Skip to content
Prev 13600 / 20628 Next

Interpretation of GLMM output in R

Dear Yvonne,

To answer Question 1, I would scale both predictors to unit standard 
deviation (ie fad.data$Volume_std<-scale(fad.data$Volume,scale=TRUE)) 
and then compare the back-transformed coefficients, as you have a 
Poisson model this would be then exp(fixef(fad)). Say that you get a 
value of 1.5 for the variable Volume, you would then interpret this as: 
an increase in one standard deviation of Volume leads to an increase of 
1.5*Pollinator, put differently every increase of one standard deviation 
in Volume increase Pollinator by 50%. Having scaled the predictor would 
also allow you to compare the strength of their effects. Also note that 
effect sizes do not give informations about how much variance in the 
response variable is explained by individuals terms.

Concerning the LRT to test the effect of predictors, it is not the best 
available option (see http://glmm.wikidot.com/faq#summary) the function 
"mixed" in the "afex" package allow you to do parametric bootstrap to 
test fixed effects (using type 2 or 3 tests). For effects with very 
small p-values (like volume) using LRT or parametric bootstrap will most 
certainly lead to the same interpretation (ie volume has a significant 
effect), for p-values close to 0.05 (like parasitoids) you might get 
different interpretations.

Sincerely yours,
Lionel
On 02/08/2015 11:31, Yvonne Hiller wrote: