Sample Size Calculator

Survey sample size calculates how many people you need to survey to estimate a population parameter (like satisfaction rate) within a given margin of error. It uses a single proportion formula.

A/B test sample size calculates how many users you need in each variant to detect a statistically significant difference between two groups (e.g., control vs. treatment). It accounts for statistical power, significance level, and the minimum detectable effect — ensuring you can reliably detect a real improvement in conversion rates or other metrics.

95% confidence level is the industry standard for most surveys and research. It means that if you repeated the survey 100 times, approximately 95 of them would yield results within the margin of error of the true population value.

Use 90% for exploratory research where precision is less critical. Use 99% for high-stakes decisions — but note that higher confidence requires larger sample sizes (about 73% larger than 95% confidence).

The proportion p = 50% (0.5) yields the maximum possible sample size for a given confidence level and margin of error — because p(1−p) reaches its peak at 0.5. Using 50% is the most conservative approach, ensuring your sample is large enough regardless of the true proportion. If you have prior knowledge suggesting a different proportion (e.g., 30% or 70%), you can adjust it — this will reduce the required sample size while maintaining the same precision.

When your target population is small (e.g., a company with 500 employees), the standard sample size formula overestimates what you need. The finite population correction adjusts the sample size downward, accounting for the fact that sampling without replacement from a small population provides more information per respondent. Our calculator automatically applies FPC when you enter a population size — the smaller the population, the greater the adjustment.

Statistical power (1−β) is the probability of correctly detecting a real effect when it exists. At 80% power, there's a 20% chance you'll miss a real improvement (Type II error). Higher power (90% or 95%) reduces this risk but requires larger samples. 80% power is widely accepted as the standard for most A/B tests — balancing resource efficiency with reliability. For mission-critical tests where missing an effect would be costly, consider 90% or 95% power.

The Minimum Detectable Effect (MDE) is the smallest improvement you want to be able to detect as statistically significant. For example, if your baseline conversion rate is 10% and you set a 20% relative MDE, you're saying you want to reliably detect an increase from 10% to 12%.

How to choose: Smaller MDEs require dramatically larger samples. Start by determining what business impact is meaningful — is a 1% absolute improvement worth detecting? Or do you only care about 5%+ improvements? Balance statistical ambition with practical constraints like traffic volume and test duration.

Sample size is inversely proportional to the square of the effect size. This means cutting the MDE in half requires approximately 4× the sample size. For example, detecting a 10% relative lift might need 15,700 per variant, while detecting a 5% relative lift could need over 62,000 per variant. This quadratic relationship is why very precise A/B tests demand high-traffic websites and why you should carefully consider the smallest effect worth detecting before launching a test.

Our calculator uses two-tailed tests by default, which is the more conservative and widely recommended approach. A two-tailed test can detect both positive and negative effects — important because sometimes changes can hurt your metrics. One-tailed tests require smaller samples but only detect effects in one direction. Most reputable testing frameworks and academic standards recommend two-tailed testing to avoid bias and ensure you catch unexpected negative results.

Once you know the required sample size per variant, divide it by your daily traffic to estimate the minimum test duration. Example: If you need 3,934 visitors per variant and your page gets 500 visitors/day, you'll need at least 8 days per variant (3,934 ÷ 500 ≈ 8 days). Always run tests for at least one full business cycle (typically 1–2 weeks) to account for day-of-week effects, even if the math says you could finish sooner. Never stop a test early just because results look significant — this inflates false positive rates dramatically.