I have a group of subjects. Each of them provided ratings on an ordinal scale for two questions. The questions themselves are not directly related (I ask for evaluations of two different scenarios).
Are the two ratings provided by each subject considered paired or unpaired? In other words, should I use the Mann-Whitney test or the Wilcoxon signed-rank test?
Hello Chen,
Treating the two ratings as paired would be the preferred approach.
Good luck with your work.
Indeed, it seems that a paired inferential test would be preferable, as you will have an increase in power (and thus, on your sample, if there is a difference among the two scenarios, you are more likely to find it as "significant" (horribile dictu these days)); in the alternative option (of using an unpaired test), you will have more statistical noise and consequently less power to detect a potential difference (assuming it does exist). This is based on the limited info you provide. I am a bit puzzled that you think that the two are not directly related, but you are interested to know if there is some difference among the two (as otherwise you would not contemplate using an inferential test for the two groups).
Thank you, Robert.
By "not directly related" I meant that it's not like a test-retest design, in which there is a treatment effect and the observations are clearly paired. Obviously I had a reason to test judgments within-subjects, I just wanted to make sure that my design can be considered paired.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics