3
$\begingroup$

Within a particular instrument, is it valid to combine multiple response models for subsets of items when estimating a latent trait?

As an example, if I had some items that were conceptually suited by an unfolding model and others by a cumulative model, would the latent traits share a common scale?

If so, under what conditions would this be true?

Many thanks,

Michael

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

This is perfectly valid to do. So long as the items measure the same construct (or multiple traits in similar ways) then the item response model does not matter. This is part of the reason why packages such as IRTPRO or mirt in R have the ability to change the respective IRT models for each item. As long as the select model provides a good description of the response behavior (i.e., provides a good fit according to goodness-of-fit criteria, among other things) then you should have little worries about mixing IRT models.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thank you, Phil. This makes sense intuitively and when thinking about response models as generative. $\endgroup$
    – Michael
    Commented Mar 5, 2019 at 11:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.