Power: Poisson Regression

Power depends on expected rate differences (effect size), α, covariates, dispersion, and sample size or exposure time.

Key Inputs

• Effect size (rate ratio or log coefficient)
• α (significance level)
• Sample size / exposure

Example

Detecting a rate ratio of 1.3 at α=0.05 typically requires larger n than for 1.5, given the same variance and exposure.

FAQs

Overdispersion?
Use robust methods or quasi-Poisson; it reduces effective power.

Offsets?
Include log exposure in the model to scale rates.