Multicollinearity in SEM

[pankaj singhi]

Hi ,

I am pretty new to SEM world and would like to understand in depth of how SEM model handles multicollinearity . Suppose I am running a structural model where in attributes within one measurement model or from different measurement models are correllated, and if I take the case of path model assuming that some attributes are significantly correllated , will this have bearing on final coefficients/fit indices ? Or SEM is smart enough to handle this?

My thinking says yes it will have impact but do not have any sound reason for it .

Please help me with some explanation on this 

[Edward Rigdon]

     I am going to guess that you are trying to approach factor-based SEM from some older texts where factors tend to be orthogonal. Take a look at the "lavaan project" tutorials:

http://lavaan.ugent.be/tutorial/index.html

I think it will be clear that which covariances are estimated and which are constrained to 0 is up to you. Generally speaking, if a nonzero covariance offers a better fit to the data, but yoor model constrains that covariance to 0, thsi will tend to be reflected in lack of fit. Collinearity has the same effects in factor-based SEM as in regression, but remember that collinearity invoves multiple predictors for the same dependent variable. In a common factor model, observed variables are often dependent on ony a single common factor (and an orthogonal unique factor or error term), so there is no collinearity.

     This is a very general SEM question, so it is better suited for a general list like semnet:

https://listserv.ua.edu/archives/semnet.html

--
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 post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

[Anonymous]

Hi,

Thanks for the response. A bit clear now on how in general collinearity is handled. I will still read some materials shared in the link and will get back if there is any confusion.

Here is my question in particular to lavaan only. After going through lot of questions in forum i got to know that its not pretty straightforward to understand Non positive definite warning of covariance matrix. My concern is when i run path model with lavaan, i never get this warning but when i run a proper structural model i very often get this warning. I believe(might be wrong) one of the reason for this warning is multicollinearity between variables. But not able to understand how multicollinearity is handled differently in case of path model and structural model.

Kindly help.