I am running a two-level model with one latent variable both with effects within and between (outcome variable is continuous). This variable is a dummy variable (4 categories, so I have 3 dummies).
As my latent variable is 0/1, a pearson correlation with the dependent variable would not be possible. So, where does the output of the correlations come from? I cannot reproduce these correlations in SAS or R (neither from the original variables nor from my proc mixed/lmer/stan_glmer multilevel regression outputs) but need to understand where they come from and why they are so different from the other regression outputs.
Sampstat gives correlations among observed variables (no latents included). It's unclear what your model is. You say "this variable is a dummy variable" - it sounds like "this" refers to your latent variable, that is, you have a mixture model.
That's right, my dummy variable is my latent variable. This variable actually causes some convergence problems (I'll have to take a look at that too).
My question is, why if it's the same dataset, do I see such different correlations between the observed variables than in SAS and R? I'm having a hard time understanding the calculations behind this output.
I have another question regarding the type of correlations in my output. I am using model estimator MLR in a model with covariates. Does this mean the correlations from the SAMPSTAT are probit residual correlations? (as suggested in the user's manual Ch.18?)