A good sample size for a national political poll is about 1,000 to 1,500 respondents, which gives a margin of error of roughly ±2.5 to ±3 points. Statewide polls usually interview 500 to 1,000 people, and House district polls 400 to 600. Past those numbers, extra interviews cost a lot and buy little precision. What matters more than raw size is whether the sample represents the people who will vote.
It seems odd that 1,000 people can stand in for 160 million voters, but the math holds as long as the sample is drawn and weighted well.
Why is 1,000 respondents the standard? #
Margin of error shrinks with the square root of the sample size, so each extra interview helps less than the one before. A quick formula for the 95% margin, in percentage points, is 98 ÷ √n:
| Sample size | Margin of error | Typical use |
|---|---|---|
| 300 | ±5.7 | Small district or early screening poll |
| 400 | ±4.9 | House district poll |
| 600 | ±4.0 | Statewide poll on a budget |
| 800 | ±3.5 | Solid statewide poll |
| 1,000 | ±3.1 | Standard national or large-state poll |
| 1,500 | ±2.5 | Large national poll |
| 4,000 | ±1.5 | Large study needing subgroup detail |
Going from 400 to 1,000 interviews cuts the margin by almost 2 points. Going from 1,000 to 4,000, four times the cost, cuts it by about 1.5 more. For most election questions that extra precision isn’t worth the money, which is why so many polls cluster around 1,000. Our guide to margin of error explains what the number does and doesn’t cover.
Does a bigger population need a bigger sample? #
No, once the population is large. The margin of error depends on the number of people you interview, not the number you’re describing. A random sample of 1,000 gives about the same precision for Wyoming as for the whole country.
The usual analogy is soup. If the pot is well stirred, one spoonful tells you how it tastes, whether it’s a small saucepan or a huge stockpot. The catch is “well stirred.” In polling, that means a sample where everyone had a fair chance of being included, or one weighted carefully to match the population.
Why does who you poll matter more than how many? #
The classic warning comes from 1936. The Literary Digest mailed about 10 million ballots and got back roughly 2.4 million, one of the biggest polls ever taken. It predicted Republican Alf Landon would beat President Franklin Roosevelt. George Gallup, using a far smaller sample, predicted Roosevelt would win. Roosevelt carried 46 of 48 states.
The Digest’s lists came partly from telephone directories and car registrations, which in the Depression skewed toward wealthier voters. Later research found an even bigger problem: the people who chose to mail back ballots were more likely to oppose Roosevelt. Either way, 2.4 million unrepresentative answers lost to a much smaller, better-designed sample.
The same lesson applies to modern online polls with huge samples drawn from opt-in panels. Size can’t fix a sample that systematically misses certain voters. Our explainer on opt-in online polls covers where web panels work and where they struggle.
What shrinks a poll’s usable sample? #
A poll’s headline sample size can overstate its precision in three ways.
Likely-voter screens #
Many election polls interview a large group of adults or registered voters and then report results for the smaller group judged likely to vote. A poll of 1,200 registered voters might report on 850 likely voters, and the margin should reflect the 850. Check which group the numbers describe. Our guide to registered vs. likely voters explains why midterm polls often differ between the two.
Weighting #
Weighting fixes imbalances in who answered, but it increases variability. The heavier the weights, the smaller the “effective” sample. Well-run polls include this design effect in the margin they publish.
Subgroups #
A poll of 1,000 might include only 120 voters under 30, with a margin near ±9 points. Anything you read about that group is far less certain than the topline.
| Subgroup in a 1,000-person poll | Rough n | Margin |
|---|---|---|
| Women | 520 | ±4.3 |
| Voters 65+ | 280 | ±5.9 |
| Voters 18 to 29 | 120 | ±8.9 |
| Suburban college-educated men | 80 | ±11.0 |
Pollsters who care about a small group oversample it, interviewing extra members, then weight them back to their real share for the overall result. Our guide to reading crosstabs shows how to spot subgroups too small to trust.
What sample size should you trust in a 2026 poll? #
Rough guidance for this cycle:
- National generic ballot or approval: 1,000 or more is standard, and 1,500 to 2,000 is better for tracking small changes.
- Senate or governor race: 600 or more likely voters is solid. Around 400 is usable but noisy.
- House district: 400 to 500 is normal because districts are expensive to poll. Treat a single district poll as a rough read.
A single poll of any size is less reliable than several polls together. Election Tracker puts the sample size and population (likely voters, registered voters or adults) on every poll card. Each race’s 30-day average also counts a pollster only once, at its strongest sample, even if it polled the race several times, so one prolific firm can’t swamp the average.
Frequently asked questions #
Why don’t pollsters interview 10,000 people? #
Cost and diminishing returns. With response rates in the low single digits for phone polls, reaching 10,000 people can take hundreds of thousands of calls or texts. It would cut the margin from about ±3 to ±1 point, but wouldn’t help at all with non-response or a bad likely-voter model, which are often bigger sources of error.
Is a 500-person poll bad? #
No. It gives a margin of about ±4.4 points, which is normal for state and district polls and fine for spotting a clear lead. Just don’t read much into its subgroups or into small changes from one poll to the next.
What’s the difference between a random sample and a representative sample? #
A random sample gives everyone in the target population a known chance of being picked. A representative sample is one whose makeup, by age, education, race, region and so on, matches the population. Pollsters aim for both: sample as randomly as they can, then weight the results so the final sample is representative.
Does a big sample make a poll accurate? #
It makes a poll precise, meaning less random noise. Accuracy also depends on reaching the right people and predicting turnout well. A 5,000-person poll that misses a group of voters can be further off than a well-designed 800-person poll.