Skip to content

Explainers / MethodUpdated September 30, 2026

Uncertainty

An estimate helps us compare players without settling every close call. A large, repeated difference deserves more attention than a tiny gap in one season. The useful question is whether the uncertainty could change the hockey decision.

A small gap needs more context

Imagine two players estimated at 2.0 and 1.9 WAR over the same season. Those are hypothetical figures. The first number is higher, but a tenth of a win is a thin reason to choose one player over the other.

Look at their roles, playing time and previous seasons. If one player fills a need and costs less, that may matter more than the narrow numerical lead. The figures alone do not tell us the probability that one was truly better.

Measuring the past and predicting the future are different jobs

Even completed games leave questions about individual credit. Players share the ice; teammates, opponents and deployment affect what happens. Adjustments help separate those effects, but the estimate is still an interpretation of the recorded play. More games provide more evidence without making every assignment of credit exact.

A forecast adds another question: what changes next? A player might improve, miss time or get a different job. Even a perfect account of last season would not tell us everything about the next one.

That is why a range around past performance and a range of possible future outcomes are different things. One cannot stand in for the other.

What a range would promise

An 80% forecast range is intended to contain the later outcome about eight times in ten across comparable predictions. It is not a guarantee for one player.

A useful range also needs to be informative. “Somewhere between barely playing and winning the scoring title” might cover plenty of outcomes without helping a team decide anything. Making the range narrow creates the opposite problem if too many players finish outside it.

Our current player estimates do not have supported measurement or forecast ranges attached. A single point therefore does not mean the uncertainty is small. The model-test explainer explains why a useful estimate and a dependable range need separate checks.

Ask what would change the decision

Suppose a contract looks attractive only if the player keeps every bit of last season’s power-play production. The important question is whether he will keep that opportunity, and how much the case depends on it.

Or suppose changing the dollar rate changes whether a past season’s WAR $ exceeds the contract price. That tells us the price judgment is sensitive to the chosen scale; it does not tell us how likely either outcome is.

Strong analysis identifies those turning points. Use the estimate to sharpen the question, then bring in the player’s record, role and realistic alternatives. A contract judgment can be persuasive without pretending the future is settled.

Full method: How we test it