My stats knowledge is less than basic, so apologies for possible confusion.
I have 2 independent groups (N=14 and N=14) who filled in a Likert-scale based standardized questionnaire, consisting of 48 items. The scale measures types of religious coping, so I have 12 questions for each one of the 4 coping styles.
I would like to compare those two groups: to see how differently they score on each one of coping styles.
My questions are:
- Is it OK to use the Mann-Whitney test? I suppose yes.
- If yes, how do I enter data in SPSS: every response (1 for "strongly disagree", 2 for "disagree", etc.) of 28 participants (which will make a total of 336 entries) for each of 4 coping styles, then code them according to group and do the test? Or should I sum up the scores (which does not feel right, as far as I use a test specifically to treat ordinal data).
- What is the best graph to present the results?
Best Answer
The assumptions of Mann-Whitney aren't met, so your inference won't quite have the properties you'd wish. It may be more or less reasonable to do in any case, depending among other things on how heavily tied it is and the exact alternatives you're interested in, but strictly speaking you know before you even get data that it doesn't meet the assumptions.
It may make sense to do some form of resampling test (permutation/randomization or bootstrapping for example).
There's no single 'best' in my opinion. It depends on many things - what questions you want to answer, your audience's background and expectations, the ability of the software you're using, and so on.
Depending on what kinds of comparisons are interesting to you, Cleveland dot-charts or stacked bar charts are reasonable choices.