A recent NPR Marketplace posting announced (or at least implied) that the “era of Big Data” had arrived for the college application process:
’As a student goes through the search and application process, many times unbeknownst to them, colleges are collecting information about everything that they do,’ says David Hawkins, director of public policy and research for the National Association for College Admission Counseling. Messages, campus visits, even social media interactions, are logged into admissions software.
At some level, this is no different from what consumer products and services companies have been doing for some time -- using “big data” and business analytics to determine which customers to target. However, the college admissions game has a special twist that sets it apart:
(Hawkins says)…that colleges don’t want to offer up a spot to a student who isn’t likely to enroll and all that data is crunched in the search for something called ‘demonstrated interest’.
At first blush, this seems reasonable: admission planning requires a good understanding of the likely attendance from the mass of students who have accepted. But, on second thought, I realized that this would actually not be that difficult to model. Why not simply offer spots to all qualified students and then use well-established techniques to forecast and manage the number who will actually enroll? After all, airlines and hotels do it all the time.
I suspect that the reluctance to “…offer up a spot to a student who isn’t likely to enroll…” may be driven by a desire to keep the so-called “yield rate” high. The yield rate for a university is the percentage of admitted students who enroll. A high yield rate is used by some ranking systems as a sign of desirability. As a New York Times article put it:
There are many ways in which colleges and universities can make their yields, and themselves, look good. Just as institutions can make themselves look more desirable by broadcasting low acceptance rates, they can also defer students who they believe are likely to enroll elsewhere. By not accepting those students, the institution is able to appear more selective, report a higher yield and, perhaps, increase its ranking in some publications and its aura of prestige among prospective students.
Yield rate is one measure of desirability – the other is acceptance rate, the percentage of total applications that are accepted. To look good in the ratings, a university would like a low acceptance rate and a high yield rate. This could be accomplished by encouraging applications from students who have little or no chance of acceptance (lowering the acceptance rate) and denying acceptance to students who might otherwise be qualified but are unlikely to attend (increasing the yield). I can see how predictive analytics can be used by colleges and universities to support this process. But this doesn’t mean that I like it – particularly the rejection of qualified candidates because of an algorithmic judgment that they are unlikely to accept.