You are here

Mxcompare Clarification

5 posts / 0 new
Last post
HomieWomie's picture
Offline
Joined: 06/12/2020 - 15:45
Mxcompare Clarification

Hi there,

This should be a quick question considering it is probably common knowledge for all of you. I am looking for some clarification regarding the mxcompare output. Just to be certain, is "diffll" synonymous with chi-squared? From the openmx user guide I cannot see where else a chi-square output would be located. If this is not the case, how can one determine the chi-square value from the usual mxcompare output?

Thank you in advance and please excuse my extreme naivety.

jpritikin's picture
Offline
Joined: 05/24/2012 - 00:35
diffll

diffll is the difference in model minus2LL. See ?mxRefModels

The chi-sq stat is generally the saturated model vs the candidate model. For example,

> mxCompare(sat.fit[['Saturated']], raw.fit)
                                         base
1 Saturated Raw Test Model to Check MxSummary
2 Saturated Raw Test Model to Check MxSummary
                         comparison ep minus2LL   df       AIC diffLL
1                               65 9186.911 4935 -683.0891     NA
2 Raw Test Model to Check MxSummary 31 9236.675 4969 -701.3251 49.764
  diffdf          p
1     NA         NA
2     34 0.03961048

Is saying the same thing as,

chi-square:  χ² ( df=34 ) = 49.764,  p = 0.03961048
HomieWomie's picture
Offline
Joined: 06/12/2020 - 15:45
Thank you!

Jpritikin,

I appreciate your response in regards to my question! This helps clear up so much for me. Best wishes.

AdminNeale's picture
Offline
Joined: 03/01/2013 - 14:09
Chi-squared also for model comparison

Asymptotically, under certain regularity conditions (Steiger et al 1985) the difference in -2lnL between a model and a submodel of it (created by equating parameters or fixing them at specific values) is distributed as chi-squared with df equal to the difference in the number of free parameters.

Various things can disrupt the regularity conditions. Mixture distributions, parameter bounds, non-linear constraints to maintain positive definiteness are some of them.

HomieWomie's picture
Offline
Joined: 06/12/2020 - 15:45
Thank you very much for the

Thank you very much for the clarification, Dr. Neale!