Market Research

How many respondents do you need? Sample size and margin of error explained

385 completed interviews give ±5% at 95% confidence. Learn when that number changes: small populations, regional results, weighting, panels and incidence, with Belgian examples.

Short answer: for a large population, 385 completed interviews give a margin of error of ±5% at a 95% confidence level, using the most cautious assumption of a 50% result. You need 1,068 for ±3% and 97 for ±10%.

That number is a starting point, not a research design. It changes as soon as you need results by region, work with a small B2B customer base, weight your data or recruit a hard-to-reach audience. You can test every scenario below with our sample size and margin of error calculator.

What is the margin of error?

The margin of error tells you how far a survey result may sit from the true value in the population, because you interviewed a sample rather than everyone. If 40% of respondents say they would switch supplier and the margin is ±5% at 95% confidence, the true value is likely between 35% and 45%.

Three things drive it:

  • Sample size: more interviews, smaller margin, but with diminishing returns. Halving the margin requires four times as many interviews.
  • Confidence level: 95% is the usual standard. Moving to 99% requires about 70% more interviews for the same margin.
  • The result itself: a 50% result has the widest margin. If you expect 20% or 80%, fewer interviews give the same precision. When you do not know, keep 50%.

Quick reference at 95% confidence

Margin of error Completed interviews
±10% 97
±7.5% 171
±5% 385
±4% 601
±3% 1,068
±2% 2,401

These figures assume a 50% proportion, a large population and no weighting. As an indicative guide, ±3% or better suits high-precision tracking, ±3% to ±5% suits most business decisions, ±5% to ±8% gives a directional read, and wider margins are exploratory.

Does the size of the population matter?

Much less than most people think. Above a few tens of thousands, the population barely changes the requirement: a survey of all Belgian adults and a survey of a large city need roughly the same sample for the same precision.

It matters when the population is small, which is common in B2B. The finite population correction then reduces the requirement:

  • a customer base of 2,000 companies needs 323 completed interviews for ±5% at 95%;
  • a base of 500 companies needs 218.

In a small population, every interview counts more, but response rates become the real constraint. Plan your invitations accordingly.

Results by region: the Belgian case

Many Belgian studies need results for Brussels, Flanders and Wallonia, or by language. This is where samples are most often under-sized.

If each region needs ±5% at 95%, you need about 385 completed interviews per region, 1,155 in total. Splitting a single national sample of 385 into three equal regions gives each region a margin of about ±8.7%: fine for a national figure, too wide to compare regions with confidence.

The same logic applies to customer segments, age groups or product lines. Decide first which groups you need to read on their own, then size each of them.

Weighting and the design effect

Weighting a sample to match the population, strict quotas or clustered fieldwork reduce the effective sample size. The design effect expresses this loss. With a design effect of 1.5, ±5% at 95% requires 577 completed interviews instead of 385.

If you plan to weight your data, budget for it from the start.

From completed interviews to invitations

The calculation gives completed interviews. The fieldwork plan also needs:

  • the response rate: 400 completes at a 20% response rate means about 2,000 invitations;
  • the incidence, the share of people who qualify: with a 25% incidence on top, you need to screen about 8,000 people.

Response rate and incidence drive feasibility, timing and cost. They do not change the number of completed interviews you need.

What the margin of error does not measure

The margin of error is a theoretical figure for a random probability sample. It does not cover coverage gaps, non-response bias, leading questions or poor respondent quality. Online access panels and quota samples rarely meet the probability assumption, so their margin of error is an indicative precision measure.

A larger sample never fixes a biased recruitment or a weak questionnaire. And if you need to compare two groups, estimate averages or run an experiment, you need a different calculation, based on the size of the difference you want to detect.

A practical checklist

  1. Write down the decision the study must support.
  2. List the groups you need to read on their own.
  3. Choose the precision each group needs.
  4. Adjust for population size, weighting, response rate and incidence.
  5. Check feasibility and budget, then simplify if needed.

Try your own scenarios in the calculator, or brief your study and we will turn the number into a fieldwork-ready design: audience, sourcing, quotas, languages, questionnaire and analysis.

Let’s turn your data into your next opportunity.