Tue, 02/14/2012 - 22:10

Hello all,

I'm trying to estimate the reliability of a sub-scale using a procedure outlined in Raykov (2004), which is attached. If I understand it correctly, I need to constrain the path from each error term to its respective indicator to be equal to the sum of factor loadings. That is, the path from E1 to Y1 must equal L1+L2+...+Lk .

Further, I need to specify dummy latent variables with variances that are equal to my constrained residual variances.

I have tried to set up part of the model (syntax attached), but I can't even get the model to accept the first dummy variance (error thrown: "Error in A_free[to, from] <- nextfree : subscript out of bounds").

The last page of the Raykov (2004) article has LISREL syntax for the same project. Am I on the right track? Does anyone know how to set up a model like this in OpenMx?

Attachment | Size |
---|---|

Raykov 2004 - Reliability with Linear Constraints.pdf | 114.1 KB |

Raykov Syntax.txt | 1.02 KB |

Hi Ian,

Your model appears to be misspecified. I defined the model as you stated it in your script. The model definition runs for me, but upon mxRun the model gives an error that the expected covariance matrix is not positive definite.

Inspecting the model, you have 5 latent variables. One is the factor and the other four are supposed to be the dummy residual variance variables (Thetas). But you did not specify any paths between the residual variances and the manifest variables. You also let the manifest variables have variances.

I created a model that runs that should be in the right direction toward what you want. Without a little more information, it's hard to tell what model you want.

Hope this helps!

Mike