Hello all-
I am using the Wilcoxon signed rank test to compare self-assessment of skills before and after a training session. Data is 1-5 Likert type scale, i.e. ordinal. I would like to compare men and women to see if one group benefits more from the training than the other, but not sure if this can be done in JMP? Naturally, I could stratify the analysis and say that the men difference for women was X and for men Y, but is there a statistical test in JMP that can be done to demonstrate significance? Anybody know?
All the best
So you have the "treatment" of before and after. You are now adding another variable, Gender. So with two "X" variables, Fit Y by X would not be appropriate. In fact, I am not sure Kruskal-Wallis was appropriate as the Before-After structure leads me to believe that you have paired observations. For each subject you have a before score and an after score. That means you have a BLOCKED design (each subject could be considered a block) and a Friedman test would likely have been more appropriate.
So given what I see (and I know I do not have all of the information, so consider my approach, but there is no guarantee that it is correct) I think your data could look like my fictitious JMP file (Original Table.JMP). I added a column called Difference that is simply the difference between the Before and After scores. The Difference score would take into account the different subjects. Notice that an After score of 3 might look low, but if the Before score for that subject was a 1, the training was still successful. That is why the analysis should take into account the subject.
To mimic your initial analysis, there are two ways it could be done. I used Distribution on the Difference score. I saved that analysis as a script named "Analysis without Gender". Instead of creating the Difference column, you could have used the Matched Pairs platform on the Before and After columns. That analysis is saved as the "Matched Pairs without Gender" script.
Now to add Gender to the analysis, I needed to stack the Before and After columns. That table is also attached and is called "Stacked Data". I did not keep the Difference column in this table, as it is not needed. I left the Data column (your scores) as Continuous. First, for the analysis, I made Data ordinal and used Fit Model. I used Gender, Before/After, and the interaction in the model. Because Data was ordinal, the modeling is Ordinal Logistic Regression. This analysis is saved as "Ordinal Model with Gender".
I also fit the same model to Data after making Data continuous. I have saved that script as "Continuous Y with Gender". The results are similar, but slightly different.
It would be your choice on if you want to leave Data as Ordinal or make it Continuous as there are arguments either way.
I hope this helps.
