nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Kruskal Wallace H test

Stuart_Little

Hi All,

I've got a customer that is looking to bring some work they've had conducted externally from Matlab into JMP. The question that they've asked that I haven't yet found the answer to is:

Can JMP do the Kruskal Wallace H test (sometimes called the one-way ANOVA on ranks)?

Can anyone shed any light on if this is possible, or if the Wilcoxon / Kruskal Wallis tests in JMP are functionally equivalent?

Thanks in advance

Stuart

Accepted answers

karencopeland

I am working on an update to the oneway analysis chapter in the help so interested to see this discussion. The current help says:

Wilcoxon / Kruskal-Wallis Test
Performs a test based on Wilcoxon rank scores. The Wilcoxon rank scores are the simple ranks of the data. The Wilcoxon test is the most powerful rank test for errors with logistic distributions. If the factor has two levels, the Wilcoxon test is equivalent to the Mann-Whitney test. If the factor has more than two levels, the Kruskal-Wallis test is performed.

From Wikipedia Kruskal- Wallis = Kruskal-Wallis H test = one-way ANOVA on ranks and from the Wadsworth Handbook On Statistical Methods for Engineers and Scientist (trusted source - go to bookshelf book as it has a little bit on lots) I have a sticky on the page for this test....and it says that the nonparametric extension of the Mann-Whitney-Wilcoxon test to k mutually independent random samples is the Kruskal-Wallis Test....

So, I think the menu item is a short cut notation as in it leads to two different sets of results. If X has 2 levels that option provides the Wilcoxon 2 sample test in addition to the one way ANOVA on ranks, if X has more than 2 levels you don't get the Wilcoxon 2 sample test (as you have more than 2) but you get the K-W test on the ranks.

Feedback more than welcome and timely as I work to improve the Oneway help doc.

All comments

ian_jmp

Not answering your question directly, Stuart, but FYI, the book by Peter G and David M is by far the best reference I've found for this kind of thing. I'm sure Malcolm would sign off a copy (and Amazon has it cheaper).

DonMcCormack

Stuart,

Is the customer asking if they are equivalent or will JMP produce the same answer as MATLAB? Even if they are equivalent (and I don't know if they are) it's possible that implementation differences between the software packages will cause the results to be different (hopefully only by a small amount). For example, how are ties dealt with? what algorithm is used to approximate the probabilities?, etc. This would be the case even if we're comparing different Kruskal-Wallace implementations.

karencopeland

I am working on an update to the oneway analysis chapter in the help so interested to see this discussion. The current help says:

Feedback more than welcome and timely as I work to improve the Oneway help doc.

Stuart_Little

Hi Don,

After some further clarification, the customer isn't trying to reproduce the exact number generated in Matlab (as you say this will inevitably end in disappointment), rather they're trying to replicate the workflow that their academic partner is conducting in JMP, and it appears that this will be achievable for them

Thanks

Stuart

Stuart_Little

Thanks Karen, this was the perfect answer to satisfy the customer. As these conversations continue, I'll try and get some feedback for you around the documentation for this.

Thanks again

Stuart