random effect variance per treatment group in lmer

Fri, Jul 13, 2007 7:44 AM

Alan,

Thanks for the suggestion.  I noticed the following error msg
for that lmer call:

+ as.numeric(drug=="D") | Patient ) + ( 0 + as.numeric(drug=="P") |
Patient ), data=dat.new ) )
Error: length(term) == 3 is not TRUE

I tried a few changes to the model but the error still exists; I'll 
keep checking.  I assume the rationale for the structure of your lmer
call, where you use as.numeric as opposed to just drug above, is to
insure that
you do not introduce a random effect for drug into the model?

Regards,
Dave

-----Original Message-----
From: Alan Cobo-Lewis [mailto:alanc at umit.maine.edu] 
Sent: Wednesday, July 11, 2007 6:40 PM
To: r-sig-mixed-models at r-project.org
Cc: " "Afshartous at basalt.its.maine.edu; Afshartous, David; 
Andrew Robinson
Subject: Re: random effect variance per treatment group in lmer


Dave,

How about using stratifying variance on level of drug using ( 
0 + as.numeric(drug=="D") | Patient ) + ( 0 + 
as.numeric(drug=="P") | Patient ) Here's some code (whose sim 
also builds in a fixed effect of time that applies only to 
the Drug condition).

set.seed(500)
n.timepoints <- 8
n.subj.per.tx <- 20
sd.d <- 5; sd.p <- 2; sd.res <- 1.3
drug <- factor(rep(c("D", "P"), each = n.timepoints, times = 
n.subj.per.tx)) drug.baseline <- rep( c(0,5), 
each=n.timepoints, times=n.subj.per.tx ) Patient <- 
rep(1:(n.subj.per.tx*2), each = n.timepoints) 
Patient.baseline <- rep( rnorm( n.subj.per.tx*2, sd=c(sd.d, 
sd.p) ), each=n.timepoints ) time <- factor(paste("Time-", 
rep(1:n.timepoints, n.subj.per.tx*2), sep="")) time.baseline 
<- rep(1:n.timepoints,n.subj.per.tx*2)*as.numeric(drug=="D")
dv <- rnorm( n.subj.per.tx*n.timepoints*2, 
mean=time.baseline+Patient.baseline+drug.baseline, sd=sd.res 
) dat.new <- data.frame(time, drug, dv, Patient) xyplot( 
dv~time|drug, group=Patient, type="l", data=dat.new ) # fit 
model treats time as a quantitative predictor ( fm.het <- 
lmer( dv ~ rep(1:n.timepoints, n.subj.per.tx*2)*drug + ( 0 + 
as.numeric(drug=="D") | Patient ) + ( 0 + 
as.numeric(drug=="P") | Patient ), data=dat.new ) )


HTH,
alan

From: "Afshartous, David" <afshart at exchange.sba.miami.edu>

 asked:

All,
I didn't receive a response to the query below sent to the general 
R-help mailing list so figured I'd try this mailing list.

Apologies in

advance if this is an overly simplistic question for this

list; I just

started w/ lmer after not using lme for awhile.
Cheers,
Dave

__________________________________________________________

All,

How does one specify a model in lmer such that say the random effect 
for

the intercept has a different variance per treatment group?  
Thus, in the model equation, we'd have say b_ij represent the random 
effect for patient j in treatment group i, with variance

random effect variance per treatment group in lmer

Thread (7 messages)