I have been playing around with the generalized mixed models and I fit two models to the salamander data using glmer.nb and glmmTMB as shown below.
library(glmmTMB)
library(merTools)
## remove the outlier count
dat <- Salamanders[which(Salamanders$count<20), ]
fm1 <- glmmTMB(count ~ cover + (1|spp),
data=dat, family=nbinom2(link="log"))
fm2 <- glmer.nb(count ~ cover + (1|spp), data=dat)
summary(fm1); summary(fm2)
Summary output for glmmTMB:
Conditional model:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 0.01037 0.24547 0.042 0.966
cover 0.41932 0.08350 5.022 5.12e-07 ***
Summary output for glmer.nb:
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -0.0007037 0.2453352 -0.003 0.998
cover 0.4180324 0.0831942 5.025 5.04e-07 ***
The model output for the intercepts are different, why is this? Is this a case like here where the two models are finding different local optima? If so how can I graphically represent this like in Ben Bolker's answer in the linked question?
