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; } } }

How do I run a Test for Rank Order Data

jo_jardina

I am a new JMP user, but not new to stats. I want to run analysis with an independent variable that has 5 levels where participants ranked their preference for the levels from 1st to 5th.

In SPSS I would do this by running a Friedman test, but everything I read in the JMP tutorials says I should run a Wilcoxon / Kruskal-Wallis Test. I have set up the data to run the Wilcoxon according to the JMP tutorials online (specifically this one: https://www.jmp.com/support/help/14-2/example-of-the-nonparametric-wilcoxon-test.shtml#1023759).

If I run the test with the ordinal data column set as ordinal (which is the correct label) I cannot run the test as the JMP tutorial has laid out; however, if I run the Wilcoxon test with the data labled continous I can run the test. I am fearful I am getting the incorrect results.

I do not know what to do. Am I supposed to run the test with the data labled as continous, even though it is ordinal? Am I not running the correct test? Any guidance is appreciated.

Accepted answers

Dan_Obermiller

I misunderstood when you said you would use a Friedman procedure. The Friedman test is a two-way ANOVA, the Kruskal-Wallis or Wilcoxon test is a one-way ANOVA.

Your response should be continuous and your independent variable is ordinal or nominal. Since you do have ranks, I agree that the modeling type could be considered as ordinal, but to get the proper test, make it continuous. Your results will still be correct.

Think of it this way. Suppose your data looked like this (only a few observations for a small example):

Group 1: 83.6, 92.1, 76.2, 84.9

Group 2: 63.2, 79.3, 77.8, 82.0

Now form the ranks:

Group 1: 6, 8, 2, 7

Group 2: 1, 4, 3, 5

Now pretend the ranks are continuous and is your response and go through the steps again. You end up with the same ranks. Therefore, JMP will create ranks based on your ranked data (essentially, unchanging the response) and performing the test. Correct methodology and correct results even though you do not have the "true continuous" response.

All comments

Dan_Obermiller

The Friedman test is not supported directly by JMP, but is quite easy to get, especially since you already have the ranks. This posting details how to do it:

Friedman Test in JMP

jo_jardina

Thanks, but the data is one-way, as there is only one independent variable with five levels, so I cannot run a Friedman test the way it is outlined in the link.

Dan_Obermiller

I misunderstood when you said you would use a Friedman procedure. The Friedman test is a two-way ANOVA, the Kruskal-Wallis or Wilcoxon test is a one-way ANOVA.

Think of it this way. Suppose your data looked like this (only a few observations for a small example):

Group 1: 83.6, 92.1, 76.2, 84.9

Group 2: 63.2, 79.3, 77.8, 82.0

Now form the ranks:

Group 1: 6, 8, 2, 7

Group 2: 1, 4, 3, 5

jo_jardina

Thank you! I was worried about if the results would be correct or not with me setting up the data as continuous.