For a market survey that estimates a percentage, such as the share of people who would buy a product, you need four inputs: a confidence level, a margin of error, an expected proportion, and the size of the population. With the usual settings of 95% confidence and a margin of error of plus or minus 5 percentage points, a large population needs about 385 completed responses. This guide shows the calculation, the adjustments that matter, and how the number changes when you report on subgroups or compare groups.
Estimating a percentage vs. comparing groups
There are two common reasons to size a survey, and they call for different calculations:
- Estimating a percentage asks “what share of people feel or do this?” You want the result to fall within a known margin, such as 42% plus or minus 5 points. The calculation in this guide applies.
- Comparing groups asks “is group A different from group B?” You need enough people in each group to detect a difference of a certain size. That calls for a power analysis, covered later in this guide.
Most market surveys do both, so decide which question matters most before you calculate.
The four inputs
- Confidence level. How sure you want to be that the true value falls within your margin. 95% is the most common choice. A 90% level needs fewer people; a 99% level needs more.
- Margin of error. How far the result may sit from the true value, in percentage points. Plus or minus 5 points is a common target. A smaller margin needs many more people.
- Expected proportion (p). Your best guess of the answer. If you have no estimate, use 50%. That gives the largest sample size, so it is the safe choice.
- Population size (N). The number of people in the group you are studying. For large populations, the size barely matters. For small populations, it changes the answer.
The formula and a worked example
For a large population, the sample size is:
n = z² × p × (1 − p) / e²
- z is the z-score for your confidence level. For 95%, z = 1.96.
- p is the expected proportion. Use 0.5 if you do not know it.
- e is the margin of error as a decimal. For plus or minus 5 points, e = 0.05.
With the usual settings: n = 1.96² × 0.5 × 0.5 / 0.05² = 384.16. Round up to 385 completed responses.
Adjust for a small, finite population
When the population is small, the sample needs less adjustment. The finite-population correction gives the adjusted size:
n_adjusted = n / (1 + (n − 1) / N)
Take a population of 5,000 people in your target market. Starting from n = 384.16, the adjusted size is about 357. The correction matters most when your sample is a sizable share of the population. For very large populations, use the uncorrected number.
What the margin of error looks like at common sample sizes
The table shows the margin of error for a percentage near 50% at 95% confidence, for a large population. The margin is widest at 50%, so these are the most conservative figures.
| Completed responses | Margin of error (± points) |
|---|---|
| 100 | 9.8 |
| 200 | 6.9 |
| 385 | 5.0 |
| 400 | 4.9 |
| 1,000 | 3.1 |
These margins assume random sampling and no non-response bias. A large sample from the wrong people, or from a self-selected group, can still give a misleading result. The margin only describes random sampling error.
Subgroups need their own sample
A total of 385 responses does not guarantee 385 people in each subgroup. If you want to report results for two age groups, each group needs its own target. A group of 100 people has a margin of about ± 9.8 points at 50%. Decide which subgroups you will report before you field the survey, and set a minimum count for each. If a group falls short, report its results as indicative or leave them out.
Comparing two groups: power analysis
If your question is whether two groups differ, the margin of error for a single percentage is the wrong tool. Use a power analysis instead. It asks how many people you need in each group to detect a difference of a given size, with a chosen significance level (usually 0.05) and power (usually 0.80).
As an example, detecting a medium difference between two group averages (Cohen’s d = 0.5) needs about 64 people per group at alpha 0.05 and power 0.80. The required number rises as the difference you care about gets smaller. Pick the smallest difference that would change your decision, not the one you hope to find.
Turn completes into invitations
The number you calculate is the number of completed surveys you need, not the number of invitations. Work backward:
Invitations = target completes / (response rate × qualify rate)
Suppose you need 385 completes, your response rate is 20%, and 60% of respondents qualify through the screener. Then you need about 385 / (0.20 × 0.60), or roughly 3,200 invitations. Response rates vary widely, so use your own history or a pilot to estimate them. The pilot testing guide explains how to measure completion and drop-off. The screener guide explains how qualify rates and quotas work.
If you use a consumer panel, ask the provider about its completion and screening rates, and about the quality checks it runs. The online survey tools guide covers what to ask.
When the number is not feasible
If the target is more than you can recruit, you have a few options:
- Accept a wider margin. A margin of ± 7 points with 200 people may be enough for a go or no-go decision.
- Narrow the question. Asking about one product or one segment needs fewer people than asking about the whole market.
- Focus on the group that matters. Report one key subgroup well, instead of several poorly.
- Add depth with interviews. A small number of in-depth interviews can explain why people answer as they do, though they cannot give a reliable percentage.
Whatever you choose, state the margin you achieved, the confidence level, and how you recruited people.
Common mistakes
- Reporting a sample size without the margin. “We surveyed 400 people” does not say how precise the result is.
- Ignoring subgroups. The overall margin does not apply to a small group within the sample.
- Forgetting invitations. Calculating completes and sending that many invitations leaves you short.
- Treating a non-random sample as precise. The margin formula assumes random sampling. Self-selected or panel samples need more caution in how you report them.
- Using a single expected proportion without thought. If you have a reliable estimate, use it, but test whether the answer is still reasonable if the true value differs.
For the broader design of a research study, including the questions and the technique, start with the market research survey questions guide.