[R] lme in R and Splus

[Peter Dalgaard BSA]

Michael Fugate <fugate at lanl.gov> writes:

> ############## BEGINNING OF CODE ###########################
> # a fake dataset to make the bumps with
> nn <- 30  # of data points
> mm <- 7   # number of support sites for x(s)
> # create sites s
> ss <- seq(1,10,length=nn)
> # create the data y
> e1 <- rnorm(nn,sd=0.1)
> e2 <- cos(ss/10*2*pi*4)*.2
> yy <- sin(ss/10*2*pi)+e2+e1
> plot(ss,yy)
> 
> # locations of support points
> ww <- seq(1-2,10+2,length=mm)
> # width of kernel
> sdkern <- 2
> 
> # create the matrix KK
> KK <- matrix(NA,ncol=mm,nrow=nn)
> for(ii in 1:mm){
> KK[,ii] <- dnorm(ss,mean=ww[ii],sd=sdkern)
> }
> 
> # create a dataframe to hold the data
> df1 <- data.frame(y=yy,K=KK,sub=1)
> df1$sub <- as.factor(df1$sub)
> 
> # now fit a mixed model using lme
> a1 <- lme(fixed= y ~ 1,
>           random= pdIdent(~KK-1),
>           data=df1,na.action=na.omit)
> 
> # obtain and plot the fitted values
> a1p <- as.vector(predict(a1,df1))
> lines(ss,a1p,lty=1)

lme in S-PLUS is older than the one in R, and some things changed. I
think you want

df1 <- data.frame(y=yy,K=I(KK),sub=1)
a1 <- lme(fixed= y ~ 1,
          random= list(sub=pdIdent(~K-1)),
          data=df1,na.action=na.omit)
lines(ss,predict(a1,df1,1))

(Apparently you can't do a level-0 prediction in a model with only an
intercept, which looks like a bit of a bug. Of course, that is just
the intercept for all observations, but...)

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907