topic Re: Pairwise analysis following Kruskal Wallis testing in Statistical Procedures

Pairwise analysis following Kruskal Wallis testing

MaximeKurtz — Tue, 03 May 2022 09:21:28 GMT

Hi all,

I have a dataset that is seperated into 4 groups (disease1, disease2, disease3, disease4). For each individual, I have several variables to test (biochemical variables such as creatinine, urea...).

I coded a npar1way procedure to perform a Kruskal Wallis test to determine if there was a statistical difference for a given variable between the 4 groups:

proc npar1way 
	data=data_descriptive_workingfile wilcoxon;
	var  creatinine ;
	class group;
run;

This worked well.

I know want to compare the data within the classes : perform a Mann Whitney U Test that compares creatinine between Disease1 and Disease2, Disease 1 and Disease3, etc. I do not want to perform a Steel-Dwass-Critchlow-Fligner test because I would like to use another type of post-hoc analysis

Is there any way to code that ?

I am running on Sas OnDemand for Academics

Many thanks in advance for your help,

Maxime

Re: Pairwise analysis following Kruskal Wallis testing

sbxkoenk — Tue, 03 May 2022 10:46:28 GMT

Hello,

Is there any particular reason you are doing a non-parametric analysis for this research question?

If the Statistical Assumptions for Using PROC GLM ( https://go.documentation.sas.com/doc/en/statug/15.2/statug_glm_details01.htm ) are met, I would do this ANOVA with PROC GLM.

Then use LSMEANS statement for performing multiple comparisons.

Thanks,

Koen

Re: Pairwise analysis following Kruskal Wallis testing

PaigeMiller — Tue, 03 May 2022 11:57:02 GMT

One way to do a pairwise comparison of means following Kruskal Wallis is called "Dunn's Test". Code for this is here: https://communities.sas.com/t5/Statistical-Procedures/Dunn-s-Test/m-p/89558#M4394 As stated at that link, there are many macros for this that can be found via your favorite internet search engine.