How to read election forecasts
What a rating, a model, a market price and a poll each tell you, and what they cannot.
Four kinds of signal
A rating is an analyst’s call placed on a scale from Safe to Toss-up. It is a category, not a probability, and each publisher defines its categories in its own way, so “Lean” at one outlet is not guaranteed to mean the same thing as “Lean” at another. Ratings change when the analyst decides they should, and they carry a date.
A model turns polls, past results, money and other inputs into a probability by simulating the election many times. A 70% chance does not mean the favourite will win: it means that in about three of ten similar situations the other side would. Models differ because their inputs and assumptions differ.
A prediction market is the price traders pay for a contract that pays out if a side wins. The price is read as a probability, but it reflects who chose to trade and how much money they put in. Thin markets, where few people trade, can move a lot on small orders; the two prices need not add up to exactly 100%.
A poll is a sample of people at one moment. The margin of error applies to each candidate’s share, so the uncertainty on the gap between two candidates is roughly twice as large. Polls also differ in who they reach and how they weight the answers.
Why they disagree
Different information (a recent poll one source has not yet used), different timing (a rating from last month next to a market price from today), different definitions, and honest differences in judgement. When sources disagree, the useful question is what each one is looking at. On our site each forecaster has a page that says who runs it, how it works and any documented tie to a party or campaign.
What none of them is
None of these is a result. A forecast describes uncertainty before the vote; it can be right or wrong in any single race, and a favoured candidate can lose. We publish no rating or probability of our own, and we endorse no candidate or party.
How to use the site
Every House, Senate and governor race page has a table that puts ratings, models and the market side by side, with the date of each. The disagreements page lists the races where sources split. The rating changes page shows who moved a race and when we first saw it. The forecasts page gives the national picture, and the sources page lists every forecaster, its owner and ties. Terms like “toss-up” and “generic ballot” are explained in the glossary.