How Many Survey Responses Do You Need?
Last reviewed: January 2026
The Sample Size Calculator is a free browser-based tool that performs this calculation instantly with no signup or downloads required. Enter your values, click calculate, and get accurate results immediately. All processing happens in your browser — nothing is sent to a server.
At 95% confidence and ±5% margin of error, you need 385 responses (infinite population). The margin of error decreases with the square root of sample size — to cut margin in half, you need 4× the responses. A common mistake: confusing confidence level with probability of a correct answer. 95% confidence means that 95% of similarly constructed intervals would contain the true value — not that you're 95% sure of your specific result. Use 50% expected proportion when unknown — this gives the most conservative (largest) sample size.
| Confidence Level | Margin of Error | Sample Size (large pop) |
|---|---|---|
| 90% | ±5% | 271 |
| 95% | ±5% | 385 |
| 95% | ±3% | 1,068 |
| 99% | ±3% | 1,849 |
| 99% | ±1% | 16,590 |
Sample size — the number of observations or participants in a study — is one of the most critical decisions in research design. A sample that is too small produces unreliable results with wide confidence intervals, low statistical power (high probability of missing real effects), and limited generalizability. A sample that is unnecessarily large wastes resources, time, and in medical research, may expose more participants to experimental treatments than ethically necessary. Proper sample size calculation balances statistical requirements against practical constraints to produce the minimum sample needed for reliable, meaningful conclusions.
The consequences of inadequate sample size are severe and pervasive. Underpowered studies frequently produce false negatives (failing to detect effects that actually exist) and, paradoxically, the effects they do detect are more likely to be inflated (the "winner's curse"). Meta-analyses of published research consistently find that small studies produce more variable and less reliable results than larger studies. The reproducibility crisis affecting psychology, medicine, and other sciences has been partly attributed to widespread use of insufficient sample sizes, leading to findings that seem significant but fail to replicate.
Four primary inputs determine the required sample size for most study designs. The significance level (α, typically 0.05) sets the threshold for declaring a result statistically significant — it represents the acceptable probability of a false positive (Type I error). Statistical power (1 - β, typically 0.80 or 0.90) is the probability of detecting a real effect — a power of 0.80 means an 80% chance of finding a significant result if the effect truly exists. Effect size quantifies the magnitude of the expected difference or relationship. Population variability (standard deviation) measures how spread out the data is.
Effect size is often the most challenging input to estimate because it requires prior knowledge about the phenomenon being studied. Cohen's conventions provide rough benchmarks: small effect (d = 0.2 for means, r = 0.1 for correlations), medium effect (d = 0.5, r = 0.3), and large effect (d = 0.8, r = 0.5). However, these conventions are arbitrary and should be replaced with domain-specific estimates whenever possible — from pilot studies, previous research, or clinically meaningful difference thresholds. Detecting smaller effects requires proportionally larger samples — detecting an effect of d = 0.2 requires roughly 16 times the sample size needed for d = 0.8 at the same power and significance levels.
Different study designs require different sample size formulas. For comparing two means (independent samples t-test), n per group = 2 × (Z_α/2 + Z_β)² × σ² / δ², where σ is the pooled standard deviation and δ is the minimum detectable difference. For comparing two proportions, n per group = (Z_α/2 + Z_β)² × [p₁(1-p₁) + p₂(1-p₂)] / (p₁ - p₂)². For estimating a proportion with a specified margin of error, n = Z² × p(1-p) / E², where E is the desired margin of error.
Survey research uses a variation of the proportion formula: for a 95% confidence level with a ±3% margin of error and an assumed proportion of 0.5 (maximum variability), the required sample size is (1.96² × 0.25) / 0.03² = 1,068 respondents. For finite populations, the calculated sample size is adjusted downward using the finite population correction: n_adjusted = n / (1 + (n-1)/N), where N is the population size. This correction becomes significant when the sample exceeds approximately 5% of the population. Our Confidence Interval Calculator complements this tool for analyzing results.
Theoretical sample size calculations provide a starting point, but practical considerations often require adjustments. Anticipated dropout/attrition rates mean the initial enrollment should exceed the calculated size — a study needing 100 completers with an expected 20% dropout rate should enroll 125 participants. Non-response rates in survey research can dramatically reduce effective sample size — a mailed survey with a 30% response rate needs to be sent to more than three times the target sample. Stratified analyses (examining results within subgroups defined by age, gender, ethnicity, or other variables) require sufficient sample within each stratum, potentially inflating the total sample requirement.
Cluster sampling (sampling groups rather than individuals, such as classrooms rather than students) requires a design effect adjustment: n_adjusted = n × DEFF, where the design effect (DEFF) depends on cluster size and the intra-cluster correlation coefficient (ICC). Typical design effects range from 1.5 to 3.0, meaning cluster-sampled studies may need 50-200% more participants than simple random samples. Matched-pair and crossover designs typically require smaller samples because they control for individual variability, but they introduce other considerations like carryover effects and matching feasibility.
Clinical trials follow particularly rigorous sample size requirements because inadequate power wastes research resources and delays the availability of effective treatments. Phase II trials typically enroll 100-300 participants, while Phase III confirmatory trials often require 300-3,000+ depending on the expected treatment effect. Regulatory agencies like the FDA review sample size justifications as part of protocol approval, requiring both statistical and clinical rationale. Adaptive trial designs allow sample size re-estimation at interim analyses, providing flexibility to increase enrollment if initial effect size estimates were too optimistic.
Market research, quality control, public opinion polling, and educational testing each have established conventions for sample size. Political polls targeting a ±2-3% margin of error at 95% confidence typically survey 1,000-1,500 respondents nationally. A/B testing in digital marketing requires sample sizes that depend on baseline conversion rates and the minimum detectable improvement — testing a 2% baseline conversion with sensitivity to detect a 10% relative improvement (2.0% to 2.2%) requires approximately 80,000 visitors per variation. For related statistical tools, explore our P-Value Calculator and Binomial Probability Calculator.
The most frequent error is using a sample size that is too small, then interpreting inconclusive results as evidence of no effect. A test with 100 users per variant can only reliably detect large differences (10%+ relative change). Detecting a 2% conversion rate improvement with 95% confidence and 80% power typically requires 10,000+ observations per group. Another common mistake is peeking at results before the test reaches the calculated sample size — this inflates false positive rates from 5% to as high as 30%. Use sequential testing methods if you need to monitor results continuously. For survey research, account for response rates: if you need 400 completed responses and expect a 20% response rate, you must contact 2,000 people. Analyze your results with our Confidence Interval Calculator.
→ Double-check with manual calculation. Use the calculator to verify your work, or work the problem by hand first and check against the calculator's result.
→ Pay attention to units and precision. Make sure your inputs use consistent units. The calculator preserves decimal precision — round the final answer to the appropriate number of significant figures for your context.
→ Use for learning, not just answers. Review the formula and step-by-step breakdown to understand the underlying math — this builds intuition for future problems.
→ Try edge cases. Test with very large, very small, or zero values to build intuition about how the formula behaves across different inputs.
See also: Percent Error Calculator · Confidence Interval Calculator · Statistics Calculator · Binomial Probability Calculator