Clicked Gallery

What is Margin of Error?

Highlighted from a real textbook passage. Explained by Clicked.

Used in a sentence

University Course Reader · STEM

The two candidates sat within the poll's margin of error, so the two-point gap between them could be nothing more than the luck of who was called.

The reader highlighted one term β€” in an article or a paper PDF. Clicked made the statistics term β€œmargin of error” easy to understand:

Explained in three depths

Same facts, different vibe β€” Slang mode 😎

The Clicked way

●○○

Overview

The margin of error is the plus-or-minus figure printed next to a poll result. A poll asks a few hundred or a few thousand people, picked at random, and reports their answer as everyone's. A different random group would produce a slightly different answer, so the result is a range, not a single number. Candidate A at 48% with a margin of 3 points means 45% to 51%, which is 48 less 3 and 48 plus 3. The margin is set at the 95% level by convention: of 20 polls run the same way, 19 land within their margin and 1 misses by more. The margin covers one thing: the difference that comes from which people happened to be picked. It does not cover the people the poll could not reach or the way the question was worded. And it applies to each figure on its own, so the gap between two candidates carries about double. Plus or minus 3 on each candidate is plus or minus 6 on the gap between them.
●○○

Overview

Margin of error is the give-or-take on a poll. Nobody asked everyone. They asked about a thousand people, and another thousand would have landed on a nearby number, so every poll figure is really a range. Candidate B on 52%, margin 3, is 49% to 55%, which is 52 minus 3 and 52 plus 3. The margin is sized so that 19 polls in 20 land inside it, and the twentieth misses by more. Here is the trap. "B leads by 4, margin 3" sounds like a lead. It is not, because the margin sits on each candidate separately, so the gap between them gets about twice as much. One more catch: the margin only counts the chance of who got picked, so if the poll missed a chunk of the public, that mistake sits on top. Give or take 3 on each name is give or take 6 on the gap, and a 4-point lead with a 3-point margin is a maybe. 😎

A quick take β€” often all you need.

●●○

Detail

The margin of error is the plus-or-minus figure reported with a poll result. It shows how far the result could sit from the true figure for everyone, because of which people happened to be picked. A poll asks a sample, a part of the group chosen at random. A different sample would give a slightly different answer, and the margin is the size of that difference. Candidate A at 48% with a margin of 3 points means the figure for all voters is probably between 45% and 51%. Probably has a set meaning: the convention is 95%, so of 20 polls run the same way, 19 land within their margin and 1 misses by more. The margin depends on how many people were asked (the sample size) and on nothing else. Divide 1 by the square root of the number asked. The square root of 1,000 is about 32, since 32 times 32 is about 1,000, and 1 divided by 32 is about 0.03, so 3 points. The size of the country plays no part, only the number asked. Halving the margin costs four times the people, so 4,000 gives about 1.6 points. The margin belongs to each figure separately. If A is at 48% and B at 43%, A could be 3 lower while B is 3 higher, so the 5-point lead carries a margin of about 6 and runs from 1 behind to 11 ahead. A slice of the poll is a smaller poll: a group making up 15% of the sample, about 160 people, carries a margin of 8 points. All of this assumes the right people were asked the right question. People the poll could not reach, people who would not answer and the wording of the question sit outside the margin, and Pew Research puts the real error of a typical poll at about double the stated one. Statisticians call the same range a confidence interval, and the margin is half its width. Read a poll in three steps: turn the figure into its range, double the margin before comparing two figures, then ask who was not asked.
●●○

