The Cohort Study Power Calculator calculates sample size and power for cohort studies using incidence rates, risk ratio, allocation, follow-up time, and significance level.
Report an issue
Spotted a wrong result, broken field, or typo? Tell us below and we’ll fix it fast.
About the Cohort Study Power Calculator
This tool computes statistical power for cohort designs that compare two groups: exposed and unexposed. Power is the probability of rejecting the null hypothesis when a true effect exists. In simple terms, higher power means a better chance of detecting a meaningful association if it is real.
You choose how effects are measured. For binary outcomes tracked over follow-up, you can model a risk difference, a relative risk, or an incidence rate ratio using person-time. The calculator uses normal or Poisson approximations that are standard in epidemiology and biostatistics.
The tool accepts common design choices: sample size per group or total sample size with an allocation ratio, expected event risks or rates, and the significance level. It reports power and, if selected, the sample size needed to reach a target power. It also displays message prompts about assumptions, such as independence between groups and the binomial or Poisson model used in the computation.
Formulas for Cohort Study Power
The calculator implements well-known approximations for two-group comparisons. It supports both risk-based and rate-based analyses. Below are the core expressions used to compute power or sample size, depending on the inputs you supply.
- Risk difference (two independent proportions): Let p1 be risk in the exposed group and p0 be risk in the unexposed group. With n1 and n0 participants, the standard error of the difference is SE = sqrt[p1(1 − p1)/n1 + p0(1 − p0)/n0]. The Wald Z under the alternative is |p1 − p0| / SE. Approximate power for a two-sided test at level α is Φ(Z − z1−α/2), where Φ is the standard normal cumulative distribution function.
- Relative risk on the log scale: Define RR = p1 / p0 and L = ln(RR). The variance of L is approximately Var(L) = (1 − p1)/(n1 p1) + (1 − p0)/(n0 p0). The Wald statistic is |L| / sqrt[Var(L)]. Power is again approximated by Φ(Z − z1−α/2).
- Unequal allocation: If r = n1 / n0, you may compute n0 from the chosen effect metric using the formulas above by expressing SE in terms of n0 and r. For example, under a risk difference, SE = sqrt[p1(1 − p1)/(r n0) + p0(1 − p0)/n0]. Solve for n0 to reach Z ≥ z1−α/2 + z1−β when planning for target power 1 − β.
- Incidence rate ratio with person-time: With counts C1 and C0 and person-time T1 and T0, the incidence rates are λ1 = C1/T1 and λ0 = C0/T0. For Poisson counts, Var[ln(λ1/λ0)] ≈ 1/C1 + 1/C0. The Wald statistic is |ln(IRR)| / sqrt(1/C1 + 1/C0) and power uses Φ(Z − z1−α/2).
- Pooled variance for planning: For sample size under the risk difference, a common planning approximation for equal allocation is n per group ≈ [z1−α/2√(2 p̄(1 − p̄)) + z1−β√(p1(1 − p1) + p0(1 − p0))]² / (p1 − p0)², where p̄ = (p1 + p0)/2.
These approximations are most accurate when expected counts are not extremely small, and when probabilities are not near 0 or 1. The tool switches to rate-based methods when you enter person-time data, which is often better for rare events.
How to Use Cohort Study Power (Step by Step)
Start by translating your study question into quantities the model can use. Decide whether you will compare risks or rates, whether your test is two-sided, and what difference is meaningful. Then set the Type I error level and any allocation constraints between exposure groups.
- Choose the effect measure that matches your design and outcome: risk difference, relative risk, or rate ratio.
- Specify the expected baseline risk or rate in the unexposed group based on prior data.
- Set the expected effect in the exposed group as a risk, relative risk, or rate ratio.
- Enter the planned sample size per group, or the total sample and allocation ratio.
- Pick a significance level α and whether the test is one-sided or two-sided.
After entering your inputs, review the power estimate and the sensitivity notes. If power is too low, adjust sample size, allocation, or effect assumptions and recalculate. Document the scenario that produces a feasible and justified result.
