Skip to content
Prev 2390 / 20628 Next

zero variance query

Le mardi 02 juin 2009 ? 10:45 +0100, Christine Griffiths a ?crit :
Aaargh ! I thought that "lmer(lncount~Treatment+(1|Month)+(1|
Block),family=gaussian)" (+/- log(area somewhere in the model or the
fit) might give good hints as a first (known bad) approximation, and
didn't made myself clear. That's what happens when you try to be
sarcastic while dog-tired...

And maybe your random effects *are* quite low. What happens with :
summary(glm(Count~Treatment+Month+Block, family=quasipoisson) (or
something to that effect) ? What says a simple crosstabulation ? or a
graph ("caterpillar", for example) ?

HTH,

					Emmanuel Charpentier