| Detailed description |
Good morning every one,
I am working in behavioural ecology for my PhD, and I am currently analysing behavioural data (number of data on a determinated period, Poisson distribution family).
I first use a GLM, which provides the predicted values and their standard error for each level of our factors (2 crossed factors, each one having 2 levels).
> model.glm<-glm(Behaviour~Habitat+Parr+Habitat:Parr,family=poisson)
> predict(model.glm,type="response",se.fit=TRUE)
But, due to the few females we can provide in this study, we used several behavioural sequences (spawnings) from each female. So we implemented a random effect (1|Female) to resolve our pseudoreplicates problem, using the lmer function in lme4 package.
> model.lmer<-lmer(Behaviour~Habitat+Parr+Habitat:Parr+(1|Female),family=poisson)
My questions are:
1) Is it correct to reduce this model until reaching the following one (as for GLM models):
> null.model.glm<-glm(Behaviour~1,family=poisson)
and then selecting the most adjusted model with AIC?
2) After selecting the "best model", is it possible (due to random effect) to have predicted values and their associated standard error for each group? We now can only get predicted values for each "individual" (each line of our dataset,not each female). I now we can estimate those predicted values from estimates in summary(model.lmer), but I can't obtain the standard error by this way.
3) Is it possible to have an estimation of percentage of the variability which is explained by our model?
As we want to publish this analysis, I would like to know if it is possible (in the last lme4 release) to get those (necessary methinks in a strong analysis).
PS: I found on nabble.com an answer from Mr Bates to a similar question:
http://www.nabble.com/-R--Predicted-values-in-lmer-modeling-td7577645.html#a7577645
But as it was in late 2006, I would like to know if this problem has been resolved.
Thank you very much for any help,
David Grimardias
PhD in Behavioural Ecology
INRA hydrobiological station
UMR INRA-UPPA ECOBIOP
Saint Pee sur Nivelle, FRANCE |
|
