Sample Size Calculator for a Finite Population
This sample size calculator (finite population correction included) shows how many survey responses you need for a chosen confidence level and margin of error.
Required sample size
- z-score
- 1.96
- Before correction (n₀)
- 384.15
- Population size
- Unlimited
How to use the Sample Size Calculator
- Choose Sample size to find how many responses you need, or Margin of error to check a sample you already have.
- Pick a confidence level. 95% is the standard for most surveys and polls.
- Enter the margin of error you can accept (for example 5 for ±5 percentage points) or your sample size.
- Enter the expected proportion. If you have no earlier data, keep 50%, which gives the largest and safest sample.
- Add the population size if the group is small and known, such as 800 employees. Leave it blank for a large or unknown population.
Formula
n₀ = z² × p(1 − p) ÷ e² n = n₀ ÷ (1 + (n₀ − 1) ÷ N)
Where:
- n₀
- Sample size for a very large (unlimited) population
- n
- Sample size after the finite population correction, rounded up to a whole response
- z
- Critical value of the standard normal distribution for the confidence level: 1.645 for 90%, 1.96 for 95%, 2.576 for 99%
- p
- Expected proportion as a decimal, e.g. 0.5 for 50%. p(1 − p) is largest at 0.5
- e
- Margin of error as a decimal, e.g. 0.05 for ±5 percentage points
- N
- Population size (number of people in the group you are studying)
Assumes a simple random sample and a yes/no (proportion) question. Margin of error mode uses e = z × √(p(1 − p) ÷ n) × √((N − n) ÷ (N − 1)), dropping the last factor when N is unknown. Source: W. G. Cochran, Sampling Techniques (3rd ed., 1977).
Worked example
A coffee chain wants to survey its 5,000 loyalty-card members at 95% confidence with a ±5% margin of error, and will give each respondent a $3 gift card. How many responses does it need, and what will the incentives cost?
- With no earlier data, use p = 0.5. For 95% confidence, z = 1.96 (more precisely 1.959964).
- n₀ = 1.959964² × 0.5 × 0.5 ÷ 0.05² = 3.841459 × 0.25 ÷ 0.0025 = 384.15.
- Correct for the population of 5,000: n = 384.15 ÷ (1 + 383.15 ÷ 5,000) = 384.15 ÷ 1.076629 = 356.80.
- Round up to a whole person: 357 responses.
- Incentive budget: 357 × $3 = $1,071.
The chain needs 357 completed responses (about 7.1% of members), and the gift cards cost $1,071. Without the population correction it would need 385.
Sample size formula for a finite population
The calculation has two stages. Cochran's formula first gives n₀, the number of completed responses needed from a very large group: n₀ = z² × p(1 − p) ÷ e². The finite population correction then shrinks it for a group of known size N:
n = n₀ ÷ (1 + (n₀ − 1) ÷ N)
Enter the population size and the result panel shows both stages: the z-score, n₀ before correction, the corrected figure and the share of the population you would survey. In that sense the tool is a Cochran formula calculator with the correction built in.
Textbooks in some regions teach Yamane's formula (often called Slovin's formula), n = N ÷ (1 + N × e²). It is the same idea with z rounded to 2 and the proportion fixed at 50%. For 1,000 people and a ±5% margin it gives 286, where Cochran's method at 95% confidence gives 278. There is no Yamane or Slovin mode here.
When the finite population correction matters
The correction matters when the sample would be a large share of the group. At 95% confidence, ±5% and a 50% proportion:
| Population | Responses needed | Share surveyed |
|---|---|---|
| 100 | 80 | 80.0% |
| 500 | 218 | 43.6% |
| 1,000 | 278 | 27.8% |
| 5,000 | 357 | 7.1% |
| 10,000 | 370 | 3.7% |
| 100,000 | 383 | 0.4% |
| Unlimited | 385 | not applicable |
Below a few thousand people the saving is large. Above 100,000 it all but disappears, which is why a national poll and a citywide poll need nearly the same sample.
How many survey responses do you need? A worked example
An HR team wants to know how many of its 2,400 employees would use a new commuter benefit. A pilot at one office suggested about 20% would. The team wants 95% confidence and a ±4% margin of error.
- For 95% confidence, z = 1.959964, so z² = 3.841459.
- With p = 0.20, p(1 − p) = 0.20 × 0.80 = 0.16.
- n₀ = 3.841459 × 0.16 ÷ 0.04² = 0.614633 ÷ 0.0016 = 384.15.
- Correct for the population: n = 384.15 ÷ (1 + 383.15 ÷ 2,400) = 384.15 ÷ 1.159644 = 331.26.
- Round up: 332 completed responses.
Had the team kept the cautious 50% instead of 20%, it would have needed 481 responses after the correction. An expected proportion well away from 50% lowers the sample, but only trust it when it comes from reasonably reliable earlier data.
