Skip to content

Error in nonlinear mixed-effects model

2 messages · Revilla,AJ (pgt), Douglas Bates

Revilla,AJ (pgt)
# 
Dear all,

I am trying to fit a mixed-effects non linear regression, but I have some trouble with it. My data are a balanced panel of 904 subjects with 8 observations (at regular periods) per subject.
The functional form of my model is Y=Aexp(-BX1)X2 +e. I want to allow parameters A and B to vary among subjects and also include an autocorrelation term. I have already fitted a standard nonlinear regression to the data, but I keep having problems with NLME.

I have defined my data as a groupedData object, and when I try to fit the model I get this error message:

dat<-groupedData(V~VAC|ID,data=dat)
attach(dat)
mod<-nlme(V~A*exp(B*YEAR)*VAC, fixed=A+B~1, random=A+B~1, correlation=corCAR1(), start=c(A=1.2,B=0.2))
Error in getGroups.data.frame(dataMix, eval(parse(text = paste("~1", deparse(groups[[2]]),  : 
        Invalid formula for groups

Do you have any clue of what?s happening? It?s the first time I fit a model like this in R, so the problem is probably pretty obvious, but I cannot see it.

Thank you very much,

Antonio
Douglas Bates
#
Revilla,AJ (pgt) wrote:
> attach(dat)

I don't think that attach'ing a groupedData object will propagate the 
grouping factors.  I suggest that you avoid the construction of the 
groupedData object and also that you avoid attaching and detaching data 
frames.  Just use

  mod <- nlme(V~A*exp(B*YEAR)*VAC, dat, fixed=A+B~1, random=A+B~1|ID, 
correlation=corCAR1(), start=c(A=1.2,B=0.2))

You will also need to specify a formula for the corAR1 constructor.  My 
guess is form = ~ YEAR | ID but you will need to decide if that is what 
you intend.  Check the help page for corAR1 to see the list of possible 
arguments.


Error in