The margin of error is the half-width of a confidence interval: the "plus or minus" attached to a survey result. It quantifies how much a result could move purely because you asked a sample rather than everyone. This calculator computes it for both proportions and means, and works the calculation backwards to tell you how many responses a tighter margin would require.
Arb Digital built the reverse calculation in because that is the question people actually have. Knowing a poll of 1,000 carries ±3.1 points is mildly interesting; knowing that halving it to ±1.55 points requires roughly 4,000 responses is what changes a research plan.
What This Margin of Error Calculator Does
In proportion mode, enter the percentage you observed and your sample size, and the tool returns the margin in percentage points along with the full confidence interval. In mean mode, enter the sample standard deviation instead and it returns the margin in your original units, using a Student t critical value because the spread was estimated from the sample.
Two optional inputs refine the result. A population size applies the finite population correction, which tightens the margin when your sample is a large share of a small group. A target margin of error triggers the reverse calculation, showing the sample size required to achieve it at your chosen confidence level.
How to Use It
- Choose proportion or mean to match the type of question. Yes/no and pick-one questions are proportions; numeric answers are means.
- Enter the observed result. If you are planning rather than reporting, leave the proportion at 50%, which produces the widest and therefore safest margin.
- Enter your sample size — the number of completed responses, not the number of invitations sent.
- Pick a confidence level. 95% is the near-universal convention in survey reporting.
- Set a target margin to see the sample size it would take, then compare that against what your budget and timeline allow.
The Formula and a Worked Example You Can Check
For a proportion the margin is MOE = z × √(p̂(1 − p̂) ÷ n). For a mean it is MOE = t × s ÷ √n. Rearranged for planning, the sample size needed for a target margin E on a proportion is n = z² × p̂(1 − p̂) ÷ E².
Take the defaults: p̂ = 0.50, n = 1,000, 95% confidence. The z critical value is 1.9600. The standard error is √(0.5 × 0.5 ÷ 1000) = √0.00025 = 0.015811. The margin is 1.96 × 0.015811 = 0.031, or ±3.10 percentage points, giving an interval of 46.90% to 53.10%. This is the familiar "±3 points" quoted alongside national polls of about a thousand people.
Now the reverse. To reach a target of ±2 percentage points at the same confidence, n = 1.96² × 0.25 ÷ 0.02² = 3.8416 × 0.25 ÷ 0.0004 = 2,401 responses. Cutting the margin from 3.1 points to 2 points — a modest-sounding improvement — costs an extra 1,400 completed interviews. The underlying interval construction follows the method documented in the NIST/SEMATECH e-Handbook section on confidence limits.
Why 50% Gives the Widest Margin
The quantity p̂(1 − p̂) reaches its maximum at p̂ = 0.5, where it equals 0.25. At p̂ = 0.9 it falls to 0.09, and at p̂ = 0.99 to just 0.0099. Because that term sits inside the square root, the margin of error shrinks substantially as a proportion moves towards either extreme.
Two practical consequences follow. When planning a survey before you know the answer, assume 50% — it guarantees your sample is large enough whatever the result turns out to be. And when reporting, remember that a single overall margin of error does not apply equally to every question on the survey: an item where 92% agreed carries a materially smaller margin than one that split 51/49, even though both came from the same respondents.
The Part the Margin of Error Does Not Cover
This is the most important limitation and the one most often ignored. The margin of error quantifies sampling error only — the variation you would expect from drawing a random sample instead of measuring everyone. It assumes the sample was drawn at random from the population you care about, and that everybody selected responded.
Real surveys violate both assumptions constantly. Non-response bias, where the people who answer differ systematically from those who do not, is not in the margin. Neither is coverage bias from an incomplete sampling frame, nor question wording effects, nor the tendency of respondents to give the answer they think is expected. These are collectively larger than sampling error in most professional surveys, and none of them shrink when you increase the sample size. A poll of 10,000 people recruited from a single website has a tiny margin of error and potentially enormous bias. Precision and accuracy are not the same thing — a distinction covered in Penn State's STAT 500 applied statistics course notes alongside the interval formulas themselves.
