Development of Short Form Scale using mirt package

[Yuen Wan Ho]

I am new to IRT. At the moment, I am developing a short scale. I have some questions in using mirt package for scale development. 

In my project, the response format of the full scale is dichotomous. There is only 1 factor in this full scale, with 7 sub-scales. Each sub-scale has 10 items (i.e. in total 70 items).

I read your paper (mirt: A Multidimensional Item Response Theory Package for the R Environment), I saw your example 4.1 with the mirt ( ) function. It appears that the results of mirt ( ) are similar to the results of traditional factor analysis. 

For the steps of developing the short scale, I wonder whether I can select items with the high item-total correlation from each sub-scale in the original  data set to form the short scale first. Then I split the original data set into two and conduct mirt ( ) in each split data set to examine the factor structure of the short scale between split data set. 

Finally, I want to examine the goodness of fit of the short scale. When I type M2, I find additional indicators for RMSEA: RMSEA_5 and RMSEA95 , what are the differences among RMSEA RMSEA_5 and RMSEA95?

Thanks alot for your help!

[Phil Chalmers]

On Thu, Mar 12, 2015 at 11:56 PM, Yuen Wan Ho <spoo...@gmail.com> wrote:

I am new to IRT. At the moment, I am developing a short scale. I have some questions in using mirt package for scale development. 

In my project, the response format of the full scale is dichotomous. There is only 1 factor in this full scale, with 7 sub-scales. Each sub-scale has 10 items (i.e. in total 70 items).

You may want to look into the bfactor() function if those subscales have residual covariation.

 

I read your paper (mirt: A Multidimensional Item Response Theory Package for the R Environment), I saw your example 4.1 with the mirt ( ) function. It appears that the results of mirt ( ) are similar to the results of traditional factor analysis. 

Indeed it is a full information item factor analysis, and when the IRT model used is the graded response model (2PL as a special case) then then MIRT and factor analysis with a polychoric matrix (with weights applied) are asymptotically equivalent. Other for IRT models this isn't true though.

 

For the steps of developing the short scale, I wonder whether I can select items with the high item-total correlation from each sub-scale in the original  data set to form the short scale first. Then I split the original data set into two and conduct mirt ( ) in each split data set to examine the factor structure of the short scale between split data set. 

You could, but I generally prefer looking at all items at once in exploratory models and weeding things out from there.

 

Finally, I want to examine the goodness of fit of the short scale. When I type M2, I find additional indicators for RMSEA: RMSEA_5 and RMSEA95 , what are the differences among RMSEA RMSEA_5 and RMSEA95?

90% confidence interval for the RMSEA version with the M2 statistic. Cheers.

Phil 

Thanks alot for your help!

--
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.