Brier Score

Definition

Brier score evaluates binary probability forecasts by measuring squared error between a predicted probability and the realized outcome (0 or 1). Lower is better.

How it works

• If the outcome happens, error is (1 minus p) squared.

• If the outcome does not happen, error is (0 minus p) squared.

• Over many forecasts, you average these errors to get your Brier score.

Why it matters

Brier score rewards honest probabilities and penalizes overconfidence. It is a practical tool for tracking whether your probabilities are improving over time, and it connects directly to calibration.

Common pitfalls

Judging from one forecast: Brier score is most useful over many events.

Ignoring calibration: A decent average score can still hide systematic bias.

Learn more

For batch evaluation, calibration tables, and deeper explanations, see BrierScore.com.