Author Topic: Model comparison with zero correlations???  (Read 433 times)

Georg II

  • Newbie
  • *
  • Posts: 21
    • View Profile
Model comparison with zero correlations???
« on: June 29, 2012, 05:43:40 AM »
Hallo Ray,
I’m comparing the fit of two models: a 5-dimensional within item model and a 4-dimensional between item model.

1.Are these two models nested? Can deviance be used to compare them?

In the 5-dimensioanl model I have set all the correlations equal to zero while in the 4-dimensional model, dimensions are allowed to correlate. Deviance shows that the 4-dimensional model fits better.
However, when in the 4-dimensional model the correlations among the dimensions are set to zero, the 5-dimensional model fits better. 

2.Which model fits better, the 4-dimensional or the 5-dimensional?

I just thought that in order to make all the conditions equal to make a fair comparison I have to set the correlations in the 4-dimensional model to zero, too. Besides, the dimensions in the  4-dimensional model are test format dimensions. We don’t expect test formats (for example, multiple-choice and gap-filling) to be correlated.

3.Does this make sense?

Thanks
Georg II

ConQuest

  • Global Moderator
  • Newbie
  • *****
  • Posts: 24
    • View Profile
    • Email
Re: Model comparison with zero correlations???
« Reply #1 on: June 30, 2012, 03:29:51 AM »
Best to post your control file so I can review it.  In general a model is a sub-model of another model if the sub-model can achieved by constraining parameter values of the super-model. Typically (but not always) the sub-model will be a special case of the super model with some parameters fixed at zero