{"site":{"name":"Koji","description":"AI-native customer research platform that helps teams conduct, analyze, and synthesize customer interviews at scale.","url":"https://www.koji.so","contentTypes":["blog","documentation"],"lastUpdated":"2026-06-14T14:26:26.834Z"},"content":[{"type":"documentation","id":"f7b3a5dc-7a98-4cd4-b7a7-e7ac858ae855","slug":"survey-margin-of-error-guide","title":"Margin of Error in Surveys: What It Means and How to Calculate It (2026)","url":"https://www.koji.so/docs/survey-margin-of-error-guide","summary":"Margin of error (MOE) is the ± range around a survey result caused by sampling. At 95% confidence, MOE = 1.96 × √(p(1−p)/n). At p=0.5, 384 responses ≈ ±5%, 1,000 ≈ ±3.1%, 100 ≈ ±9.8%. MOE shrinks with the square root of sample size, so halving it requires 4× the responses. The finite population correction lowers it for small populations. MOE measures only random sampling noise — not selection bias, non-response, or bad questions — and applies to the full sample, not sliced sub-groups.","content":"# Margin of Error in Surveys: What It Means and How to Calculate It (2026)\n\n**Answer-first (BLUF):** Margin of error (MOE) is the plus-or-minus range around a survey result that tells you how far the number could be from the truth simply because you asked a sample instead of everyone. For a typical survey at **95% confidence**, the formula is **MOE = 1.96 × √(p(1−p) / n)**. At the worst case (p = 0.5), **384 responses give roughly ±5%**, **1,000 responses give about ±3.1%**, and **100 responses give about ±9.8%**. MOE shrinks with the square root of sample size — so cutting it in half costs four times the responses. It says nothing about bias, bad questions, or a skewed sample; it only quantifies sampling noise.\n\n## The one-paragraph version\n\nIf a survey of 1,000 people reports that 60% prefer Option A with a ±3% margin of error at 95% confidence, the honest reading is: \"We are 95% confident the true figure is somewhere between 57% and 63%.\" The margin of error is the width of that cushion. It depends on three things — your sample size, your confidence level, and the result itself — and on almost nothing else once your population is reasonably large. The single most common mistake is treating a tight margin of error as proof the survey is *accurate*. It is not. MOE measures only one kind of error (random sampling), and a beautifully precise number drawn from the wrong people is still wrong.\n\n## What margin of error actually measures\n\nWhen you survey a sample instead of the entire population, your result is an estimate. Run the same survey again with a fresh random sample and you would get a slightly different number. Margin of error captures that wobble: it is the maximum expected gap between your sample's answer and the true population answer, at a stated confidence level.\n\nAccording to [Pew Research Center](https://www.pewresearch.org/), a national poll of around 1,500–2,000 adults typically carries a margin of error of about ±2.5 to ±3 percentage points — which is why political polls cluster around those sample sizes. As the [Encyclopedia of Survey Research Methods](https://methods.sagepub.com/) notes, the margin of error is \"a measure of the precision of a survey estimate\" and reflects sampling variability alone, not the many other ways a survey can go wrong.\n\nThree inputs drive every calculation:\n\n1. **Confidence level** — how often the true value falls inside your interval if you repeated the survey many times. **95% is the research standard** (19 times out of 20). The matching z-score is 1.96.\n2. **Sample size (n)** — the number of completed responses. More responses, smaller margin.\n3. **The proportion (p)** — the result itself. MOE is widest when a result is near 50/50 and narrows as it approaches 0% or 100%.\n\n## The formula, step by step\n\nThe standard margin of error for a proportion at 95% confidence is:\n\n**MOE = z × √( p × (1 − p) / n )**\n\nWhere:\n- **z** = 1.96 for 95% confidence (use 1.645 for 90%, 2.576 for 99%)\n- **p** = the proportion as a decimal (use 0.5 when you do not know it — this is the most conservative, largest-MOE assumption)\n- **n** = your sample size\n\n### Worked example\n\nYou survey **n = 600** customers and **45% (p = 0.45)** say they would recommend you.\n\n- p(1 − p) = 0.45 × 0.55 = 0.2475\n- 0.2475 / 600 = 0.0004125\n- √0.0004125 = 0.0203\n- 0.0203 × 1.96 = **0.0398 → ±4.0%**\n\nSo your honest finding is: **41% to 49% would recommend you**, at 95% confidence. If a rival's score is 47%, you cannot claim you are different — the intervals overlap.\n\n### Quick reference (worst case, p = 0.5, 95% confidence)\n\n| Sample size (n) | Margin of error |\n|---|---|\n| 100 | ±9.8% |\n| 250 | ±6.2% |\n| 384 | ±5.0% |\n| 500 | ±4.4% |\n| 1,000 | ±3.1% |\n| 2,000 | ±2.2% |\n| 5,000 | ±1.4% |\n\nNotice the diminishing returns: going from 1,000 to 2,000 responses only tightens the margin from ±3.1% to ±2.2%. Because MOE falls with the *square root* of n, **halving your margin of error requires quadrupling your sample**.\n\n## The finite population correction (for small populations)\n\nThe textbook formula assumes a very large (effectively infinite) population. When your population is small — common in B2B, where you might have 800 total customers — you can apply the **finite population correction (FPC)**, which legitimately *lowers* the required sample or tightens the margin:\n\n**FPC = √( (N − n) / (N − 1) )**, where N is the total population.\n\nIf you survey 200 of 800 customers, the correction is √((800−200)/799) ≈ 0.866, shrinking a ±6.9% margin to about ±6.0%. The practical takeaway: above a population of ~20,000 the correction is negligible (which is why national polls ignore it), but for small, finite audiences it meaningfully helps. See our [survey sample size guide](/docs/survey-sample-size-guide) for the full sample-size math.\n\n## What margin of error does NOT tell you\n\nThis is where most teams go wrong. MOE is a measure of *precision*, not *accuracy*. It is silent on:\n\n- **Coverage and selection bias** — if your sample over-represents power users, no sample size fixes it. As survey methodologists put it, a precise estimate from a biased sample is \"precisely wrong.\"\n- **Non-response bias** — the people who ignore your survey may differ systematically from those who answer. See [how to increase survey response rates](/docs/how-to-increase-survey-response-rates).\n- **Question wording and order** — a leading or poorly sequenced question corrupts the data before MOE ever applies. See [question order bias](/docs/question-order-bias-guide) and [survey question wording](/docs/survey-question-wording-guide).\n- **Sub-group slicing** — the headline MOE applies to the *full* sample. The moment you filter to \"enterprise users in EMEA,\" your effective n collapses and the real margin balloons. This is the most common analysis error in product research.\n\nA margin of error also only applies cleanly to **probability samples**. Most product and market research uses convenience or panel samples, where the ± figure is best read as a \"modeled\" or indicative margin rather than a strict statistical guarantee — a nuance covered in our [statistical significance guide](/docs/statistical-significance-survey-research).\n\n## How Koji helps: precision and depth, not a trade-off\n\nMargin of error exists because surveys force a trade-off — to shrink the cushion you need more responses, and more responses traditionally meant blander, shallower data. Koji collapses that trade-off.\n\n- **Real-time confidence as responses land.** Koji's reporting updates aggregate results live, so you can watch a result stabilize and stop collecting once the interval is tight enough — instead of guessing your sample size up front or over-collecting \"to be safe.\"\n- **Quant rigor *and* qualitative why.** A traditional survey gives you a precise number with no explanation. Koji's [AI-moderated interviews](/docs/ai-interviews-vs-surveys) ask the same [structured questions](/docs/structured-questions-guide) — including scale, single-choice, and multiple-choice types you can aggregate statistically — and then probe each answer in the respondent's own words. You get the percentage *and* the reasoning behind it.\n- **Better samples, not just bigger ones.** Because MOE is meaningless on a skewed sample, Koji emphasizes targeted recruiting and screening so your tight margin actually describes the right population. Teams using AI-assisted research report substantially faster time-to-insight, letting you reach decision-grade sample sizes in days rather than weeks.\n- **When 20 interviews beat 2,000 responses.** For discovery questions — understanding *what* to measure — a large survey with a tiny margin of error answers the wrong question precisely. Koji lets you run 15–30 depth interviews at survey-like speed when the goal is understanding, then switch to structured scale questions when the goal is quantification.\n\nYou do not need a statistics degree to use this well. Koji surfaces the confidence and sample context alongside every result, so the margin of error stops being a footnote you forget and becomes a guardrail you actually act on.\n\n## Practical rules of thumb\n\n- **Default target:** ±5% at 95% confidence (n ≈ 384) for decision-grade product surveys.\n- **For directional reads:** ±10% (n ≈ 100) is fine to spot a signal, not to set a price.\n- **For comparing two groups:** each group needs its own sample — compute MOE per segment, never off the combined total.\n- **Report it every time:** \"60% (±3%, 95% CI, n=1,000)\" is honest; \"60% of users\" alone is not.\n- **Fix bias first:** a representative sample of 200 beats a skewed sample of 5,000.\n\n## Related Resources\n\n- [Survey Sample Size: How Many Responses Do You Really Need?](/docs/survey-sample-size-guide)\n- [Statistical Significance in Survey Research: A Plain-English Guide](/docs/statistical-significance-survey-research)\n- [How to Analyze Survey Data: A Step-by-Step Guide](/docs/how-to-analyze-survey-data)\n- [Structured Questions Guide: The 6 Question Types in Koji](/docs/structured-questions-guide)\n- [Question Order Bias: How Sequencing Skews Your Data](/docs/question-order-bias-guide)\n- [How to Increase Survey Response Rates](/docs/how-to-increase-survey-response-rates)\n","category":"Research Methods","lastModified":"2026-06-14T03:15:50.946722+00:00","metaTitle":"Survey Margin of Error: Formula & How to Calculate (2026)","metaDescription":"Margin of error explained in plain English: the 95% confidence formula, a worked example, a sample-size reference table, and what MOE does (and does not) tell you.","keywords":["margin of error","survey margin of error","margin of error formula","how to calculate margin of error","confidence level survey","margin of error 95 percent","sampling error"],"aiSummary":"Margin of error (MOE) is the ± range around a survey result caused by sampling. At 95% confidence, MOE = 1.96 × √(p(1−p)/n). At p=0.5, 384 responses ≈ ±5%, 1,000 ≈ ±3.1%, 100 ≈ ±9.8%. MOE shrinks with the square root of sample size, so halving it requires 4× the responses. The finite population correction lowers it for small populations. MOE measures only random sampling noise — not selection bias, non-response, or bad questions — and applies to the full sample, not sliced sub-groups.","aiPrerequisites":["Basic familiarity with surveys","Comfort with simple arithmetic"],"aiLearningOutcomes":["Define margin of error and explain what it does and does not measure","Calculate MOE for any proportion at 95% confidence using the standard formula","Apply the finite population correction for small populations","Read a sample-size-to-MOE reference table and recognize diminishing returns","Avoid the sub-group slicing error and report results honestly"],"aiDifficulty":"beginner","aiEstimatedTime":"12 min read"}],"pagination":{"total":1,"returned":1,"offset":0}}