Dear All,
I have a series of Poisson counts taken over six timepoints from ten
subjects, with a single covariate that is 0,1,0,1,0,1 over the six
timepoints. As I want to condition out the effects between individuals,
I'm doing a multinomial logistic rather than a Poisson regression,
conditioning on the total counts per individual. It's like the Alligators
example, but with I=1,J=10 and K=6, with an additional factor.
Ignoring the role of the factor at the moment (so I just have alpha[2] to
alpha[6], and gamma[2,2]-gamma[10,6]), when I plot out the gamma
parameters, transforming them into deviations from the mean, it appears
that I have correlated residuals - for my system, this makes biological
sense. I'd like to model the host residuals using a multivariate normal
distribution ~ dmvnorm(mu.g[],Omega[,]). Does anyone know how to
implement this kind of model (or whether it's a statistically reasonable
one?).
Best wishes
Simon
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|