Getting your sample size right is one of the most critical decisions in any survey research project. Pick a number too small, and your findings lack the statistical power to be trusted. Pick one too large, and you burn through budget and time collecting data you do not need. If you have ever stared at a blank spreadsheet wondering how to determine an appropriate sample size for a survey study, this guide walks you through every factor, formula, and practical adjustment you need.
Sample size is the number of completed responses your survey needs to represent the target population accurately. Calculating it requires four inputs: population size, confidence level, margin of error, and estimated standard deviation. Once you understand how these variables interact, the math becomes straightforward.
Our team has broken down the entire process into clear sections. You will find the formulas, a z-score reference table, power analysis guidance, attrition adjustments, survey-type recommendations, and answers to the questions researchers ask most. Whether you are running a customer satisfaction poll, a healthcare study, or a dissertation survey, this guide gives you the tools to arrive at a defensible number.
Table of Contents
What Is Sample Size in Survey Research?
Sample size refers to the number of individual respondents who complete your survey and whose answers you analyze. Instead of surveying an entire population (say, all 50,000 employees at a company), you survey a smaller group whose responses represent the whole. The goal is to gather enough data to draw accurate conclusions without surveying every single person.
Think of it like tasting soup. You do not need to drink the entire pot to know if it needs salt. One well-stirred spoonful tells you what you need. But if you only taste a single drop from the edge of the bowl, you might miss the salt sitting at the bottom. Your sample size is that spoonful. It needs to be large enough and representative enough to reflect the whole.
Why Sample Size Matters More Than You Think
A sample that is too small leads to Type II errors. That means you fail to detect a real effect or difference that actually exists in your population. Your survey says there is no difference in customer satisfaction between two product versions, when in reality there is. You make business decisions on faulty data.
A sample that is too large creates a different problem. You waste money, time, and respondent goodwill collecting more data than you need. Worse, an oversized sample can detect differences that are statistically significant but practically meaningless. A massive sample might tell you that Version A has a 0.3% higher satisfaction rating than Version B. The math says it is real. The reality is nobody cares.
The sweet spot gives you enough statistical power to detect meaningful effects without overcollecting. That sweet spot is what the rest of this guide helps you find.
Key Factors That Determine Sample Size
Four primary variables drive sample size calculations. Change any one of them, and your required number of respondents shifts. Understanding each factor individually makes the formula feel logical rather than arbitrary.
1. Population Size
Population size is the total number of people in the group you want to study. If you are surveying employees at a 500-person company, your population size is 500. If you are surveying adult consumers across the United States, your population is roughly 260 million.
Here is the counterintuitive part. Population size has a relatively small effect on sample size once you pass about 10,000 people. A sample of 385 respondents gives you 95% confidence with a 5% margin of error whether your population is 10,000 or 10 million. This is why national political polls can predict election outcomes with samples of around 1,000 to 1,500 people.
For small populations (under 1,000), population size matters a lot. You need a higher percentage of the total population to achieve the same confidence level. We cover this in the formula section with a finite population correction.
2. Confidence Level
The confidence level tells you how certain you want to be that your sample accurately reflects the population. A 95% confidence level means that if you ran the same survey 100 times with different random samples, 95 of those surveys would produce results within your margin of error.
Most survey research uses a 95% confidence level as the standard. Academic and scientific studies sometimes push to 99% for greater certainty. Exploratory or internal business research might accept 90%.
Higher confidence levels require larger samples. Moving from 95% to 99% confidence increases your required sample size by roughly 70%. That is a significant jump, so only choose 99% when the stakes justify it.
3. Margin of Error (Confidence Interval)
Margin of error is the range within which you expect your true population value to fall. If your survey shows 62% of respondents are satisfied with a product, and your margin of error is 5%, then you can say the true satisfaction rate in the population is somewhere between 57% and 67%.
Smaller margins of error require larger samples. Cutting your margin of error in half (from 5% to 2.5%) roughly quadruples your required sample size. Most surveys use 5% as a standard, but precision-critical studies may tighten it to 3% or even 1%.
