Sample Size Calculator

Sample size refers to the number of individual observations or data points collected from a larger group for use in a study or survey. Rather than measuring every member of a group, researchers draw a smaller subset and use it to make inferences about the whole. When that subset is chosen and sized correctly, its findings can reliably represent the full population.

The distinction between a sample and a population is foundational here. The population is the entire group you want to understand, whether that is all registered voters in a state, every customer who has purchased from your store, or all patients with a given diagnosis. The sample is the portion of that population you actually measure. Because studying an entire population is rarely practical or cost-effective, sample size determination becomes one of the most consequential decisions in research design.

Size matters in both directions. A sample size that is too small increases sampling error, widens your confidence interval, and produces results that may not hold up when tested against the broader population. A sample that is far larger than necessary consumes resources without meaningfully improving precision. Beyond a certain threshold, adding more respondents yields diminishing returns on accuracy.

Sample size decisions come up across virtually every field that relies on data. Common applications include:

Understanding what sample size means, and why it matters, is the starting point for using any sample size calculator effectively.

Before the calculator can return a result, four inputs need to be defined. Each one shapes the final number in a different way.

Confidence Level

The confidence level expresses how certain you want to be that your results reflect the true population value. It is stated as a percentage and corresponds directly to a z-score, which the formula uses to set the width of your confidence interval.

The three most commonly used confidence levels in US research are:

A 95% confidence level is the standard for most surveys and academic studies. Choosing 99% produces a narrower margin of uncertainty but requires a noticeably larger sample. Dropping to 90% reduces the required sample size but also reduces how much confidence you can place in the results.

Margin of Error

The margin of error defines the maximum acceptable difference between your sample estimate and the true population value. It is expressed as a percentage, and the most widely used figure in general population surveys is plus or minus 5%.

A margin of error of 5% means that if the true population value is 60%, your survey result is expected to fall somewhere between 55% and 65%. Tightening the acceptable margin of error, for example moving from 5% to 3%, requires a substantially larger sample to achieve that additional precision. This is one of the most direct levers researchers have when balancing accuracy against cost.

Population Proportion

The population proportion is your best estimate of how the population splits on the question being measured. For example, if prior research suggests that 40% of your target audience holds a particular view, you would enter 40%.

When no prior estimate exists, the conservative default is 50%. This is intentional. A proportion of 50% produces the largest possible required sample size for any given confidence level and margin of error, which means it offers the most protective estimate when uncertainty is high. Using 50% ensures you do not underestimate how many responses you need.

Population Size

For most large-scale surveys where the population runs into the tens of thousands or more, the population size field can be left blank. In those cases, the population is treated as effectively infinite, and the correction for finite populations does not meaningfully change the result.

When the population is small and known, entering that figure allows the calculator to apply a finite population correction, which adjusts the required sample size downward. For example, if you are surveying employees at a company with 400 staff, accounting for that finite population size produces a more accurate and typically smaller required sample than the unlimited population formula would.

The Core Formula

The sample size formula used by this calculator is:

Each variable in the sample size formula serves a specific role:

The relationship between these variables is direct and predictable. A higher z-score, driven by a higher confidence level, increases the required sample size. A smaller margin of error, because it appears squared in the denominator, has a particularly strong effect on how large the sample needs to be. A population proportion closer to 50% also increases the required sample size, which is why 0.5 is used as the conservative default when the true proportion is unknown.

Finite Population Correction

The formula above assumes an infinitely large or unknown population. When the total population size is known and relatively small, a second step is needed to avoid overestimating how many responses you actually need.

The finite population correction formula is:

Where n₁ is the sample size calculated from the core formula, N is the total population size, and n₂ is the corrected minimum sample size.

This adjustment reflects the fact that as your sample approaches a meaningful share of the total population, each additional observation contributes less new information. The finite population correction factor becomes most consequential when the uncorrected sample size is more than 5% to 10% of the total population. For large populations, the correction is negligible and the two formulas produce nearly identical results.

What the Output Means

The number the calculator returns is the minimum number of completed, usable responses required for your results to meet the statistical constraints you set. It is not the number of people to contact or invite. If your expected response rate is 30%, you would need to reach roughly three times the calculated sample size to end up with enough completed responses.

The following example walks through a realistic US research scenario using the formula covered in the previous section.

The Scenario

A retail company wants to survey its customer base of 20,000 people to find out what proportion prefers curbside pickup over in-store shopping. No prior data exists on this preference split, so the team uses the conservative default of 50% for the population proportion. They want results they can act on with confidence, so they set a 95% confidence level and an acceptable margin of error of 5%.

Step 1: Identify the Inputs

Step 2: Apply the Core Formula

Step 3: Apply the Finite Population Correction

Because the population of 20,000 is known, the finite population correction is applied to refine the result.

