Last updated: 7 October 2026
Quick Answer: Sample size is the number of people or observations a study includes, written as n and set before fieldwork with a formula or online tool. For a survey estimating a percentage, about 385 completed responses give a margin of error of plus or minus 5 points at 95% confidence. Group comparisons, subgroups and qualitative work follow different rules.
Sample size is the number of completed responses or observations your conclusions rest on. It decides how much chance can move a result, which is why the first question in most research briefs is "how many do we need?"
There is no single right answer. The NIST engineering statistics handbook puts it bluntly: there is no correct sample size without assumptions about the precision you need, the risks you accept and how variable the answers are.
How Do You Calculate Sample Size?
To calculate sample size for a percentage, use the formula n = z² × p × (1 − p) ÷ e². Here z is 1.96 for 95% confidence, p is the expected proportion and e is the margin of error. With p set to 0.5, the safest assumption, and a 5-point margin, the sample size formula gives 384.2, rounded up to 385.
Small populations need fewer. Apply the finite population correction, n ÷ (1 + (n − 1) ÷ N), and a population of 2,000 needs about 323 completes for the same precision, while 10,000 needs about 370.
What Sample Size Gives Which Margin of Error?
Precision improves with the square root of the sample, so halving the margin of error takes four times the completes. The table shows the 95% margin of error on a 50% result for a large population, from smallest sample to largest.
| Completed responses | Margin of error (±) |
|---|---|
| 100 | 9.8 points |
| 200 | 6.9 points |
| 385 | 5.0 points |
| 1,000 | 3.1 points |
| 1,067 | 3.0 points |
| 2,401 | 2.0 points |
These figures assume a random sample. They describe chance alone, not who was reached.
How Do You Determine Sample Size for Comparing Groups?
To determine sample size for a comparison, decide the smallest difference worth detecting, the confidence level and the power, the chance of spotting a real difference. Detecting a 10-point gap between 50% and 60% at 95% confidence and 80% power takes about 385 completes in each group, not 385 in total.
Four inputs set the number:
- The smallest difference worth detecting.
- The confidence level, usually 95%.
- The power, usually 80%.
- The expected baseline proportion.
That is why A/B tests and concept tests with several cells grow quickly. Each extra cell adds another full group.
What Is a Good Sample Size for a Survey?
A good sample size for a survey is one that gives enough precision for the smallest group you need to read on its own. A total of 400 looks solid at ±4.9 points, but if four regions matter, each gets about 100 and a margin near ±10 points.
Working norms vary by study. Alchemic's concept testing typically uses 200 or more respondents, within a range of 100 to 400, while early usability work can run 8 to 15 people per persona, because it hunts for problems rather than measuring percentages. A minimum sample size is therefore set by the decision, not by a universal number.
What Sample Size Does Qualitative Research Need?
Qualitative research needs enough interviews to stop hearing new themes, a point called saturation. A 2022 systematic review by Monique Hennink and Bonnie Kaiser found that studies reached saturation within 9 to 17 interviews, or 4 to 8 focus groups, especially with fairly homogeneous groups and focused aims.
More varied audiences, several segments or several markets push the number up, because each segment needs its own run toward saturation.
Why Is Sample Size Important, and When Does It Not Help?
Sample size matters because too few responses leave results at the mercy of chance, while too many waste money and respondents' time. Psychiatrist Chittaranjan Andrade argues in a 2020 article on sample size that a sample too large is unnecessary and unethical, and one too small is unscientific and also unethical.
What size cannot fix is bias. Every margin of error above assumes the sample was drawn at random from the right list, the basis of probability sampling. A large opt-in sample can be precise and still wrong, so the source of the sample matters as much as its size.
Once it has the client's brief, Alchemic designs and tailors the sampling plan to handle these risks before fielding, rather than leaving that work to the buyer.
Who Do You Lose Between Invitations and Completes?
You lose everyone who is invited but never responds, and they are rarely a random slice. Pew Research Center reported that its telephone response rates fell to 6% in 2018, so hitting a target sample size can mean inviting many times more people, with the risk that those who answer differ from those who do not.
Raising response across groups protects the sample as well as the count. WhatsApp-native interviews and AI phone interviews reach people in channels they already use. Alchemic publishes 60+ languages including Spanish and Hindi, with managed fieldwork or bring your own sample across 14 markets including the USA and the UK.