RealClearPolitics
Forecaster quality, broken out by office and by cycle. Scored on the Brier score of its final pre-election win probability (lower is better), over 219 races, 2016–2024.
Head-to-head vs the VotePredictor model
Apples-to-apples: on the 219 races RealClearPolitics forecast, its win-probability Brier vs VotePredictor's calibrated win-prob on those same races.
| Model | Brier | Called right |
|---|---|---|
| RealClearPolitics | 0.148 | 71% |
| VotePredictor | 0.094 | 87% |
Every race (219)
Each race RealClearPolitics scored, with its final pre-election call vs who actually won, and the Brier score for that call. Click a race for its full detail.
By office
| Office | Races | Brier | Called right |
|---|---|---|---|
| House | 110 | 0.193 | 76% |
| Senate | 58 | 0.130 | 93% |
By cycle
| Year | Races | Brier | Called right |
|---|---|---|---|
| 2020 | 81 | 0.089 | 98% |
| 2022 | 100 | 0.207 | 73% |
| 2024 | 38 | 0.122 | 100% |
Why no “time to election” curve? Unlike pollsters — where we have every poll with its field dates — our forecaster archive (JHK Forecasts) stores only each outfit's final pre-election forecast per race, not its earlier snapshots. So a lead-time breakdown genuinely isn't in the data for forecasters. Pollster pages do show it (e.g. the leaderboard).