The margin of error in a poll is shaped by several factors, but sample size is the biggest one. A national poll of 1,000 adults typically carries a margin of error around plus or minus 3 percentage points at the 95% confidence level. Increase that sample to 2,000 and the margin shrinks, but not by half. The relationship between sample size and precision follows a curve of diminishing returns, which is why most major polls land in that 1,000-to-1,500 range as a practical sweet spot.
Sample Size Has the Largest Effect
The more people you survey, the smaller the margin of error. But the gains taper off quickly. Going from 100 respondents to 400 cuts the margin of error roughly in half. Going from 1,000 to 4,000 cuts it in half again. To halve the margin of error at any point, you need to roughly quadruple your sample size. This is why polling organizations don’t just keep adding respondents indefinitely: the cost grows much faster than the precision.
For national surveys of a large population (say, 260 million U.S. adults), the total population size barely matters. Whether you’re polling a country of 10 million or 300 million, a random sample of 1,000 produces essentially the same margin of error. The population only starts to matter when your sample represents a large fraction of it, like surveying 500 people out of a town of 2,000.
The Confidence Level You Choose
Every margin of error comes paired with a confidence level, which describes how sure you want to be that the true value falls within your margin. Most polls use 95% confidence, meaning that if you repeated the poll 20 times under the same conditions, you’d expect the true value to land within the stated range about 19 of those times.
Choosing a higher confidence level widens the margin. At 90% confidence, the multiplier used in the calculation is 1.65. At 95%, it rises to 1.96. At 99%, it jumps to 2.6. So a poll that reports a margin of error of plus or minus 3 points at 95% confidence would have a margin closer to plus or minus 4 points if you wanted 99% confidence from the same data. Pollsters almost always use 95% because it balances precision with practicality.
How Close the Result Is to 50/50
The margin of error is largest when opinion is evenly split. A poll showing a candidate at 50% has more statistical uncertainty than one showing a candidate at 90%. This is because the underlying math involves multiplying the estimated proportion by one minus that proportion, and that product peaks at 0.5 times 0.5, or 0.25. When the proportion is 90%, the product drops to 0.9 times 0.1, or 0.09, which produces a notably smaller margin.
This is why pollsters planning a survey before they know the results use 50% as their assumption. It gives the most conservative (widest) margin of error, ensuring the poll will be accurate enough no matter how the results turn out. In practice, this means that if a race is a blowout, the reported margin of error is slightly wider than it technically needs to be.
Subgroups Expand the Margin Dramatically
The margin of error reported for a poll applies to the full sample. The moment you start breaking results down by demographic groups, the effective sample size for each group shrinks, and the margin of error balloons. Pew Research Center offers a useful illustration: in a national poll of about 1,000 adults, Hispanic respondents (roughly 15% of the U.S. adult population) would account for only about 160 people in the sample. That pushes the margin of error for that subgroup to around plus or minus 8 percentage points for a single candidate’s support, and plus or minus 16 points for the gap between two candidates.
It gets worse. Some demographic groups, particularly minorities and younger adults, respond to surveys at lower rates and need to be statistically “weighted up” to reflect their true share of the population. That means the effective number of interviews driving the estimate can be even smaller than the raw count suggests. So when you see a headline like “Young voters favor Candidate X by 10 points,” the margin of error behind that number may be wide enough to make the lead statistically meaningless.
Weighting and Survey Design
Most polls don’t use simple random sampling, where every person in the population has an equal chance of being selected. Real-world polls use stratified designs, cluster sampling, or online panels, and then apply statistical weights so the final sample matches the population on characteristics like age, race, education, and geography. These adjustments are necessary, but they come with a cost to precision.
The American Association for Public Opinion Research (AAPOR) notes that these “design effects” can substantially increase the margin of error beyond what a simple random sample of the same size would produce. A poll of 1,000 people that requires heavy weighting may effectively behave like a simple random sample of 700 or 800. High-quality surveys factor these design effects into the margin of error they report, but not all polls do, which can make some surveys look more precise than they actually are.
What the Margin of Error Doesn’t Cover
The reported margin of error in a poll accounts for only one type of error: the randomness inherent in surveying a sample instead of the entire population. It says nothing about several other sources of error that can be just as large or larger.
- Coverage error occurs when certain groups in the population have no chance of being reached. A phone-only poll, for example, misses people without phones entirely.
- Non-response bias happens when the people who agree to take the survey differ systematically from those who refuse. If politically engaged people are more likely to answer, the poll skews toward their views.
- Measurement error comes from the questions themselves. Leading wording, confusing answer choices, or the order in which questions appear can all push responses in a particular direction.
Researchers studying total survey error have found that these components, including coverage gaps, item non-response, and measurement problems, can individually rival or exceed the margin of sampling error. A poll might report a margin of plus or minus 3 points while carrying an additional 3 to 5 points of unmeasured bias from these other sources. This is one reason polls sometimes miss badly despite having a reported margin of error that should have captured the true result.
Putting It All Together
When you see a poll’s margin of error, think of it as a best-case floor for how uncertain the results are. The actual uncertainty is almost always higher. Sample size is the most controllable factor, but the confidence level, how close results are to a 50/50 split, the size of any subgroup being analyzed, how much weighting was needed, and the survey’s design all play a role. And the kinds of errors that matter most in real-world polling, like who refuses to participate and how questions are worded, don’t show up in that plus-or-minus number at all.