PDA

View Full Version : Nonparametric Post-hoc tests?

this_barb
06-27-2008, 10:28 AM
I've done Kruskal-Wallis test on a 9 groups and I found statistical significance. The SPSS help file says to use the Mann-Whitney U test for pairwise comparisons. Is this the correct method? What about Bonferroni correction? I don't see a method to change the alpha level in the 2-sample nonparametric test section. Are there any other methods to do nonparametric posthocs? Thanks for your help.

Dragan
06-27-2008, 12:36 PM
An appropriate post-hoc test for the Kruskal-Wallis one-way ANOVA would be:

|RBari - RBarj| >= Zcrit * Sqrt [ ((N*(N + 1))/12) * (1/ni + 1/nj) ]

where:

RBari, Rbarj, ni, nj are the average of the Ranks and sample sizes from groups i and j

N is the total sample size

Zcrit is the critical value from the unit normal distribution.

Zcrit is determined as:

1. take the alpha rate (say 0.05) and divide it by k*(k-1) where k is the number of groups.

2. Find the value of Z that corresponds to adjusted alpha rate in Step 1.

Example,

if k=9, then 0.05/(9*8) = 0.00069444....

Thus, Zcrit = 3.19695.

I don't know of any software packages that will do this post-hoc analysis.

this_barb
06-29-2008, 04:04 PM
An appropriate post-hoc test for the Kruskal-Wallis one-way ANOVA would be:

|RBari - RBarj| >= Zcrit * Sqrt [ ((N*(N + 1))/12) * (1/ni + 1/nj) ]

where:

RBari, Rbarj, ni, nj are the average of the Ranks and sample sizes from groups i and j

N is the total sample size

Zcrit is the critical value from the unit normal distribution.

Zcrit is determined as:

1. take the alpha rate (say 0.05) and divide it by k*(k-1) where k is the number of groups.

2. Find the value of Z that corresponds to adjusted alpha rate in Step 1.

Example,

if k=9, then 0.05/(9*8) = 0.00069444....

Thus, Zcrit = 3.19695.

I don't know of any software packages that will do this post-hoc analysis.

Ah, alright, thanks. I'll just do a macro in Excel or something for that. What's the name of this procedure so I can read up on it more?

Dragan
06-30-2008, 01:50 PM
Ah, alright, thanks. I'll just do a macro in Excel or something for that. What's the name of this procedure so I can read up on it more?

See: pp 213-214, Section 8.3.3, Multiple Comparisons Between Treatments
in,

Siegel, S., & Castellan, N. J. (1988). Nonparametric Statistics for the Behavioral Sciences (2nd). McGraw-Hill: New York.

song_number_two
03-26-2012, 02:17 AM
I don't know of any software packages that will do this post-hoc analysis.

Hello! You can do it in R. http://www.r-project.org/

The post hoc results for the Kruskal-Wallis test you can get by using "pgirmess" package: