Margin of Error in Surveys: What It Means and How to Calculate It (2026)

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.

The one-paragraph version

If 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.

What margin of error actually measures

When 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.

According to Pew Research Center, 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 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.

Three inputs drive every calculation:

1. 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.
2. Sample size (n) — the number of completed responses. More responses, smaller margin.
3. The proportion (p) — the result itself. MOE is widest when a result is near 50/50 and narrows as it approaches 0% or 100%.

The formula, step by step

The standard margin of error for a proportion at 95% confidence is:

MOE = z × √( p × (1 − p) / n )

Where:

• z = 1.96 for 95% confidence (use 1.645 for 90%, 2.576 for 99%)
• p = the proportion as a decimal (use 0.5 when you do not know it — this is the most conservative, largest-MOE assumption)
• n = your sample size

Worked example

You survey n = 600 customers and 45% (p = 0.45) say they would recommend you.

• p(1 − p) = 0.45 × 0.55 = 0.2475
• 0.2475 / 600 = 0.0004125
• √0.0004125 = 0.0203
• 0.0203 × 1.96 = 0.0398 → ±4.0%

So 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.

Quick reference (worst case, p = 0.5, 95% confidence)

Notice 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.

The finite population correction (for small populations)

The 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:

FPC = √( (N − n) / (N − 1) ), where N is the total population.

If 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 for the full sample-size math.

What margin of error does NOT tell you

This is where most teams go wrong. MOE is a measure of precision, not accuracy. It is silent on:

• 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."
• Non-response bias — the people who ignore your survey may differ systematically from those who answer. See how to increase survey response rates.
• Question wording and order — a leading or poorly sequenced question corrupts the data before MOE ever applies. See question order bias and survey question wording.
• 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.

A 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.

How Koji helps: precision and depth, not a trade-off

Margin 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.

• 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."
• Quant rigor and qualitative why. A traditional survey gives you a precise number with no explanation. Koji's AI-moderated interviews ask the same structured questions — 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.
• 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.
• 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.

You 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.

Practical rules of thumb

• Default target: ±5% at 95% confidence (n ≈ 384) for decision-grade product surveys.
• For directional reads: ±10% (n ≈ 100) is fine to spot a signal, not to set a price.
• For comparing two groups: each group needs its own sample — compute MOE per segment, never off the combined total.
• Report it every time: "60% (±3%, 95% CI, n=1,000)" is honest; "60% of users" alone is not.
• Fix bias first: a representative sample of 200 beats a skewed sample of 5,000.

Related Resources

• Survey Sample Size: How Many Responses Do You Really Need?
• Statistical Significance in Survey Research: A Plain-English Guide
• How to Analyze Survey Data: A Step-by-Step Guide
• Structured Questions Guide: The 6 Question Types in Koji
• Question Order Bias: How Sequencing Skews Your Data
• How to Increase Survey Response Rates
• Error Propagation in Research Metrics - what happens to that margin when the number is calculated from two others.
• Every Input Was Accurate and the Difference Was Not - why the margin of error on a difference is larger than on either estimate.