The company needs a minimum of 378 completed survey responses to meet its statistical requirements. Because the population is large enough that the correction produces only a modest adjustment, either figure, 385 or 378, represents a practical and defensible threshold.

How Changing the Inputs Shifts the Result

The required sample size is sensitive to changes in confidence level and margin of error. Using the same scenario with the population proportion held at 50% and population size at 20,000:

Two patterns stand out. First, tightening the margin of error from 5% to 3% more than doubles the required sample size at every confidence level. Because the margin of error is squared in the denominator of the formula, small reductions in the acceptable error range have a disproportionately large effect on how many responses you need. Second, moving from a 95% to a 99% confidence level increases the required sample by roughly 70% when all other inputs stay fixed.

These tradeoffs are worth working through before launching any survey. Deciding upfront how much precision the study actually requires, and what confidence level the findings need to support, prevents both undersampling and unnecessary oversampling.

Sample size is not a fixed number you look up in a table. It is the product of several interacting variables, and understanding how each one pulls on the final figure helps you make deliberate tradeoffs rather than arbitrary ones.

Confidence Level

A higher confidence level demands a larger sample. Moving from 90% to 95% to 99% raises the z-score used in the formula, and because that z-score is squared, the effect on sample size is amplified. Researchers who need findings strong enough to inform high-stakes decisions, such as a product launch or a policy change, typically set their confidence level at 95% or higher. Exploratory work with lower stakes can often justify 90%.

Margin of Error

The acceptable margin of error is one of the most powerful levers in sample size calculation. Because it appears squared in the denominator of the formula, even modest improvements in precision require substantially more responses. Tightening the margin of error from 5% to 3% more than doubles the required sample size, as shown in the worked example above. Before setting this value, consider what level of precision the study's conclusions actually require. A difference of a few percentage points matters enormously in a close election poll but may be irrelevant in an internal employee satisfaction survey.

Population Variability

Population variability refers to how spread out the characteristic being measured is across the population. In the sample size formula, this is captured by the proportion p and its complement (1 − p). The product p × (1 − p) reaches its maximum value when p equals 0.50, which is why using 50% as the default population proportion gives the most conservative, and therefore most protective, sample size estimate. As the true proportion moves further from 50% in either direction, the required sample size decreases. If reliable prior data suggests the proportion is closer to 20% or 80%, using that figure rather than 50% will produce a smaller, more efficient required sample.

Finite vs. Infinite Population

For most large-scale surveys where the population size runs into the tens of thousands or beyond, the finite population correction has little practical effect. The uncorrected formula and the corrected formula produce nearly identical results. The correction becomes meaningful when the calculated sample size represents a significant share of the total population, generally above 5%. In those cases, applying the finite population correction reduces the required sample size and produces a more accurate estimate. Surveying a known group of 500 employees, for example, will yield a meaningfully different result with and without the correction applied.

Study Type

Different research contexts carry different expectations for how large a sample needs to be, and why.

Understanding which category your study falls into helps set realistic expectations for how many responses you need and what level of confidence those responses can actually support.

These three concepts appear in nearly every discussion of sample size, yet they are frequently misunderstood or conflated. Getting them right matters because each one captures a different dimension of how reliable a study's findings actually are.

What a 95% Confidence Level Actually Means

The confidence level is one of the most misread figures in survey reporting. A 95% confidence level does not mean there is a 95% probability that the true population value falls within your interval. It means that if you repeated the same study many times under the same conditions, drawing a new random sample each time, approximately 95% of the confidence intervals produced by those studies would contain the true population value.

Any single study either captures the true value in its interval or it does not. The 95% refers to the reliability of the procedure across repeated sampling, not to the probability attached to any one result. This is a subtle but important distinction, particularly when communicating findings to stakeholders who may interpret confidence level as a straightforward accuracy guarantee.

How to Interpret Margin of Error in Survey Results

When a survey reports a result with a margin of error of plus or minus 5%, it means the true population value is expected to fall within 5 percentage points of the reported figure, given the stated confidence level. If a poll finds that 54% of respondents support a particular policy, and the margin of error is 5%, the actual population support could reasonably range from 49% to 59%.

Margin of error applies to the overall sample. Subgroup results, such as responses broken out by age group, income bracket, or region, carry a larger effective margin of error because they are based on a smaller portion of the total sample. This is a common source of overinterpretation in survey reporting.

Statistical Power and the Probability of Detecting a Real Effect

Statistical power is the probability that a study will detect a real effect when one actually exists. It is expressed as 1 minus beta (1 − β), where beta is the probability of a type II error, also called a false negative. A study with 80% power has an 80% chance of detecting a true difference between groups if that difference genuinely exists, and a 20% chance of missing it.

