Structural equation modelling - estimate means and intercepts?

My research question is related to the determinants of organizational commitment, and I am using survey data to estimate a latent variable SEM model in Stata. In AMOS, the default is to not estimate means and intercepts, however, in Stata the default is to estimate both. In my model, indicators for the latent variables are unstandardized answers from survey questions (either low to high satisfaction, or low to high agreement). Using factor analysis, I have confirmed the theorized structure of the data, and now am interested in the relationships between the latent variables.

Should I be estimating means and intercepts in this situation?


Phineas Packard
Hi Jesse,

If you are only interested in covariance based models (i.e. latent variable A regressed on latent variable B; or their correlation) then it does not really matter and the AMOS approach is fine.

If however you want to know if the latent mean for latent variable A is different to the Latent mean for latent variable B then yes you need to estimate the intercepts and latent means.