How to Calculate Sample Size for a Cross-Sectional Study

Q: What formula is used for sample size in a cross-sectional study?

The Cochran formula is standard: n = Z2 - p - (1-p) / d2. Where Z = 1.96 for 95% confidence, p = expected prevalence (use 0.5 if unknown for maximum sample size), and d = acceptable margin of error (typically 0.05 for 5%). This gives the minimum sample size for estimating a proportion in a large population.

Q: What if I don't know the expected prevalence?

Use p = 0.5 (50%). This is the most conservative estimate because p - (1-p) is maximized when p = 0.5, giving the largest (most conservative) required sample size. This approach is recommended when there is no reliable prior estimate of prevalence from the literature.

Q: What is the design effect and when do I need it?

The design effect (DEFF) accounts for the reduced statistical efficiency of cluster or multistage sampling compared to simple random sampling. If you are using cluster sampling (e.g., sampling hospitals, then patients within hospitals), multiply your Cochran formula result by the DEFF typically 1.5 to 2.0 for medical surveys. You do NOT need a design effect for simple random sampling or systematic sampling of a single population.

Q: How do I adjust for non-response in sample size calculation?

Divide the calculated sample size by the expected response rate. Formula: n_adjusted = n / response_rate. If you expect an 80% response rate: n_adjusted = n / 0.80. For questionnaire surveys, expected response rates are typically 6085%. Always justify your assumed response rate based on similar studies or pilot data.

Q: Is the Cochran formula different for an analytical cross-sectional study?

Yes. The Cochran formula estimates sample size for descriptive studies (estimating prevalence). For analytical cross-sectional studies that test associations (e.g., comparing prevalence between groups, or testing correlation), you need a different formula based on the expected effect size and the specific test you will use. For comparing two proportions, use the two-proportion chi-square sample size formula. For correlation studies, use Fisher's Z transformation.

Cross-sectional studies including prevalence surveys, KAP studies, and needs assessments are among the most common research designs in medical theses and public health research. Yet sample size calculation for cross-sectional studies is frequently done incorrectly: wrong formula, wrong parameters, or missing adjustments. This guide covers the descriptive (Cochran) formula; for case-control study designs, see the dedicated guide on sample size for case-control studies. Below you will find three fully worked examples you can adapt for your own study.

What Is a Cross-Sectional Study?

A cross-sectional study collects data from a population at a single point in time (or over a short period that represents one snapshot). It is used to:

Estimate the prevalence of a condition, disease, risk factor, or behavior in a defined population
Assess knowledge, attitudes, and practices (KAP studies) among healthcare workers, patients, or the general public
Examine associations between exposures and outcomes at a single time point
Evaluate the quality of care or services at a health facility

Cross-sectional studies cannot establish causality (because exposure and outcome are measured simultaneously), but they are efficient, relatively inexpensive, and ideal for generating hypotheses and measuring disease burden.

Descriptive vs AnalyticalA descriptive cross-sectional study estimates prevalence. An analytical cross-sectional study tests associations between variables. The sample size formula differs between these two objectives this guide covers the descriptive (prevalence estimation) formula, which is the most commonly needed formula for medical theses.

The Cochran Formula Standard Formula for Cross-Sectional Studies

For descriptive cross-sectional studies where the primary objective is estimating a proportion (prevalence, percentage, frequency), the Cochran formula is the standard:

Cochran Formula Large or Unknown Population

n = Z2 - p - (1 p) / d2

n = required sample size Z = Z-value for desired confidence level (1.96 for 95% CI; 2.576 for 99% CI) p = expected prevalence or proportion (from literature or prior studies) (1 p) = also written as q; the complement of p d = acceptable margin of error / precision (most commonly 0.05 = +/-5%)

What if you don't know p?Use p = 0.5 (50%). This gives the maximum (most conservative) sample size because p - (1-p) is maximized when p = 0.5. The Cochran formula with p = 0.5 and d = 0.05 gives n = (1.962 - 0.5 - 0.5) / 0.052 = 384 the famous "384" used in many social science and public health studies when prevalence is unknown.

Z-Values and Margin of Error Reference

Confidence Level	Z-value (Z_+/-/2)	Most common use
90%	1.645	Exploratory studies, some health surveys
95%	1.960	Standard for most medical and public health research
99%	2.576	High-stakes research, drug safety studies

Margin of error (d): Most studies use d = 0.05 (+/-5%), meaning the true prevalence may differ from the estimated prevalence by up to 5 percentage points. For more precise estimates, use d = 0.03 (+/-3%) but this substantially increases sample size. For less precise estimates (acceptable in large heterogeneous populations), d = 0.10 (+/-10%) may be used.

