Fitting lme I need some clarifications about the use of pdMatrix.
My model is simple: yi=b0+bi+ei where bi are the random effects (and ei
residuals). I have as many random effects as observations identified by
the parameter 'idx' which simply enumerates my observations.
When fitting lme(fixed=y~1, random=list(idx=pdDiag(~1)) I think the fitted
model assumes that bi ~ N(0,PSI) where PSI is proportional to an identity
NxN matrix, multiplied by the 'random effects
variance' sigma^2.
My goal is the following: I want to impose a more complicated structure to
this PSI matrix, for example I want to put on PSI[1,2] and PSI[2,1]
covariance equal to 0.3*sigma^2 ( I always want to put covariance equal to a fraction of the overall
sigma^2).
I know there are lots of pdMatrix classes, but not sure which one to use
and how.
The problem with pdMatrix is that there is not (or better I could not
find) a way to display somehow the selected structure
(the 'initialize' function that works so nicely for
corMatrix is not very helpful for pdMatrix) , so I'm
not always sure about how a selected pdMatrix looks like.
Many thanks for any help
Georgia
______________________________________
Dr. Georgia Salanti
Medical Statistician
MRC - Biostatistics Unit
Institute of Public Health
University Forvie Site
Robinson Way
Cambridge CB2 2SR
Telephone number: ++44 (0)1223 763853
Fax number: ++44 (0)1223 330388
---------------------------------------
|