item fit statistics for bifactor model

[Matthew Sunderland]

Hi Phil,

Thank you for putting together a fantastic program for IRT and adaptive testing. I have a question about a particular analysis I am conducting. 

I am trying to fit a graded response bifactor model to an item pool of 133 items (5 point response options) with one general and five specific factors for the purpose of a CAT. The bifactor model provides better overall fit in comparison to a unidimensional model (and other bifactor models with fewer specific factors) according to the AIC and BIC. Furthermore, the residual matrix for the bifactor model indicates relatively low (close to zero) residual correlations and SRMSR = 0.047. 

However, when I test for item fit using the following code: 

ifitbi<-itemfit(b2mod2, fit_stats="S_X2", QMC=TRUE) 

The results suggest that virtually all of the items have poor fit p<0.0001. 

I'm a little confused about what to make of the results and whether i'm using the correct code to generate the item fit statistics?

When I ran the item fit statistics for the unidimensional model they indicated that all the items fit well p>0.05 but the overall fit was worse in comparison to the bifactor model. Do these results point towards a unidimensional model, bifactor model, or neither?

Many thanks, 

Matthew. 

[Phil Chalmers]

Hi Matthew,

That does sound strange, I wouldn't expect the item-fit to get worse either. Could you provide a reproducible example and script file? It will help me chase down the problem a bit faster (or determine if one exists at all). I know that the S-X2 statistic can get a little weird for longer tests, but for multidimensional tests less is known. 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+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.