DIF with two-dimensional 2PL or Unidimensional or ....

[Leonardo Carlucci]

Hi all,

I am conducting a DIF on a scale with two correlated dimensions (r= .73; somatic and cognitive anxiety), I wonder if I should test DIF on the two-dimensional or one-dimensional model or DIF for each scale separately.
Which is the best way?

Best,

Leo

[Matt Diemer]

Hello,

I'd lean toward testing for DIF in two distinct one-dimensional models, to have a simpler model and (perhaps) more power to detect for DIF. However, I'm interested to hear what others would suggest as well.

Thanks,

Matt

[Phil Chalmers]

Good question. If the model demonstrates  simple structure (isolated unidimensional patterns), and there's no real dependence between the simple structures (e.g., no constrained slopes or something like that) then it technically makes little difference asymptotically, though if the correlation between the traits is non-zero and reasonably strong then you could gain some additional power by utilizing the multidimensional structure. However, numerical integration becomes a limiting time/computational factor, so there's that to consider too.

Basically, if your sample size is large enough then I don't see any real harm though in treating the structure as effectively unidimensional packets. 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...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/mirt-package/347d2bac-4b46-42bb-8fc5-2e30e9491fccn%40googlegroups.com.