[R-sig-ME] Set individual variance components to zero in lme

Gustaf Granath gustaf.granath at gmail.com
Mon Nov 18 20:50:25 CET 2013


Hi,
Is it possible to set individual variance components in lme() to zero?

For example, say that you have 4 different locations and and you 
estimate individual variance components at each location + separate 
residual variances.

lme( y ~ location, random = ~ 0 + location|plot/plant, data = dat, 
weights = varIdent(form = ~1|location) )

Now I want to test if the plot variance component at location x is 
greater than zero. I can do this by fitting individual models.
lme( y ~ 1, random = ~ 1|plot/plant, data = dat, subset = location=="x" 
)  versus
lme( y ~ 1, random = ~ 1|plant, data = dat, subset = location=="x" )
using a LRT test. However, it would be convenient if it is possible to 
fix individual components to zero in the first model with all locations, 
like the G.param argument can do in asreml-r. It doesnt seem like there 
is an easy way with lme() but maybe someone has tried it?

Cheers

Gustaf

-- 
Gustaf Granath (PhD)
Post doc
McMaster University
School of Geography & Earth Sciences



More information about the R-sig-mixed-models mailing list