choose Penn over Cornell (lol)
66% choose Penn over Duke
15% choose Penn over MIT
According to the website, 53% of students pick Penn over Yale, Penn gets about a 50/50 split with Stanford, Penn takes 71% of cross-admits from Brown, Penn wins 62% of cross admits from Columbia, and Penn takes 72% of cross admits from Dartmouth. Moreover, Penn takes 88% of cross admits from Cornell, and 72% of cross-admits choose Penn over Duke. Penn also takes 60% of cross-admits from MIT.
What data are you looking at? That sounds completely off. This data is from 2006, but I doubt the numbers have changed that much since then. Penn only takes 11% from Yale, 15% from Stanford, etc.
As has been discussed ad nauseam here, that chart from the New York Times (a) never reflected actual choices -- it was a mathematical projection from fairly limited data -- and the original academic article wasn't really about which colleges students would choose, but rather the authors' proposed technique for predicting choices based on limited data, and (b) the limited data used in that study is 10-15 years old now, and has gotten more than a little stale.
The study was interesting, but no one should be looking at that chart as representing more than a rough guess of what students were doing around the turn of the century.
Some of the numbers seem a bit skewed to me as well. I didn't look too closely, but on first glance it appears to me that parchment.com relies on self selection as to who finds the site and inputs their data. If this is true and they don't norm the data, then you could have a lot of selection bias that significantly skews the results. This is true regardless of sample size, statistical validity and other measures that infer validity.
For example, if there is more word of mouth about the site amongst freshman at Penn, and a disproportionate number of Penn students have entered their data (vs. yale, Stanford, etc.) relative to the size of the accepted applicant pools, that would skew the results in favor of Penn vs. all "competitors". Similarly, if a disproportionate amount of mid-Atlantic kids are in the data base, relative to say West Coast kids, then one would expect some bias in selection of Penn vs. Stanford.
I took a moment to look at the parchment.com data, and compared scatterplots for Penn and Stanford with the filter "accepted greater >= 1" (just to de-clutter the graphs by only showing accepted students). Far more green dots for Penn, so a lot more attending Penn students have their data included, and since they obviously chose Penn there is indeed significant selection bias affecting the parchment.com data.
As others have noted above (see Princetongrad2000's post) the Hoxby data, which is the source for the NYT chart linked above, is also suspect.
Unfortunately, there is no good publicly available source of data for this that I am aware of.
"'Christoph Guttentag, dean of undergraduate admissions, said data indicates that for the classes of 2008 and 2009, about 85 percent of students accepted to Duke and one of the five institutions that fall in the first group of competing schools--a group for which Guttentag uses the acronym "H-Y-P-S-M"--did not choose Duke.
"The numbers vary from year to year but stay in fairly defined ranges," Guttentag wrote in an e-mail.
Yet by many measures, Duke ranks on par with the rest of the Ivy League-in particular, Brown, Columbia, Dartmouth and Penn.
During the last two years, Guttentag said data indicates that between 40 and 60 percent of students accepted to Duke and one of the second group of competing schools chose Duke.
Duke enrolls between 75 and 90 percent of cross-admitted students from the third group of top competitors, which includes Cornell, Northwestern and Georgetown."'
So, Duke loses 85% of cross admits to HYPSM.
Duke enrolls between 40-60% of cross admits to the remaining 5 Ivies.
Duke enrolls between 75 and 90% of cross admits to the next tier of competition-Georgetown, Cornell and Northwestern.
If anyone else knows articles which cite admissions yield estimates from other deans of admission, that would give us a very clear idea of revealed preferences.
according to it, yale loses to harvard 34/66 (compared to princeton's 20/80 and stanford's 22/78), wins princeton 80/20, wins stanford 67/33, but it's fourth overall behind princeton and stanford? doesn't make sense to me.