Recommended Free Tools
A poll’s margin of error describes uncertainty from sampling, not the total amount a poll could be wrong. A small lead or a change between polls is meaningful only when the uncertainty in that difference and the comparability of the surveys support that conclusion.
What a poll’s margin of error means
A poll asks a sample of people to represent a larger population. Another properly conducted sample could produce a somewhat different estimate, and a margin of sampling error expresses the range of sampling variation associated with a particular estimate under the poll’s design and confidence procedure.
It is not a guarantee that the population’s true value falls within the range, nor does it account for every source of error. The American Association for Public Opinion Research (AAPOR) explains that a margin applies to sampling error, not problems such as nonresponse bias or an incorrect turnout model: AAPOR’s “Polling Accuracy” explainer.
Use percentage points to describe movements in support. A rise from 48% to 51% is a 3-percentage-point increase; a 3% increase would instead describe a relative change.
#1 Best Overall
Is a 2-point lead meaningful?
Not automatically. AAPOR’s explanatory example has Candidate A at 48% and Candidate B at 46%, each with a margin of error of ±3 percentage points. AAPOR characterizes the race as a statistical tie: the 2-point difference is too small to establish that one candidate is ahead. The example is a rule-of-thumb illustration, not a universal test for every poll.
The margin shown for each candidate’s share is not the uncertainty of the lead itself. In a Pew Research Center example, individual candidate estimates with 3-point margins produce approximately 6 points of uncertainty for the difference between candidates. The calculation depends on the estimates and survey design; readers should use a pollster’s direct comparison when available. See Pew’s explanation of margins in election polls.
Did support really change between two polls?
A change in headline percentages does not, by itself, show that opinion changed. For two survey waves, look for a pollster-provided significance test or confidence interval for the difference, calculated using the uncertainty of both estimates and the survey design. The UK Office for National Statistics describes significance testing as a way to assess whether a difference between survey estimates reflects population change rather than sample variation; it notes that a 5% threshold is often used. That threshold is a convention, not proof that a result matters in practical terms. Read ONS guidance on significance testing.
Do not treat visual overlap—or lack of overlap—between two reported margins as a definitive test unless the pollster’s methodology supports that interpretation. If the necessary design information or test is not published, the headline figures alone may not establish whether the apparent change is statistically meaningful.
Rank #3
Statistical significance and practical importance are separate questions. A small difference may be detectable with enough data but have little real-world consequence; a larger apparent difference can remain uncertain when estimates are imprecise.
Check whether the polls are comparable
Even a valid statistical comparison can mislead if the polls measured different things or used different methods. Before describing a trend, compare the following:
Rank #4
- Population and geography: Adults, registered voters, and likely voters are different populations; national and state-level results are not interchangeable.
- Field dates: A later survey may reflect intervening events or campaign movement as well as sampling variation.
- Question wording and answer options: Changes in wording or available responses can change what people report.
- Mode, recruitment, and sample construction: How respondents are contacted and selected can affect the estimates.
- Sample size and subgroup size: Smaller samples generally yield less precise estimates. Subgroup estimates use fewer respondents than the full poll and therefore tend to have greater uncertainty; check the subgroup base and its reported precision.
- Weighting and design effects: Weighting, clustering, and other adjustments can affect estimates and their uncertainty. Check whether the published precision accounts for them.
- A direct test of the difference: This addresses whether the change is larger than expected from sampling variation more appropriately than comparing point estimates alone.
AAPOR’s transparency guidance describes the methodological details readers need to assess survey results. Its election-polling resources also advise reporting subgroup sample sizes and recognizing that subgroup margins are larger than those for the full sample.
Why some polls do not have a conventional margin of error
Probability samples
When people are selected through a probability-based design, a design-based margin of sampling error can be estimated. Real survey designs may require adjustments for weighting, clustering, and other departures from a simple random sample. AAPOR says reports should indicate whether sampling-error estimates have been adjusted for design effects; see its standard definitions and reporting guidance.
Best Value
Nonprobability samples
Opt-in panels and other nonprobability samples do not have a simple conventional margin-of-error calculation. Their precision estimates depend on a statistical model and its assumptions, which should be disclosed. A reported “credibility interval” is model-based and is not interchangeable with the conventional margin of sampling error for a probability poll. AAPOR discusses these limitations in its task-force reports on nonprobability sampling.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A poll is a snapshot, not an election forecast
A poll estimates opinion for a defined population during a particular field period; it does not determine the election result or guarantee who will win. AAPOR’s 2024 pre-election guidance says polls can offer an approximate picture of where things stand but are not predictive and may not identify who is ahead in a very close election. For close contests, look at multiple polls and broader trends rather than treating one point estimate as a forecast. See AAPOR’s 2024 pre-election polling guidance.
For historical context only, AAPOR’s 2020 polling task-force report found that 223 of 348 state-level polls (64%) had a candidate margin more than twice the margin of error; under the report’s standard, the other 36% could not statistically distinguish the margin from a tie. Those figures describe that specific set of 2020 polls, not current polling generally. The report is available in AAPOR’s task-force reports.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems
