Skip to content
Prev 15680 / 20628 Next

MCMCglmm with a AR(1)

Hi,

MCMCglmm cannot fit AR1 models but it can fit 1st-order antedependence 
models (an ar1 process sampled from time t=0 rather than from the 
stationary process). That being said, it only really works when you have 
short-time series where all observations have been sampled at the same 
time. I'm not sure if this is your case? Its also not clear how you 
would set up the AR1 for a 5-category multinomial  - something like a 
vector autoregressive might make sense but this is not implemented in 
MCMCglmm except for Gaussian traits and at the observation-level rather 
than the residual level. For non-Gaussian traits, particularly 
multinomial, you're probably going to have to code the model yourself in 
stan/JAGS if you want autocorrelation at the observation-level.

Cheers,

Jarrod



(except for Gaussian traits and at the observation-level).
On 15/08/2017 10:39, Kim, S. wrote: