Skip to content
Prev 168583 / 398502 Next

convergence problem gamm / lme

Geert,

Sorry for slow reply... I don't see any obvious problems with what you've 
done, so I guess it's the usual problem that PQL just doesn't *have* to 
converge, and the bit of extra flexibility of using a smooth is too much for 
it in this case. If you send me the data offline I can dig a little bit more 
if you like (I'll only use the data for this purpose etc. etc.) 

You are right that PQL does the same thing for Poisson and quasi-poisson. I 
don't think there is an easy way to use the values for the reduced dataset 
fit in the full dataset fitting, unfortunately. 

Another option is to use `gam' to fit the random effects. It'll be a bit slow 
with 70+ random effects, as you have, and it's a bit more work to set up, but 
it should converge. See ?gam.models which has some examples showing how to do 
this.

best,
Simon
On Thursday 29 January 2009 08:20, geert aarts wrote: