Some comments on grading

Students are often concerned about grading. Here are a few comments to help you understand where I am coming from.
  1. Students will sometimes ask if the class is “curved”. I never know how to answer. There are no letter grades corresponding to fixed percentages, yet neither do I see students as competing with one another. I compute total scores and try to draw in lines that feel like they make sense for the particular term. I’ll look at a final exam and say to myself: this looks like a B. The distribution on letter grades usually ends up looking customary for a CSE course, despite the numerical averages often being low. While I think my classes are harder than average (maybe even much harder), I don’t think my grading is particularly easy or hard relative to other people in the Department.

  2. My grading of problems is intended to be rather bimodal: I try to give more than half the possible points to any answer that is basically correct, and fewer than half the possible points otherwise. This translates to less partial credit than many students are used to. In the “extreme” version of this grading convention, your score is simply the number of problems for which you provide a well-written, fully correct solution. Indeed that is how I would like to grade, but I have found it infeasible because it would result in too many students getting zero or near-zero scores. Still, I do try, to the extent that it is workable, to approximate this ideal. The upshot is that if you’re a student who survives on partial credit, you will probably end up doing poorly in my classes.

    The reason for bimodal grading is that I prefer people to be clear about what they do and do not know. If you can’t solve a problem, just say so. Explain what you’ve tried. It doesn’t bother me when a student can’t solve problems I ask—after all, I can’t solve most of the problems I consider, either. What you shouldn’t do is to bullshit an answer, copy an answer, try to adapt an answer you don’t fully understand, or say things that fundamentally don’t make sense. In the ideal world, any of these things would earn you a worse score than saying nothing at all.

  3. On true/false questions that don’t ask for justifications, I will penalize for wrong answers, but never to the extent that you are wrong (in terms of maximizing expectation) to guess. The purpose of taking off for wrong answers on true/false questions is just to normalize scores so that “knowing nothing” doesn’t give you an expected 50% score.

  4. Exams might ask for a short essay. I started doing this a few years ago, after reading a piece by a physicist who explained that, when he was a graduate student, he had classmates who were great at solving the highly technical kinds of problems the profs gave, but who couldn’t actually explain anything, and had no qualitative understanding of the physics. I’m sure we have the same sort of thing among CS/CSE undergrads: people who can write a working program, say, but can’t cogently explain a thing. A genuine understanding of a subject, any subject, includes the ability to synthesize and explain what you’ve learned, doing so in an organized, articulate, grammatical fashion.

  5. Finally, I will remind you what you hopefully figured out years ago: that grades are a necessary evil. Beyond encouraging people to focus on the wrong things, they are inherently noisy: plenty of noise mixed in with the signal. I urge you not to stress out over grades—not half as much as I stress out over having to assign them.

Phillip Rogaway