Modification indices?

[Jeanne Sinclair]

Hello Phil and thanks so much for your help. I am finally getting my models to converge with MIRT :) I'm comparing with results from lavaan and I was wondering if mirt has modification indices similar to lavaan's lavTestScore or modindices?

As of now my configural model in mirt has this model fit:

                          M2 df p RMSEA RMSEA_5 RMSEA_95 SRMSR.Eng_Not_L1 SRMSR.L1_Eng TLI CFI stats 10465.98 2288 0 0.04228521 0.04145549 0.04309736 0.1472131 0.1181285 0.8173768 0.8248777

with this factor structure:

Q2 <- 'FEELGLOBVOC = 13 , 16-21, 23-25 

GRAPH = 26-30

GRAM = 15 , 22 , 31-38  

LONGWSHORTW = 47-50, 43-46

NON = 1-12

RESP2READ = 39-42 '

I'd love to see how I can improve model fit without resorting to trial and error. Any advice would be very much appreciated!

Jeanne

[Phil Chalmers]

From the model alone, adding things like co-variance terms between the traits will help improve the model fit, as constraining traits to be orthogonal can be a little strict in the psychological sciences. 

Regarding modification indicies, these are generally much more difficult with IRT. Due to the marginal estimation nature of the estimation techniques we typically don't get efficient consequence statistics as a result after estimation (even when using the EM algorithm obtaining the asymptotic co-variance matrix itself is no trivial computation......something I wrote about very recently in the British Journal of Mathematical and Statistical Psychology). mirt has the lagrange() function, which is basically what these SEM statistics are based on, but they are in no way as cheap to obtain. HTH.

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.

[Jeanne Sinclair]

Thanks again. I'll check out the article. I ran itemfit() on the multidimensional model and got the following output, but perhaps itemfit() isn't to be used for multidimensional measures? 

     item    S_X2 df.S_X2 p.S_X2

1   rmc01  51.571      41  0.125

2   rmc02  76.663      39  0.000

3   rmc03  38.882      40  0.521

4   rmc04  56.868      38  0.025

.....

Phil

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.

[Phil Chalmers]

itemfit() can be used for multidimensional IRT models. 

Phil

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package+unsubscribe@googlegroups.com.