How to read a 2028 primary poll without getting fooled
Margin of error, sample type, house effects and undecideds — the four numbers that matter more than the topline.
Published July 30, 2026 · Updated August 5, 2026
Key takeaways
- A single poll is a data point; an average is evidence.
- Margin of error applies to each candidate separately, so a gap needs to be roughly twice the MoE to be meaningful.
- A large undecided share is the single most underrated indicator of a volatile race.
Margin of error is bigger than you think
A survey reported with a three-point margin of error means each candidate's number carries that band. When you compare two candidates, the uncertainty on the gap between them is close to twice that. A four-point lead in a three-point-MoE poll is not a lead you should treat as real.
Subgroups are worse. When a pollster reports how a candidate performs among voters under thirty, that number is drawn from a fraction of the sample and carries a margin of error two or three times the headline figure.
Who was actually asked
Primary polls sample very different universes: all adults, registered voters, self-identified partisans, or modeled likely primary voters. Likely-voter screens this far from an election are guesswork, because nobody knows who will show up for a caucus eighteen months out.
Open primaries make this harder still. In states where independents can vote in either primary, the pollster's assumption about crossover turnout can move a candidate's number several points on its own.
House effects and averages
Pollsters make consistent methodological choices, and those choices produce consistent lean — a house effect. It is not bias in the pejorative sense; it is the residue of weighting decisions. The fix is not to discard the pollster but to average across many of them, which is what an aggregate does.
Averages also smooth timing. A single survey taken during a news cycle captures that cycle; an average taken over weeks captures the underlying level.
Undecideds are the real story
When a fifth or more of respondents are undecided, the race is not what the topline says it is. Those voters break late and they break together, which is how a stable-looking primary turns over in the final ten days.
Second-choice questions are the best available window into where they go. A candidate who is nobody's first choice but everybody's second is one consolidation event away from leading.
Frequently asked
- Are polls this far from 2028 accurate?
- They are accurate measures of current opinion, not forecasts. Early primary polling mostly measures name recognition, which is why organization and fundraising are better early indicators.
- Why do different 2028 polls disagree?
- Different pollsters sample different universes and weight them differently. Those choices produce consistent house effects, which is why averaging across many surveys is more reliable than any single one.