4. Standard Deviation (Expected Variance)
Standard deviation measures how spread out your responses are likely to be. If you are asking a question where most people will answer similarly (low variance), you need a smaller sample. If responses are all over the map (high variance), you need more respondents to capture that spread accurately.
The challenge is that you often do not know your standard deviation before running the survey. The standard workaround is to use 0.5 (50%) as your estimated population proportion. This represents maximum variance and gives you the largest, most conservative sample size. If your actual variance turns out lower, your survey will be more precise than expected, which is a good problem to have.
How to Determine an Appropriate Sample Size for a Survey Study: Step-by-Step
This five-step process works for the majority of survey research projects. Follow it in order and you will arrive at a defensible sample size for your study.
Step 1: Define Your Population and Research Question
Start by clearly identifying who you are studying and what you want to learn. Are you surveying all customers who purchased in the last 90 days? All registered voters in a specific state? Every employee in your organization?
Write down your population size. If you do not know the exact number, estimate it. For large or unknown populations (over 100,000), you can treat the population as effectively infinite for calculation purposes. The math works out the same.
Also define your research question clearly. Are you measuring a single proportion (what percentage of customers are satisfied)? Comparing two groups (do men and women respond differently)? Testing a correlation? Your analysis type affects which formula and approach you use.
Step 2: Choose Your Confidence Level
Select a confidence level based on how much certainty your study requires. Here are the common choices.
- 90% confidence: Acceptable for internal, exploratory, or low-stakes research.
- 95% confidence: The gold standard for most survey research, business studies, and social science.
- 99% confidence: Reserved for high-stakes research where errors carry significant consequences (medical, policy).
For most surveys, 95% is the right call. It balances precision with practicality.
Step 3: Set Your Margin of Error
Decide how much wiggle room you can accept in your results. The standard margin of error is 5%, meaning your results could be off by up to 5 percentage points in either direction.
- 5% margin of error: Standard for most surveys and polls.
- 3% margin of error: Used when precision matters more, such as competitive market research.
- 1% margin of error: Reserved for census-level precision or high-budget national studies.
Tighter margins of error require dramatically larger samples. A 5% margin might need 385 respondents. A 1% margin for the same population could require nearly 10,000.
Step 4: Estimate Your Standard Deviation
If you have data from a previous similar survey, use the standard deviation from that study. If not, use 0.5 as your estimated population proportion. This conservative estimate assumes maximum variance and ensures your sample size calculation will not underestimate what you need.
This is the approach most statisticians recommend for first-time surveys with no prior data. It is better to overestimate your sample size than to discover after data collection that you do not have enough respondents.
Step 5: Apply the Sample Size Formula
Now plug your values into Cochran’s formula, the most widely used sample size formula for surveys with large or infinite populations.
Cochran’s Formula: N = (Z squared times p times (1 minus p)) divided by E squared
Where:
- N = required sample size
- Z = z-score for your chosen confidence level (1.96 for 95%)
- p = estimated population proportion (use 0.5 if unknown)
- E = margin of error expressed as a decimal (0.05 for 5%)
Worked example: You want to run a customer satisfaction survey with 95% confidence, 5% margin of error, and no prior data on variance.
Z = 1.96, p = 0.5, E = 0.05
N = (1.96 squared times 0.5 times 0.5) divided by 0.05 squared
N = (3.8416 times 0.25) divided by 0.0025
N = 0.9604 divided by 0.0025
N = 384.16
Round up to 385. You need 385 completed survey responses.
If your population is small (under 10,000), apply the finite population correction factor to reduce this number. The correction formula is:
Adjusted N = N divided by (1 + ((N minus 1) divided by Population))
For a population of 1,000 with an initial sample size of 385:
Adjusted N = 385 divided by (1 + (384 divided by 1000)) = 385 divided by 1.384 = 278
You would need 278 respondents from a population of 1,000 to achieve the same confidence and precision.
