I'm running a glmmTMB model with various truncated count distributions (truncated_poisson
, truncated_compois
, truncated_nbinom1
, truncated_nbinom2
). When I predict from the model, the values seem to be lower than expected, as if the prediction is not accounting for the truncation. Where am I going wrong? A toy example is provided, showing that predicted values are lower than observed means.
Any advice would be appreciated. Extra points if the advice can extend to the other truncated count distributions (see above) and if it shows how to correctly get the 95% confidence band around the estimated values in these cases.
library(dplyr)
library( extraDistr)
library(glmmTMB)
set.seed(1)
df <- data.frame(Group = rep(c("a", "b"), each = 20), N = rtpois(40, 1, a = 0), ran = "a") %>%
mutate(N = ifelse(N == 0, 1, N))
m <- glmmTMB(N ~ Group + (1|ran), data = df, family = "truncated_poisson")
df %>% group_by(Group) %>% summarize(mean(N))
predict(m, newdata = data.frame(Group = c("a", "b"), ran = NA), type = "response")