Skip to content

bivariate-response mixed model (MCMCglmm) versus mixed-RMA regression options

1 message · Ned Dochtermann

#
Hello all,
I'm currently working on a project where I'm interested in the 
relationship between two variables that are measured with error, 
suggesting the need for reduced major axis regression. However, the data 
structure also necessitates the inclusion of random effects for both 
variables so I initially thought to use a bivariate-response mixed 
model. Unfortunately the relevant covariance/correlation isn't quite 
what I'm interested in for the biological question of interest.

The tentative solution I've come up with is to use the variances and 
covariances to estimate a slope (COVx,y/VARx) and the slope and variable 
means to calculate the intercept. Since I'm doing this on the posteriors 
I'm able to get credibility intervals and mode estimates and not have to 
run a regression on the BLUPs. This gets directly at the question in 
which I'm interested and does so at the level that is relevant.

Does this seem like an appropriate approach? Are there mixed versions of 
RMA (google didn't reveal anything to me) or other alternatives that 
seem preferable?

Thanks for any feedback and sorry for a bit of rambling and the open 
ended nature of the query,
Ned


(slope.me<-posterior.mode(ests.trunc$VCV[,2]/ests.trunc$VCV[,4]))
HPDinterval(ests.trunc$VCV[,2]/ests.trunc$VCV[,4])
intercept.me<-ests.trunc$Sol[,1]-slope.me*ests.trunc$Sol[,2]
posterior.mode(intercept.me)
HPDinterval(intercept.me)