What You Need to Use the Cohort Study Power Calculator
Before you begin, gather a few practical numbers that reflect your study population and expected outcomes. These inputs should come from pilot data, literature, or credible expert estimates.
- Baseline risk in the unexposed group (p0) or baseline rate per person-time.
- Expected effect in the exposed group, entered as p1, a relative risk (RR), or a rate ratio (IRR).
- Planned sample size per group, or total sample with an allocation ratio r = n1/n0.
- Significance level α and sidedness (usually two-sided α = 0.05).
- Planned follow-up and expected loss to follow-up, or available person-time.
Ensure risks are between 0 and 1, rates are ≥ 0, and sample sizes are positive integers. Very low counts and extreme probabilities can reduce accuracy, so check edge cases with a rate-based approach or simulation when needed.
How to Use the Cohort Study Power Calculator (Steps)
Here’s a concise overview before we dive into the key points:
- Select the comparison type: risk difference, relative risk, or rate ratio.
- Enter baseline risk or rate for the unexposed group.
- Enter the expected risk or rate for the exposed group, or enter RR/IRR to auto-compute it.
- Enter sample sizes (n0 and n1) or total N with an allocation ratio.
- Set α and choose one-sided or two-sided testing.
- Optionally enter loss to follow-up or overdispersion to adjust effective sample or counts.
These points provide quick orientation—use them alongside the full explanations in this page.
Worked Examples
Preventive exposure with equal allocation (risk difference approach). Suppose an unexposed group has a 10% six-month risk of infection (p0 = 0.10). You expect the exposure to reduce risk by 30%, so p1 = 0.07 (RR = 0.70). You can enroll 2,700 participants total with 1:1 allocation, so n0 = n1 = 1,350. The standard error of the difference is SE = sqrt[0.07×0.93/1,350 + 0.10×0.90/1,350] ≈ sqrt[(0.0651 + 0.0900)/1,350] ≈ sqrt(0.0001149) ≈ 0.0107. The absolute difference is 0.03, so Z = 0.03 / 0.0107 ≈ 2.80. With a two-sided α = 0.05, z1−α/2 ≈ 1.96, and power ≈ Φ(2.80 − 1.96) = Φ(0.84) ≈ 0.80. Interpretation: with 1,350 per group, you have about 80% power to detect a 3 percentage-point reduction from 10% to 7%. What this means: your planned sample is adequate to detect the expected effect at the chosen α.
Unequal allocation with rates and person-time (rate ratio approach). Assume the unexposed rate is 2 events per 100 PY, and the exposed rate is 1.4 per 100 PY (IRR = 0.70). You expect 15,000 PY in the unexposed group and 7,500 PY in the exposed group. Expected counts are C0 = 300 and C1 = 105. The standard error for ln(IRR) is sqrt(1/C1 + 1/C0) = sqrt(1/105 + 1/300) ≈ sqrt(0.00952 + 0.00333) ≈ 0.113. The effect on the log scale is |ln(0.70)| ≈ 0.357, giving Z ≈ 0.357 / 0.113 ≈ 3.15. With two-sided α = 0.05, power ≈ Φ(3.15 − 1.96) = Φ(1.19) ≈ 0.88. Interpretation: the planned person-time yields high power to detect the 30% rate reduction. What this means: your follow-up targets are sufficient for a strong test of the rate ratio.
Accuracy & Limitations
These computations use large-sample approximations. They are reliable for many cohort studies, yet several conditions can affect accuracy. Be mindful of the model-to-data fit before relying on a single number.
- Very small counts or extreme risks (near 0 or 1) can bias normal approximations.
- Unequal or variable follow-up complicates risk-based methods; rate models may fit better.
- Misclassification of exposure or outcome reduces effective effect size and power.
- Clustering, matching, or repeated measures require design effects not covered by simple formulas.
- Overdispersion in event counts inflates variance beyond Poisson assumptions, lowering power.
Use pilot data, published estimates, or simulations to check the assumptions. When in doubt, run sensitivity analyses across plausible risks, rates, and sample sizes. Always report which formulas you used and how they map to your design.
