Vote share estimates how much support a candidate has or may receive. Win probability estimates how often a candidate would win under a model's assumptions about possible outcomes.
These numbers answer different questions. A candidate can have a high modeled chance of winning with a predicted vote share only slightly above half.
An illustrative model
Imagine a simplified forecast running 100 possible election outcomes. Candidate A wins in 70 and loses in 30. The model would assign A a 70% win probability within that simulation framework.
This does not mean A receives 70% of all ballots. It does not mean the model knows which of the possible outcomes will occur.
The example illustrates probability only; it is not an actual forecast for a 2026 race.
Where probability comes from
A forecast may combine polls with assumptions about polling errors, turnout, past elections, and race conditions. The uncertainty assumptions are essential.
If a model allows larger errors, the same polling lead may produce a lower win probability. If it assumes related errors across states, its chamber-control probabilities can change substantially.
Poll aggregation and forecasting are therefore distinct activities. Background: AAPOR's guide to aggregators.
Why a favorite can lose
A 70% probability leaves a 30% probability of losing in that model. A loss is not automatically inconsistent with the number.
Evaluation requires more than checking one favored candidate. Across many appropriately assessed forecasts, calibration asks whether events assigned similar probabilities occur at roughly the corresponding frequency.
A model can also be poorly specified. Probability notation does not establish that its assumptions are realistic.
What to inspect
Read the update time, input sources, uncertainty treatment, and explanation of the modeled event. Winning the popular vote, winning one seat, and controlling a chamber are different events.
Be wary of precise probabilities with no reproducible methodology. A percentage presented without assumptions can be an opinion formatted to look statistical.
How to quote a forecast
Say who produced it, when it was updated, what outcome it models, and what uncertainty remains. Never rewrite a model's “favored” status as a certified result.
Related reading: polling averages and projections vs. certification.