Residuals as predictors
Message/Author
 Amanda Hugan-Kiss posted on Tuesday, April 14, 2009 - 5:51 am
Good morning -- I am attempting to estimate a non-standard SEM in which I use the residuals of a manifest variable as a predictor instead of the variable itself (e.g., like an AR1 residual structure, but predicting another variable, not another residual). I have found how to do this in several other programs, but would very much like to stay within Mplus. Is there a way I might attempt this here?

Thank you -- Amanda
 Linda K. Muthen posted on Tuesday, April 14, 2009 - 9:00 am
Try the following to obtain the residual for y4:

MODEL:
f BY y1-y4;
y4res BY;
y4 ON y4res@1;
y4@0;
z ON y4res;
 Emanuel Jauk posted on Thursday, February 20, 2014 - 7:29 am
I have a related question:
I would like to use the residual of a latent endogenous variable as a predictor of a different latent variable.

Is this the default in using an ON statement with endogenous variables (like it would be the case for error correlations specified using WITH) or do I have to specify it similar to the method desribed above?

[The goal of the analysis is to disentangle the effects of shared and unique portions of variance in a higher-order CFA model. I would like to use the higher order factor as well as the residuals of the lower order factors as simultaneous predictors of a criterion variable.]
 Emanuel Jauk posted on Thursday, February 20, 2014 - 7:41 am
I reconsidered my problem and it is now apparent to me that this cannot be the default function of ON with endogenous variables, but I'm still not sure how to specify the residual.

Thank you!
 Linda K. Muthen posted on Thursday, February 20, 2014 - 10:20 am
See the following FAQ on the website:

Regressing on a residual
 Emanuel Jauk posted on Tuesday, February 25, 2014 - 5:06 am
Dear Linda,

Thank you very much for your quick response!
I was actually wondering how to get the residual of a latent variable - does the same procedure apply to latent variables, too?
Wouldn't it be a problem to fix the variance to zero in this case?

Thank you!
 Bengt O. Muthen posted on Tuesday, February 25, 2014 - 12:19 pm
Same procedure. See if it works. You may get a non-pos def matrix warning, but that's ok.