Multicollinearity - std.all - what can I do

[Daniel Groß]

Hey group,

I calculated a path analysis with two interactions (sample size 155). Howeve,r the std.all variable was for some variables higher than 1 (see graphic). Is this right that this might be a result of multicollinearity? How can I test this in lavaan? Does this mean the model was not quite good even though the model fit. I know that this is more of a general question but I am grateful for any help. was ok? 

Thank for your help.

Daniel

[Edward Rigdon]

Two things:

1. Standardization and interaction do not go together. IF A and B are normal, then variance(A*B) = variance(A) * variance(B) + 2 coviance(A, B) squared (going back to Kenny and Judd 1984). So the variance of the product cannot be 1, even if the variances of the components are 1, unless the components are mutually orthogonal. So, for certain, collinearity is involved.

2. Your results suggest a wide range of variances of the observed variables. That can cause problems in SEM. I suggest that you try to reduce these differences by multiplying small-variance variables by some constant, like a power of 10.

     I notice that your standardized estimates are the same as the raw estimates--only the "std.all" estimates change. Makes sure that you don't have large differences in variance across those variables,a nd make sure that standardization makes sense for those variables.

--
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.