Three Essential Adjustments

1. Finite Population Correction (FPC)

The Cochran formula assumes an infinitely large population. If your study population is small (generally < 10,000, and especially < 1,000), apply the finite population correction:

Finite Population Correction

n_FPC = (n - N) / (n + N 1)

n = sample size from Cochran formula N = total population size (the entire group you want to study) n_FPC = corrected sample size (will always be smaller than n)

The FPC only meaningfully reduces the required sample size when n/N > 5%. For large populations (N > 10,000), the correction is negligible and can be ignored.

2. Design Effect (DEFF) for Cluster Sampling

If you are using cluster sampling (selecting groups first, e.g., hospitals, clinics, schools, villages then individuals within those groups), the observations are not fully independent. Subjects within the same cluster tend to be more similar to each other than to subjects in other clusters. This clustering reduces statistical efficiency.

The design effect corrects for this: n_adjusted = n - DEFF

For simple random sampling: DEFF = 1.0 (no adjustment needed)
For cluster sampling in health facilities: DEFF typically 1.5 to 2.0
For community cluster surveys (WHO EPI method): DEFF = 2.0 is standard
If DEFF is not known, use 1.5 as a conservative estimate and cite the justification

3. Non-Response Adjustment

Not all contacted participants will complete your questionnaire or agree to participate. You must recruit more subjects than your minimum required sample to account for expected non-response:

Non-Response Adjustment

n_final = n_adjusted / expected_response_rate

Example: If you need 200 participants and expect 80% response rate: n_final = 200 / 0.80 = 250 participants to contact Typical response rates: 7590% for face-to-face interviews; 6080% for self-administered questionnaires

Three Fully Worked Examples

Worked Example 1 Prevalence Study

What is the prevalence of hypertension among adults aged 3060 in a city of 50,000 adults?

Set parameters: 95% confidence Z = 1.96. Expected prevalence from literature: 30% p = 0.30. Acceptable margin of error: +/-5% d = 0.05. Sampling method: simple random sample (DEFF = 1). Expected response rate: 85%.

Apply Cochran formula: n = (1.962 - 0.30 - 0.70) / 0.052 = (3.8416 - 0.21) / 0.0025 = 0.8067 / 0.0025 = 322.7 323

Finite population correction: N = 50,000 n/N = 323/50,000 = 0.65% less than 5%, FPC correction is negligible. Skip it.

Non-response adjustment: n_final = 323 / 0.85 = 380.0 380

Final sample size: 380 adults to be recruited

Worked Example 2 KAP Study with Unknown Prevalence

Assessing knowledge of COVID-19 prevention among healthcare workers at 12 hospitals (cluster sampling). No prior data on knowledge level.

Set parameters: 95% CI Z = 1.96. Unknown prevalence p = 0.50 (conservative maximum). Margin of error: +/-5% d = 0.05. Cluster sampling across 12 hospitals DEFF = 1.5. Expected response rate: 90%.

Cochran formula: n = (1.962 - 0.50 - 0.50) / 0.052 = (3.8416 - 0.25) / 0.0025 = 0.9604 / 0.0025 = 384.2 384

Apply design effect: n_DEFF = 384 - 1.5 = 576

Non-response adjustment: n_final = 576 / 0.90 = 640 640. Divide across 12 hospitals: 640 / 12 = ~54 per hospital.

Final sample size: 640 healthcare workers (approximately 54 per hospital)

Worked Example 3 Small Population (Finite Population Correction)

Prevalence of burnout among physicians at a single hospital with 180 medical staff. Expected prevalence 45% based on previous local study.

Set parameters: 95% CI Z = 1.96. Prevalence: p = 0.45. Margin of error: +/-7% d = 0.07 (wider margin acceptable for small population). No cluster sampling DEFF = 1. Response rate: 90%.

Cochran formula: n = (1.962 - 0.45 - 0.55) / 0.072 = (3.8416 - 0.2475) / 0.0049 = 0.9508 / 0.0049 = 194.0 194

Finite population correction: N = 180. n_FPC = (194 - 180) / (194 + 180 1) = 34,920 / 373 = 93.6 94. The small population reduced sample size from 194 to 94!

