Calculating difficulty parameters per individual polytomous item in IRT using the Graded Response Model

[Leon Botter]

Dear Stata experts,

I am just learning to use Stata for my PhD and have a question I can't find an answer to after reading the manual, basic IRT literature or previous forum topics. I hope it's not all too elementary. I also assume it is not necessary to provide examples of my dataset, considering my question pertains to a general analysis method.

For a polytomous personality questionnaire (a 6-point likert scale) I want to place individual items on the theta axis, as to inspect for redundancy and possible item deletion. However, using the GRM models and graphs, I only receive output of difficulty levels (and ICCs) per answer category, opposed to per item. Is there a way to either modify the GRM analysis per commands, or manually calculate the difficulty per item using my output? I've consered just calculating the mean of the difficulty levels, but something tells me it's to simple to be true.

[ericmelse]

Dear Leon,

Besides consulting the Stata manual about its command irt:

h irt

I also can recommend the user contributed command uirt which has been published by Bartosz Kondratek (2022). uirt: A command for unidimensional IRT modeling. The Stata Journal, 22(2), 243–268.

Model results between these two commands are comparable if not identical but uirt has additional options that are particularly useful when you are dealing with graded response models.
To install use:

net describe st0670, from(http://www.stata-journal.com/software/sj22-2)

and use the mouse click event trigger (click here to install).
Should you also want to save the ANCILLARY FILES by using: (click here to get), you better first set your working folder to a location where you can save documentation and the do file, something like (or any other location on your system):

. cd "C:/D_Stata18/Tutorial IRT uirt"

For installation (of only the command) from the ssc server use:

ssc install uirt, replace

[Leon Botter]

Dear Eric,

Apologies for my late response, I wrongfully expected an e-mail notfication when my topic received a reply. Thank you kindly for making me aware of the existence of uirt. I will install it and explore its use for my analyses.

[ericmelse]

Great Leon,

You will enjoy uirt, and, indeed, you should get an email notification from the Statalist server when somebody creates a new response (I did).