[Rasch] compare nested models in Facets
rmt at rasch.org
Wed Dec 21 18:18:26 EST 2011
Iasonas, you are describing a conventional statistical technique for
comparing nested models. Please Google "testing nested models" for references.
In a measurement context, it often makes more sense to investigate whether
the elements of the 4th facet have substantively and statistically
different measures. This is one reason that Facets reports a fixed-effect
chi-square for each facet.
At 12/21/2011, you wrote:
>"The reason I need this, is to do the following: I run a model with 4
>facets and then
>I used the same data but removed one facet from the analysis. Presumably,
>if I deduct
>the "Data log-likelihood chi-square" of the second model from that of the
>second, then this gives me the improvement of the model fit for df1-df2
>where the df
>is the "Approximate degrees of freedom" for each model. Then, I assume that
>this is a chi-square with df1-df2 degrees of freedom in order to test if I
>model the 4th facet or not. Is this procedure correct to test the two
>models? Do we havea reference for this?"
Editor, Rasch Measurement Transactions
rmt at rasch.org www.rasch.org/rmt/ Latest RMT: 25:3 Winter 2011
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Rasch