[R-sig-ME] autocorrelation by group in mixed model

[Mike Dunbar]

Dear Irene

How about making a new column in your dataset which is the response variable lagged by 1 time step. Then use this term as a predictor in your model and look at the BLUPs. This would assume that the group by group autocorrelation would follow a normal distribution, I'm not sure if this is valid from theory or not. Is this what you need?

Further suggestions. 
Emailing both R-help and R-sig-ME with the same question does not encourage replies.
Read the posting guide and post a reproducible example, or at least give some more information on what you are trying to do. We know nothing about your data so it is difficult to give meaningful help. For example you probably need quite alot of data within each group in order to estimate AR terms well.
When you say some groups don't display autocorrelation how did you determine this? If there are sufficient data to determine this precisely then you probably don't need a mixed model, just model each group separately. Alternatively the differences in group-specific AR could be sampling variability, in which case a mixed model would be appropriate, but you can't say "not all my groups display auto-correlation"

regards

Mike


>>> "Irene Mantzouni" <ima at difres.dk> 11/01/2008 16:41 >>>
Hi all!
 
(How) is it possible to fit a mixed model with group specific auto-correlation structure ? For instance, not all my groups display auto-correlation so I would like to use a corMatrix (corAR1) only for the auto-correlated ones. If I construct manually a
the corMatrix, is it possible to use it  as input somehow?
 
thank you!
 
Irene

_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models


-- 
This message (and any attachments) is for the recipient ...{{dropped:6}}