glmmTMB errors for anovaPB

[mmcgillycuddy]

Hi David, I am getting some errors when using glmmTMB objects for anovaPB.

library(ecostats)
library(glmmTMB)
fm1 <- glmmTMB(count~mined+(1|spp),
               data=Salamanders,
               family=nbinom2)
fm0 <- update(fm1, ~ . - mined)
anovaPB(fm0, fm1, n.sim = 100)

Error in checkForRemoteErrors(val) : 
  100 nodes produced errors; first error: object 'nbinom2' not found
In addition: Warning message:
In Ops.factor(1, spp) : ‘|’ not meaningful for factors

[dwarton]

drats... it seems that the nbinom2 family is not getting passed to the nodes when using multiple processors. You can beat this issue for the moment by using the argument ncpus=1 in your anovaPB call.

[mmcgillycuddy]

Ok, thanks.

Meanwhile I came across another problem where I seem to get very different results if I run a bootstrap myself. For a LR = 545 and df =1 I would expect a small p-value but I get 0.941.

library(ecostats)
library(glmmTMB)
fm1 <- glmmTMB(count ~ mined + (1|spp), data = Salamanders, family=poisson())
fm0 <- update(fm1, ~ . - mined)
LRobs <- anova(fm1, fm0)[2, 6]
LRobs
LRstat <- rep(NA, 100)
for(i in 1:100){
  simy <- simulate(fm0)[[1]]
  ft_sim1 <- glmmTMB(simy~mined+(1|spp),
                     data=Salamanders,
                     family=poisson() )
  ft_sim0 <-  update(ft_sim1, ~ . - mined)
  LRstat[i] <- anova(ft_sim1, ft_sim0)[2, 6]
}
(sum(LRstat >= LRobs, na.rm = T) + 1)/(100+1)
anovaPB(fm0, fm1, n.sim = 1000)

[dwarton]

OK, the issue there is that you need to set colRef=6, I'll change the default next update so that it does that

[mmcgillycuddy]

Oh nice! That's easy. Thanks David!