[R] lmer multilevel

[Spencer Graves]

Hi, Doug:

	  I just confirmed a discrepancy (reported by Paul Cossens) between the 
output of lme and lmer using the Oxide example in Pinheiro and Bates, 
pp. 167-170.  It appears that coef(lmer(...)) adds only the ranef for 
Wafer and omits the one for Lot.

	  Consider the following:

library(nlme)
Oxide <- Oxide
#data(Oxide,package="nlme")
fit.lme <- lme(Thickness~1,Oxide)
coef.lme <- coef(fit.lme)

<exit R, restart>
library(lme4)
fm3Oxide<-lmer(Thickness~ (1|Lot)+(1|Lot:Wafer),data=Oxide)
coef.lme4 <- coef(fm3Oxide)

The following matches the first 3 numbers in coef.lme but not in 
coef.lme4:

fixef(fm3Oxide)+ranef(fm3Oxide)[[2]][1]+ranef(fm3Oxide)[[1]][1:3]

	  Best Wishes,
	  spencer graves

Paul Cossens wrote:

> My question relates to problems that I'm having matching lme and lmer
> examples in P&B. 
> using Matix 0.995
>  
> In the Oxide example in p167-170 I can't get the level 2 coefficient
> estimates to match 
> the fm1Oxide model in lme is 
>  
> data(Oxide,package="nlme")
> lme(Thickness~1,Oxide)
> 
> which I translate in Lmer syntax to 
>  
> fm3Oxide<-lmer(Thickness~ (1|Lot)+(1|Lot:Wafer),data=Oxide)
> #or alternatively which gives the same result
> Oxide$LW<-with(Oxide,Lot:Wafer)[drop=TRUE]
> fm4Oxide<-lmer(Thickness~ (1|Lot)+(1|LW),data=Oxide)
>  
> however if you look at say Lot 8, lme gives
>  
> 8    1993.767
> 
> 8/1    1993.677
> 8/2    1995.170
> 8/3    1990.693
> 
>  
> while  lmer gives 
> coef(fm3Oxide)
> ....
> 8    1993.767
> 8:1 2000.062
> 
> 8:2 2001.555
> 
> 8:3 1997.078
> 
> To me this looks like lmer in not including the lot random effect (8=
> -6.385129, Intercept 2000.153 ). Is this because I'm not specifying the
> model correctly?
> 
> Thanks Paul
> 
>  
> 
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html