WLSMV and reverse coding of variables

[Kiki Ganesan]

Hi everyone, 

I'm looking to conduct a CFA, constraining across timepoints with 9 measurements. I had a question about using WLSMV with mostly continuous variables. One variable in my data span is memory span (e.g. 1, 2...7). I was building a model and had a model converge well. 

I proceeded to reverse code one of my measures and oddly to my surprise - the model no longer converged. I've never quite encountered this with mlr for example and was wondering if anyone had any insights regarding this. 

Best, 

Keertana 

[Edward Rigdon]

The impact of reverse coding should be to reverse the signs of observed variable correlations. If sample size is low and model fit is somewhat marginal, it is possible that flipping the signs of correlations could produce an ill-conditioned correlation matrix. But there may also be an interaction with the specific algorithm that lavaan uses to optimize fit. This kind of pattern may not have been tested as heavily during software development.

Try setting starting values for the parameter estimates related to the reverse coded item. You might choose as starting values the parameter estimates from the model that converged but with reversed signs. You might also try a maximum likelihood solution, just to see if the behavior is specific to WLSMV.

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/1d586305-0f42-4d4c-aa53-eddc8352ae34n%40googlegroups.com.