
Message/Author 


When dealing with latent variables that are also dependent variables in MPlus, is there any way to set their variances equal, as opposed to their RESIDUAL variances? I have an autoregressive model where a series of latent variables predict each other. It seems that I am only able to set their unexplained variances equal, whereas what I would like to do is set their overall (explained plus unexplained) variances equal. I can do this before I add the autoregressive pathways (since all of their variance is unexplained), but unable to figure out how to do it afterward. Thanks for the help! 


You can accomplish this by using the Model Constraint command. In your Model command you label the parameters that determine the variances and then in your Model Constraint command you express the variances in terms of these labels and hold the variances equal. For an example of how to use Model Constraint, see the V5 UG ex 6.17. 


Thank you for the reply. I know how to use the constraint command, but am unable to figure out how to apply it here without being able to access a nonresidual variance. I have higherorder latent variables partitioning the variance of lowerorder ones. In order to do this, I fix the lower order residual variances to zero, therefore all of this variance is partitioned into the higherorder latent variables. Although I can set the pathways from higher to lower equal, since the variance of the lowerorder latent variables is set by the variance (or stddev) of the manifest variable with a loading of one, and these variances are different, the std loadings from lower to higherorder are not equal. I would like to set the std pathways equal (or the higherorder latent variables variances), but feel I can not do this without accessing the actual variance (of either the manifest variable or the higherorder latent variable). Is this possible in Mplus? Thank you! 


Maybe UG ex 5.20 can give you ideas. Here the "rel2" line of Model Constraint expresses the variance of the dependent variable y2 in terms of model parameters. Note that "vf1" is what you call a "nonresidual variance" and so is V(y2). This example also shows standardized expressions and how these can be held equal. 

Back to top 

