Retrieving variables from gsem estimation

[Vinicius Lima]

Hi Statalist,

consider example 31g from the gsem help files:

use https://www.stata-press.com/data/r16/gsem_cfa, clear

gsem (MathAb -> q1-q8, logit) (MathAtt -> att1-att5, ologit)

Is there any way to get the data used to calculate var(MathAb), var(MathAtt) and cov(MathAb,MathAtt) so that I could reproduce these figures outside the estimation process?

I was thinking about something similar to getting the xb/residuals after a regression, but I couldn't find a postestimation option to do such operation after gsem.

Thanks in advance.

[Weiwen Ng]

Vinicius, I’m not clear what you mean. The data used to calculate the variances and covariance of the two latent variables are the dataset provided. I’m not clear how you can hand-calculate the estimates yourself, unless you mean to do MLE by hand?

Im also not clear how the above would relate to residuals. I can say this: latent variables are, by definition, unobserved. Their level is inferred from the other variables. A residual is observed - expected. So, there’s no residual for the latent variable. There are residuals as to the indicator variables, e.g. q1-q8, which are the ones you used to infer the latent variables. For ordered categorical questions, I’m not sure exactly how a residual is calculated, because you have the probability of responding in each response category (e.g. P(q1=0), P(q1=1), etc)