Backtest · Senate and governors, 2016–2024

When the Forecast Disagreed With the Polls

Where FutureBallot's rebuilt forecasts and the polls differed by more than three points, the rebuilt forecast came closer more often: the forecast in thirty-three races, the polls in seventeen

Every Senate and governor race from 2016 to 2024 with at least two polls in its final three weeks, placed by the polls' average margin (across) and FutureBallot's forecast (up). Where the two differed by more than three points (fifty races), the mark shows which ended closer to the result.

Swipe the chart sideways to see all of it.

How to read it. On the diagonal, the forecast and the polls agree; the grey band is within three points. These are not forecasts anyone published at the time. They are rebuilt by today's model (version 0.8.7) from only what was known on each election eve, the final snapshot of each race. A model built after these elections has been tested against them, so this is a kinder test than a forecast made in real time. Calibration across all races is on the Accuracy page.

Source. FutureBallot backtest runs of the current race model, each race's last run before its election; polls from FutureBallot's archive that finished in the last twenty-one days, each poll's margin between the forecast's two candidates averaged equally; results as recorded, November round. Margins in points; the Democrat or independent is on the right. Drawn November 2, 2026, 2:32 AM Pacific.

Link to this chartChart image pageDownload the data (CSV)Free to republish with credit to FutureBallot.