Relative effects
Message/Author
 JOEL WONG posted on Tuesday, July 30, 2013 - 7:26 am
I have a simple SEM model involving two latent variables X1 and X2, each with paths leading to several outcome variables Y1, Y2, Y3, etc.

Is it possible to compare the collective effects of X1 and X2 on the various outcome variables? If so, should I be looking at the R-square in the Mplus output? For example, if the R-square for Y1 is .45 and the R-square for Y2 is .29, could I conclude that overall, X1 and X2 are more strongly related to Y1 than to Y2?

Is this comparison appropriate given that I've different outcome variables?
 Bengt O. Muthen posted on Tuesday, July 30, 2013 - 1:43 pm
This seems reasonable since R-2 is not dependent on the outcome scale.
 JOEL WONG posted on Wednesday, July 31, 2013 - 7:23 pm
Thank you very much, Bengt.

Is R-2 in Mplus/SEM similar to R-2 in multiple regression (proportion of variance in an outcome accounted for by a group of predictors)? Ideally, I would like to convert the R-2 to Cohen's f2 for the purposes of interpreting effect sizes.

Cohen's f2 = R2/(1 - R2)

Could this formula still be used to convert the R2 in Mplus/SEM to f2?
 Bengt O. Muthen posted on Thursday, August 01, 2013 - 8:25 am
Yes, R-2 in Mplus/SEM is the same as in linear regression. You can compute Cohen's f2 in MODEL CONSTRAINT.
 Jan posted on Sunday, December 07, 2014 - 3:12 pm
May I ask you how to define R2 in the MODEL CONSTRAINT (to compute Cohen's f2)?
I know how to specify the label for a beta, mean, or variance:
e.g.
model:
x on y (a) y2 y3; [x](b); x(c);

but how to label the R-squared?
 Bengt O. Muthen posted on Sunday, December 07, 2014 - 7:47 pm
Model:

y on x (b);
y(resvary);
x (vx);

Model Constraint:
New(r2);
r2 = b*b*vx/(b*b*vx+resvary);