What Is Sample Size Determination?
Sample size determination is the process of calculating how many respondents or observations a study needs to produce results with acceptable precision and statistical power. It's the bridge between "we need to do a survey" and "we need to survey exactly this many people." Too few respondents and your confidence intervals are too wide to support decisions. Too many and you've spent budget and time that didn't meaningfully improve your data quality. Getting the number right, not just picking a round number that feels sufficient, is one of the most consequential decisions in any quantitative research design.
Why Sample Size Determination Matters in Research
Underpowered studies waste resources by producing results too imprecise to act on. Overpowered studies waste budget by collecting far more data than the analysis requires. Both happen constantly in market research because teams either skip the calculation entirely ("let's do 500, that seems like enough") or run the formula without understanding what the inputs mean. A sample of 200 might be perfectly adequate for one study and completely inadequate for another, depending on what you're measuring and how much precision you need.
How Sample Size Determination Works
The Core Formula for Estimating Proportions
The most common sample size calculation in survey research estimates a population proportion (e.g., "what percentage of customers prefer feature A?").
n = (Z^2 x p x (1 - p)) / E^2
Where:
- n = required sample size
- Z = z-score for the desired confidence level (1.96 for 95%, 2.576 for 99%)
- p = expected proportion (use 0.5 if unknown, which gives the largest sample)
- E = margin of error (acceptable range around the true value)
Worked Example
You want to estimate the percentage of customers who'd recommend your product, with 95% confidence and +/- 4% margin of error. You don't know the expected proportion, so you use p = 0.5.
n = (1.96^2 x 0.5 x 0.5) / 0.04^2 n = (3.8416 x 0.25) / 0.0016 n = 0.9604 / 0.0016 n = 600.25
Round up: 601 respondents.
Quick Reference Table
| Confidence Level | Margin of Error | Required n (p = 0.5) |
|---|---|---|
| 95% | +/- 5% | 385 |
| 95% | +/- 4% | 601 |
| 95% | +/- 3% | 1,068 |
| 95% | +/- 2% | 2,401 |
| 95% | +/- 1% | 9,604 |
| 99% | +/- 5% | 664 |
| 99% | +/- 3% | 1,844 |
Notice the diminishing returns. Going from +/- 5% to +/- 3% adds 683 respondents. Going from +/- 3% to +/- 1% adds 8,536. For most commercial research, +/- 3 to 5% is the practical sweet spot.
Finite Population Correction
If you're sampling a significant portion of a small population (more than 5%), the standard formula overestimates the required sample size. Apply the finite population correction:
n_adjusted = n / (1 + (n - 1) / N)
Where N is the total population size. For a company with 2,000 employees, a standard calculation calling for 385 respondents would adjust to:
n_adjusted = 385 / (1 + 384 / 2,000) = 385 / 1.192 = 323
You'd need 323 instead of 385.
Factors That Increase Required Sample Size
Subgroup analysis. Each subgroup you plan to analyze independently needs its own adequate sample. If you're comparing four customer segments, you need 385 per segment for +/- 5% precision within each, that's 1,540 total, not 385.
Cluster sampling designs. Cluster sampling inflates sampling error by a factor called the design effect (DEFF). Multiply the base sample size by DEFF. A design effect of 2.0 means you need twice as many respondents as the basic formula suggests. See our cluster sampling guide for DEFF calculation.
Expected non-response. If you expect a 40% response rate, you need to invite 2.5 times the required sample. For 600 completes at 40% response: invite 1,500.
Comparing means rather than proportions. If your primary analysis compares average scores (satisfaction, willingness to pay), the formula changes. You'll need the expected standard deviation and the minimum difference you want to detect (minimum detectable effect). Power analysis tools handle this calculation.
Multiple comparisons. Testing many hypotheses from the same data inflates false positive rates. Adjustments like Bonferroni correction effectively raise the bar for significance, which means you need more respondents to detect real differences.
