Hi all,

I've read through some other forum posts similar to this topic, but didn't think they were clearly mapped onto the situation I'm facing. I am analyzing psychotherapy data where some variables are collected only at intake and others (distress) are collected every single session. Data all come from US university counseling centers, so I'm curious about type of institution (specifically, if the institution is an HSI or not). So, I'm looking into a multilevel model with data nested by session, client, therapist, and counseling center.

My struggles are the following:
(a) Am I specifying the model correctly with time-invariant variables (e.g., gender and HSI-status)?:

mixed distress c.session c.social_support i.gender i.HSI_status || center: || therapist: || client:, mle

(b) When I add the time invariant variables (gender, HSI_status, social_support) to the model, it drastically reduces the number of observations noted in my output - goes from 194,000 observations (in a model without the time-invariant variables) to just under 12,000. I'm guessing this is due to the time-invariant variables, but I'm not certain what the implications of this reduction in observations is and if it is something I should be correcting for in another way.

(c) I'm also wondering about the interpretation of the time-invariant variables. I have one continuous (social_support) and two categorical (gender and HSI_status) - are these interpreted as a typical fixed effect (i.e., coefficient represents the difference between groups for the categorical predictors and the linear relationship for the continuous predictor)?

Thanks very much for your help.