mixed mirt, fixed effects with polytomous IRT model

[Gregor Liegl]

Hi everyone,

I aim to estimate a polytomous IRT fixed effct model with the mixedmirt() -command. My data set: 20 variables with different number of response categories (2 to 6), 8000 responders, incomplete block-design: all participants answered at least 3 items, none answered the full item set, more than 3000 persons responded to at least 15  items. The tested fixed effect is an item-level covariate (4 different item-formats).

I tried several modifications of the following command, always with the same Error message:

>mixedmirt(data, model = 1, itemtype="2PL",fixed = ~ 0 + itemorder, itemdesign = id)
Error in LoadPars(itemtype = itemtype, itemloc = itemloc, lambdas = lambdas,  : 
  Item 1 requires exactly 2 unique categories

Same command works with dichotomized items. Is there a way to estimate polytomous fixed effect models?

I'm really looking forward to answers...

Thank you in advance,
Gregor

[Irshad Mujawar]

Hello Gregor,

Yous should try using polytomous itemtype, e.g. 'gpcm' or 'graded' as your data contains more than 2 response categories. 


Try this:

mixedmirt(data, model = 1, itemtype="gpcm",fixed = ~ 0 + itemorder, itemdesign = id)

I hope it helps you.


Regards,
Irshad

[Gregor Liegl]

Hi Irshad,

thank you for your answer. These itemtypes you suggested where the first I entered, unfortunately I got the same Error. Have you ever estimated a mixedmirt model with polytomous data successfully? I really wonder, what I do wrong...


Thanks,
Gregor

[Irshad Mujawar]

Hello Gregor,

As I do not have the same data which you are using I used Science data (which has 4 response categories for each of the item) and I did not use itemorder and id which I also do not have.


my output is:

mixedmirt(Science, model = 1, itemtype="gpcm")
Stage 3 = 664, LL = -1921.4, AR(5.60) = [0.30], gam = 0.0043, Max-Change = 0.0008


Calculating information matrix...


Calculating log-likelihood...


Call:
mixedmirt(data = Science, model = 1, itemtype = "gpcm")


Full-information item factor analysis with 1 factor(s).
Converged within 0.001 tolerance after 664 MHRM iterations.
mirt version: 1.9
M-step optimizer: NR


Information matrix estimated with method: MHRM
Condition number of information matrix = 782.6456
Second-order test: model is a possible local maximum


Log-likelihood = -1614.083, SE = 0.034
AIC = 3260.166; AICc = 3261.616
BIC = 3323.706; SABIC = 3272.939

It worked fine for me.

I again repeat that error can only occur if you are using a dichotomous model on polytomous data (more than 2 response categories)

[Phil Chalmers]

Irshad is correct here, mixedmirt supports polytomous item types (see the documentation). However, what it doesn't support is polytomous items with the itemdesign argument. The itemdesign is meant to define/explain clusters of items with systematic changes in intercepts/difficulty, which for polytomous items is largely ambiguous. The only items that satisfy this type of reasoning are rating scale models, which currently are not supported in mixedmirt (perhaps one day they will, but IMHO good rating scale items are extremely hard to come by, let alone adding further design constraints). Cheers.

Phil

--
You received this message because you are subscribed to the Google Groups "mirt-package" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Gregor Liegl]

Thank you very much Phil, that definitely clears up my issue...

Regards,
Gregor