Factors That Decrease Required Sample Size
Stratified sampling. Stratification reduces within-group variance, which means you can achieve the same precision with a smaller total sample compared to simple random sampling. The gain depends on how different the strata are from each other.
Prior knowledge of p. If you know the true proportion is likely near 0.1 or 0.9 (rather than the worst-case 0.5), the required sample drops substantially. N for p = 0.1 at +/- 5% confidence 95% is 139, versus 385 for p = 0.5.
Relaxed precision requirements. If +/- 7% margin of error is acceptable for your decision (and it often is for early-stage concept tests), the sample drops to 196.
When to Use Sample Size Calculations
- Before fielding any quantitative survey: even a rough calculation prevents the two most common errors: too few respondents for the analysis you planned, or too many for the precision you needed
- During budget planning: sample size drives fieldwork cost, so the calculation informs whether a study is financially feasible at the desired precision
- When planning subgroup analysis: the overall sample might be fine, but specific segments may be too small. Run the calculation per segment.
- For longitudinal studies and trackers: each wave needs its own adequate sample, and detecting change between waves has its own power requirements
- When justifying methodology to stakeholders: "we surveyed 500 people" means nothing without context. "We surveyed 500 people, giving us +/- 4.4% margin of error at 95% confidence" is a defensible statement.
Common Mistakes to Avoid
- Picking a round number without running the formula. "Let's do 1,000" might be overkill or inadequate depending on the analysis plan. Run the math first, then round.
- Using the overall margin of error when subgroup analysis is the goal. n = 385 gives you +/- 5% for the total sample. If your smallest subgroup has 50 people, the margin of error for that subgroup is +/- 14%. That's rarely precise enough for any decision.
- Ignoring non-response in the invitation count. Calculating that you need 400 completes and sending exactly 400 invitations guarantees you'll fall short. Divide by the expected response rate to get the invitation target.
- Confusing sample size with data quality. A sample of 5,000 from a biased recruitment source isn't better than 500 from a well-designed sampling plan. Size doesn't fix bias.
- Not revisiting the calculation when the study design changes. Adding subgroups, switching from proportionate to disproportionate allocation, or changing the primary metric all affect the sample size requirement. Recalculate when the plan evolves.
How Quali-Fi Supports Sample Size Planning
Quali-Fi's platform includes built-in sample size calculation tools that take your confidence level, margin of error, and expected variability and return the required sample, with adjustments for finite populations and subgroup analysis. For studies using external panel respondents, the CINT integration shows estimated feasibility (how many respondents matching your criteria are available) alongside the calculated requirement, so you know before fielding whether your target is achievable.
For a quick calculation, try our sample size calculator.
Plan your sample size with Quali-Fi
Frequently Asked Questions
What's the minimum sample size for a survey?
There's no universal minimum, it depends entirely on the precision you need and the analysis you plan. For estimating a single proportion with +/- 5% margin of error at 95% confidence, 385 is the standard. For comparing two groups with a medium effect size at 80% power, about 64 per group (128 total). For qualitative research, 12-30 interviews typically reach saturation.
Does population size affect sample size?
Only when you're sampling a large fraction of the population. For populations above 20,000, the required sample size is essentially the same whether the population is 50,000 or 50 million. The finite population correction only matters when n/N > 0.05.
How do I calculate sample size for comparing two groups?
Use a power analysis. You need four inputs: significance level (usually 0.05), desired power (usually 0.80), the minimum difference you want to detect, and the expected standard deviation. For comparing proportions, Cohen's h provides the effect size measure. For comparing means, Cohen's d. Most statistical software includes power analysis modules, and free online calculators are widely available.
Should I adjust sample size for expected dropout or non-response?
Always. If your analysis requires 500 complete responses and you expect 60% of invited respondents to complete the survey, you need to invite at least 834 people (500 / 0.60). Build the non-response adjustment into your fielding plan, not your analysis plan.