Degrees of freedom in Onyx
I'm exploring Onyx with great interest as a possible recommendation to my students (instead of AMOS).
In order to familiarize myself with Onxy, I am comparing the results of the Tutorial CFA example with those from AMOS and Mplus. The results in AMOS and Mplus are very similar to each other. Results in Onyx are similar for parameter estimates, but very different for model fit.
It appears that Onyx is calculating the degrees of freedom differently.
In the example, there are three indicators [for which I added one latent factor]. If I remember correctly, that should give me 6 observed (co-)variances: 3*(3+1)/2 = 6. However, Onyx tells me that there are "Observed Statistics: 9.0".
Consequently, the model isn't just-identified (as I expected), but has 3 df.
Why is that the case? Is there a way to make these settings more comparable across software packages?
Any help would be greatly appreciated!
Heiko
Means
I'm not an Onyx expert, but my guess would be that Onyx is including the means as observed statistics. The covariances (6) plus the means (3) would give 9 observed statistics. If you add a model for the means where all of them are freely estimated, I'm betting you'll get the same fit as other programs. Of course this is just my educated guess.
Cheers,
Mike Hunter
Log in or register to post comments
Hi Heiko! Thanks for noticing
Thanks for noticing this. Mike is right insofar as Onyx always assumes a means model. We have introduced the distinction between "saturated means" and "explicit means" to allow users to do covariance-only modeling without caring about the means (saturated means) and explicit modeling of the means (explicit means). In the former case, Onyx still estimates the mean structure internally but hides all related estimates. Still, the saturated means structure uses up degrees of freedom (matching your intuition) and they need to be subtracted from restricted dfs. In consequence this means that you have found a bug in Onyx. It seems that the dfs are not correctly calculated in your case. We will fix this asap.
Best regards,
Andreas
Log in or register to post comments