topic Re: post-hoc test for Kruskal-Wallis-Test in New SAS User

post-hoc test for Kruskal-Wallis-Test

AndreaSch — Mon, 07 Jan 2019 20:26:21 GMT

Dear Sir or Madame,

I want to perform a post-hoch test for the Kruskal-Wallis test. I have already written my code for this test and it showed differences between the groups. So now, I red, that you can do a Bonferroni-post-hoc test. However, I cannot find a procedure for that. Does anybody know how to do it? Or can anybody show my another post-hoc test for my problem?

This is the used code for the Kruskal-Wallis test. Mabey it’s possible to extend it?

PROC NPAR1WAY DATA = Statistik WILCOXON;
CLASS Gruppe; 
EXACT; 
RUN;

I am running SAS version 9.4. on Windows.

I appreciate any help. Thanks in advance.

Regards

Re: post-hoc test for Kruskal-Wallis-Test

PaigeMiller — Mon, 07 Jan 2019 20:50:56 GMT

PROC GLM can do Bonferroni t-tests, using the MEANS statement and the BON option.

PROC MULTTEST will perform the post-hoc adjustments for Bonferroni (and other) tests.

I'm not aware of a non-parametric way to do this.

Re: post-hoc test for Kruskal-Wallis-Test

PGStats — Mon, 07 Jan 2019 20:52:42 GMT

Proc multtest is exactly for that purpose. Check the INPVALUES= option.

Re: post-hoc test for Kruskal-Wallis-Test

AndreaSch — Mon, 07 Jan 2019 21:51:10 GMT

Thanks both to you for your answers. I tried your advice. But I still have an error and I don't know how to solve the problem. My code is below. Besides I wrote something about raw p-Values. Can' I just use my measuring results? Do I have to compute new ones?

DATA Statistik;
INPUT Gruppe Raw_P @@; 
CARDS;
1	1320.527281	1	1294.800111	1	1242.468356
1.5	1211.42937	1.5	1418.558864	1.5	1129.583849
2	725.241616	2	1436.191316	2	1216.12421
2.5	1251.782715	2.5	710.0275351	2.5	1251.788858
3	1230.328215	3	1384.634321	3	1300.866933
3.5	1396.716718	3.5	1457.118658	3.5	1426.917688
4	1300.747017	4	730.510824	4	1236.251115
5	1404.1575	5	1354.503454	5	1330.99349
10	1476.694468	10	1416.509109	10	1380.101564
15	1401.714976	15	1364.866794	15	1408.147304
20	1148.021385	20	1239.254325	20	1239.254325
25	1217.440692	25	1193.010811	25	1112.880134
RUN;
PROC MULTTEST INPVALUES=Statistik bon;
RUN;

Thanks in advance.

Re: post-hoc test for Kruskal-Wallis-Test

ballardw — Mon, 07 Jan 2019 22:59:24 GMT

@AndreaSch wrote:

Thanks both to you for your answers. I tried your advice. But I still have an error and I don't know how to solve the problem. My code is below. Besides I wrote something about raw p-Values. Can' I just use my measuring results? Do I have to compute new ones?

I don't think p-values are allowed to be greater than 1 since they are probabilities.

This runs though I make no claim for appropriate values for the p-values.

DATA work.Statistik;
INPUT Gruppe Raw_P @@;
Raw_P = Raw_P/10000; 
CARDS;
1	1320.527281	1	1294.800111	1	1242.468356
1.5	1211.42937	1.5	1418.558864	1.5	1129.583849
2	725.241616	2	1436.191316	2	1216.12421
2.5	1251.782715	2.5	710.0275351	2.5	1251.788858
3	1230.328215	3	1384.634321	3	1300.866933
3.5	1396.716718	3.5	1457.118658	3.5	1426.917688
4	1300.747017	4	730.510824	4	1236.251115
5	1404.1575	5	1354.503454	5	1330.99349
10	1476.694468	10	1416.509109	10	1380.101564
15	1401.714976	15	1364.866794	15	1408.147304
20	1148.021385	20	1239.254325	20	1239.254325
25	1217.440692	25	1193.010811	25	1112.880134
RUN;
PROC MULTTEST INPVALUES=work.Statistik bon;
RUN;

Re: post-hoc test for Kruskal-Wallis-Test

AndreaSch — Tue, 08 Jan 2019 16:52:02 GMT

Thank you very much. Now my code is running. However, how can I interpret the result according to the post hoc Kurskal-Wallis test I want to perform? Now I get raw p-values and Bonferroni p-values. Every Bonferroni value is 1.0000.

Thanks in advance