Should I switch from Poisson to negative binomial for overdispersed count data?

[Justin_B]

I’m trying to decide if I should use a Poisson regression for my count data on website errors per day, but the variance is almost double the mean. I’ve read that this overdispersion means the model assumptions are violated, but I’m not sure what to use instead.

[NoraEG]

I ran into this last year with daily website errors. Poisson fed me numbers that looked fine but the residuals told a different story and the standard errors were blown up.

[James92]

My go to fix was a negative binomial model which lets the variance be bigger than the mean. It tended to fit better and the inference felt more reasonable.

[Ava.S]

Could the zeros be the hidden problem instead of dispersion?

[Anthony_H]

I also tried a couple of tweaks like adding an offset for site traffic and checking a quasi Poisson alternative, but the improvements were small and I stayed unsure about the right path.