How to Choose a Sample Size: A Practical Guide for Beginners

"How many people do I need?" is perhaps the most common question in applied statistics, and the honest answer is: it depends on what you want to learn. A sample that is too small cannot detect real differences, and one that is too large wastes time and money. Fortunately the logic is simple, and a few formulas cover most beginner needs.

Step one: decide what you are estimating

Sample size depends on the goal. If you want to estimate a percentage, such as the share of customers who are satisfied, you need enough respondents to keep the margin of error small. If you want to compare two groups, you need enough people to detect the smallest difference that matters. Generally, sample size depends on the expected effect size, the variability of the data, the significance level and the power (Kadam and Bhalerao, 2010).

Estimating a percentage

For a survey, the classic formula is n = z² × p(1 − p) / E², where E is the margin of error and p is the expected proportion. When p is unknown, 50% is the cautious choice, since it gives the largest sample. At 95% confidence, a margin of ±5 points needs 385 respondents, ±3 points needs 1,068, ±2 points needs 2,401, and ±1 point needs 9,604. The sample size calculator for proportions does this instantly.

Notice the square in the denominator: halving the margin of error multiplies the required sample by four. That is why precise surveys get expensive quickly.

Small populations need fewer people

If you are surveying a small, finite group, such as 2,000 employees, the finite population correction shrinks the requirement: n = n₀ / (1 + (n₀ − 1) / N). For ±5 points, 385 becomes 322. For ±3 points, 1,068 becomes 696. For a population of 10,000, the numbers are 370 and 964, and for very large populations the correction disappears. Our article on poll margins of error explains why a poll of 1,000 can describe a whole country.

Comparing two groups

For two group means, a common approximation is n = 2(z₁₋α/₂ + z₁₋β)² σ² / Δ² per group, where σ is the standard deviation and Δ is the smallest difference worth detecting. With σ = 10 and Δ = 5, at 5% significance and 80% power, you need about 63 per group. At 90% power it is 85. If the difference of interest shrinks to 3, you need about 175 per group; at 2, about 393.

Halving the effect you want to detect roughly quadruples the sample, the same square-root logic as before. For two proportions, such as 40% against 50%, the answer is about 388 per group. Once the data arrive, the independent samples t-test calculator and the difference of two proportions calculator analyze them.

Do not forget dropouts

Sample size formulas give the number you need to analyze. If 15% of participants are likely to drop out or not respond, divide by 0.85: 63 becomes 74.

The most common mistake is working backwards: collecting whatever number is convenient and then asking whether it is enough. A small pilot study to estimate the standard deviation, and a clear decision about the smallest difference that matters, are the two inputs that matter most. Both come before the formula.