Exam Grading In grading the exam, do you use an analytic or a holistic scoring scheme (assigning points for individual steps or overall "level" of solution)? Do students start at 0 points and earn points for correct statements or do they start at 100% and lose points for mistakes? Do you give partial credit for answers and if so, how do you assign partial credit? Do you have an expected "average" score on the exam for each class? Do you "curve" exam scores?
Do you use a straight total or percentage correct to be added to other points in the course to contribute towards a total grade, or do you give a grade to each exam?

In assigning course grades, do you "curve" student scores? Based on what mechanisms?

I'd never thought explicitly about the analytic/holistic polarity; I guess I use a mixture. A few of my exam questions are straightforward to grade. For example, if I give a linear decomposition of a two-way ANOVA and ask students to fill in an ANOVA table, I'll devise a way to assign points to each entry in the table -- so many points for getting all the df right, for example, with part credit for getting some but not all df correct. This sounds analytic, but the student who gets all but two df right and has them all add up to the correct total has recognized a principle that has escaped the student who gets the same number of df right but doesn't have them add correctly, so the first student gets a higher partial score. This strikes me as more holistic. For other kinds of questions (e.g, I give an ANOVA table and cell means for two-way ANOVA, and ask for an interaction graph and discussion of what the results mean), I start holistic and work toward analytic: I read through a dozen answers, comparing solutions with my own sense of what a complete and correct solution should be. Then I try to devise a way to assign points to parts of solutions that will give the best answers the top scores. Typically this means that it is possible to score more points on the question than the point-value I've assigned to the question as a whole. I don't actually award the extra points, but the system does provide more than one way to get full credit for a solution.

I always assign letter grades, sometimes with a waffle attached (e.g., A-/B+), and I always curve the grades. I explain to students in advance of the exam, and again when I return them, that my exams tend to be hard, and that I think hard exams are important for their learning: the hard exams reflect high standards for what I want them to learn. But I add that giving hard exams means that raw scores don't give a fair indication of how well they've done, and that the scores need to be rescaled to adjust for the level of difficulty of the exam. I adjust the scores by comparing the distribution of raw scores with my sense of which papers should be As, which Bs, etc. Then I create an ad hoc formula that converts As to the 90-100 range, Bs to 80-89, etc. (I tell the formula when I return the exams.)

See all of George's answers