Friday, November 14, 2014

Can We Find a Better Way to Rank Students?

I have been complaining about GPA for years. I don't care that it's an inconstant measure -- it has been drifting upwards pretty much since the day it was invented, and there is very little reason ever to compare GPAs of today's students against GPAs of students a decade ago since they are almost never running against each other for anything GPA is used to calibrate. I don't even care much that it is compressed; in fact, that has some benefits, is it makes it easier to justify ignoring it to focus on other criteria instead. The problem I have with GPA is that even if it were constant over time, it would be almost meaningless as a measure of academic excellence, much less any of the more important kinds of excellence. 

Giving GPA the official status it has disincentives ambition. It discourages the pursuit of excellence by encouraging the pursuit of grades in a curriculum that is largely elective. When faced with a choice between two courses, the decision strategy that tends to maximize GPA, which we say we value, is clear: take the course from which you will learn less because you already know more of the material it teaches. 

I am in a very fortunate position as Director of Undergraduate Studies in Computer Science. I can tell students with complete honesty that they are better off ignoring GPA and not worry about getting Bs and Cs. Unless you apply to graduate school, I say, no one will ever see your transcript. In interviews, tech employers may just give you a problem to work on and see if you can solve it. Whether you can solve the problem is less correlated with your GPA than with whether you took challenging courses. You shouldn't let the pursuit of credentials get in the way of getting an education. And by the way, even if you DO apply to graduate school, a faculty letter praising your senior thesis is going to be more useful than a straight-A record.

I can't do anything about the way law school and medical school admissions committees screen applicants and I am not about to try. But maybe we can do something about the honors the university itself gives to high-GPA students. Just including the GPA on the official student record signals our institutional reverence for it. I can't object to supplying students with that information since we keep using it in the various ways we use it -- for example, in the award of graduation honors (cum laude, etc.). But the stupidity of the metric really hit me this fall as I sorted students for two prizes, one for freshmen and one for seniors, both aimed at rewarding true academic excellence.

These prizes are not given just for high GPA (we do have one of those too, the Sophia Freund Prize, given to the highest GPA summa -- in recent years it has been snared among multiple 4.0s). For the prizes I am talking about, GPA is used to create a pool several times larger than the number of prizes to be awarded, and then the committee reads transcripts, letters, and other supporting material to pick out the real intellectuals. The process works pretty well because the faculty on the committees take the job seriously. But several things have become evident to me.
  • Very high GPA is highly correlated with good pre-college preparation. That is, the vast majority of the pool seems to consist of students who had the good fortune to go to excellent high schools, public or private. The best public schools are either in high-income zip codes, or they are exam schools. Some of the independent school are graduating well-prepared low-income graduates, but you don't see many students in these pools from public schools in low-income zip codes.
  • Because of the compression, any one B will knock you out of contention, so freshman-year grades are among the major criteria on which, de facto, these honors are awarded. Freshman year grades tend to be lower not just because students are adjusting, but because freshmen take more large courses and grades in large courses tend to be lower.
  • Most of the transcripts were pretty easily classifiable as "hard core" or "elementary," with only a few that required more serious scrutiny. By "elementary," I mean some perfectly good course programs -- let's say, Math 1, Ec 10, Spanish A, Expository Writing, and a Freshman Seminar. Nothing wrong with that program if you landed here from one of the many American high schools that does not teach AP math and from a part of the country where they don't think foreign languages are important. But not the sort of program that should win you any prizes for superior intellectual achievement, even if you got a 4.0. The increasing variance in socioeconomic background of the Harvard student body may be making those transcripts more common, I'm not sure.
What do I mean by "hard core"? I got that phrase from Ballmer's CS50 talk. One of several pieces of good advice he offered students was to be hard-driving, intense, focused, hardworking, passionate about things. He talked about Taking Physics 55 (then, as I recall, a physics analog of the legendary super-honors Math 55 course that still exists) and getting a 33/100 on the first exam.

Fact is, the committees looking over student records can judge, reasonably well, which are hard core and which are not. Faculty at least can make that judgment for the courses in their own area. But those judgments are not easily automated. Some courses with graduate numbers are not hard core and some courses numbered less than 100 are hard core. "Everybody" knows that Math 55 is hard core but CS 20 is not. (CS 20 is a great and important and highly educational course. But it's not the course to take if you want to convince me that you are going to win the Turing Award some day.) "Everybody" knows that CS 161 is hard core (that is so well known in the tech industry that even interviewers who didn't go to Harvard listen up when interviewees say they took it) and CS 171 isn't (it's just a hugely educational course that EVERYBODY should take!).

Not every course a hard-core student takes is going to be hard core. In fact one of the blessings of guts at Harvard is that they make it possible for normal students to be hard-core some of the time, and   taking just one hard-core course can be life-changing. So when I see a transcript, I sniff at courses taken pass-fail, but I don't mind seeing well-known guts if the student took something hard core at the same time. 

So my question is, rather than fruitlessly trying to normalize grading (as Princeton just gave up doing) or trying to compute GPAs in a way that takes into account the grading curve in a course or the grades in other courses of the students taking the course, can we come up with something better, that incentivizes ambition as demonstrated by a hard-core transcript -- or even a "beautiful transcript," as Professor Elaine Scarry put it to me once? I don't want to automate the whole process of rewarding students; letters, essays, and so on are important. But I don't like the idea that students with only basic coursework are crowding out of the pool other students who have wound up with blemished records because they really stretched themselves to the max. Can we socially engineer a "hard coreness" rating for courses? What would be the incentive for students to rate courses honestly, for the Lampoon not to troll the rating system, and so on? Would faculty refuse to go along with this because they would find it too stigmatizing to have their courses classified as not-so-hard-core?

Of course the other way to handle this would be to stop giving those prizes. That ain't going to happen, but I'll leave all that for another day.


  1. 1) Sounds like this particular award is handled pretty well with the human element there to figure out what courses are hardcore and not.
    By contrast, ugrad and grad school admissions is likely harder since GPA from diff places really is diff, and less clear what the hard-core courses are.

    2) When I am on such comittees I ignore the GPA and just look at the transcript to get s sense of what the student has done. Look for hard courses and somewhat forgiving of low grades as a freshman. A bigger problem might be that GPA is used as a cutoff before the transcript even gets to me.

    3) I know a case where a HS student took Linear Algebra as a Sophmore at a local college and was in a bit over her head and got a B-. This went into her GPA and was held against her in college admissions. Thats GPA-ism run amok.

    4) A better test is can I pass the `are you a human' test for making this comment. That often gives me a problem for some reason.

  2. This comment has been removed by a blog administrator.

  3. This is probably one of the hottest debates going on in academic circles nowadays. I think trying to bring in a better ranking system partially justifies the ideology of ranking itself. No matter how great a system we come up with, our students are smart enough to find a loophole and abuse it to their heart's content. Its like research in computer-security. Keep finding robust protocols which need to be replaced every few years. I think a better approach could be to leave the ranking system the way it is but not give it the importance it gets. This again cannot be done at the institute level but as individual faculty members, we can clearly convey to students in our class that our opinion of their abilities has no correlation with their grades. I think this itself would be a powerful change agent! And trust me, the best students are better not given the best ranks!! It will just kill the fire that burns within them.

  4. This comment has been removed by a blog administrator.