I agree it's not perfect, and there are certainly outlying cases where it breaks down (just as with GPA).
But I think that, statistically, class rank would probably end up better overall. Or maybe an aggregate of class rank and GPA would be better than either alone.
The point is, as you said, reducing an UG transcript/experience to a number necessarily involves a huge loss of information. It's just hard for USNWR to plug a person's transcript into their formula. Reducing it down to one number makes things simple, quantifiable, and of course questionable/debatable.
You think wrong, because you can't control the basis of class rank. It is entirely plausible that the range of student qualities at many institutions do not over lap.
And you can't control the basis of GPA either. What class rank, at least, does is to control for more difficult majors and harsh/lenient grading. With GPA an engineer who's class had some harsh professors may get a 2.9 but still place in the top 20% of his/her class.
Class rank still fails to account for differences in the quality of the overall student pool. However, this problem is intractable because you are using a measure to assess the "quality" of individual students that itself depends on the "quality" of the pool of students.
r6_philly wrote: Even if they do, there is no way they even come close to matching, so class rank will be VERY statistically wrong. It is possible that the 1 percentile student at MIT's EE program is doing better work and has more potential than the 99% student at any of the 100% acceptance rate colleges around here.
And of course it's possible that the 2.5 GPA MIT EE student is doing better work than the 3.9 underwater basketweaving from Podunk College. Or maybe students entering Podunk College almost all come from a local magnet school where the brightest local students go for high school and they all have IQs of at least 130+. Again, we're trying to use an assessment of student quality that depends on the quality of the pool of students.
One possibility. We could LSAT distributions to assess the quality of a student pool (group of students graduating a particular year in a particular major from a particular school). Correlate them statistically with their class rank and produce an LSAT/rank-normalized GPA for each student. Now students with an LSAT/rank-normalized GPA that falls below their LSAT can be seen as "smart, but didn't have it all together or was lazy etc" those with an LSAT-normalized GPA above their LSAT can be seen as hard-working, dedicated, mature, etc. Of course this assumes the underlying validity of the LSAT, and that's another issue.
This way an EE with a 167 LSAT who's top 20% with a 2.9 GPA may have something like a 3.5 LSAT/rank normalized GPA. On the other hand, an underwater basketweaver who's top 20% with a 160 LSAT and a 3.7 GPA may have something like a 3.1 LSAT/rank normalized GPA.