The standard thresholds used across US research literature are:

Higher power requires a larger sample. If detecting a small but meaningful difference is important to the study's purpose, power should be set at 90% and the sample size calculated accordingly. Under-powered studies are a leading contributor to results that fail to replicate.

Type I and Type II Errors

These two error types represent the two ways a study can reach a wrong conclusion.

A type I error, also called a false positive, occurs when a study concludes there is a significant difference or effect when none actually exists. The probability of a type I error is controlled by the alpha level, set at 0.05 in most US research, meaning researchers accept a 5% chance that their p-value reflects a difference that is actually due to chance.

A type II error, also called a false negative, occurs when a study fails to detect a real difference or effect. The probability of a type II error is beta, with 0.20 being the standard cutoff, corresponding to 80% power.

Both error types involve tradeoffs. Lowering alpha to reduce false positives requires a larger sample size to maintain the same level of power. Increasing power to reduce false negatives also requires more observations. Balancing these two error rates against practical constraints is a core part of research design.

How These Concepts Interact

Confidence level, margin of error, and statistical power are not independent settings. They pull on the same sample size calculation from different directions, and adjusting any one of them affects what the others can deliver given a fixed number of observations.

A study designed with a 95% confidence level, a 5% margin of error, and 80% power will require a specific minimum sample to satisfy all three constraints simultaneously. Tightening any one of them, raising confidence to 99%, narrowing the margin of error to 3%, or increasing power to 90%, increases the required sample size. Loosening one constraint can free up room to tighten another without increasing the total number of responses needed.

Understanding how these inputs interact is what allows researchers and survey designers to make deliberate decisions rather than defaulting to arbitrary numbers.

While every study should calculate its required sample size based on its specific inputs, a handful of benchmarks appear consistently across research practice in the US. Knowing what these numbers represent, and when they apply, helps set realistic expectations before running the full calculation.

The 385 Benchmark

A sample size of approximately 385 is the standard result for estimating a population proportion when the population is large or unknown, the confidence level is set at 95%, the margin of error is 5%, and the population proportion is assumed to be 50%. This figure appears so frequently in survey methodology that it functions as a practical default for general population research.

It is worth being precise about what this benchmark assumes. The moment any of those inputs change, the required sample size changes with it. A population proportion that differs meaningfully from 50%, or a margin of error tightened below 5%, will produce a different number. The 385 figure is a starting reference, not a universal answer.

The 1,067 Benchmark

Tightening the margin of error from 5% to 3%, while holding confidence at 95% and proportion at 50%, produces a required sample size of approximately 1,067. This is the threshold commonly used in national polls and studies where a higher degree of precision is needed to distinguish meaningful differences between groups or over time.

The jump from 385 to 1,067 illustrates how sensitive the sample size formula is to changes in the acceptable margin of error. Nearly three times as many responses are needed to gain less than two additional percentage points of precision.

Fewer Than 30: Where Reliable Inference Breaks Down

A sample size below 30 is generally considered too small to support statistically reliable conclusions for most research purposes. With so few observations, the margin of error becomes wide, the confidence interval loses practical utility, and the results are highly sensitive to individual responses. Studies operating with fewer than 30 data points are typically limited to pilot work, qualitative exploration, or contexts where the population itself is extremely small and nearly fully sampled.

When 100 to 400 Responses Is Sufficient

For many practical research applications, a sample size somewhere between 100 and 400 is adequate. This range tends to apply when:

Within this range, the exact number should still come from the formula rather than from convention. Picking 200 because it feels reasonable is not the same as calculating that 200 meets the study's actual statistical requirements.

When Larger Samples Are Necessary

Larger samples become necessary when the margin of error needs to be tight, when the study must support subgroup comparisons, when the population proportion is expected to be close to 50%, or when the findings will be used in high-stakes contexts such as clinical research, regulatory filings, or large-scale policy decisions.

Studies that need to detect small differences between groups, or that require high statistical power to avoid false negatives, will also need substantially more observations than general surveys.

Diminishing Returns Above Large Thresholds

One of the least intuitive aspects of sample size is how quickly additional responses stop adding meaningful precision. For population-level proportion estimates, the confidence interval narrows quickly up to a few hundred responses and then begins to plateau. Moving from 1,000 to 10,000 responses produces a far smaller improvement in precision than moving from 100 to 1,000.

This is a direct consequence of the sample size formula. Because sample size appears under a square root in the relationship between sample size and margin of error, doubling the sample size does not halve the margin of error. It reduces it by a factor of roughly 1.4. For most large-population surveys, collecting beyond 1,500 to 2,000 responses offers increasingly marginal gains in accuracy, and the resources required to go further are rarely justified by the statistical improvement achieved.