Hello all,
I have a simple path analysis model that I planned to expand on. This
is a multivariate regression-type model, but Y1 also predicts Y2 via the
coefficient beta1[1]. All variables are mean-centered.
The problem is that the beta1[1] coefficient is highly autocorrelated
and does not converge. I tried over-relaxing but this did not
noticeably diminish the autocorrelation. (The inclusion of the inits
did not change the situation appreciably.)
Any thoughts on how to get this model to run more smoothly would be most
appreciated!
Thanks,
Gene Hahn
George Washington University-graduate student
model path;
{
for (i in 1:N) {
Y[i,1:M] ~ dmnorm(mu[i,], Omega[,]);
}
for(i in 1:N) {
mu[i,1] <- alpha1[1]*x1[i];
mu[i,2] <- alpha1[2]*x1[i] + beta1[1]*Y[i,1];
}
for (j in 1:M) {
alpha1[j] ~ dnorm(0.0, 0.0001);
}
beta1[1] ~dnorm(0.0, 0.0001);
beta1[2] <- 0.0;
Omega[1:M,1:M] ~ dwish(R[,], 2);
for (i in 1:M) {
for (j in 1:M) {
Sigma2[i,j] <- inverse(Omega[,],i,j);
}
}
}
Inits
list(alpha1 = c(.5,.5),
beta1 = c(0.5,NA),
Omega = structure(
.Data = c(.1, .1, .1, .1),
.Dim = c(2, 2)
)
)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|