👥 Sample Size Calculator

Why Your Survey Probably Has the Wrong Number of Respondents

Picture this: a product manager at a mid-size software company sends a satisfaction survey to 200 customers and confidently reports that "78% of users love the new dashboard." A week later, leadership makes a pricing decision based on that number. Nobody asks the obvious question — was 200 even the right number to survey in the first place?

This scenario plays out constantly, in businesses, academic studies, local government initiatives, and journalism. The sample size is either pulled from thin air ("let's aim for a hundred"), borrowed from a previous project with different parameters, or inflated out of anxiety into something far larger than necessary. Both mistakes are costly: too few respondents and your conclusions are statistically meaningless; too many and you've burned time and budget on data you didn't need.

Getting the sample size right isn't a matter of instinct — it's a matter of three concrete inputs and one formula.

The Three Numbers That Drive Everything

Confidence level is how certain you want to be that your survey results reflect the true population. A 95% confidence level — the standard in most research — means that if you ran the same survey 100 times, 95 of those results would fall within your stated margin of error. A 99% level is more demanding and requires a larger sample. A 90% level is acceptable when resources are tight and the stakes are lower.

Margin of error (also called the confidence interval) is the wiggle room you're willing to accept. If 78% of respondents say they're satisfied, and your margin of error is ±5%, the true population figure is somewhere between 73% and 83%. A ±3% margin is more precise but demands roughly three times as many respondents as ±5%. This inverse-square relationship catches many researchers off guard — halving your margin of error roughly quadruples your required sample size.

Response distribution describes how you expect the population to split on the key question. If you genuinely don't know, use 50% — this is the most conservative assumption, because a 50/50 split produces the maximum variance and therefore the largest required sample. If prior data suggests 80% of customers are satisfied, you can plug in 80% and the formula rewards you with a smaller required sample. Most researchers stick with 50% when in doubt, which is the statistically safe choice.

The Formula Under the Hood

The core calculation for an unlimited (or very large) population is straightforward:

n = Z² × p × (1 − p) ÷ E²

Where Z is the z-score corresponding to your confidence level (1.96 for 95%, 2.576 for 99%, 1.645 for 90%), p is the response proportion as a decimal, and E is the margin of error as a decimal. For the classic 95% confidence, ±5% margin, 50% proportion scenario, this gives exactly 385 respondents — a number that appears so frequently in research papers it has become almost folklore.

The formula assumes you're sampling from an infinitely large group, which is a safe approximation when your population numbers in the tens of thousands or more. But what happens when you're surveying the employees of a 400-person company, the residents of a small town, or the registered users of a niche application?

The Finite-Population Correction: Where Smaller Populations Get a Break

Here is the insight that many survey designers miss: when your population is small relative to the sample size the infinite formula demands, you do not need as many respondents. Intuitively, if you're surveying a population of 500 people and the formula says you need 385, you're already capturing 77% of the entire group — the statistical uncertainty is far lower than if you were sampling 385 from a million.

The finite-population correction (FPC) adjusts for this:

n_adjusted = n₀ ÷ (1 + (n₀ − 1) ÷ N)

Where n₀ is the sample size from the infinite formula and N is the actual population. Applied to the classic case with a population of 1,000, the required sample drops from 385 to 278 — a savings of 107 interviews. With a population of 500, it falls to 217. These are not trivial differences. For organizations conducting phone surveys at ₹200 per completed call, or fieldwork at even higher cost, the FPC can cut a project budget by 30%.

The correction matters most when your sample would otherwise represent more than 5–10% of the total population. When the population is large (say, above 10,000) and the raw sample is only a fraction of a percent, the FPC changes the answer by only a handful of respondents and can safely be ignored.

Practical Pitfalls to Avoid

Confusing sample size with response rate. The calculator tells you how many completed responses you need. If you expect only 30% of people you contact to actually fill out the survey, you need to reach out to roughly three times the calculated sample size. The required sample is a floor on completions, not on invitations sent.

Using one sample size for multiple subgroups. If you plan to analyze results separately for different age brackets, regions, or customer tiers, each subgroup needs its own sample size calculation. A total of 385 respondents sounds robust until you realize your youngest cohort has only 22 people in it — far too few for any meaningful conclusion about that segment.

Assuming 95% is always the right confidence level. It is the conventional choice, but it isn't sacred. A product team doing quick iterative research might comfortably work at 90%. A medical device manufacturer testing a safety feature should be at 99% or higher. Match the confidence level to the consequences of being wrong.

Ignoring response distribution when you have prior data. Defaulting to 50% is conservative and safe, but if you have historical data — say, last year's survey showed 72% satisfaction — plugging in 72% will give you a more efficient sample size without sacrificing accuracy.

A Quick Real-World Walkthrough

Suppose a city council wants to survey residents about a proposed park renovation. The city has 8,500 registered households. The council wants 95% confidence with a ±4% margin and has no prior data, so they'll use 50% distribution.

The infinite formula gives: (1.96² × 0.5 × 0.5) ÷ 0.04² = 600 respondents. But with the FPC applied to a population of 8,500, the adjusted figure drops to 554. That's 46 fewer household surveys — meaningful when the fieldwork involves door-to-door canvassing.

If the council could live with a ±5% margin instead, the infinite formula gives 385, and the FPC brings it to 370. Loosening the precision requirement by just one percentage point saved 184 interviews. The trade-off between precision and cost is always explicit when you run the numbers.

The Bottom Line

Sample size is not a guess or a budget decision dressed up in statistical clothing. It is the output of a precise calculation built on three deliberate choices: how confident you need to be, how precise your estimate must be, and how the population is likely to respond. When the population is small enough, the finite-population correction reduces your workload without compromising validity.

Running these numbers before fieldwork begins is one of the simplest, highest-leverage improvements any research process can make. It turns "we surveyed about 200 people" into "we surveyed exactly the number of people needed to draw conclusions with 95% confidence" — and that distinction matters every time someone acts on the data.