# How do opinion polls work?

> Written with AI · [Communications: social media, public opinion and media literacy](https://lessoons.com/en/social-media-and-public-opinion) · Grade 12 · 2026-11-12 · https://lessoons.com/en/social-media-and-public-opinion/how-do-opinion-polls-work

Opinion polls work by asking a random sample, often about 1,000 people, and using the answers to estimate the views of a whole population. Because chance alone moves the result, every poll has a margin of error: for 1,000 respondents about ±3 percentage points. A larger sample shrinks it slowly, but no sample size fixes a biased sample, as the Literary Digest learned in 1936.

## The topic explained

### A sample that stands for millions

Opinion polls cannot ask everyone, so they ask a sample and generalize to the population. The key is a random sample: every member of the population has a known, equal chance of being chosen, so that the sample resembles the population in every way, including ways no one thought to check. A well-drawn random sample of about 1,000 people can describe a country of millions.

How can so few speak for so many? Think of tasting soup: one spoonful tells you about the whole pot if the soup is well stirred, whether the pot is small or huge. The accuracy of a random sample depends mainly on how many are asked, not on what share of the population they are. The common mistake is to think a poll of 1,000 Israelis is ten times less accurate than a poll of 1,000 people in a city a tenth the size. It is about equally accurate.

### The margin of error

Even a perfect random sample differs from the population by chance. The margin of error says by how much. For a result near 50%, at the usual 95% level of confidence, it is about 0.98 ÷ √n, where n is the number of respondents. For 1,000 people: 0.98 ÷ 31.6 ≈ 0.031, about ±3.1 percentage points. For 500: about ±4.4. For 4,000: about ±1.5. Four times the sample gives only half the error, which is why most polls stop near 1,000: more respondents cost a lot and add little.

A made-up poll of 1,000 voters finds party A at 26% and party B at 24%. With ±3.1 points on each number, the true values could easily be the other way round; the poll cannot say who leads. Reporting it as “A ahead” is the classic mistake.

### When size is not enough: 1936

In 1936 the American magazine The Literary Digest mailed ballots to about 10 million people and received about 2.38 million back. It predicted that Alf Landon would win 57% of the vote. Franklin D. Roosevelt won with almost 61% and lost only two states. George Gallup, with a sample of about 50,000, called the result correctly.

The Digest's sample was huge but biased. Its lists came from its own readers, car owners and telephone users, wealthier than most Americans in the Depression, and those who chose to answer were more often Roosevelt's opponents: non-response bias. A biased sample does not improve with size; it only gives a more confident wrong answer.

### Other sources of error

Sampling error is the only error the margin measures. Others can be larger: a sample frame that leaves people out, non-response, the wording and order of questions, social desirability (answering what sounds good), and timing, when opinions change after the poll. Media reports also shape polls' effect: stories about who is ahead may help the leader (a bandwagon effect) or the trailer (an underdog effect), and may feed the spiral of silence.

A careful reader asks five things of any poll: who was asked and how, how many, when, what exactly was asked, and what the margin of error is. A strong exam answer explains random sampling, calculates a margin of error, uses the 1936 case to show bias, and names two other sources of error.

## Practice: how opinion polls work

For 12th grade communications students. The polls and parties in the examples are made up. Use the formula 0.98 ÷ √n for the margin of error. Answer from what is printed here, then check yourself against the answers.

1. **What is a random sample, and why does it allow a poll to describe a whole population?** — A sample in which every member of the population has a known, equal chance of being chosen. It then resembles the population in every respect, including ones no one checked, so its results can be generalized within the margin of error.
2. **Calculate the margin of error for a poll of 2,000 people. Show your work, and check it against the graph.** — √2,000 ≈ 44.7; 0.98 ÷ 44.7 ≈ 0.022, about ±2.2 percentage points, between the values for 1,000 (±3.1) and 4,000 (±1.5) on the graph.
3. **A pollster wants to halve the margin of error of a poll of 1,000. How many people must she ask?** — About 4,000: the error falls with the square root of n, so four times the sample gives half the error.
4. **A made-up poll of 1,000 shows party A at 26% and party B at 24%. A headline says “A takes the lead”. Is this justified?** — No. The margin is about ±3.1 points on each figure, so the true values could be reversed; the poll shows the two parties about level.
5. **Read the chart. Why was the larger 1936 poll wrong and the smaller one right?** — The Literary Digest's sample was biased: it came from readers, car owners and telephone users, wealthier than average, and those who answered were more often Roosevelt's opponents. Gallup's smaller sample was better chosen. Size does not fix bias.
6. **Fill in the missing row of the table: give the source of error and a made-up example.** — Question wording: a loaded phrase such as “Do you back a fair deal?” pushes respondents toward yes.
7. **A made-up news site asks readers to vote online on a question and reports that 80% said yes. Why is this not a poll of the public?** — The respondents chose themselves (a self-selected sample): only readers of that site who cared enough answered, so the result says nothing reliable about the population, and no margin of error applies.
8. **Should the media report election polls in the final days before a vote? Write a short balanced answer and a conclusion.** — A good answer gives the case for (voters have a right to information, and polls are news) and against (polls have errors beyond the margin and can create bandwagon or underdog effects or silence some voters), and concludes with a reasoned view, for example reporting polls with their margin and method.

## Common questions

### What is the margin of error in a poll?

The range within which the true value probably lies because of chance in sampling. For a result near 50% at 95% confidence it is about 0.98 ÷ √n: about ±3 points for 1,000 respondents. It does not cover errors from bias or wording.

### How can 1,000 people represent a whole country?

If they are chosen at random, the sample resembles the population, and its accuracy depends mainly on how many are asked, not on the size of the population, just as one spoonful of well-stirred soup tells you about the whole pot.

### What went wrong with the Literary Digest poll of 1936?

Its 2.38 million responses came from a biased list of readers, car owners and telephone users, wealthier than most voters, and from the people who chose to reply. It predicted the loser would win; Gallup's much smaller sample was right.

### Why do polls sometimes get elections wrong?

Because of errors the margin does not measure: people who refuse to answer, samples that miss groups, loaded questions, respondents who say what sounds good, and voters who change their minds or decide late.

### What is a self-selected poll?

A poll in which people choose to take part, such as an online vote on a news site. It reflects only those who chose to answer and cannot be generalized to the public.
