using -simirt- to simulate ordinal data

[Ariel Linden]

I am interested in generating a simulated dataset of ordinal items using the user-written command simirt (findit simirt). However, it is not clear from the help file how one goes about generating ordinal items (I contacted the author but have not received a reply).

I would be grateful if someone who is familiar with this command would provide me with some direction. For an example, assume 13 items, each item can be scored 1-4 (worst to best), and a single latent trait.

Thanks in advance!

Ariel

[Joe Canner]

Ariel,

I don't know anything about simirt and very little about IRT, but your question intrigued me so I gave it a shot anyway...

To get items with four levels you need to use the Rating Scale Model (RSM). Based on the example given in the help file, here is one possibility:

simirt , dim(13) rsm1(1 .5) clear

This generates 13 items which range from 0 to 3, so you will have to add 1. I'm not sure what the parameters in the rsm1() option signifiy, but putting one parameter results in three levels, two parameters result in four levels, etc. By default, the distribution for each item shifts from being skewed towards the low end in the early items to skewed towards the high end in the later items. This can be changed by using the diff() option. The diff() option needs one parameter for each item. A diff() parameter of 0 gives a unifiorm distribution, negative values skews it towards the high end and positive values skews it towards the low end. For example:

simirt , dim(13) rsm1(1 .5) diff(3 2.5 2 1.5 1 0.5 0 -0.5 -1 -1.5 -2 -2.5 -3) clear

For what it's worth, the Partial Credit Model (PCM) can also be used, but it's not clear how the difficulty parameters that are specified for each item are incorporated. Here is an example with four items:

matrix A=(0,1,2\2,1,0\-1,-2,-3\2,3,4)
simirt , dim(4) pcm(A) clear

Hope that helps a little....

Regards,
Joe

[Ariel Linden]

Thank you, Joe!!! This is very helpful.