Sunday, April 4, 2010
2010 Rankings of HYPSM
From Comments of 2010 Intel Talent Search Finalists
1. MIT -- 19
2. Harvard -- 12
3. Stanford -- 10
4. Princeton -- 5
4. Yale -- 5
From Projected Yield/Admit Ratios
1. Harvard -- 76.5/6.9 = 11.08
2. Stanford -- 69.9/7.2 = 9.71
3. Yale -- 66.8/7.5 = 8.91
4. Princeton -- 58.9/8.2 = 7.18
5. MIT -- 63.9/9.7 = 6.59
Note: The calculations are based on last year's yields over this year's admit rates. This year's yield could be grossly lower for each school.
From Yields of Sample Cross-Admits of Class of 2013
1. Yale -- 47%
2. Harvard -- 42%
3. Princeton -- 40%
4. MIT -- 35%
5. Stanford -- 30%
Overall
1. Harvard -- 5
2. Yale -- 8
3. Stanford/MIT --10
5. Princeton -- 11
________________
Subscribe to:
Post Comments (Atom)
16 comments:
Your "sample of cross admits" is statistically invalid to a gross degree, due to its pitifully small sample size and the absence of any evidence that the sample was random. Furthermore, it is at substantial variance with such legitimate data as has been reported.
Why did you switch from your earlier metric in re Intel finalists?
See:
http://mathacle.blogspot.com/2010/02/where-do-intel-talent-search-finalists.html
http://www.mychances.net/blog/2009/12/06/college-rankings-4-college-preference-matchups/2009-college-preference-matchup/
http://www.mychances.net/blog/2009/12/06/college-rankings-4-college-preference-matchups/2009-college-preference-matchup/
The Intel comments are for this year only to measure the current trend. The other one is for recent trend.
The samples are fine as they are close to 40%, and once I have enough data over several years, the mean should give a good indicator.
"40%" of WHAT? From what source(s)?
"40%" of WHAT?
Sorry, I meant the measure is around 40%, not the sample size. If the the yield is 1% for Harvard, then the measure is way off. The total cross-admits should be around 1500-2000. Give me sometime and I will convince you once I find out how to put everything together.
I doubt you will convince me. You don't have the numbers and I doubt you can get from any reliable source.
The Revealed Preference number used by the NY Times are from a very large, random sample.
The "MyChances" data is also from a very large sample, tho not as scientifically selected.
I know the exact from one school, and I can assure your projections are way off in that case.
Revealed Preference number was not the real thing and "myChances" data were not consistent for useful analysis. Since you have an exact number from Harvard, I will try to find out from my side.
The rankings need not to be precise, just like a stock market index, they are just "linear" approximation to the real thing. Daily or yearly changes should not affect the underlying values, they are just a reflection of people's "pink noise" minds v.s. the "brown noise" of the reality.
I'm not sure what you said - or meant - in that last point. Data is either sufficiently reliable or it is not. The Revealed Preference formula - while a projection - was based on a lot of carefully selected data, and they explained how it was collected and what its limitations were. The "My Chances" numbers are at least based on a large sample. Your claimed cross admit percentages do not meet either test.
Compare:
http://college.mychances.net/college/tools/college-cross-admit-comparison.php?compare=Harvard&with=Stanford
"MyChances" does not tell you how big the samples are. Wisconsin v.s Stanford is 40:60. The way they try to put everything together to make money could be garbage at core.
what does your overall score or ranking mean? how was it derived?
Add the numbers together. e.g., Stanford is 3rd on the Intel comments, 2nd on the Yield/Admit, and 5th on the cross-admits, so its number is 3+2+5=8. The lower the number the better.
So all of these random categories - several statistically dubious - are given equal weight?
Keep it simple. After all, we are living in a deterministic chaos world which is made of simple cyclical patterns. If people were asked, back to 2000 years ago, what would be possible: sending people to the moon or telling the next day's grain prices, the answer would be obvious. The linearization of our "faster-than-normal" thinkings fails to solve every problem. If I am convinced, I would rather listen to a monk in a remote mountain in China to tell me how to predict by counting his toes than to go to Harvard to get a MBA.
The same thing for ranking HYPSM
Post a Comment