We want to conduct a survey where participants will be asked to rank the importance of ten options representing the causes of a target event. We want to know what are the top three causes the participants believe that result in the event. What would be the statistical test for this?
Thank you so much for any help!
Hello Bo,
Descriptively, you could simply count the instances of each option being chosen as #1 (where #1 implies the most important of the causes), #2, etc., and create an ersatz ranking summation for each (e.g., if ranked by one respondent as #4 in the list, that option is given 4 points). At the end, the lower the summation (across all participants) for an option, the more important the option was perceived as being among your sample.
For the second part, are there significant differences among option rankings, the Friedman two-way anova by ranks could be used. This would identify only whether any differences exist among the set of 10 options. You could elect to follow up a significant result by running subsequent Wilcoxon signed-rank sum tests for a given pair of options, if your aim was to try to pinpoint differences more precisely.
Good luck with your work.
I agree with David Morse, except for the selection of a post-hoc test for the Friedman test. A few thoughts on this:
- The Wilcoxon signed rank test handles data differently than does the Friedman test, and so doesn't match this test well for a post-hoc test.
- The Wilcoxon signed rank test treats data as interval. You probably want to treat your rank data as ordinal. The Friedman test can handle ordinal data.
- There are post-hoc test designed specifically for the Friedman test. These have names like Conover, Nemenyi-Wilcoxon-Wilcox, Siegel-Castellan, and Eisinga-Heskes-Pelzer-Te-Grotenhuis.
- Badges
- Science topic
- Similar topics
- Computer Science
- Computer Communications (Networks)
- Internet