singularities in fitting lme

["Steffen Barembruch" <steffen@barembruch.de>]

Hello!

Sorry to bother you again!

 

When fitting an LME model to my data, everything works fine. here is the code:

lme(y ~ weeks + weeks:conditions + weeks:package, data="" random= ~weeks | batch)

where y and weeks are numeric, and package, conditions and batch are factors.

 

But now I played around a little for implementing an algorithm which will use the lme function.

mydata$weeks1 <- mydata$weeks

if i now replace the effect "weeks" by "weeks1" as following

lme(y ~ weeks1 + weeks:conditions + weeks:package, data="" random= ~weeks | batch)

i get an error: singularity in backsolve

 

furthermore

if i replace the constant term in the model by a vector of one's i will get the same error.

The code would now look

mydata$intercept <- rep(1, dim(mydata)[1])

lme(y ~  intercept - 1 + weeks + weeks:conditions + weeks:package, data="" random= ~weeks | batch)

 

I don't understand that, because the design matrices look the same for all models, don't they. And as far as I understood the maximum likelihood estimation of these parameters, the design matrices are the only thing that matters.

I'd appreciate any hints on what S-plus, Version 7.06 on Windows XP,  is doing.

 

Thanks a lot

Steffen Barembruch