For those of us who have to grade on a curve, a problematic scenario is that you grade your exams and the scores are flat. Of course, if all of the answers are really good, you'll feel better about ourselves than if they're all bad. Hey, you actually taught people something!
But, if they're all good, not only will it make curving the results very difficult, it might also mean that the question was too easy.
However, does that really ever happen? A long-time professor told me before I gave my first exam ever that we could give the students the answers the day before the exam and we'd still get a really nice curve. She said this as a cynical way of telling me to make my questions easy, so as to cause the least amount of pain and stress among the students.
Now, I won't go that far, but I have yet to give a fact-pattern based question where the answers aren't well differentiated. My normal grading experience is that after reading the first few answers I panic, thinking I'm not going to get a curve because they're all great/mediocre/bad. But then, by the end of grading 70+ of the same answer, inevitably there's a wide spread.
Maybe I just can't write an easy question (although I try sometimes!), but maybe this is just natural human variation that will always show itself on an exam.
Do you find the same thing? If not, and you've come across a situation where you haven't had the differentiation your curve might need, what have you done?
Yes, this has happened to me. The practical result was that most of the scores to that question were quite similar, and that any differences in scores to other more difficult questions were the ones that made the difference in grade when I put the raw scores into the curve.
Posted by: Orin Kerr | December 22, 2010 at 03:01 PM
Did you consider trying to normalize the distributions of the various exam questions? I've never done that, but I've heard of folks who do so as to make each question actually count the amount that the prof said it would count. If one question winds up have much less differentiation than another, that one question becomes minimized whereas the other takes on much more significance. As I said, I've never done this, but there's something that seems theoretically sound about it.
Posted by: David S. Cohen | December 22, 2010 at 03:04 PM
I think it is an unsound approach, actually. The goal of my grading is to report my assessment of a student's lawyerly ability. If a question is too easy, and the answers are all pretty similar, any slight differences in the scores are not generally related to lawyerly ability. Rather, the slight differences in score are likely to be caused by random chance. If you normalize the distributions of the exam questions, you take the random chance that has no connection to lawyerly ability and make that random chance a significant part of the grade. That seems backwards to me, as it increases the randomness of the grade.
Normalization is very important when you're putting together mixed types of responses: For example, if you have a multiple choice section, and you've announced that the section is X percent of the grade, you need to normalize the curves between the multiple choice and essay sections. But I don't think it works within a particular type of response such as essays.
Posted by: Orin Kerr | December 22, 2010 at 08:50 PM
I like the services that amazingly surprised me that this kind of work should be done in the proper way and also thanks for the information you shared!
http://www.ukwritingexperts.co.uk
Posted by: Account Deleted | December 23, 2010 at 03:51 AM
If one question winds up have much less differentiation than another, that one question becomes minimized whereas the other takes on much more significance. As I said, I've never done this, but there's something that seems theoretically sound about it.
Posted by: Moncler Jackets | January 04, 2011 at 09:29 PM