topic Re: Getting group specific model fit values out of proc irt in SAS Programming

Getting group specific model fit values out of proc irt

Daniel_Paul — Mon, 02 Mar 2015 11:00:12 GMT

Hello,

i would like to perform a multi-group analyses with proc irt. In addition to get the overall fit values of the model, i like to get the group specific model fit values out of proc irt (like in CALIS Output 25.25.5; SAS/STAT(R) 9.22 User's Guide). Has anybody an idea how i can reach my goal?

Many thanks.

Bye, Daniel

Re: Getting group specific model fit values out of proc irt

SteveDenham — Mon, 02 Mar 2015 20:04:26 GMT

Would the multiple group example at

http://support.sas.com/documentation/cdl/en/statug/67523/HTML/default/viewer.htm#statug_irt_examples03.htm

for version SAS/STAT 13.2 be a start in the right direction?

Steve Denham

Re: Getting group specific model fit values out of proc irt

Daniel_Paul — Tue, 03 Mar 2015 19:54:42 GMT

Absolutely right, but running this example with the additional statement

ods output FitStatistics=sm.asgash_sgfdfd_mft_test

just produce fit statistics for the overall model, not for the different groups separately.

Re: Getting group specific model fit values out of proc irt

SteveDenham — Wed, 04 Mar 2015 19:40:49 GMT

I think you want:

ods output ParameterEstimates=<some file name>

to get the estimates for the two groups.

In order to get separate fit statistics, you will have to run the model with BY GroupVar replacing group GroupVar. However, this will not allow enforcing the equality constraints on the intercept between groups. To enforce the within group equality constraint, you would probably have to run two completely separate analyses.

The point is that separate fit stats don't apply to the joint fit that employs the constraints listed--there are connections between the two groups that make the separate log likelihoods and any IC's not equivalent to two "parts".

Steve Denham

Re: Getting group specific model fit values out of proc irt

Daniel_Paul — Fri, 27 Mar 2015 15:17:50 GMT

I see; so the fitting function is over the group, right? i know the multi-group fitting equation by Bock & Zimowski (1997; pp. 433-448), but this guy is only for completely restricted item parameters estimates between groups. no free parameters across groups are allowed in this equation. do you know a resource were the underlying fitting equation of proc irt in the case of multi-group is depicted? And: Are there a possibility in proc irt to fix the item parameters to some giving values? i know how i can do this for the slop parameters (just using the factor statement), but how can i fix the item difficulty/threshold parameters as well?

Bock, R. D. & Zimowski, M. F. (1997). Multiple Group IRT. Chap. 25, pp. 433-448. In: W. J. van der Linden & R. K. Hambleton (Eds.). Handbook of modern item response theory. Springer: New York.