Leah posted on Tuesday, February 09, 2016 - 12:55 pm
I am fitting a two-group (MZ/DZ) independent pathway model, with 3 binary variables and one continuous variable. I am employing the WLSMV estimator. I am encountering an issue in which the residual correlation for one of the binary variables (which can be decomposed to derive specific genetic and environmental factors) is greater than 1.0 in MZs. The variable has a reasonable endorsement rate and the model has good covariance coverage. I have attempted to constrain the estimate to lower values (e.g., corr < .20), but it adopts the boundary value. I have fit alternate multivariate models with common and specific factors (e.g., common pathway model) using these variables without any problems. Alternate models also indicate that there is significant residual variance in the variable to be explained. Any thoughts on what may be causing this issue?
I don't know. It sounds like you have successfully used WLSMV with a combination of binary and continuous variables in other data, but perhaps not using the independent pathway model? Perhaps other multivariate models fit better. Maybe you want to ask on SEMNET.