|
|
 |
| Modeling L1 variables at L2 |
 
|
| Message/Author |
|
|
| Erik Ruzek posted on Wednesday, June 03, 2020 - 2:31 pm
|
|
|
I am running a twolevel model with most of the covariates at level 1. In my level 2 model, I am only interested in using a handful of these same variables as predictors of the outcome. How would you suggest handling the level 1 variables not in the level 2 model?
I'm considering:
1) To preserve the latent mean centering at level 1, I could estimate their variances.
2) To preserve the latent mean centering at level 1, I could estimate their variances and covariances.
Model fit (SRMRb) is improved if I do #2, but I'm interested in which is the more statistically sound approach. Which would you recommend and why? |
|
|
|
|
| 2) is probably best - assuming you correlate them also with the level 2 DVs. You could also do observed-variable centering for those variables not included in level 2 and declare them as Within=. |
|
|
| Back to top |
|
|
