Read an election poll without overclaiming
Election polls can describe current preferences, but the outcome depends on who votes, how views change, and how well the sample represents the electorate. Small reported leads require careful wording.

How large is the apparent lead
Registered versus likely voters
Read the population first: all adults, registered voters, or modeled likely voters. Each answers a different question. Check how respondents were contacted, the field dates, exact candidate wording, undecided share, and treatment of people who say they may not vote. A conventional margin of error captures only some sampling uncertainty; nonresponse, weighting choices, and turnout modeling add other sources. Poll averages can smooth individual noise but may still reflect shared bias or stale surveys. A national poll may say little about a contest decided in smaller geographic units. Compare multiple reputable polls and their methods, then describe the range of plausible interpretations instead of announcing a winner. A poll can be useful even when the final result differs; it was a snapshot of measured preference, not a ballot count.
Who is being polled
One poll shows a candidate ahead by two points. Its uncertainty and likely-voter method leave room for a different order, and the fieldwork ended before a major debate. Calling the race decided from that result would overstate the evidence.
Timing and turnout assumptions
For a poll headline, find the methodology and record sample definition, field dates, sample size, undecided share, and question wording. Compare at least one other poll of the same contest and period before describing a shift.
An estimate with uncertainty
Rewrite a headline about a narrow lead as a precise statement about the surveyed population and dates. Add one sentence explaining why the result does not settle the eventual vote.
Look across several measures
Do not add the reported margin of error to a lead mechanically and call that a complete forecast. Poll uncertainty includes factors that a single published interval cannot capture.


