At medical schools, pharma companies, and even at the FDA, biostatisticians and clinical researchers recognize that retrospective data when treated with appropriate bias-controlling statistical methods, can in fact be reliable, particularly as it points to direction of influence and the general magnitude of effect.
Sternberg (provost and senior vice president and professor of psychology at Oklahoma State University), which recently appeared in InsideHigherEd and raises a number of points which are worth additional comment, particularly since I think Sternberg misses the real reason behind college rankings in the first place.