[R] logLIk(lme(...))?

[Spencer Graves]

Hello, all:


	  I have a dataset with 2 groups.  I want to estimate 2 means and 2 
standard deviations.  I naively think I should be able to use lme to do 
that, e.g., lme(y~gp, random=y~1|gp, method='ML').  I think I should get 
the same answer as from lm(y~1, ...) within each level of group.  I can 
get the same means, but I don't know how to extract the within-gp 
standard deviations, and the sum of logLik for the latter two does not 
equal the former.


TOY EXAMPLE:


library(nlme)


set.seed(1)


lmePblm <- data.frame(y=c(rnorm(5, 1, 2), rnorm(5,3,5)),
                       gp=factor(rep(1:2, each=5)))


fit22 <- lme(y~gp, lmePblm, random=~1|gp, method='ML')


fitGp1 <- lm(y~1, lmePblm[lmePblm$gp==1, ])


fitGp2 <- lm(y~1, lmePblm[lmePblm$gp==2, ])


(ll22 <- logLik(fit22))


(llGp1 <- logLik(fitGp1))


(llGp2 <- logLik(fitGp2))


# Why isn't (ll22 = llGp1+llGp2)?


(ll22 - llGp1-llGp2)


# And secondarily, how can I get the residual standard deviations
# within each gp from fit22?


	  Thanks,
	  Spencer Graves