I'm doing repeated measurement ANOVA, but one of my conditions does not meet the normal distribution, so can I use Scheierer Ray Hare test instead?
I'm not familiar with this test, but it looks like it is for independent designs - so probably not.
However, the good news is that repeated measurement ANOVA does not assume a normal distribution of the outcome variable. It assumes normality of the errors of the model. This is assessed by looking at the residuals of the model not the raw scores. So it may be that RM ANOVA is just fine.
If not you might consider a transformation or (more principled) a multilevel generalised linear model among other options. It would require more information about the data and distributional issues to assess the best approach.
Hello Song,
No.
The Scheifer-Ray-Hare test is an extension (to two factors) of the Kruskal-Wallis test, which is intended for comparing independent groups/batches. Repeated measures designs involve dependent scores, so an alternate approach would be called for.
If you have a mixed design (e.g., one between-subjects factor and one within-subjects/repeated measures factor), then consider the aligned ranks procedure. Here's a link which explains: https://rcompanion.org/handbook/F_16.html
If all factors are repeated measures, then consider an exact test or resampling/bootstrap procedure. No need for distributional assumptions with either. Here's a link that explains bootstrap: http://math.furman.edu/~dcs/courses/math47/R/library/Hmisc/html/rm.boot.html
Good luck with your work.
Samuel Oluwaseun Adeyemo, your statement that the Scheirer-Ray-Hare test is "a nonparametric alternative to repeated-measures ANOVA" contradicts statements from both Thom Baguley and David Morse. The Wikipedia page describes it as an extension of the Kruskal-Wallis test "to the application for more than one factor". There is no mention of repeated measures. Can you direct us to another resource that discusses use of this test for repeated measures or mixed designs? Thanks.
- https://en.wikipedia.org/wiki/Scheirer%E2%80%93Ray%E2%80%93Hare_test
Thank you. I want to supplement my research design. My research design is a 2 * 2 mixed between- within subject design .The S-W test results show that one group does not meet the normal distribution, so I can't do ANOVA.
"The S-W test results show that one group does not meet the normal distribution, so I can't do ANOVA." That's not what you need to check - look at the distribution of residuals not the "group" scores. Its best to do that graphically as the Shapiro-Wilks test doesn't indicate the degree of violation.
Information about the type of data might be helpful as (e.g., an ordinal or count model might be a better option for those data types).
Dear professor David Morse
Thank you so much for your assistance.I have already implemented the aligned ranks and the results were comparable to the RM-ANOVA.Once again, thank you for your help.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude