NPS margin of error calculator
Enter promoters, passives and detractors, or the total and percentages. Get the margin of error of NPS, its confidence interval, and whether two scores really differ.
- Checked against reference values
- Free, no sign-up
- Nothing leaves your browser
Your NPS answers
Wald is the textbook formula. Adjusted Wald (Rocks, 2016) is more accurate with small samples.
Assumptions and limitations
- The answers are a random sample of the customers you want to describe. The margin covers sampling error only, not a biased invitation list or low response.
- Standard groups: 9–10 promoters, 7–8 passives, 0–6 detractors.
- Two waves or segments are compared as independent samples. The same people answering twice need a paired test.
Collect real answers in a SurveyNinja survey and see them in charts in the report.
Use this template See the questions firstData is not stored — calculation runs in your browser.
Ready-made template
Do it with your own respondents
The “NPS survey” template already has the questions. Put in your company name, share the link and watch the answers come in.
- Copy the template: the questions are written
- Share it by link, QR code or a widget on your site
- The report counts NPS and the share of each group: enter those counts here for the margin
New accounts get 14 days of Premium features, no credit card.
What is the margin of error of NPS?
The margin of error of NPS is how far the score from your sample can be from the score of all your customers. Count each promoter as +1, each passive as 0 and each detractor as −1: NPS is their mean × 100, and its standard error gives an interval such as +45 ± 14.5 points at 95% confidence.
How the margin of error of NPS is calculated
-
Variance of NPS
With p the share of promoters and d the share of detractors, the variance of one answer is p + d − (p − d)². The standard error is √(variance / n), and the margin is z × SE × 100 NPS points, with z = 1.645, 1.960 or 2.576 for 90, 95 or 99% confidence.
-
Adjusted Wald for small samples
Rocks (2016) found that adding a few pseudo-answers gives intervals closer to the stated confidence on small samples. In the version MeasuringU uses, add 3 to n, 0.75 to promoters and 0.75 to detractors, then compute the score and the variance from the adjusted counts. The interval is centred on the adjusted score.
-
Comparing two scores
The difference of two independent NPS has the standard error √(var₁ / n₁ + var₂ / n₂). Divide the difference by it to get z, and the two-sided p-value comes from the normal distribution. If the interval of the difference includes 0, the change is not significant at that level.
-
Why NPS needs more answers than a percentage
NPS subtracts two shares, so their errors add up. With 100 answers, a single percentage has a margin of about ±10 points at 95%, while NPS has ±14 to ±16. To halve the margin, collect four times as many answers.
Examples
-
One score
- 100 answers: 60 promoters, 25 passives, 15 detractors. NPS = +45.
- Variance = 0.60 + 0.15 − 0.45² = 0.5475. SE = √(0.5475 / 100) = 0.0740.
Margin at 95% = 1.96 × 0.0740 × 100 = ±14.5 points, so the interval runs from +30.5 to +59.5. With the adjusted Wald method it is +29.3 to +58.0.
-
Two waves
- Wave A: 50 promoters, 30 passives, 20 detractors (NPS +30). Wave B: 55, 28, 17 (NPS +38).
- SE of the difference = √(0.61 / 100 + 0.5756 / 100) = 0.1089.
The difference is +8 points ± 21.3 at 95%, z = 0.73 and p = 0.46. The score went up, but these samples cannot tell the change from noise.
-
Two products, small samples
- Product 1: 31 answers, 7 promoters, 12 passives, 12 detractors. Product 2: 36 answers, 15, 13 and 8.
- Adjusted Wald, as in MeasuringU's worked example.
The difference is 32.7 points, z = 1.81 and p = 0.07: significant at 90% but not at 95%.
Margin of error of NPS by number of answers
NPS 20 means 45% promoters and 25% detractors, NPS 45 means 60% and 15%. 95% confidence, Wald interval, in NPS points.
| Answers | NPS 20 | NPS 45 |
|---|---|---|
| 30 | ±29.1 | ±26.5 |
| 50 | ±22.5 | ±20.5 |
| 100 | ±15.9 | ±14.5 |
| 200 | ±11.3 | ±10.3 |
| 500 | ±7.1 | ±6.5 |
| 1,000 | ±5.0 | ±4.6 |
Answers needed for a margin of error
n = 1.96² × variance / margin², with the same shares as the table above. To compare two segments, each needs its own sample of this size.
| Margin at 95% | NPS 20 | NPS 45 |
|---|---|---|
| ±15 points | 113 | 94 |
| ±10 points | 254 | 211 |
| ±7 points | 518 | 430 |
| ±5 points | 1,015 | 842 |
| ±3 points | 2,818 | 2,337 |
What to avoid
-
Reading a few points as a trend
A move from +30 to +38 with 100 answers per wave is well inside the margin. Check the difference in the second mode before reporting a change.
-
Using the margin of a percentage
The usual ±1.96 × √(p(1 − p) / n) is for one share. NPS combines two shares and needs its own variance, which gives a wider interval.
-
Comparing overlapping intervals by eye
Two intervals can overlap while the difference is still significant. Test the difference itself, with its own standard error.
-
Forgetting the size of each segment
A company NPS from 1,000 answers can be precise while each of five regions has only 200 and a margin of ±11 points.
Next step
Read the answers in a report, not a spreadsheet
Charts are built for you from every answer. Share the report by link or export the answers to Excel, CSV or SPSS.
- Ready questions or AI from a description
- A link for your email or chat, a QR code or a widget on your site
- Charts in the report, export to Excel, CSV and SPSS
FAQ about the margin of error of NPS
How do you calculate the margin of error of NPS?
Take the shares of promoters p and detractors d. The variance is p + d − (p − d)², the standard error is √(variance / n), and the margin is z × SE × 100 points. With 60% promoters, 15% detractors and 100 answers, the margin at 95% is ±14.5 points.
What is the adjusted Wald interval for NPS?
A correction for small samples studied by Brendan Rocks (The American Statistician, 2016). In the form MeasuringU uses, you add 3 answers, 0.75 of them to promoters and 0.75 to detractors, and compute the interval around the adjusted score. It matters most below about 100 answers; with large samples both methods agree.
How do I know if a change in NPS is significant?
Switch to “Compare two NPS” and enter both waves. The calculator divides the difference by its standard error, √(var₁ / n₁ + var₂ / n₂), and gives z and the p-value. A p-value below 0.05 means the difference is significant at 95%.
How many responses do I need for NPS?
For a margin of ±10 points at 95%, plan about 200 to 250 answers, and about 850 to 1,000 for ±5 points. The exact number depends on the shares of promoters and detractors, as the table above shows.
Can I enter percentages instead of counts?
Yes. Choose “Total and %” and enter the number of answers with the percentages of promoters and detractors. Passives are the rest. The total matters: the same percentages from 50 or 500 answers give very different margins.
Can the interval go beyond 100?
The Wald formula can, when almost everyone is a promoter or a detractor and the sample is small. The calculator cuts the interval at −100 and +100. In such cases the adjusted Wald method gives a more realistic interval.
Does SurveyNinja show the margin of error of NPS?
The report of the NPS question counts NPS from −100 to 100 and the share of each group on every plan, including Free. It does not show the margin of error: enter the counts from the report here. Responses also export to CSV, XLSX or SPSS.
Learn the method
Read how the method works and where its numbers come from.
Turn the numbers into a survey
Start from the “NPS survey” template: the questions are ready.
Google Forms Limits: Responses, File Uploads, Timers and Logic, With Workarounds