[R] maintaining specified factor contrasts when subsetting in lmer

[Afshartous, David]

I'm familiar w/ relevel but that doesn't seem to help here 
(e.g., tried setting the reference level to 3 but the second
model w/ the subset argument doesn't seem to estimate the 
desired model).

Possibly another contrasts statement needs to be provided prior 
to the second model, this time w/ a reduced number of levels, but
that seems to introduce problems w.r.t. the total number of levels
prior to subsetting.   

Is the only way around this to define a new dataframe for the subset
and define the contrasts for that dataframe accordingly?

 

-----Original Message-----
From: Francisco J. Zagmutt [mailto:gerifalte28 at hotmail.com] 
Sent: Friday, July 06, 2007 6:40 PM
To: Afshartous, David
Cc: r-help at stat.math.ethz.ch
Subject: Re: maintaining specified factor contrasts when subsetting in
lmer

See ?relevel

Regards,

Francisco

Afshartous, David wrote:
> All,
>  
> I'm using lmer for some repeated measures data and have specified 
> the contrasts for a time factor such that say time 3 is the base.
This
> works fine.  However, when
> I next use the subset argument to remove the last two time values, the

> output indicates that
> the specified contrast is not maintained (see below).   I can solve
this
> by creating a new dataframe
> for the subset of interest and redefining the constrasts, but I was 
> wondering if there is a direct method that allows me to continue w/ 
> the subset argument?  (perhaps via supplying a contrast argument to 
> lmer directly, but this doesn't seem possible based on the defintion 
> of this argument in model.matrix.default).
>  
> Thanks,
> Dave
>  
>  
> z <- rnorm(24, mean=0, sd=1)
> time <- factor(paste("Time-", rep(1:6, 4), sep="")) Patient <- 
> rep(1:4, each = 6) drug <- factor(rep(c("D", "P"), each = 6, times = 
> 2)) ## P = placebo, D = Drug dat.new <- data.frame(time, drug, z, 
> Patient)
>  
> ### specify the contrast as time 3:
> contrasts(dat.new$time) <- contr.treatment(6, base=3)     
> dimnames(contrasts(dat.new$time))[[1]] <-
> as.character(levels(dat.new$time))    
> dimnames(contrasts(dat.new$time))[[2]] <-
> as.character(levels(dat.new$time)[-3])
>  
> fm1 <- lmer(z ~ drug + time + (1 | Patient), data = dat.new ) Fixed 
> effects:
>              Estimate Std. Error  t value
> (Intercept) -0.182774   0.464014 -0.39390
> drugP       -0.281103   0.352309 -0.79789
> timeTime-1   0.150505   0.606462  0.24817
> timeTime-2   0.612016   0.606462  1.00916
> timeTime-4   0.775342   0.606462  1.27847
> timeTime-5   0.093741   0.606462  0.15457
> timeTime-6   0.452442   0.606462  0.74604
>  
> ## time 3 is the base as specified
> 
> fm2 <-  lmer(z ~ drug + time + (1 | Patient), data = dat.new, subset =

> dat.new$time !="Time-6" & dat.new$time != "Time-5")
> 
> Fixed effects:
>              Estimate Std. Error  t value
> (Intercept)  0.052975   0.500675  0.10581
> drugP       -0.451593   0.447818 -1.00843
> timeTime-2   0.461511   0.633310  0.72873
> timeTime-3  -0.150505   0.633310 -0.23765
> timeTime-4   0.624837   0.633310  0.98662
>  
> ### time 3 no longer the base; was expecting to see the fixed effects 
> for time-1, time-2, and time-4, w/ Intercept ### representing time-3
>  
>  
> 
> 	[[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
> and provide commented, minimal, self-contained, reproducible code.
>