Pollsters · 2026
Who is polling, and how good are they
Every poll on a race page comes from a firm with a record. This is what the forecast knows about each one, and what it does with their polls.
The field
The scorecard
Each firm's page lists its polls and what the average did with each.
What the average subtracts from the firm's polls. Blue right of the axis is Democratic, red left is Republican; the track runs six points either way.
Mean miss on the margin, zero to ten points. The gray tick is the field median; green is at or under it.
Distance from the other polls in the race, zero to 1.6. The gray tick is independence, 1.00; amber is closer to the field than chance allows.
Left out of the averages
These polls stay visible, marked excluded, but count in no average: not a race, the generic ballot or approval.
Reading the columns
- This cycle
- The firm's polls in the database this cycle, and the races they cover. Primaries and hypotheticals count here; only general-election questions enter an average.
- House effect now
- The lean the average removes from the firm's polls today. Beneath it: the raw gap to other firms in the same races, over how many polls, and the published lean it is shrunk toward.
- Our error, 2018 to 2024
- Mean miss on the margin, with signed bias, for the firm's House and Senate polls in the final 21 days, sponsored ones included. A firm with fewer than 5 scored polls is marked thin and is not ranked or plotted.
- Against the field
- The gap to other firms' polls in the same race, in units independent polling would show: about one is independent, well under one stays close to the field.
- Not in the table
- Silver Bulletin's own error, margin-of-error share, herding penalty and pollscore are on each firm's page.
The forecast reads two of these: the published lean a house effect is shrunk toward, and whether a firm is excluded. The methodology page says why the pollscore carries no weight. Ratings are Silver Bulletin's 2026-01 edition, used with attribution.