What Race Ratings Mean

The scale, in plain English

• Safe — the race is not expected to be close. One party has a structural advantage large enough that a competitive finish would be a genuine surprise.
• Likely — one party has a clear advantage, but an upset is conceivable. These races deserve watching, not assuming.
• Lean — a competitive race in which one side holds a modest edge. Either party can realistically win.
• Toss-Up — no clear leader. The evidence does not support naming a favorite.

Each label except Toss-Up carries a party direction — “Lean D” or “Likely R” — indicating which side holds the edge, not which side we favor. The same evidentiary bar applies to races of every partisan lean.

A description, not a prediction

A rating summarizes each race's current standing given the evidence today: recent election results in that state, incumbency, and polling where enough of it exists. It is not a forecast of the outcome, and it is deliberately not a win probability — we don't publish those, for reasons explained in How We Handle Uncertainty. When the evidence changes, the rating changes with it; a rating in June describes June, not November.

Favorites lose sometimes — that's the point of the scale

An honest reader should expect rated favorites to lose now and then. Across past cycles — under our scale and everyone else's — some Lean races flip to the other side, the occasional Likely race produces an upset, and even Safe seats very rarely surprise. If every Lean favorite won, the label would be miscalibrated: “Lean” is supposed to mean genuinely competitive. So a Lean D rating reads best as “the Democrat has a modest edge in a race either side can win,” never as “the Democrat will win.” We don't attach precise flip rates to these labels, because any such number would depend on whose ratings, which offices, and which years you count.

Where the rating comes from

The Decision Labs rating is computed the same way for every race, from a published formula: a blend of fundamentals (the state's recent presidential lean, the last result for the same office, incumbency) and our weighted polling average once enough polls exist. When polling is sparse or stale, the formula widens its uncertainty band and shrinks the rating toward Toss-Up — we'd rather under-claim than overstate a lead we can't support. Races rated on fundamentals alone are tagged “fundamentals only” on the ratings page, where we also show Cook Political Report, Sabato's Crystal Ball, and Inside Elections for comparison.

Go deeper

The exact formula — the fundamentals blend, the uncertainty band, and the thresholds between labels — is published in full in the rating methodology. Our nonpartisanship and sourcing commitments are in the editorial standards.