Post
by rayiner » Tue Oct 04, 2011 7:31 pm
Good comment from that ATL thread that sums it up quite nicely: "What this demonstrates is that the USN&WR rankings are prepared by statistical illiterates.
If you have several scores that vary very little between top schools (like GPA and LSAT scores), and a few that vary very much, simply multiplying the scores by different weights and adding them up will make most of the variation between schools be accounted for by the scores that vary the most. It would be better to adjust the raw scores in each cattegory into percentile rankings, the number of standard deviations away from the mean or in some other way - maybe the negative logarithim of (100 minus the percentile rank) - to make the variability of the different scores more comparable before adding them up. I'm no statistical wizz, I've forgotten most of what I once knew, but it shouldn't be impossible to make the variability of the socres equivalent before combining them. Otherwise, you get BS like spending per student outweighing hypothetical perfect GPA, LSAT and repuation scores."