Detail

A margin of error is the give-or-take a poll admits to. It exists because the poll talked to a sample, a random few hundred or thousand, and reported their answer as everybody's. Pick a different random few hundred and the number shifts a bit, and the margin says how much. Where it comes from is simpler than it looks: the more people asked, the smaller it gets, but slowly. Ask 100 people and the margin is about 10 points, because 1 divided by the square root of 100, which is 10, comes to 0.10. Ask 400 and it is 5 points, 1 divided by 20. So four times the people buys half the margin, every time, which is why pollsters stop near a thousand and live with 3, whether the country holds 5 million people or 300 million. Then there is the 19-in-20 rule. The margin is sized so 19 polls out of 20 land within it, and the twentieth is off by more despite being run just as carefully. Now the headline that gets everyone: "A leads B, 50 to 46, margin 3." Each number can be 3 off in either direction, so A could really be 47 and B could be 49. A 4-point lead on paper is a 2-point deficit inside the range. The gap carries about double the margin, 6 points, and 4 sits inside 6. Not a lead. Subgroups make it worse. "Voters under 35" might be 250 people out of that thousand, and 250 people carry a margin of about 6 points, so a subgroup swinging 5 points is a headline about noise. Last, the margin counts one thing only: the chance of who happened to be picked. People with no landline, people who hang up and a question worded to nudge are not in the 3, and Pew Research reckons a real poll's error runs about double the printed margin once those are counted. Stats people call the range a confidence interval. The margin is half of one. Read a poll in three moves: turn the number into a range, double the margin before you compare two numbers, then ask who never picked up. 😎

Want more? One click digs deeper.

●●●

Analogy

The margin of error is how far a poll's figure can sit from the figure for everyone, because of which people happened to be picked, and a bag of M&Ms shows the same thing. The factory mixes the colours so that 20% are blue, and each bag is a random handful of that mix. Open one bag of 50 and the blue share can land anywhere from 9% to 31%, 19 bags in 20. That is a margin of 11 points, and it is why one bag can hold hardly any blue. Pour four bags together, 200 sweets, and the share settles between 14% and 26%: four times the sweets, half the margin. The gap between two colours works the same way: 13 blue against 7 brown in one bag says nothing about which colour the factory makes more of, because each count has its own margin and the gap carries both. A poll of 1,000 people is a bag drawn from everyone, and its margin is the same kind of number. Where the picture breaks: a filling line that has run out of blue gives bags with no blue at all, and opening more of them never puts it back. A poll that cannot reach part of the public has the same fault, and the margin does not show it.
●●●

Analogy

A margin of error says how much a count from a random handful can miss the real share by, and you can watch it from a café window. Count 100 cars going past and 12 are electric. So the town is 12% electric, give or take 6 points, because 100 cars is not many cars. Stay for 400 and the give-or-take drops to 3, at the cost of four coffees: four times the cars, half the margin. That is a poll. The traffic is everyone, the cars you counted are the people asked, and the 12% is the headline number with its margin printed next to it. Now the catch. If the café is next door to the EV dealership, you can count 10,000 cars and still be miles off, because the cars going past were not a fair sample of the town. The margin of error has no way to spot that. It measures how much a fair sample wobbles, not whether your sample was fair. 😎

Unfamiliar concept? A real-world example makes it click β€” fresh analogies on tap.

AI explanations may contain errors · Not professional advice

Formal definition β€” The same term, explained the usual way

The margin of error (margin of sampling error) is the half-width of a confidence interval around a survey estimate, conventionally at the 95% confidence level: for a proportion p from a simple random sample of n, it equals z Γ— √(p(1 βˆ’ p) Γ· n), with z = 1.96, and reaches its maximum at p = 50%, where it approximates 1 Γ· √n (about Β±3 percentage points for n β‰ˆ 1,000). The margin for the difference between two estimates from the same sample is roughly twice the margin for either; margins for subgroups are larger because their n is smaller; and weighting inflates the margin by the design effect. The statistic quantifies random sampling error only and excludes noncoverage, nonresponse and measurement error, which may be as large or larger; Pew Research Center (Mercer, 2016; Keeter and Kennedy, 2024) reports total error for a typical election poll at roughly twice the stated margin. Margins quoted for nonprobability (opt-in) samples are not interpretable in the same way.

Want Clicked to explain terms like “margin of error” directly in your browser β€” including on PDFs?

Add to Chrome β€” Free

50 free Explanations · No credit card required