Tutorial 4: Handling Overdispersion

Negative Binomial regression for overdispersed count data

Continuing from Tutorial 3: The Overdispersion Problem

In Tutorial 3, we fit a Poisson model to bike rental counts. But we discovered a serious problem: overdispersion (deviance/df = 524, should be ~1). The Poisson model assumes Variance = Mean, but our data has Variance >> Mean. This makes our standard errors way too small and our p-values unreliable.

Solution: Negative Binomial regression adds a dispersion parameter to properly model the extra variance.

Step 1: Select the Response Variable

Drag Rental Count to the Response zone on the right.

Same setup as Tutorial 3 - the difference will be in distribution choice.

Why this variable? We're modelling the same counts, but with a better distribution.

Available Variables

Your Model

Predictors (X)

Variables that help predict the count

Drop predictors here

Response (y)

What we predict

Drop response here

Systematic component ready! This is identical to Tutorial 3.
The key difference: we'll choose Negative Binomial instead of Poisson.

← Back to Tutorials