we are facing some problems with matched pairs analysis right now. Our situation: 4 groups with 16 patients each, the 1st patient of group 1 is matched to the 1st patient of group 2,3 and 4 and accordingly patients 2 -16 are matched.
For each patient we have 58 metabolic regions that need to be compared within the 1st patients of all groups, the 2nd patients of all groups, and so on...
We were wondering how to do a matched pair analysis for 2 or more groups for parametric as well as nonparametric data?
We already used the matched pair platform but we can only analyze two groups and only parametric data. We also know about the Wilcoxon signed rank test (Comparison of two groups with nonparametric data), but dont know how to exactly perform it: Is it correct to always assume the hypothesized mean as 0? Which standard deviation should we choose?
For the parametric case, you have a mixed model that generalizes the paired t-test (matched pairs). Say you have a column (character/nominal) for "group" (takes on values 1, 2, 3, or 4), and a column (character/nominal) for "patient" (takes on values 1, 2, 3, ..., 16 for each group), and then 58 columns for your responses (continuous). In the Fit Model platform, you should have group and patient as Model Effects, with patients as Random (via the Attributes options). And be sure to leave Unbounded VC option checked. This will give you what you are asking for. And you can check that for two groups this method gives the exact same result as the Matched Pairs platform.