[Rasch] Rasch vc CFA

Svend Kreiner svkr at sund.ku.dk
Sat Sep 29 00:18:17 EST 2012


There are two answers to your question depending on how you would do the CFA.

If you treat the items as continuous variables during the analysis you are simply using the wrong model. Items are categorical and should be treated as such. So never do that.

If you regard the items as discrete variables during the analysis using polychoric or tetrachoric correlation coefficients you will use a model that is close to the graded response model if item are polytomous and close to a 2-parameter model if items are dichotomous. There are technical differences concerning how you estimate the parameters and how you test the fit of items to the model, but there are no fundamental differences between CFA models for categorical data and other IRT models.

The answer to your question is therefore the answer to a question of why and when one should prefer another model to the Rasch model. Or, perhaps, why we prefer the Rasch model to all other models.


At 9/27/2012, you wrote:
>Dear all,
>I have spent many years using various Rasch models in order to 
>investigate the unidimensionality of my scales, tests etc, and this 
>list supported me reliably (thank you!). The Rasch model served me 
>faithfully and produced a very descent number of publications over many 
>years. I understand the strengths very well (and some limitations). 
>However, the more I get to know CFA (confirmatory factor analysis), the 
>more I am puzzled as to why and when one should prefer CFA over Rasch (and vice-versa).

Rasch mailing list
Rasch at acer.edu.au
Unsubscribe: https://mailinglist.acer.edu.au/mailman/options/rasch/svkr%40sund.ku.dk

More information about the Rasch mailing list