|
|
Estimating interactions with sig-w |
|
Message/Author |
|
Samuel posted on Monday, January 30, 2006 - 1:08 pm
|
|
|
Hello Linda, hello Bengt, is it somehow possible to estimate interactions between latent (exogenous) variables when analyzing sig-w (since xwith requires raw data)? For instance, would it be appropriate to create product-terms and include them when estimating sig-w? I should say that I'am only interested in the within-model, but because of too many free model parameters compared to the cluster-level sample size, it is not possible to use type=twolevel with no between-model. Thanks for your time. |
|
|
I don't believe this is possible. You may have to ignore the clustering and realize that your standard errors will be too small and chi-square too large if you have clustering. |
|
Samuel posted on Tuesday, January 31, 2006 - 8:00 am
|
|
|
OK, I was afraid that this is impossible. Would it be appropriate to use type=complex to estimate an interaction (either with XWITH or product-terms (e.g. Ping or Kenny & Judd, etc.))? |
|
|
You can use XWITH with COMPLEX RANDOM. You cannot estimate a latent variable interaction in one step without XWITH. |
|
Back to top |
|
|