Mike Zyphur posted on Saturday, July 02, 2011 - 7:23 am
Hi Linda and Bengt, I am trying to translate the Stealth and Cascade extended twin family models into Mplus. In Mx they can work with the covariance of parents directly to define the "copath" mu (in Keller et al., 2009). Under an assumption of primary phenotypic assortative mating this is:
CV(spouses) = Signma^2*mu*Sigma^2
Under different assumptions about mating this changes to be dependent upon genetic and environment factors (pp. 14-15).
However, when working with raw data in Mplus, I can't figure a quick way to estimate covariance among parents for use in Model Constraint to define mu. It can be done "manually" by estimating it in a separate model and then entered in Model Constraint. What do you recommend?
Keller et al (2009). Modeling extended twin family data I: Description of the cascade model. Twin Research and Human Genetics, 12, 8-18.
Drs. Muthen, Muthen and Asparouhov, I am trying to use a Bayesian approach to estimate A,C,E components of a continuous phenotype in a sample of MZ and DZ twins. I understand that Bayesian estimation currently is not available for multiple group analysis. Moving into mixture framework (using v6.11 and data from example 5.18 of the manual), I used the KNOWNCLASS option with TYPE=MIXTURE and MLR estimation to replicate the results of the multiple group analysis as is shown in example 5.18. When I tried Bayesian estimation in the mixture framework, the model would not run because the PSI matrix is not positive definite (due to A1 and A2 being correlated at 1.0 in the MZ twin class, and C1 and C2 being correlated at 1.0 in the MZ and DZ twin classes). Is there a solution or work-around for this?