Response rate: how many people to invite
The result is a count of completed responses, not invitations, and there is no response rate field. Divide by the rate you realistically expect. Internal surveys at this company usually get a 40% response rate, so the team should invite at least 332 ÷ 0.40 = 830 employees. Add a cushion for answers you will discard as incomplete.
Sample size without a known population
Leave the population box blank for a large or unknown group, and the answer is n₀ rounded up. Two inputs then do most of the work. Raising the confidence level from 95% to 99% swaps z = 1.96 for 2.576 and enlarges the sample by about 73%. Tightening the margin from ±5% to ±4% enlarges it by about 56%, and ±3% nearly triples it, because the margin is squared in the denominator.
Margin of error for the sample you have
Responses rarely arrive exactly as planned. Switch to Margin of error mode, enter the number you actually received and read the precision you achieved. At 95% confidence and a 50% proportion, 400 replies from a large population give ±4.90%. The margin shrinks with the square root of the sample, so 1,600 replies are needed to halve it to ±2.45%.
Size the survey for the smallest group you will report on. A total of 385 split across four regions leaves about 96 per region, a margin of roughly ±10% each, not ±5%.
Assumptions and limits
- Proportions only. The question must have a yes/no or percentage answer. There is no input for a mean or a standard deviation, and no power, effect size, A/B test or clinical trial design.
- Fixed confidence levels. Choose from 80%, 85%, 90%, 95%, 98%, 99%, 99.5% and 99.9%.
- Simple random sampling. Ten thousand replies to a self-selected website poll can be less accurate than 400 well-chosen ones. Stratified, clustered or weighted designs need a design effect, which is not applied here.
- Percent, not decimal. Type 5 for a ±5% margin. Entering 0.05 asks for ±0.05%, which needs millions of responses.
- Always round up. Stopping at 384 instead of 385 no longer guarantees the target margin.
For studies that feed a medical, legal or major financial decision, have the design checked by a statistician.
Frequently asked questions
How many responses do I need for a survey to be valid?
It depends on the confidence level and margin of error you choose. Use Cochran's formula n₀ = z² × p(1 − p) ÷ e². For 95% confidence (z = 1.96), p = 0.5 and a ±5% margin, n₀ = 3.8415 × 0.25 ÷ 0.0025 = 384.15, which rounds up to 385. If your population is small and known, divide by 1 + (n₀ − 1) ÷ N to get a smaller number.
What is a good sample size for a survey?
385 completed responses is the usual benchmark. It is the answer for the most common settings: 95% confidence, a ±5% margin of error, p = 50% and a large population. It works for a city, a country or the whole world, because once a population is large, its size barely changes the sample you need. A small, known group needs fewer responses, and a tighter margin needs more.
What sample size do I need for 95% confidence?
That depends on the margin of error you can accept. At 95% confidence with p = 50% and a large population, ±5% needs 385 responses, ±3% needs 1,068, ±2.5% needs 1,537 and ±1% needs 9,604. Because the margin of error is squared in the formula, halving it roughly quadruples the sample.
How do I determine sample size for an unknown population?
Leave the population size blank and use n₀ = z² × p(1 − p) ÷ e² with no correction. An unknown or very large population needs barely more than a moderately large one: at 95% confidence and ±5%, a population of 1,000 needs 278 responses, 10,000 needs 370, and an unlimited population needs 385. The finite population correction only makes a real difference when your sample would be a sizeable share of the whole group.
Why use 50% for the population proportion?
The term p(1 − p) is largest when p = 0.5, so 50% gives the biggest sample size. If you do not know what answer to expect, 50% is the safe choice: your margin of error will be no wider than planned whatever the true proportion turns out to be.
Does this work for averages, such as average spend?
No. This calculator is for proportions (yes/no or percentage answers), which is what most surveys report, and it has no field for a mean or a standard deviation. Estimating a mean uses n = (z × σ ÷ E)², where σ is the standard deviation and E the margin in the same units, so you need an estimate of the spread first.
Related calculators
- Confidence Interval CalculatorThis confidence interval calculator for mean values (t or z) and proportions turns summary statistics into lower and upper limits with the margin of error.
- Standard Deviation CalculatorThis standard deviation calculator (sample vs population) turns a pasted list of numbers into s or σ, with the variance and mean alongside.
- Z-Score CalculatorZ-score calculator: percentile, tail probabilities and p-value for any raw score or z-score, or the z-score behind any percentile.
- Probability CalculatorThis probability calculator with multiple attempts finds the chance of at least one success in n tries, or combines two independent events.
- Percentage CalculatorA percentage calculator: of a number, as a share of a total, or back to the whole from a part, with the formula shown.
Last updated: