How to compare election polling coverage

Before you compare

1. Identify the poll and release date; record the interview dates separately from the article’s date.
2. Match adults, registered voters, or likely voters, along with geography and question wording.
3. Ask whether the headline describes a candidate’s support, the gap between candidates, or a change since an earlier poll.
4. Read the methodology and the pollster’s explanation of uncertainty, including any subgroup or model assumptions.

Fictional practice example

A lead can narrow without establishing a trend

Fictional practice example: two releases from the same invented pollster survey the same population with the same question. In the earlier release, Candidate A has 48% and B has 44%. In the later release, A has 47% and B has 45%. These figures are not real election data. No uncertainty assessment is supplied in this exercise.

A headline about A’s numerical advantage and one about a narrower gap can both describe these figures. The arithmetic alone does not show that opinion really shifted or that A has a statistically clear lead. That requires the survey’s uncertainty assessment and appropriate analysis of the comparison.

Start with the underlying poll

Open the pollster's release or methodology linked by the article. Record the pollster, sponsor, interview dates, population, sample size, and question wording. Articles published on the same day may discuss different surveys. AAPOR's reporting guidance explains why methods, wording, timing, and the population surveyed matter when interpreting results.

Separate the result from the claimed change

Consider this invented example: one headline says Candidate A leads a new survey, while another says Candidate B closes the gap. The first describes a level; the second describes movement relative to an earlier measure. Both need evidence. Open both reports and identify which earlier poll the second uses. If the comparison changes pollsters, populations, or question wording, record that before interpreting the movement. This exercise concerns what the reporting supports, not which candidate you prefer.

Read uncertainty in the source's terms

A poll measures opinion at a particular time; it is not an election result. Check how the sample was obtained and how the source describes uncertainty. A sampling margin does not capture every possible source of survey error. Small subgroups can have greater uncertainty than the full sample, so a subgroup headline may not have the same precision as the overall result.

Why a small numerical lead is not a clear advantage

Pew Research Center cautions against presenting uncertain poll differences as firm leads. A reported sampling margin describes only part of a survey’s uncertainty. Also distinguish uncertainty around one candidate’s support from uncertainty around the difference between two candidates; the printed margin is not automatically the right threshold for the lead. Use the pollster’s analysis rather than deciding significance by eye or treating one poll as an election forecast.

Compare emphasis without guessing motives

Use the paired-headline view to notice the subject, action, timeframe, and qualifications. Then ask what the lead paragraphs add. Does one report foreground the numerical leader while another discusses turnout assumptions or movement over time? Write a descriptive observation: 'This report emphasizes the current lead; that report emphasizes change.' The wording highlights identify differences in vocabulary. They cannot establish that an outlet intended to mislead, and a spectrum position is not an accuracy score.

Save evidence you would want to revisit

Share the selected pair so another reader can inspect the same comparison. Follow the story to revisit added reporting. On your next visit, check whether a new article brings a new survey or repeats an existing one. Several outlets citing the same poll are several presentations of one measurement, not several independent measurements. If Parallax groups unrelated polls or events, use the correction link on the story rather than treating the mismatch as a difference in framing.