Testing whether the rank ordering of items between two group

Discuss statistics related things
Post Reply
Whirly123
Posts: 31
Joined: Mon May 06, 2019 3:07 pm

Testing whether the rank ordering of items between two group

Post by Whirly123 »

Putting this really simply, I will have data from two different groups of participants where they rank order the "moral worth" of a set of non-human animals. I want to be able to test whether the rank ordering between the two groups is significantly different.

Is there an appropriate test in Jamovi for this sort of analysis? A cursory look at different ways to test this leads me to State-trace analysis which might be too in-depth for what I need (or maybe that is exactly right, just wanted to check if there was a simpler approach).
User avatar
MAgojam
Posts: 421
Joined: Thu Jun 08, 2017 2:33 pm
Location: Parma (Italy)

Re: Testing whether the rank ordering of items between two g

Post by MAgojam »

Hi, @Whirly123.
I think a simpler approach than STA is to try the Friedman test you find in jamovi [Analysis-> ANOVA-> Repeated Measures ANOVA (Non-parametric)].

Regards,
Maurizio


Update
Maybe Jonathon can say more for STA.
Here's something of hers.
https://www.researchgate.net/publicatio ... e_analysis
Whirly123
Posts: 31
Joined: Mon May 06, 2019 3:07 pm

Re: Testing whether the rank ordering of items between two g

Post by Whirly123 »

That doesn't seem quite right (unless I doing it wrong).

My data could look like this:
dat.jpg
dat.jpg (42.58 KiB) Viewed 6178 times
And Friedman's won't let me test between groups 1 and 2.
User avatar
MAgojam
Posts: 421
Joined: Thu Jun 08, 2017 2:33 pm
Location: Parma (Italy)

Re: Testing whether the rank ordering of items between two g

Post by MAgojam »

Whirly123 wrote:That doesn't seem quite right (unless I doing it wrong).
My data could look like this:
Hi, @ Whirly123.
With the recorded data (wide type) as in the screenshot you attached, you can't do much with the Friedman test available in jamovi, but most of all you can't check if the order of the ranks between the two groups is significantly different.
The data should be organized in long type, and aggregated on the mean of the ranks before a Friedman test with the verification of the two groups.
For simplicity, if you find it useful, I am attaching a jamovi file with fake data similar to yours and a small R script in the Rj module.
The R code will create a data frame from wide to long mode and aggregate them with the rank averages and then the friedman.test() function will do the required verification for the two groups.
moral_worth.omv
(4.97 KiB) Downloaded 306 times
Cheers,
Maurizio

Suggestion:
You may be interested in the book "Statistical Methods for Ranking Data" by Mayer Alvo and Philip L.H. Yu, Springer 2014.
Millet
Posts: 2
Joined: Tue Apr 19, 2022 3:13 pm

Re: Testing whether the rank ordering of items between two g

Post by Millet »

Hello,
I'm trying to do a non-parametric repeated measures ANOVA, but I want to add a "group" variable to compare the difference in means between two different groups.

I can't do it with a Friedman ANOVA, how can I do it?

Thank you in advance for your answers.
Millet
Posts: 2
Joined: Tue Apr 19, 2022 3:13 pm

ANOVA Friedman

Post by Millet »

Hello,
I'm trying to do a non-parametric repeated measures ANOVA, but I want to add a "group" variable to compare the difference in means between two different groups.

I can't do it with a Friedman ANOVA, how can I do it?

Thank you in advance for your answers.
User avatar
jonathon
Posts: 2613
Joined: Fri Jan 27, 2017 10:04 am

Re: Testing whether the rank ordering of items between two g

Post by jonathon »

i don't think this is possible ... at least not with the basic stats tools we typically use.

perhaps someone else can suggest something.
Bobafett
Posts: 76
Joined: Thu Jul 18, 2019 11:33 am

Re: Testing whether the rank ordering of items between two g

Post by Bobafett »

I would have thought a more straightforward option would be to calculate the sum total of each person's ranking across all animals - then just use a Mann-Whitney to determine if these summed totals (or means of ranks if you'd prefer) differ across the two groups.

Or are you are trying to determine if there are differences between the two groups for each individual animal? If so, then you'd need to do Mann-Whitney for each animal, but there are issues with running analyses based on a single measure - it'd be akin to separately analysing each item from a questionnaire rather than the total score....
Bobafett
Posts: 76
Joined: Thu Jul 18, 2019 11:33 am

Re: ANOVA Friedman

Post by Bobafett »

Millet wrote:Hello,
I'm trying to do a non-parametric repeated measures ANOVA, but I want to add a "group" variable to compare the difference in means between two different groups.

I can't do it with a Friedman ANOVA, how can I do it?
What you are suggesting could be achieved a factorial ANOVA, but this not possible with non-parametric analyses. Do you absolutely have to use non-parametrics? You could try running the Friedman's ANOVA and then a separate Mann Whitney to look for overall group differences?
User avatar
reason180
Posts: 274
Joined: Mon Jul 24, 2017 4:56 pm

Re: Testing whether the rank ordering of items between two g

Post by reason180 »

I would take the following approach:

(1) Identify some standard against which each person's ranks can be compared--perhaps some natural ranking of the items on some objective dimension.

(2) Compute some sort of correlation coefficient, for each participant, describing the relationship between each participant's ranks and the standard rankings. (They could be Kendall's Tau-b coefficients, for example.)

(3) Perform some sort of non-parametric, two-group comparison of the coefficients in the Group A participants against the coefficients in the Group B participants. A good comparison would be a non-parametric bootstrap of the difference between the two medians of coefficients. Alternatively, you could do a Mann-Whitney U, which would pick up any differences between the two distributions (not just differences in central tendency).
Post Reply