Explainer · 7 min read

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.

Keep reading