Model comparison for invariance testing
Hi everyone
I'm conducting invariance testing using SEM in JASP. I'm using the robust MLR estimator. As far as I can tell, I cannot compare models using by adding model 2, etc (as any change I make to one model seems to replicate in the other models. . . instead of being independent from the other models).
That's fine. I can just duplicate the model and add constraints. But, I would like to compare the fit of the models using the Santora-Bentler chi-square. However, the formula for doing so requires knowing the scaling factor - which JASP doesn't seem to produce.
I'm relatively new to JASP so I may be missing something. Thanks in advance for any suggestions!
Seth
Comments
I'll forward this to our expert. This seems like a bug though, in which case a report on GitHub would be most welcome. But I'll have our expert look at this first.
E.J.
Thanks. I appreciate it.
Seth
Dear @sethakap,
As far as I can tell, I cannot compare models using by adding model 2, etc (as any change I make to one model seems to replicate in the other models. . . instead of being independent from the other models).
That is indeed strange! It should be possible to change the models independently by specifying different syntax. Could you provide a JASP file/screenshots that demonstrate this issue? You can send it to s.kucharsky@jasp-stats.org if you want to keep the data confidential.
Attached is an example jasp file that follows the examples from https://lavaan.ugent.be/tutorial/groups.html showing equality constraints in a multigroup SEM - which hopefully shows that the model comparison is possible.
It is true though that you cannot place different equality constraints using the checkboxes in the "Multigroup SEM" section - these are passed globally to each model using the group.equal argument of lavaan (this is shown at the bottom of the lavaan tutorial linked above). Note that this option is just a shortcut for the lavaan syntax and you should be always able to use custom constraints per individual models using the lavaan syntax. But I do see that it's not the most practical to do it that way, perhaps we can improve this functionality.
Let me know if this clarifies the issue!
Hi
Ah! Got it. That makes sense. Yes, I'd been using the checkboxes to do this. Sorry I missed that in the tutorial. Yeah, I do think that if you were able to allow for testing models using the checkboxes going forward, that would be even easier. I'm teaching an SEM class right now, and I'm sure students would benefit from that. But, using the lavaan syntax certainly works too and is straightforward enough Thanks again for your help!
Seth