Opinion polls can clarify what survey respondents say at a particular moment, but they are not forecasts of election results. This guide explains how to read political polls, compare surveys fairly, understand margins of error and turnout assumptions, and decide whether an apparent change is meaningful enough to track.
Overview
A poll is a measurement of opinion gathered from a sample of people. It usually reports how that sample answered a question about a candidate, party, policy, or election matchup. Because the survey does not normally ask every eligible voter, its result includes uncertainty. Because voters can change their minds, campaigns can develop, and turnout is difficult to model, a poll should be treated as a snapshot rather than a prediction.
The most useful way to read election polling is to examine several elements together: who was surveyed, when the interviews took place, how respondents were contacted, how likely voters were identified, what question was asked, and how the results compare with other credible surveys. A single percentage rarely tells the whole story.
Polling averages can reduce the influence of an unusually high or low result, but an average is only as sensible as the polls included in it. Different aggregators may use different rules for weighting pollster quality, sample size, recency, methodology, and whether a survey covers registered voters or likely voters. Treat an average as a useful summary of available evidence, not as a guarantee of the outcome.
What to track
1. The population being measured
First identify whether the poll surveyed all adults, registered voters, likely voters, or another defined group. These populations are not interchangeable. A poll of adults may describe broad public opinion, while an election poll based on likely voters attempts to estimate the people most likely to cast ballots. The change from one population to another can affect the result even when public sentiment has not changed.
2. The field dates
Record when interviews were conducted, not merely the date the poll was published. A survey released after a major event may have been completed before that event. When comparing polls, field dates usually provide a more accurate timeline than publication dates.
3. Sample size and margin of error
Sample size describes how many people provided usable responses. A larger sample can reduce random sampling error, but it does not automatically make a survey accurate. The margin of error is an estimate of the uncertainty associated with random sampling under stated assumptions. For example, if a candidate receives 48 percent in a poll with a margin of error of plus or minus 3 percentage points, the result should not be read as an exact 48 percent measure.
Margins of error also do not cover every possible source of error. Nonresponse, question wording, weighting decisions, coverage gaps, and turnout assumptions can affect results. Some online or opt-in surveys may use different uncertainty measures from traditional probability samples, so read the pollster's methodology rather than assuming every margin is calculated in the same way.
4. Question wording and the candidate field
Small changes in wording, response options, order, and the inclusion of third-party candidates can influence answers. Compare like with like whenever possible. A two-candidate matchup should not be directly compared with a poll that offers several candidates or asks about a different stage of the election process.
5. Contact method and weighting
Note whether respondents were reached by telephone, online panel, text invitation, in-person contact, or a combination of methods. Pollsters may weight responses to align the sample with characteristics such as age, education, geography, or past voting behavior. Weighting can improve comparability with the target population, but each pollster's choices are part of its methodology and can produce different results.
| Item to compare | Question to ask | Why it matters |
|---|---|---|
| Field dates | When were interviews completed? | Shows which events respondents could have considered. |
| Population | Adults, registered voters, or likely voters? | Different groups may have different preferences and turnout patterns. |
| Sample size | How many respondents answered? | Helps assess the amount of sampling uncertainty. |
| Method | How were people contacted and selected? | Reveals how the sample was assembled. |
| Question | What exactly was asked, and what options were offered? | Wording and candidate lists can affect responses. |
| Uncertainty | What margin or confidence information is provided? | Prevents small differences from being treated as precise. |
A reusable polling worksheet
For each survey, record: pollster; source link; field dates; release date; geography; population surveyed; sample size; contact method; question wording; candidate or policy options; reported percentages; margin of error or other uncertainty measure; and any weighting or turnout notes. Then add the survey to a simple comparison sheet with the newest field date at the top.
For a basic unweighted average, add the reported percentages from comparable polls and divide by the number of polls. This is a descriptive calculation, not a formal forecast. Do not combine polls that ask materially different questions or measure different populations without labeling the difference. If using a published polling average, read its methodology and note how it handles recency, pollster frequency, sample size, and partisan or house effects.
Cadence and checkpoints
During a quiet period, checking polling monthly or quarterly is usually enough for an evergreen record. During an active campaign or a period of frequent releases, review the comparison sheet whenever a new survey appears, but avoid rewriting the overall conclusion after every individual result.
Use three checkpoints:
- At publication: confirm the field dates, population, sample size, method, wording, and uncertainty.
- After several comparable polls: compare the range and central tendency rather than focusing on the newest number alone.
- After a major methodological change: separate the new series from older results if the population, question, candidate field, or polling method has changed substantially.
Keep a dated log. A record of what was known at each point prevents hindsight from making old uncertainty appear obvious.
How to interpret changes
Begin by asking whether the movement is larger than the uncertainty in the individual surveys. A one-point change may be ordinary variation, especially when the polls use different samples or methods. Even a change larger than a stated margin of error is not automatically evidence that voters have changed their minds, because the margin does not capture every source of error.
Look for persistence across multiple comparable polls. A pattern is more informative when several pollsters, using recent field dates, point in the same direction. Also check whether the apparent movement is caused by changes in undecided responses, turnout screens, third-party options, or the composition of the sample.
Undecided voters should not automatically be allocated to one candidate. Their eventual choices may matter, but polling alone cannot establish where they will go or whether they will vote. Similarly, a candidate's lead is not the same as a probability of winning. Polls measure preferences; election results depend on turnout, ballot rules, vote counting, and the distribution of support across relevant places.
When reporting a close contest, use careful language such as “the surveys are within the stated uncertainty” or “recent comparable polls show a narrow range.” Avoid turning a small lead into a claim of certainty. For background on following official election information, see How to Follow an Election: Deadlines, Ballots, Results, and Official Sources.
When to revisit
Revisit this guide and your worksheet whenever a new poll is released, a pollster changes its method, a candidate enters or leaves a race, or the survey question changes. Refresh the definitions and comparison notes monthly or quarterly during quieter periods, and more often during an election campaign. Replace outdated links, verify that field dates and methodology details remain available, and label any discontinued series.
Before sharing a polling claim, complete this short check: Is the source identifiable? Are the field dates clear? Is the population comparable with the other polls being discussed? Are the question wording and candidate options the same? Is the difference larger than ordinary uncertainty? Does the claim describe a measurement rather than a prediction?
If the answer to any of these questions is no, qualify the statement or wait for better evidence. Polls are most useful when read as a body of imperfect, time-stamped information. A consistent tracking habit—rather than attention to the most dramatic headline—helps readers understand what the numbers show, what they leave unknown, and when a real change may be emerging.
