Hi Colleagues,
As you know, the homogenous subsets are automatically formed in Post hoc analysis (Duncan, Tukey etc.) using One-way ANOVA in SPSS. But, when data are classified as non-parametric the Kruskal-Wallis analyses should be done to compare means of the groups. After this, when the difference between the groups is statistically significant, non-parametric post hoc analysis, such as Bonferroni, Dunnet, Tamhane etc. should be performed. But, we can't see the homogenous subsets in the result of the post hoc analysis, the all groups are compared with each other in Pairwise Comparisons tables.
Is there any standard method for creating the homogenous subsets according to the result of the post hoc analysis after the Kruskal-Wallis test?
Thanks for your answers in advance.
A Post hoc test identifies the differences between means. In ANOVA, if the null hypothesis is rejected, then it is concluded that there are differences between the means. I am not sure what kind of homogeneity you expect from comparing means. If you have several variables you want to use to subgroup, you could use a cluster analysis.
Hope that helps.
As a preliminary note, the K-W test does not test for a difference in means.
* * *
I'm not an SPSS user, but it looks like the Homogeneous Subsets output is essentially what people would think of a Compact Letter Display. You can calculate this by hand.
The following is the procedure.
First order your groups by median or mean. Assign the largest group the letter A. Look at the comparison between that group and the next largest group. If they are not different, assign that second group an A; if they are different, assign no letter. Continue comparing each group to the first group.
Now go to the second largest group, assign it the letter B, and follow the same procedure of comparing each of the groups to this second largest group.
Continue for all groups.
When you are done, you will need to remove letters that are subsumed by another letter. For example, if all B's also have A's, you can get rid of all B's.
Each letter will indicate a "homogeneous subset." And you can have a group in more than one subset. For example, you can end up with a group with AB, which is in both the A group and the B group.
* * *
Dear Salvatore S. Mangiafico,
First of all, thank you for your interest. If you interpret the results of the attached files, I'd be grateful.
You may see the pairwise comparisons table with the file name "Pairwise-Comparisons-table.png", and the created homogeneous subsets taking into account the adjusted significance values, determined in the previous table, with the file name "Procedures.png".
Yours sincerely.
Let me propose a more general solution.
The attached .txt file is R code. You can run it by copying and pasting it in the lower window of this website. http://www.r-fiddle.org/#/
Double check that I have the data correct.
The input is a little clunky, but works. The threshold can be changed. I used the adjusted p-values for this example.
The result obtained from "R" is given below. It is suggested that the lettering is done correctly in the attached file. I have understood the method for creating the homogenous subsets. Dear Dr. Mangiafico, thank you so much for your helping.
June a
July ab
August abc
May bc
September abc
April c
Dear researchers, from the messages I received from private, I concluded that it is necessary to simplify the subject and produce the codes to be pasted into the interface of the R program through an Excel file. If you follow the process steps defined in the attached file, you will reach the result.
Please write to me at [email protected] for feedbacks.
Best regards.
- Badges
- Science topic