I have a set of 15 perception questions asked before and after a course to the same group of 13 students. The responses are 0%, 20%, 40%, 60%, 80% and 100%. What is the best way to analyze them to test for significant difference?
2 Answers
I would use the Wilcoxon signed-rank test. That test does not depend on the distribution type, that is, it is a nonparametric test. Thus, that the data categories are ordinal should make no difference as the ranking procedure is also ordinal.
Other questions you might want to answer include nonparametric significance of difference of variance for which I would use Conover's squared ranks test.
-
1$\begingroup$ Wilcoxon's signed-rank test as usually applied requires at least interval-level data - so that you can calculate differences before ranking them. (I imagine the OP might be happy to consider these as interval-level.) $\endgroup$ Jul 30, 2017 at 15:55
-
-
$\begingroup$ But you say in your answer "the data categories are ordinal". (I'm not sure why - there's not much information given about what they represent.) $\endgroup$ Jul 30, 2017 at 20:26
-
$\begingroup$ @Scortchi The responses are 0, 0.2, 0.4, 0.6, 0.8, and 1. Looked like ordinal, equidistant responses to me. In other words, they are also interval; $0.2*(i-1)$, where $i$ is 1 to 6. $\endgroup$– CarlJul 31, 2017 at 0:18
-
$\begingroup$ @Scortchi Reason for my equivocation is that I think that ordinal is required for Wilcoxon, although interval is fine. I think that all that is required for Wilcoxon is one one mapping, i.e., a mapping that doesn't fold onto itself in a sort of not functional way. $\endgroup$– CarlJul 31, 2017 at 0:24
Use paired t test to find whether there is significant difference between pre and post course
-
$\begingroup$ The data is ordinal. t-testing is not applicable. $\endgroup$– CarlJul 25, 2017 at 20:57