I am trying to determine if a Wilcoxon signed rank test is appropriate for my data. The experiment tested how pairs of participants interacted with 2 versions of the same software. The software was collaborative in nature, hence why each pair of participants tested each version together. Each participant's behavior was to some extent influenced by their partner's behavior. To my knowledge, the pairs of participants were assigned randomly drawing from the same sample.
During analysis, each participant's interactions were coded individually. Researchers counted the frequency of certain behaviors. The result was a number of behaviors for each participant: one during the condition with v1 of the software, and the other during the condition with v2 of the software.
The data looks something like this:
| participant | condition 1 | condition 2 |
|---|---|---|
| Pair 1; Participant 1 | 18 | 6 |
| Pair 1; Participant 2 | 11 | 10 |
| Pair 2; Participant 3 | 12 | 7 |
| Pair 2; Participant 4 | 9 | 9 |
| Pair 3; Participant 5 | 11 | 7 |
| Pair 3; Participant 6 | 15 | 10 |
Ultimately, the goal of this work is to compare differences in how all participants interacted with v1 of the software and v2 of the software. I don't believe that the effects from how participants were paired is of interest (nuisance variable?).
Based on my understanding of the Wilcoxon signed rank test, my data fulfills the requirements of being 1) non-parametic and 2) paired. I believe my question is similar to this previous question, for which an ordinal logistic regression was suggested. I am not sure if that approach is appropriate for my case.
I'm struggling with how to account for the fact that each paired observation is dependent on another paired observation. I have considered adding the totals together, so that each pair of participants would be a single observation, but I am curious if there is an alternative approach.
