Hello everyone,
What kind of post-hoc tests are appropriate for Friedman tests?
I am doing analysis of the variables which are repeatdly measures for over 5 times (at baseline, 3 months, 6 months, 12 months and 18 months) hence Friedman test is applied having a significant result. There is a drop out of subjects at different time intervals. Next is how to compare the time intervals pair wise?
Convetionally many suggest Wilcoxon test as post hoc test. Whether Wilcoxon test is appropriateaa? As while running Wilcoxon test it includes the entire samples without drop out?
I am using SPSS version 23 having an option of Dunn post hoc test. Dunn post hoc analysis for the drop out too as oppose to Wilcoxon test.
If we are going for Dunn post hoc test, which value of P should be considered from the output? Whether its Sig. or Adj. Sig.?
Thank you....
Neither the Wilcoxon test nor the Dunn (1964) test analyzes the data in a manner similar to Friedman test. And so neither really makes sense to use as post-hoc for Friedman.
That being said, if a mismatch between the omnibus test and the post-hoc doesn't bother you, the Wilcoxon signed-rank test is meaningful. Note you still need the same participants for each of the time interval. There's also a paired-samples sign test.
There are several tests that are developed specifically for post-hoc tests for Friedman. The best source I can offer is to look at the following pdf, and read the descriptions for functions beginning with frdAll . You'll see there are several with names like Conover or Nemenyi. You have to be careful though, because there may be different tests ascribed to these authors. https://cran.r-project.org/web/packages/PMCMRplus/PMCMRplus.pdf
Note also that the Friedman test will have problems if you have participants dropping out at different time points. You might look at the Skillings–Mack test, which might be more appropriate for your situation. But finding a post-hoc test for this test is challenging.
- Badges
- Science topic
- Similar topics
- Statistics
- Survey