Obtaining the predicted residuals from a regression after SEM.

[Vinbamba Boris Laurence]

"Hello everyone, can someone help me obtain the predicted residuals of each individual after my SEM regression? I need to use these residuals to construct internal instruments following Lewbel's approach (2012).
The command "predict resid_MI, residuals" is not working.
Here is my syntax. sem (MI <- $covlist) (MI -> $indicators), nocapslatent latent(MI) method(ml) log
estat residuals, normalized (displays the residuals matrix)

Thanks

[Rich Goldstein]

I'm a little confused; can't you get the predicted values and then just subtract them from the observed? see

help sem_predict

[Vinbamba Boris Laurence]

Yes, that's exactly it. The predict command doesn't work after sem. I'm using another command (estat residuals, normalized ) to obtain the normalized error matrix, but I'm having trouble creating a variable to store the errors for each individual.

[Andrew Musau]

The predict command doesn't work after sem

This is not accurate. Look at

help sem_postestimation

Per Rich's advice in #2, you need something like

sysuse auto, clear
sem (mpg <- weight turn foreign)
predict mpghat, xb
gen res= mpg-mpghat

[Joseph Coveney]

I'm using another command (estat residuals, normalized ) to obtain the normalized error matrix, but I'm having trouble creating a variable to store the errors for each individual.

Aren't those residuals of the fitted (implied) covariance matrix?.It doesn't seem that "errors for each individual" would be meaningful in that context.

And I'm not sure what kind of residuals "Lewbel's approach (2012)" requires, but if you want to store the matrix returned by the postestimation command in your dataset, then you can

matrix define Residuals = vech(r(nres_cov))'
svmat double Residuals

[Vinbamba Boris Laurence]

Thank you Joseph . I have find solution . Thanks

[Vinbamba Boris Laurence]

Hello, I have over 20 covariates in a regression with the continuous variable Y representing mental health. I want to use the Lasso linear regression to select the most relevant covariates. Which is the best model to use (CV, adaptive, plugin, BIC)?