Units & Conversions
Power depends on consistent units for risks, rates, and person-time. Mixing percentages with proportions or person-months with person-years will distort your result. Use this table to convert commonly used measures into the calculator’s expected units.
| Quantity | From | To | Conversion |
|---|---|---|---|
| Risk as a proportion | Percent (%) | Proportion (0–1) | Divide by 100 (e.g., 12% → 0.12) |
| Incidence rate scale | Per PY | Per 1,000 PY | Multiply by 1,000 (e.g., 0.02/PY → 20/1,000 PY) |
| Person-time | Person-months | Person-years | Divide by 12 (e.g., 6,000 person-months → 500 PY) |
| Follow-up duration | Days | Years | Divide by 365.25 (e.g., 730 days → 2.0 years) |
| Confidence level | α (two-sided) | CI level | CI = 1 − α (e.g., α = 0.05 → 95% CI) |
Check which format the calculator expects before entering values. For example, enter risk as 0.12, not 12, and verify whether your rate input is per PY or per 1,000 PY to avoid scaling errors.
Common Issues & Fixes
Most power hiccups come from unit mismatches or inconsistent assumptions. Watch for these patterns and correct them before finalizing your plan.
- Entering percentages instead of proportions: convert 10% to 0.10.
- Forgetting the allocation ratio: a 2:1 design needs r = 2, not r = 1.
- Using risk formulas for rate data: switch to person-time and the rate ratio method.
- Zero expected events in one group: add continuity adjustments or increase person-time.
- Ignoring loss to follow-up: reduce effective sample or person-time accordingly.
If power is lower than expected, adjust the sample size, extend follow-up, or revisit the expected effect size. Re-run the calculator after each change and document the assumptions tied to each scenario.
FAQ about Cohort Study Power Calculator
Is power the same as required sample size?
No. Power is the probability of detecting the effect with your current design. Required sample size is how many participants you need to reach a target power under the same assumptions.
Should I use a one-sided or two-sided test?
Use a two-sided test unless a one-directional effect is scientifically justified and a harmful opposite effect would be uninterpretable. Many journals expect α = 0.05 two-sided.
Can I handle matched or clustered cohorts with this tool?
Not directly. For matching or clustering, apply a design effect or an effective sample size that accounts for intracluster correlation, then recalculate power with adjusted inputs.
What if participants have varying follow-up times?
Use the rate ratio option with person-time. Summarize total person-time per group, estimate rates, and compute power with the Poisson-based formulas.
Glossary for Cohort Study Power
Power
The probability that a study will detect a true effect of the size specified by your assumptions at the chosen significance level.
Significance level (α)
The tolerated Type I error rate. It is the probability of a false-positive result when no effect exists, often set to 0.05.
Risk (incidence proportion)
The proportion of participants who experience the outcome during a defined follow-up period.
Incidence rate
The number of events divided by total person-time at risk, often reported per 100 or 1,000 person-years.
Relative risk (RR)
The ratio of the risk in the exposed group to the risk in the unexposed group, often summarized on the log scale for calculations.
Rate ratio (IRR)
The ratio of incidence rates between exposed and unexposed groups, frequently estimated under a Poisson model with person-time.
Allocation ratio (r)
The planned ratio of exposed to unexposed sample sizes, r = n1/n0, used to distribute a fixed total sample.
Design effect
A multiplier that inflates variance to account for clustering, matching, or other design features that reduce independent information.
References
Here’s a concise overview before we dive into the key points:
- OpenEpi: Sample Size for Two Proportions
- Dupont WD, Plummer WD. Power calculations for health studies
- Fleiss JL et al. Sample size calculations in clinical research
- Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic Research: Principles and Quantitative Methods
- OpenEpi: Sample Size for Poisson (Person-Time) Rates
- Hsieh FY, Lavori PW. Sample-size calculations for the Cox proportional hazards regression model
These points provide quick orientation—use them alongside the full explanations in this page.