I'm trying to analyze data that violates the assumptions of an ANOVA (homogeneity of variances), thus I think I have to conduct a non-parametric test. My study follows a factorial completely randomized design: 3 sources 4 treatments 4 replicates per treatment*source combination (replicates are equal in size)

Dependent variable is seed germination.

The Kruskal-Wallase test was suggested to me, but is there a way to do it and test for both source, treatment, and source*treatment interaction? Furthermore, can I draw pair-wise comparisons between source*treatment combinations much like Tukeys LSD would do?

Could I use PROC GLIMMIX in SAS instead? If so, what are the assumptions that need to be met for that model and how do I make sure that I'm meeting them?



1 Answer 1


I will give a partial answer to this question even tought it's old, as I found it while looking for assumptions tests with GLIMMIX.

No, please don't do non-parametric unless you are really desesperate.

When not following assumptions of ANOVA you can: 1-Transform your data 2-Use a different model with different assumption.

Whitout more details on the data-type, I am not sure GLIMMIX is appropriate. GLIMMIX is good when you have a poisson or negative binomial distribution, such as encountered with count data.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.