topic Re: Data about rankings in Discussions

Data about rankings

gav2013 — Sat, 10 Jun 2023 23:32:12 GMT

Hi there,

what are the options to analyse rankings data in JMP? Do they implement the PlackettLuce approach?

I would like to see the conditional ranking to other covariates.

Thank you

Joseph

Re: Data about rankings

P_Bartell — Tue, 22 Jun 2021 13:21:57 GMT

A basic JMP design concept is that one has to define a data type for variables as a precursor to analysis. Some example data types are numeric, nominal, ordinal, text, multiple response, and more. So for rank data, ordinal is probably your best option. From there, then one typically is looking to either describe or explain something about the variables. These actions are supported by what are called 'analysis platforms' in JMP. Many, many JMP analysis platforms can work with ordinal data so the options for analysis are many and widespread. My initial suggestion is to peruse the JMP online documentation, perhaps searching for specific analytical methods. Here is a link to the online documentation 'home page':

JMP Online Documentation

Re: Data about rankings

gav2013 — Tue, 22 Jun 2021 15:05:09 GMT

Thank you for reply.

I am aware of the basic aspects of JMP or statistics. What I was asking if there is the possibility of doing modeling with rank data in JMP?

I have 4 options ranked one to four by respondents and I want to see the conditional probabilities of ranking, the optimal ranking and the distance metrics for rank data. I can't find anything of analysing rank data in the JMP manuals.

Thanks.

Re: Data about rankings

Kevin_Anderson — Tue, 22 Jun 2021 15:54:16 GMT

Hi, gav2013!

If you consider each choice in the ranking as a multinomial observation, with one item observed out of a possible set, you can apply a multinomial-Poisson transformation and express the model as a log-linear model, where the response is the count (1 or 0) of each possible outcome within each choice. Then, the model can then be fitted using Generalized Linear Models, which is a personality available in JMP Fit Model platform.

This approach can run into problems when you have lots of items to rank (lots > 10?) because you have to set up dummy variables to represent the presence or absence of each item in each choice and set up a factor to identify each choice. The combinations explode pretty quickly and you can challenge memory capacity. You don't mention how many items you have ranked to model.

It might be easier for you to utilize JMP's ability to run R. R has several packages that model ranks, including Plackett-Luce models.

Good luck!

Re: Data about rankings

gav2013 — Wed, 23 Jun 2021 07:42:19 GMT

Where can I find the addin to interface with R within JMP?

Thank you

Re: Data about rankings

Kevin_Anderson — Wed, 23 Jun 2021 15:07:55 GMT

It's best done with scripting. Try to understand the JMPtoR_bootstrap.jsl script in the Sample Scripts directory, accessible under the Sample Data selection under Help. This will give you an idea about how to run R from JMP.

The ability to easily move data and reports back and forth between the two products is really powerful!!