sjstats- Model is overfitting zero-counts

Question

I have a glmmTMB model and I used the overdisp and the zero_count functions with these results:

overdisp(glmmtmb_poisson)

# Overdispersion test
# 
# dispersion ratio = 0.2953
# Pearson's Chi-Squared = 3104.5094
# p-value = 1.0000
# 
# No overdispersion detected.


zero_count(glmmtmb_poisson)
# Observed zero-counts: 9946
# Predicted zero-counts: 9953
# Ratio: 1.00
# 
# Model is overfitting zero-counts

I am not sure what this means. I understand that the model is not overdispersed nor zero-inflated, but I am not sure what this overfitting means. Also, there was a notes saying that in the case of glmm a p value larger than 0.05 indicates overdispersion, yet the p value does not seem to indicate overdispersion in my case.

Should I just go ahead and run a simple glmmTMB model as my final model without any further worry about zero inflation and overdispersion?

Thank you very much!

---

Thank you very much for your answer. This makes sense. I was wondering if you could provide some references so I can learn more about overfitting as I am not clear what that means. Thanks again!

Answer 1

The output is correct: Since the p-value is > 0.05, there is no over-dispersion. That the p-value > 0.05 indicates no overdispersion, has changed recently due to a change in the computation, however, I did not update the docs (and will do this right now).

Regarding the zero-counts: This is due to no tolerance and the exact split at the value 1. The ratio is almost 1, so there's is actually no overfitting. I will increase the tolerance for this function, to not confuse users, and to accept a certain range around the ratio of 1 (e.g. +/- .05) as OK.