R help - Jan 2008 - How to run interaction between to categoric variables in lme()?

En innebygd og tegnsett-uspesifisert tekst ble skilt ut...
Navn: ikke tilgjengelig
Nettadresse:
https://stat.ethz.ch/pipermail/r-help/attachments/20080130/78d2a14c/attachment.pl

Falco tinnunculus <kestrel78 <at> gmail.com> writes:
> How do I run interaction between to categoric variables in lme()?
> 
> I tried this:
> 
> > lmefit1<- lme(Handling ~Mass+factor(Prey)+factor (Decap)+
> factor(Prey)*factor(Decap), random = ~ 1 |Place, data=nestling1)
> 
> Error in MEEM(object, conLin, control$niterEM) :
> Singularity in backsolve at level 0, block 1
First, I would suggest that you make Prey, Decap factors in your data frame.
Makes for a much easier reading, and fewer errors if you test

nestling1$Prey = as.factor(nestling1$Prey)
nestling1$Decap = as.factor(nestling1$Decap)
lmefit1<- lme(Handling ~Mass+Decap*Prey, random = ~ 1 |Place, data=nestling1)

You should not write 

Prey+Decap+Prey*Decap. 

I you want to do it explicitly, used the colon instead

Prey+Decap+Prey:Decap

or, let lme do the dirty (i.e. in most cases correct) work:

Prey*Decap

which will expand to what you need. If you still get the error message, it's
your data: you want to much from them. Check if you have enough degrees of
freedom left, or if the data set is degenerate by not including a sufficient
number of pairing.

Dieter