Data

Confidence Intervals & Sample Size: How Much Data Do You Actually Need?

Every survey, experiment and customer-research project hinges on one decision: how many people do I need? Too few and you make decisions on noise; too many and you waste budget. Confidence intervals and sample size are two sides of the same question — how precise is my estimate, and what does it take to make it more precise?

What a confidence interval tells you

A confidence interval is a range that is likely to contain the true value you are trying to estimate — the real average customer rating, or the real share of people who would buy. It has two parts: the estimate in the middle and the margin of error on each side.

For an average, the formula is:

CI = x̄ ± z × (s ÷ √n)

where x̄ is the sample mean, s the standard deviation, n the sample size and z the critical value for your confidence level (1.96 for 95%, 2.576 for 99%). The term s ÷ √n is the standard error.

Worked example: a customer satisfaction score

200 customers rate you on a 1–5 scale. The average is 4.2 with a standard deviation of 1.1.

• Standard error = 1.1 ÷ √200 ≈ 0.078
• Margin of error at 95% = 1.96 × 0.078 ≈ 0.15
• 95% confidence interval: 4.05 to 4.35

You can reproduce this in seconds with our confidence interval calculator.

Margin of error for percentages

Most surveys report a percentage (“47% would recommend us”) rather than an average. For a proportion p and sample size n, the margin of error is z × √(p(1 − p) ÷ n).

A poll of 1,000 people finds 47% support. At 95% confidence the margin of error is ±3.1 points, so the interval is roughly 43.9% to 50.1%. Check it with the margin of error calculator.

What “95% confidence” really means: if you repeated the same survey many times and built an interval each time, about 95% of those intervals would contain the true value. It is a statement about the method’s reliability, not a 95% probability that this particular interval is right.

Sample size cheat sheet (95% confidence, worst case p = 0.5)

• ±10% margin → 97 responses
• ±5% margin → 385
• ±3% margin → 1,068
• ±2% margin → 2,401
• ±1% margin → 9,604

The formula behind these is n = z² × p(1 − p) ÷ E², always rounded up. Plug in your own numbers with the sample size calculator.

The diminishing-returns trap

Margin of error shrinks with the square root of the sample size, so precision gets expensive fast:

• 100 responses → ±9.8%
• 400 responses → ±4.9%
• 1,600 responses → ±2.5%

Every time you want to halve the margin of error, you need four times the sample. Decide how precise the decision actually needs to be before paying for more responses.

Choosing a confidence level

• 90% — quick, low-stakes reads such as internal polls (z = 1.645).
• 95% — the standard for most business and academic work (z = 1.96).
• 99% — high-stakes decisions where being wrong is costly (z = 2.576). At ±5%, this raises the required sample from 385 to 664.

Practical adjustments

Plan for your response rate

The sample size is the number of completed responses. If 20% of invited customers usually respond and you need 385, invite about 1,925.

Small populations

If you are surveying a small, fixed group, apply the finite-population correction: n_adj = n ÷ (1 + (n − 1) ÷ N). For a team of N = 500 employees, 385 becomes about 218.

Sub-groups need their own samples

If you want to report results separately for, say, new and returning customers, each group needs to be large enough on its own.

Common mistakes

• Ignoring who responds. A large but self-selected sample can be far more misleading than a smaller representative one. Margin of error only covers random sampling error, not bias.
• Quoting one margin for every question. The ±3% headline applies to questions answered by the full sample at around 50%; sub-group results are much less precise.
• Confusing confidence level and margin of error. Higher confidence widens the interval; a bigger sample narrows it.
• Treating overlapping intervals as “no difference”. To compare two groups properly, test the difference directly with the statistical significance calculator.

FAQs

Why is 385 such a common sample size? It is the number needed for ±5% at 95% confidence when you assume the worst-case 50% proportion: 1.96² × 0.25 ÷ 0.05² ≈ 384.2, rounded up.

Does a bigger population need a bigger sample? Hardly, once the population is in the thousands. Precision depends mainly on the sample size, not the population size.

What if my survey is already done? Use the margin of error calculator with your actual response count to see how precise the results are.