Course contentsShow
← Data Science
Lesson 691 of 2,14518. Generalized Linear ModelsFree lesson

Interpreting Poisson Coefficients

Understanding coefficients as multiplicative effects on the rate, and converting log-scale estimates to interpretable ratios.

Interpreting Poisson Coefficients

You'll learn: How to translate Poisson regression coefficients from log-scale into meaningful multiplicative effects on count rates.

The Core Idea

In Poisson regression, we model count data (like the number of customer calls per hour or accidents per month). The model works on the log scale of the rate, which means our coefficients represent logarithmic changes. To interpret them in a practical way, we need to convert them back to multiplicative effects.

Here's the key: when a coefficient is β, a one-unit increase in that predictor multiplies the expected count by e^β (where e ≈ 2.718).

From Log-Scale to Ratios

Imagine you're predicting daily website visits. If your model gives a coefficient of 0.5 for "marketing_campaign" (a binary variable):

  • The raw coefficient (0.5) is on the log scale
  • Exponentiate it: e^0.5 ≈ 1.65
  • Interpretation: Running a marketing campaign multiplies expected daily visits by 1.65, or increases them by 65%

If the coefficient were -0.3:

  • e^-0.3 ≈ 0.74
  • The predictor reduces the count to 74% of baseline (a 26% decrease)

Why Multiplicative?

Unlike linear regression where effects are additive ("+5 units"), Poisson models deal with rates that compound multiplicatively. This makes sense for counts: doubling your staff might double calls handled, not just add a fixed amount.

The link function (log) ensures predictions stay positive, and exponentiation reverses it for interpretation.

Key Takeaway: Poisson coefficients live on the log scale; exponentiate them (e^β) to get interpretable rate ratios. A coefficient of 0 means no effect (e^0 = 1), positive coefficients increase the rate, and negative ones decrease it.