Identifying test questions that were done poorly/well

I have a test, and want to identify which questions were answered particularly well or poorly on a statistical basis. I know who answered which question correctly, and therefore have frequencies of the number of people who answered each question right. I also have a total score for each person, and an average score for the test.

I am trying to figure out what statistical test (if any) I might be able to use to confirm which questions were answered particularly well or poorly. Does anyone have an idea?

Chi-square and binary logistic regression have been suggested, but I'm unsure how to implement either appropriately.

brittany812 said:

I have a test, and want to identify which questions were answered particularly well or poorly on a statistical basis. I know who answered which question correctly, and therefore have frequencies of the number of people who answered each question right. I also have a total score for each person, and an average score for the test.

I am trying to figure out what statistical test (if any) I might be able to use to confirm which questions were answered particularly well or poorly. Does anyone have an idea?

Chi-square and binary logistic regression have been suggested, but I'm unsure how to implement either appropriately.

Click to expand...

Sorry, no idea and, unless you can find evidence to support a choice, I think you should read each individual answer and grade it as well or poorly answered, then maybe see if there is some sort of relationship between how each individual question was answered and the test score, i.e for question number one 80% of the 'well answered' were for test scores above 85%. If there then seems to be any consistency in those results, it might suggest a distribution.

A lot of questions? Start with a few and continue until possibly something suggests itself. The problem with deciding before hand to use, for example, a Chi-Squared distribution is that you have no idea of the validity of that use until you at least have some experimental evidence to support that use.