A candidate leads by two percentage points in a new poll. Does that mean they’re ahead in the race?
Not necessarily.
Political polls offer a snapshot of public opinion at a particular moment, but understanding what those numbers mean requires looking beyond the headline. Sample size, margin of error, the people surveyed and the questions they’re asked can all affect how poll results should be interpreted.
Michael Adams, a political science professor at Texas Southern University (TSU), says voters should understand what a poll measures before drawing conclusions from it.
One of the first distinctions is between approval ratings and election polls.
An approval rating measures how people evaluate an elected official’s job performance. It essentially asks voters how they think a leader is doing in office.
Election polls, meanwhile, measure voter preferences in a particular contest. Primary polls gauge support among voters choosing a party’s nominee, while general election polls measure preferences in the contest between candidates running in the general election.
Exit polls serve a different purpose, surveying voters after they have cast their ballots to learn more about their choices and the factors that influenced them.
Each type of poll answers a different question, which is why the numbers should be considered in context.
When a poll shows a candidate leading by just two or three percentage points, the race may effectively be a toss-up.
Adams says the margin of error is an important factor in understanding how much confidence to place in a poll’s results.
For example, if one candidate has 48% support and another has 47%, that one-point difference does not necessarily establish a clear leader. If the poll’s margin of error is larger than the gap between the candidates, the results may be statistically consistent with either candidate having the advantage.
“The margin of error is basically the way we look at the surveys,” Adams said, explaining that researchers also use weighting and sample design when analyzing their results.
The margin of error is not a guarantee that the actual result will fall within a particular range. It is a measure of sampling uncertainty under specified assumptions, and it does not account for every possible source of polling error.
A poll’s sample size can affect the precision of its estimates. Generally, a larger random sample produces a smaller margin of sampling error, assuming other factors remain comparable.
Adams said researchers conducting statewide polling in Texas may survey roughly 1,800 to 2,000 likely voters, depending on the study.
But sample size alone does not determine whether a poll is reliable. Researchers must also consider whom they survey, how participants are selected and whether the sample reflects the population they’re trying to measure.
That is particularly important in elections, where the opinions of people who are likely to vote may be more relevant than those of the general population.
Adams said researchers can use voter files and participation in previous elections to help identify likely voters rather than relying solely on people’s stated intentions.
Another important question is who commissioned the poll.
Adams said some surveys are conducted internally by candidates, campaigns or political action groups. These polls can help campaigns assess their standing and may also be used to support fundraising efforts.
That does not automatically make a campaign-sponsored poll inaccurate. However, readers should consider the source and examine how the survey was conducted before accepting its conclusions.
Adams also cautioned against polls that use leading questions or wording designed to influence respondents.
For example, a question that presents negative claims about a candidate before asking whether the respondent supports that candidate can influence how people answer.
Polls that invite people to call in or voluntarily submit their opinions also have limitations. Because participants choose whether to take part, those results do not necessarily represent the broader voting population.
Adams recommends paying attention to a poll’s methodology, including how participants were recruited, how questions were worded and how the results were presented.
Readers should look for more than the headline numbers. Reputable polling reports may also publish detailed breakdowns, often called cross-tabs, that show how different groups of respondents answered questions.
Those breakdowns can help readers understand differences in opinion among demographic groups or across other characteristics. However, smaller subgroups often have greater uncertainty, so those figures should be interpreted carefully.
Adams also pointed to the importance of research oversight and protecting survey participants’ privacy. At universities, research involving human participants may require review to ensure appropriate safeguards are in place.
Even carefully conducted polls can miss the mark. Public opinion can change, some groups may be harder to reach than others, and survey methods have limitations.
Adams referenced historic polling failures, including the widely cited 1936 presidential election polling that incorrectly suggested Republican candidate Alf Landon would defeat President Franklin D. Roosevelt, and the 1948 election, when some polls incorrectly pointed toward Thomas Dewey defeating President Harry Truman.
Modern polling methods have evolved, but no survey can eliminate uncertainty.
The takeaway for voters is simple: Before deciding what a poll says about a race, check who conducted it, who was surveyed, the sample size, the margin of error and how the questions were asked.
A poll can provide useful information about public opinion at a particular moment. It cannot guarantee what will happen on Election Day.
