NewsBeagle Blog · Practical guides
How to Read an Opinion Poll Beyond Its Headline Percentage
Use a poll-reading worksheet to check the population, fieldwork, exact question and undecided answers before interpreting a headline result.
Find the people behind the percentage
A poll percentage needs a population attached to it. Adults, registered voters, likely voters, residents and a website's voluntary respondents are not interchangeable groups. Open the methodology or results table and write down who the poll is intended to describe. Then distinguish that target population from the people who actually answered. A large response count alone cannot tell you whether the two groups align.
For a useful reading note, complete this sentence: ‘The poll estimates the views of ___, using responses from ___, collected during ___.’ Leave a blank when the published material does not establish the answer. Do not replace an unknown sampling method with an assumption that everyone had an equal chance of responding. This worksheet evaluates what a claim supports; it does not certify a pollster.
Read the question before interpreting the answer
AAPOR's journalist guidance recommends checking the question wording, question order, sponsor, interview dates, sampling approach and any subgroup being reported. Its guidance also distinguishes probability samples from non-probability samples and warns against presenting election polling as an actual election result. Those checks are more informative than accepting a chart merely because it includes a precise-looking percentage.
Copy the wording of the particular question into your private notes, including its answer options. Was the respondent asked whether a proposal sounds useful, whether they support its cost, or whether they personally would use it? These are different questions. A headline about support for a policy should not silently substitute for an answer about one attractive feature of that policy.
Worked example: the missing undecided group
Imagine a hypothetical survey of 1,000 residents: 460 support a transport proposal, 340 oppose it and 200 are undecided. Support is 46 percent of all respondents. If someone removes undecided respondents, support becomes 460 divided by 800, or 57.5 percent of those expressing a preference. Both calculations can be described accurately, but they answer different questions.
A careful summary would say ‘46 percent support, 34 percent oppose and 20 percent are undecided in this fictional sample.’ If discussing the reduced denominator, say explicitly that 57.5 percent refers only to respondents who chose support or opposition. Neither figure establishes that most residents will use the proposed service, and neither is a real poll about Pattaya or another community.
Keep interview dates and uncertainty visible
A result published today may describe interviews conducted before a recent announcement. Record the fieldwork interval separately from the article's publication time. If a second poll uses different questions, respondents or collection methods, do not treat the difference as a clean measure of opinion changing. First establish what remains comparable; otherwise label the comparison as limited.
Do not apply a whole-sample error figure automatically to a small subgroup. Ask whether the report supplies uncertainty for that estimate and explains its method. AAPOR's best-practice material separates missing responses from options such as ‘don't know’ and discusses transparent reporting of adjustments. An uncertainty label does not account for every possible problem, such as people misunderstanding a question or some groups being missed.
Save a reproducible poll-reading card
Keep the original table link, target population, sample description, fieldwork dates, question wording, answer categories and denominator in one note. Add the sponsor and the stated uncertainty method. For an article quoting only one subgroup, record its response count separately. This small card lets you explain what you know without turning a poll into a forecast or a referendum.
When comparing headlines through NewsBeagle, follow their links to the original poll whenever it is available. If several articles cite one survey, they are not several independent measurements. Use the existing coverage-comparison guide for that source question; use this worksheet for the separate issue of what the survey actually asked and whom its percentages describe.
Our editorial approach
This guide is written by the NewsBeagle Editorial Desk to help readers compare attributed coverage and find original reporting. See our editorial policy and publisher directory.