Non-response adjustment: n_final = 94 / 0.90 = 104.4 105

Final sample size: 105 physicians to be contacted (out of 180 total staff)

Sample Size for Analytical Cross-Sectional Studies

If your cross-sectional study has an analytical objective testing an association between two variables rather than just estimating prevalence the Cochran formula is NOT appropriate. You need a formula specific to your analysis:

Comparing two proportions (Chi-Square): Use the two-proportion formula with expected proportions in each group and desired power (usually 80%)
Correlation analysis (Pearson or Spearman): Use Fisher's Z transformation to determine n needed to detect a minimum correlation coefficient r
Multiple logistic regression: Use 1020 events per predictor variable as the rule of thumb

For these analytical objectives, use StatClinic's Sample Size Calculator, which covers all these scenarios with automatic parameter input.

Common Mistakes in Cross-Sectional Sample Size Calculation

Using p = 0.5 when a reliable prevalence estimate exists this unnecessarily inflates sample size. Use the best available estimate from published literature
Forgetting the design effect for cluster samples if you sample by cluster (hospitals, clinics, wards), ignoring DEFF severely underestimates the required sample
Not inflating for non-response this is the most common omission; always add at least 1015% for expected non-response
Using the wrong formula for an analytical study Cochran estimates sample size for prevalence estimation, not for testing associations between groups
Using a margin of error that is too wide d = 0.10 (+/-10%) gives a small sample but yields imprecise estimates that may not be publishable
Not citing the formula and its source in your methods section, always state the formula used, the source (Cochran 1977 or equivalent), and all parameter values with their justifications

Methods section wording template"Sample size was calculated using the Cochran formula: n = Z2pq/d2, where Z = 1.96 (95% confidence level), p = 0.35 (expected prevalence based on [citation]), q = 1 p = 0.65, and d = 0.05 (margin of error). The minimum required sample was 350. Accounting for a design effect of 1.5 (cluster sampling) and an expected 15% non-response rate, the final target sample size was 607 participants."

Frequently Asked Questions

What formula is used for sample size in a cross-sectional study?+

The Cochran formula is standard: n = Z2 - p - (1-p) / d2. Where Z = 1.96 for 95% confidence, p = expected prevalence, and d = acceptable margin of error (typically 0.05). This gives the minimum sample size for estimating a proportion in a large population. Apply finite population correction for small populations, design effect for cluster sampling, and inflate for non-response.

What if I don't know the expected prevalence?+

Use p = 0.5 (50%). This is the most conservative estimate because p - (1-p) is maximized at 0.5, giving the largest and therefore safest required sample size. With Z = 1.96 and d = 0.05, this gives the famous n = 384 used widely in survey research.

What is the design effect and when do I need it?+

The design effect (DEFF) corrects for reduced efficiency when using cluster or multistage sampling. If you sample hospitals and then patients within hospitals, multiply your Cochran result by the DEFF (typically 1.52.0). For simple random sampling from a single list, DEFF = 1 and no adjustment is needed.

How do I adjust for non-response in sample size calculation?+

Divide by the expected response rate: n_final = n / response_rate. Example: n = 300, expected response rate = 80% n_final = 300 / 0.80 = 375 subjects to contact. Justify your assumed response rate based on similar published studies or your pilot data.

Is the Cochran formula different for an analytical cross-sectional study?+

Yes. Cochran's formula estimates sample size for descriptive (prevalence) objectives. For analytical studies testing associations between groups, use the two-proportion formula (for Chi-Square analysis) or Fisher's Z transformation (for correlation). Use StatClinic's Sample Size Calculator for these calculations.

Need help analyzing your study?

Use StatClinic AI Statistical Assistant free sample size calculator, chi-square, Mann-Whitney, and all major statistical tools. Includes auto-generated methods text.

Use StatClinic AI Statistical Assistant

How to Calculate Sample Size for a Cross-Sectional Study: Complete Guide with Worked Examples

What Is a Cross-Sectional Study?

The Cochran Formula Standard Formula for Cross-Sectional Studies

Z-Values and Margin of Error Reference

Three Essential Adjustments

1. Finite Population Correction (FPC)

2. Design Effect (DEFF) for Cluster Sampling

3. Non-Response Adjustment

Three Fully Worked Examples

Sample Size for Analytical Cross-Sectional Studies

Common Mistakes in Cross-Sectional Sample Size Calculation

Frequently Asked Questions

Need help analyzing your study?