“Candidate A leads by two points.” A headline like that sounds like a fact. It is closer to a careful estimate, built from the answers of a small group of people and carrying a known amount of uncertainty. Understanding where that uncertainty comes from makes it much easier to read poll coverage without either swallowing it whole or dismissing it.
How can a thousand people stand in for millions?
The idea behind a poll is sampling. If every member of a population had an equal chance of being picked, the group that answers should look, on average, like everyone else. That is called a probability sample. Pollsters cannot interview everyone, but they can use mathematics to say how far a random sample is likely to stray from the truth.
The practical difficulty is the word “random.” There is no complete list of every American adult with a working phone number and a willingness to talk. Pollsters build samples from address lists, phone numbers, online panels recruited in various ways, or a mix of these, and each method has its own strengths and gaps.
What the margin of error says
The margin of error describes sampling uncertainty only. For a well-designed probability sample, the usual rule of thumb is that a survey of about 1,000 people has a margin of roughly plus or minus 3 percentage points at the standard 95 percent confidence level. That means if the same survey were repeated many times, about 95 times out of 100 the result would land within three points of the true value for the whole population.
Three consequences follow:
- A two-point lead is not clearly a lead. If one candidate is at 48 percent and another at 46 with a margin of 3, the true picture could be a small lead for either, or a tie.
- The margin applies to each number, and the gap between two numbers is more uncertain than either one. That is why a close race is described as a “statistical tie.”
- Subgroups carry larger margins. If 1,000 people were surveyed and 150 are young adults, the figure for that subgroup rests on 150 answers and is much less precise than the headline number.
From interviews to published numbers
Raw responses rarely match the population. Some groups answer more often than others, and pollsters adjust the data through weighting. If a sample has too few people without a college degree compared with census benchmarks, those respondents count a little more; if it has too many, they count a little less. Common weighting targets include age, sex, race and ethnicity, education and region, and some pollsters weight by party identification or past vote as well.
Weighting helps, but it can only correct for traits the pollster knows about and measures. It also adds uncertainty of its own, so a weighted poll’s true margin of error is usually larger than the simple formula suggests. Reputable pollsters say how they weighted and why.
Why polls miss, and what they do not measure
The margin of error covers luck in sampling. It does not cover everything that can go wrong:
- Nonresponse. Response rates for phone surveys have fallen sharply over the decades. If the people who decline differ systematically from those who answer, the poll leans.
- Question wording and order. Asking about “tax cuts” versus “reducing government revenue” can change answers, and an earlier question can color a later one.
- Likely voters. Election polls try to guess who will actually vote, and that guess is a judgment, not a measurement.
- Timing. A poll is a snapshot of a moment. Opinions can move before an election, and a poll published today describes interviews done days earlier.
- Social pressure. Some people give answers they think are acceptable rather than what they believe.
When a polling average or a single poll differs from an election result, the cause is usually one of these systematic errors rather than the arithmetic of the margin.
How should you read a poll story?
A few questions are worth asking each time:
- Who ran it, and who paid for it? Look for the pollster’s name and the sponsor.
- Who was surveyed, and how? Adults, registered voters or likely voters? Phone, text, mail or online?
- When, and how many? Dates and sample size should be stated, along with the margin of error.
- What exactly was asked? The full wording matters more than the summary in the headline.
- What do other polls show? A single result that stands apart from the rest deserves caution. Averages of many polls tend to be steadier than any one survey.
Responsible pollsters publish their methods. The American Association for Public Opinion Research runs a transparency initiative that encourages disclosure of exactly this information, and Pew Research Center explains its own methods in detail for readers who want to see how the work is done.
Polls matter beyond election season. They shape how coverage frames the Electoral College and who is ahead in the states, and knowing their limits is part of reading political news well. The habit of asking where a number came from is the same one that helps with reporting that blends fact and opinion.
Sources: Pew Research Center; American Association for Public Opinion Research; U.S. Census Bureau
