Wilcoxon Signed- Rank Test vs Kruskal Wallis vs One Way Anova? Which one to use.

Dear forum Professors,

I am conducting a research to rank the importance of certain elements within a property management framework. Would like to know which test is best to help me rank the items. The methodology is as follows:

(n = 300)
I will be sending out questionnaires with roughly 20 questions(item) asking people to rate how important they believe each item was in the management framework. (e.g of items are Security, Customer Service, Cleanliness and etc.)

The rating would be done through a 5 point likert scale.

When i get back the questionnaire, i would like to rank the items and determine which item is perceived as the most important element. After the test is done, I would need the result to tell me which one is the most important, second most important, third most important element and so on.

Would like to know what test would be the best for the situation above which involves (n = 300) and roughly 20 items to rank.

I only have SPSS. Am welcome to suggestions of any other tests that you believe will be better suited :)

I'm fairly new to this and am still trying to grasp the essence of the tests.

Thanks in advance to all those who read this thread :)


Less is more. Stay pure. Stay poor.
Don't have much time, but this definitely seems like one of those survey questions that comes up in those arenas. Off-hand not sure any of those three approaches out-and-out are you best approach. Pairwise comparisons would get fairly high given 20 items, thus a corrected alpha would be very small.
I have considered doing a conjoint analysis by restructuring the questionnaire. However like you said, the comparisons would be quite huge. Do you have any suggestion on what test to use for this situation?


TS Contributor
Seemingly the signed rank test will work. But
there are 190 comparisons (even if it won't be
190 tests, all possible comparions have to
be taken into account here). As mentioned before,
a Bonferroni correction of the significance level
would be very conservative. Maybe yu use 1%
instead of 5% so you have a better protection
against type-1-erors while preserving power.

With kind regards