Head-to-head
We forecast every race; forecasters and pollsters cover only some. Averages over different races aren't comparable, so here each one is paired with our model on only the races it called, one opponent at a time. Lower is better. Click a name to see the two of us side by side on every one of those races.
vs forecasters
The Brier score of each final win probability, 2012–2024. We were closer is the share of races where our probability beat theirs, leaving out races where the two were identical.
VotePredictor Elections beats 15 of 16 forecasters on their own races.
| forecaster | Races | Their Brier | Our Brier | We were closer |
|---|---|---|---|---|
| Sabato's Crystal Ball | 764 | 0.083 | 0.073 | 63% |
| Cook Political | 748 | 0.081 | 0.071 | 66% |
| FiveThirtyEight | 683 | 0.072 | 0.074 | 57% |
| Inside Elections | 538 | 0.074 | 0.060 | 66% |
| DDHQ/Decision Desk | 531 | 0.095 | 0.080 | 73% |
| The Economist | 501 | 0.088 | 0.080 | 52% |
| JHK Forecasts | 391 | 0.083 | 0.072 | 60% |
| RealClearPolitics | 342 | 0.118 | 0.069 | 79% |
| Elections Daily | 293 | 0.082 | 0.059 | 59% |
| CNalysis | 282 | 0.084 | 0.060 | 63% |
| Race to the WH | 276 | 0.072 | 0.061 | 49% |
| Fox News | 242 | 0.082 | 0.074 | 64% |
| Split Ticket | 196 | 0.089 | 0.075 | 58% |
| Princeton Election Consortium | 131 | 0.060 | 0.038 | 63% |
| New York Times | 51 | 0.081 | 0.059 | 78% |
| Silver Bulletin | 40 | 0.045 | 0.013 | 85% |
vs pollsters
The margin miss in points: a pollster's final poll against our final forecast, 2004–2025. Our model averages the polls, so it is expected to beat most single pollsters; the table shows by how much. Pollsters with fewer than 10 shared races are hidden. Where we have no forecast for some of a pollster's races (mostly older cycles), the races column shows scored / covered.
VotePredictor Elections beats 123 of 132 pollsters on their own races. (12 with fewer than 10 shared races not shown.)
What this does and doesn't show. Their numbers were published before each election. Ours are our walk-forward record: for each cycle the model is fit only on earlier elections and forecasts that cycle, but it is run today, with today's code. That is the standard way to test a model, but it isn't the same as a forecast published at the time. Our 2026 forecasts are published before the election, so they will be.
This pairs raw scores on identical races. For a single ranking that adjusts for race difficulty, see the forecaster and pollster boards.