Subgroups: Where Margins Quietly Explode
A survey of 1,000 people has a margin of about ±3.1 points overall. Break it into six regional subgroups of roughly 167 each and every one of those subgroups carries a margin of about ±7.6 points. Split by age within region and you may be reporting on cells of 40 people with margins beyond ±15.
This is the single most common misuse of survey data in business reporting. The headline number is quoted with its margin, then the analysis moves to a comparison between two subgroups — where the relevant margin is far larger, and larger still for the difference between two of them, since both carry their own error. A gap of 8 points between two subgroups of 167 is not evidence of anything. Before designing the survey, decide the smallest group you intend to report on and size the sample for that group, using the sample size calculator to work backwards from the precision you need.
The Square Root Rule and What Precision Costs
Sample size enters the formula under a square root, so precision improves slowly and expensively. Quadrupling the sample halves the margin. Going from 1,000 to 2,000 responses improves the margin only from ±3.1 to ±2.2 points, and from 2,000 to 4,000 gets you to ±1.55.
This is why almost every national poll lands somewhere between 800 and 1,200 respondents. That range is the point where the margin is acceptable and the cost per additional point of precision starts rising sharply. It is also why raising the confidence level is not a free improvement: moving from 95% to 99% confidence multiplies the margin by 2.576 ÷ 1.960, about 31% wider, from exactly the same data. You have not become more certain, only vaguer. The confidence interval calculator shows the same trade-off from the interval side.
When the Finite Population Correction Applies
The standard formula assumes an effectively infinite population. When you survey a large share of a small closed group — 300 of a company's 400 employees, or 80 of 120 customers — that assumption overstates your uncertainty, because you have already measured most of the population.
The correction multiplies the margin by √((N − n) ÷ (N − 1)). Sampling 300 from 400 gives a factor of about 0.50, halving the margin. Sampling 1,000 from 250,000 gives 0.998, which is no change worth mentioning. As a rule the correction only matters once your sample exceeds roughly 5% of the population, which is why it appears in internal employee and customer surveys and essentially never in national polling.
Arb Digital publishes a growing library of free calculators for statistics, maths, and analysis — no sign-up, no limits, and every formula documented on the page.
Browse All Free Tools Contact Arb DigitalCommon Mistakes to Avoid
- Applying the overall margin to subgroups — a subgroup of 150 carries roughly three times the margin of a full sample of 1,000.
- Treating the margin as total survey error — non-response and question wording are usually larger, and no sample size fixes them.
- Comparing two results without allowing for both margins — the margin on a difference is wider than the margin on either figure alone.
- Using the number invited rather than the number who responded — only completed responses count towards n.
- Raising the confidence level to look more rigorous — 99% confidence widens the margin by about a third without adding any information.
Related Free Tools From Arb Digital
Plan the study with the sample size calculator, then express the finished result as a range using the confidence interval calculator. Check the spread of numeric answers with the standard deviation calculator, compare two groups with the t-test calculator or, for categorical splits, the chi-square calculator. Everything else lives in the free online tools hub.
Frequently Asked Questions
It means that if the same survey were repeated many times with fresh random samples, about 95% of the resulting intervals would contain the true population value. A reported 52% with a ±3 point margin is consistent with anything from 49% to 55%.
At 95% confidence with a conservative 50% proportion, about 1,068 responses. For 5% you need roughly 385, and for 1% around 9,604. The relationship is quadratic, so each improvement costs disproportionately more.
Barely, unless your sample is a large share of it. Sampling 1,000 people gives almost the same margin whether the population is 100,000 or 100 million. The finite population correction only bites once the sample exceeds roughly 5% of the total.
Because the quantity p times 1 minus p is largest at 50%, producing the widest margin. Planning with it guarantees the sample is big enough regardless of how the answers actually fall.
No. Increasing the sample reduces random sampling error only. If the people responding differ systematically from the population, a larger sample simply measures the wrong group more precisely.
No. The margin is the half-width added and subtracted from your result, while the interval is the full range between the lower and upper limits. The interval is twice the margin wide.
This tool is provided for education and general analysis. It quantifies random sampling error only, and is not a substitute for professional advice on survey design or research methodology.