Z-Score Reference Table for Common Confidence Levels
Every sample size formula requires a z-score that corresponds to your chosen confidence level. Use this reference table to find the correct z-score for your calculation.
| Confidence Level | Z-Score | When to Use |
|---|---|---|
| 80% | 1.282 | Informal or exploratory research with low stakes |
| 85% | 1.440 | Internal pilot studies or preliminary surveys |
| 90% | 1.645 | Business research where some uncertainty is tolerable |
| 95% | 1.960 | Standard for most survey research and social science |
| 99% | 2.576 | High-stakes studies, academic publication, medical research |
| 99.9% | 3.291 | Extremely rare, used in quality control and safety-critical fields |
The z-score represents the number of standard deviations a value is from the mean in a normal distribution. A 95% confidence level uses 1.96 because 95% of the area under a normal curve falls within 1.96 standard deviations of the mean.
If you are working with small samples (under 30), use t-scores from a t-distribution table instead of z-scores. The t-distribution accounts for the additional uncertainty that comes with small samples.
Power Analysis and Effect Size for Research Surveys
If your survey involves hypothesis testing (comparing groups, testing relationships, evaluating interventions), you need more than Cochran’s formula. You need power analysis. This approach calculates sample size based on your ability to detect a real effect if one exists.
Most survey-only studies do not require power analysis. But if you are running an A/B test, comparing employee satisfaction before and after a policy change, or testing whether a new training program improved performance scores, power analysis gives you a more precise sample size than the simpler formula above.
What Is Statistical Power?
Statistical power is the probability that your study will detect an effect when that effect truly exists in the population. A study with 80% power has an 80% chance of finding a real difference and a 20% chance of missing it (a Type II error).
The accepted standard for statistical power is 80%, though some researchers push for 90% in high-stakes studies. Higher power requires larger samples.
What Is Effect Size?
Effect size measures the magnitude of the difference or relationship you are trying to detect. The smaller the effect you want to catch, the larger your sample needs to be. This is why detecting a 1% difference in customer satisfaction requires a massive sample while detecting a 20% difference needs far fewer respondents.
Researchers typically use Cohen’s d as a standardized measure of effect size. Cohen’s guidelines provide benchmarks for interpreting effect sizes.
| Effect Size (Cohen’s d) | Classification | Practical Meaning |
|---|---|---|
| 0.2 | Small | Visible only with careful measurement, subtle difference |
| 0.5 | Medium | Noticeable to an observer, moderate difference |
| 0.8 | Large | Clearly apparent, substantial difference |
For a two-sample t-test with 80% power, 95% confidence, and a medium effect size (0.5), you need approximately 64 respondents per group, or 128 total. Drop to a small effect size (0.2) and you need about 394 per group, or 788 total. This illustrates why effect size is the single biggest driver of sample size in hypothesis-testing research.
When to Use Power Analysis vs. Simple Formula
Use Cochran’s formula when your survey measures proportions, percentages, or descriptive statistics (what percentage of customers are satisfied, what proportion of employees prefer remote work). Use power analysis when your survey involves comparing groups, testing hypotheses, or evaluating interventions (does satisfaction differ between departments, did the new policy change behavior).
Power analysis can be performed using statistical software like G*Power (free), R with the pwr package, or Python with statsmodels. Online calculators from SurveyMonkey and Qualtrics also offer power analysis tools for common study designs.
Adjusting Sample Size for Non-Response and Attrition
Your calculated sample size tells you how many completed responses you need. But not everyone you invite will respond. If you need 385 completed surveys and your expected response rate is 25%, you need to invite far more people than 385.
The Non-Response Adjustment Formula
Required invitations = Required sample size divided by expected response rate
If you need 385 completed surveys and expect a 25% response rate:
Required invitations = 385 divided by 0.25 = 1,540
You would need to send your survey to approximately 1,540 people to end up with 385 completed responses.
Adjusting for Dropout in Longitudinal Studies
If your study spans multiple survey waves (pre-survey, intervention, post-survey), some participants will drop out between rounds. Research typically sees 15% to 30% attrition between survey waves depending on the topic and time gap.
Adjusted initial sample = Required final sample divided by (1 minus expected attrition rate)
If you need 200 respondents at the final wave and expect 20% attrition:
Adjusted sample = 200 divided by (1 minus 0.20) = 200 divided by 0.80 = 250
Recruit 250 participants initially to account for the expected 20% dropout.
Why This Step Matters
Researchers frequently skip this adjustment and end up with too few completed responses at the analysis stage. There is no way to fix a too-small sample after data collection ends. Planning for non-response and attrition upfront is the single most effective way to protect your study from underpowered results.
Sample Size Recommendations by Survey Type
Different types of surveys have different sample size expectations based on the precision required, the size of subgroups you need to analyze, and the stakes of the findings. Use this table as a starting point, then adjust based on your specific needs.
| Survey Type | Minimum Sample | Recommended Sample | Key Consideration |
|---|---|---|---|
| Customer Satisfaction (CSAT) | 100 | 300 to 500 | Enough to segment by product or region |
| Market Research | 384 | 500 to 1,000 | Subgroup analysis by demographic segments |
| Employee Engagement | 100 | 200 to 500 | Depends on total workforce size |
| Healthcare Survey | 200 | 400+ | Often requires ethics board approval and power analysis |
| Education Survey | 100 | 300+ | May need separate samples for students, parents, staff |
| Political Poll | 500 | 1,000 to 1,500 | National polls need demographic weighting |
| Academic Research | 200 | 350+ | Journal reviewers expect power analysis justification |
| Exploratory or Pilot Study | 30 | 50 to 100 | Used to refine questions, not for final conclusions |
These numbers assume a 95% confidence level and 5% margin of error. If you need to analyze subgroups (for example, comparing satisfaction between new and long-term customers), calculate sample size for each subgroup separately. A total sample of 385 might be enough for overall results but too small to compare subgroups of 30 to 50 people each.
Sample Size Best Practices and Common Mistakes to Avoid
Even with the right formula, researchers make predictable mistakes that undermine their sample size calculations. Here are the practices to follow and the pitfalls to avoid.
Best Practices
Calculate before you collect. Decide your sample size before launching your survey, not after. Post-hoc sample size justification is a red flag for reviewers and stakeholders.
Plan for subgroups. If you intend to compare results across demographic groups, departments, or segments, each subgroup needs an adequate sample. A study comparing five departments needs enough respondents in each department, not just in the total sample.
Over-recruit by 10 to 20%. Some completed surveys will be incomplete, rushed, or low quality. Build a buffer so you still have enough valid responses after cleaning your data.
Document your assumptions. Record your confidence level, margin of error, estimated variance, and any power analysis parameters. This documentation matters for academic publishing, stakeholder trust, and replicating your study.
Use pilot data when available. A small pilot survey (30 to 50 respondents) gives you real standard deviation data to use in your main sample size calculation. This is more accurate than defaulting to 0.5.
Common Mistakes
Confusing sample size with response rate. Your sample size is the number of completed responses you need. Your response rate determines how many people you must invite to get there. These are different numbers.
Ignoring non-response bias. If the people who respond to your survey differ systematically from those who do not, your results will be skewed regardless of sample size. Always consider whether your respondents represent your population.
Treating any sample as statistically valid. A convenience sample of 50 friends is not equivalent to a random sample of 50 from your target population. Sampling method matters as much as sample size.
Forgetting attrition in multi-wave studies. If you collect data at multiple time points, your sample shrinks at each wave. Plan for attrition from the start.
Over-collecting without purpose. Collecting 10,000 responses when 400 would suffice wastes resources and can produce statistically significant but practically meaningless results. More data is not always better.
Tips for Small Populations
When your total population is under 100, standard formulas produce sample sizes that represent a huge percentage of your group. For a population of 50 at 95% confidence and 5% margin of error, you need approximately 44 respondents. That is 88% of your population.
In these cases, consider surveying the entire population (a census) rather than sampling. The effort difference between surveying 44 out of 50 people and surveying all 50 is minimal, and you eliminate sampling error entirely.
When Smaller Samples Are Acceptable
Not every survey needs statistical significance. Qualitative surveys, exploratory research, internal pilots, and user experience studies often work with small samples intentionally. If you are trying to understand why a problem exists rather than measuring how widespread it is, 20 to 30 thoughtful responses can be more valuable than 400 superficial ones.
The key is transparency. Be clear about what your sample can and cannot tell you. Do not claim generalizability from a small convenience sample. Match your claims to your methodology.
Quick Reference: How Changing Variables Affects Sample Size
This table shows how adjusting each variable changes your required sample size. All examples assume a baseline of 95% confidence, 5% margin of error, and 0.5 estimated proportion, yielding 385 respondents.
| Variable Change | Effect on Sample Size | New Approximate Sample |
|---|---|---|
| Confidence level 95% to 99% | Increases by about 71% | 664 |
| Confidence level 95% to 90% | Decreases by about 30% | 271 |
| Margin of error 5% to 3% | Increases by about 178% | 1,068 |
| Margin of error 5% to 10% | Decreases by about 75% | 96 |
| Population from infinite to 1,000 | Decreases by about 28% | 278 |
| Population from infinite to 500 | Decreases by about 45% | 218 |
| Effect size from medium (0.5) to small (0.2) | Increases by about 516% (per group) | 394 per group |
Use this table to make trade-off decisions. If your budget allows for 300 respondents and you need 95% confidence, you can either widen your margin of error slightly or accept a lower confidence level. Understanding these relationships helps you make informed compromises rather than arbitrary cuts.
FAQs
How do I determine the appropriate sample size for a survey?
To determine sample size, identify your population size, choose a confidence level (typically 95%), set a margin of error (typically 5%), estimate your standard deviation (use 0.5 if unknown), and apply Cochran’s formula: N = (Z squared times p times (1 minus p)) divided by E squared. For most surveys with large populations, this produces a sample of approximately 385 respondents.
What is the 10% rule in sampling?
The 10% rule states that when sampling without replacement from a finite population, your sample size should not exceed 10% of the total population. This rule matters most for populations under 10,000. For larger populations, the 10% rule becomes negligible because standard sample size formulas (which typically yield 385 to 1,500 respondents) already fall well below 10% of populations over 15,000.
Is a 10% sample size acceptable?
A 10% sample is acceptable and often more than sufficient for populations under 10,000. For example, sampling 100 people from a population of 1,000 provides strong statistical power. However, for large populations, a 10% sample is unnecessary. A sample of 385 respondents provides 95% confidence with a 5% margin of error for populations of any size over 100,000.
Is 100 respondents enough for a survey?
Yes, 100 respondents is enough for exploratory research, pilot studies, or surveys of small populations. However, 100 is below the standard threshold of 385 needed for 95% confidence with a 5% margin of error in larger populations. For publishable academic research, subgroup analysis, or statistically significant findings, aim for at least 200 to 385 respondents depending on your precision requirements.
What is the minimum sample size for statistically significant results?
There is no universal minimum, but common benchmarks are 30 respondents for basic statistical inference (central limit theorem threshold), 200 for medium-complexity models per Kline’s guidelines, and 385 for 95% confidence with 5% margin of error. For hypothesis testing, calculate sample size using power analysis with 80% power and your expected effect size.
How do I calculate sample size for an unknown population?
When population size is unknown or very large (over 100,000), treat it as infinite and use Cochran’s formula without the finite population correction. Plug in your z-score (1.96 for 95% confidence), estimated proportion (0.5 for maximum variance), and margin of error (0.05 for 5%). This yields 385 respondents for standard parameters regardless of actual population size.
What is a good sample size for quantitative research?
For quantitative survey research, 385 respondents is considered the standard good sample size because it provides 95% confidence with a 5% margin of error for large populations. For studies requiring subgroup analysis, 500 to 1,000 respondents is better. For academic publication with hypothesis testing, calculate using power analysis targeting 80% power and a medium effect size of 0.5.
Conclusion: Getting Your Sample Size Right
Determining how to determine an appropriate sample size for a survey study comes down to understanding four variables, applying one formula, and making practical adjustments for the real world. Define your population, pick your confidence level and margin of error, estimate your variance, and calculate. Then adjust upward for non-response, attrition, and data quality.
The magic number for most large-population surveys is 385. That gives you 95% confidence with a 5% margin of error. For small populations, apply the finite population correction. For hypothesis-testing studies, use power analysis instead. For multi-wave research, build in attrition buffers from day one.
Your next steps are to document your assumptions, plan your recruitment strategy to hit your target completed-response count, and design your survey to maximize response rate. With a defensible sample size in hand, you can collect data with confidence that your findings will hold up to scrutiny.