
Message/Author 


Dr. Muthen, I am running a Bayesian Panel SEM. Two questions: 1. I have panel data, and I want to run a fixed effect instead of a random effect. In other words, I would like to include a time invariant component of the error that is correlated with the independent variables. I'm not interested in transforming the model with GLS as in random effects. I want to assume that the error is not necessarily just white noise. The command I have found is (TWOLEVEL), but it seems to be generally be paired with RANDOM. If I run twolevel by itself is that the same as a fixed effect? 2. Are there any theoretical issues with integrating panel into a Bayesian SEM model? Thank you so much for your help! Jessica 


We are probably coming from areas with different notation, but let me try to decipher and ask what you are asking. 1. When you say "a time invariant component of the error that is correlated with the indep vbles", I think of a random intercept growth factor model (which can easily be done). That is, a random (personspecific) effect that influences the outcome equally at all time points. But that is a random effect, not a fixed effect which you seem to want. TWOLEVEL works with random effects, that is, effects that vary across Level2 units. 2. No  Bayesian SEM of panel data is straightforward. 

Back to top 

