I have a within-groups study with 4 groups (400+ users). Each user answered 10 questions for each group. The answers were scored as right/wrong (1/0). Can the percentage correct for each user/group be calculated to use in a statistical test such as Kruskal-Wallis or Mann-Whitney? Or should I just stick with chi-squared analysis using raw totals? Or Cochran's Q?
To further complicate things, the questions were multiple choice with only two choices. So completely random answers would still get about 50% correct. Perhaps this won't matter if the statistical test compares groups on a user by user basic, which is what I think Mann-Whitney test does (as opposed to chi-squared analysis).
I apologize in advance if this is unclear at all and will try to clarify. There are some similar questions on here, but there are no solid answer. Thanks!