Skip to content
Prev 13691 / 20628 Next

glmm question

Dear Joaquin,

I did not look in your script but here are my thoughts on your question:
- concerning residuals, the choice of accepting a model as fitting to 
the data or not based on residuals plot (ie residuals vs fitted value) 
is at the discretion of the analyst/scientist. Sometime when I am unsure 
I simulate data from the model (ie simulate for merMod object), fit 
these simulated data to the model (ie function refit for merMod) and see 
if the resulting residual plot differ from the one I got from my 
empirical data. If yes my model is wrong, otherwise the model is not so bad.

- Two-way interaction between categorical variables or between a 
categorical and continuous variable are rather easy to understand, it 
gets trickier for interaction between two continuous variables, I wrote 
an article on this a while back: http://rpubs.com/hughes/15353, this was 
for lm but for glmm the interpretation would be the same (except that we 
have a link function, log in the case of poisson, between the observed 
data and the modeled response). There are certainly many more resource 
out there, I found the UCLA pages nicely written: 
http://www.ats.ucla.edu/stat/r/

Yours,
Lionel
On 02/09/2015 20:45, Joaqu?n Aldabe wrote: