Content
Summary
The Poisson distribution is defined by a single parameter where the mean and variance are identical. This equality serves as a diagnostic tool for count data, though real-world data often exhibits overdispersion requiring alternative models like quasi-Poisson or negative binomial.
Intuition
Think of the Poisson distribution as a strict rule for count data: the average number of events must equal the spread of those events. If you observe data where the variance is much larger than the mean, the strict Poisson assumption is violated. This concept helps researchers decide if simple Poisson models are sufficient or if more complex models are needed to handle extra variability.
Key equations
The mean and variance of a Poisson random variable are both equal to the parameter .
The variance in a quasi-Poisson model is a dispersion parameter phi times the mean.
Derivation
- The Poisson distribution is defined by a single parameter .
- By definition, the expected value equals .
- By definition, the variance equals .
- Therefore, E[Y_i] = Var(Y_i) = mu_i.
- In quasi-Poisson models, this equality is relaxed to Var(Y_i) = phi * mu_i to account for overdispersion.
Worked example
Assess if polio deaths (1968-1977) fit Poisson assumptions given mean 18.0 and variance 23.1.
- Calculate sample mean: 18.0.
- Calculate sample variance: 23.1.
- Compare mean and variance: 23.1 is close to 18.0.
- Conclude that the proximity suggests Poisson is a reasonable approximation.
Answer The data suggests a reasonable Poisson fit due to the closeness of mean and variance.
Common mistakes
- Assuming mean and variance must be exactly equal in real data.
- Ignoring overdispersion when variance significantly exceeds mean.
- Using standard Poisson regression when overdispersion is present.