Sample Size Calculation for Clinical Research: A Step-by-Step Guide + G*Power Tutorial

Introduction: Why Sample Size Matters in Clinical Research

Most projects hit the same wall at the planning stage of their research project. The question and the study design are decided and  the ethics application is written. And then someone asks the most crucial question, “ how many participants you plan to recruit”?

It’s a fair question, and guessing will not work here. A study with too few participants can miss an effect that was genuinely there. A study that over-recruits burns funding and participant time it never needed, and ethics committees do notice that. Sample size calculation gives you a number you can defend. What feeds into it: your study design, the effect size you expect, your alpha, your power, the primary outcome you actually care about, and the participants you will almost certainly lose along the way.

This guide walks through how the number is calculated, what power and alpha really mean, how to run the calculation in G*Power, and how to adjust for attrition.

What is Sample Size Calculation in Research?

Sample size calculation estimates how many participants you need for your results to be precise enough, and powerful enough, to answer your research question. It runs off your planned analysis. Picking a round number that sounds serious is not the same thing.

Do it before recruitment starts because that part is simply not negotiable. Once data collection is over, a sample that turned out too small can’t be rescued, and reviewers, ethics committees, and journals will all want to know where your number came from.

Say you are testing whether a new intervention lowers blood pressure more than standard care does. The calculation tells you how many patients you need in order to pick up a difference big enough to matter in clinic.

Method follows design here, so a cross-sectional survey and a randomized trial will not use the same approach. 

What is Power Analysis?

Power analysis is what connects your sample size to your chances of finding a real effect. Statistical power is that chance, expressed as a probability: if the effect is out there, how likely is your study to catch it?

Clinical studies are usually planned at 80% power. Read that as an 80% chance of detecting the effect, assuming it’s real and roughly the size you assumed. Some studies go to 90%, and that tends to happen when missing a true effect would be costly or dangerous, such as a trial of treatment for a serious condition.

What is Alpha Error?

Alpha (α) is the chance of a Type I error. You report an effect that isn’t actually there.

Nearly everyone uses α = 0.05, which is the same as a 5% significance level.

What is Beta Error?

Beta (β) is the chance of a Type II error, and it’s the opposite problem. The effect is real, and your study walks straight past it.

Beta and power come from the same place:

Power = 1 − β

Set β at 0.20 and you are working at 80% power.

Type I and Type II errors sit underneath almost every statistical decision you will make, not just this one. Medical statistics for beginners covers both alongside p values and confidence intervals.

Why Do Alpha and Power Affect Sample Size?

Ask for more power and you pay for it in participants. Moving from 80% to 90% means you want a better chance at the same effect, and that costs data. Tightening alpha from 0.05 to 0.01 pushes the sample up too, for a slightly different reason: you’re now demanding stronger evidence before you’ll call anything significant.

What is Effect Size in Sample Size Calculation?

Effect size is your estimate of how big the difference or association is going to be. Tests need it as an input for a practical reason: spotting a subtle difference takes far more data than spotting an obvious one.

When you’re comparing two means, the standard measure is Cohen’s d. The familiar benchmarks:

• 0.2 = small effect
• 0.5 = medium effect
• 0.8 = large effect

Treat those as conventions rather than rules. Plenty of “small” effects matter a great deal at the bedside.

Why Does Effect Size Matter?

The relationship works in one direction:

Smaller expected effect → larger sample required Larger expected effect → smaller sample required

Suppose your new drug is expected to drop systolic blood pressure by 10 mmHg against standard care. You’ll need a fraction of the sample you would need to prove a 2 mmHg difference.

So where should the number come from? Published studies, your own pilot data, a meta-analysis, or a difference that clinicians agree is worth having. What you shouldn’t do is run the logic backward, starting from the sample you can afford and reverse-engineering an effect size that produces it. That inflates your assumption and leaves the study underpowered before it even begins.

How Do You Calculate Sample Size in Research?

There is no single formula that covers every study. The formula follows the design, the primary outcome, and the statistical test you plan to use.

Study Design	Common Sample Size Approach
Cross-sectional study	Expected prevalence/proportion + precision
Case-control study	Expected exposure/odds ratio
Cohort study	Expected incidence/risk ratio
Randomized controlled trial	Difference between groups
Comparing two means	Mean difference + SD
Comparing two proportions	Proportion difference
Correlation study	Expected correlation coefficient
Regression	Number of predictors/events and model assumptions

Decide your statistical test and primary outcome before you calculate anything. Running the calculation first and choosing the test afterward gives a number that does not match your analysis.

Choosing between those designs is the decision that shapes everything downstream. Cross sectional study design in medical research covers what each design can and cannot answer.

Why the Primary Outcome Matters

Base the calculation on your primary outcome and primary hypothesis. Studies usually measure several outcomes, and each would generate a different number. Choosing whichever secondary outcome gives the smallest sample is a common way to end up with an underpowered study.

Sample Size for Cross-Sectional Study

For a cross-sectional study, sample size usually depends on how precisely you want to estimate a prevalence, rather than on detecting a difference between groups.

The inputs are:

• Expected prevalence
• Confidence level
• Desired margin of error (precision)
• Population size, when the population is small and finite
• Design effect, for cluster or multistage sampling
• Expected non-response

The standard formula is:

n = Z² × p × (1 − p) / d²

Where:

• n = required sample size
• Z = Z-score for the chosen confidence level (1.96 for 95%)
• p = expected prevalence, as a proportion
• d = desired margin of error

Sample Size Calculator for Cross-Sectional Study

An online calculator asks for exactly the same four things: expected prevalence, confidence level, margin of error, and population size where it applies. The math behind the box is the formula above.

Try it with realistic numbers. You expect a prevalence of around 30%, so p = 0.30. You want 95% confidence, which fixes Z at 1.96. And you’ll accept a margin of error of 5%, so d = 0.05.

n = (1.96² × 0.30 × 0.70) / 0.05² = 322.7

Round up and you’re recruiting about 323 people. Now tighten that margin of error to 3% and watch what happens: the requirement jumps past 800. Precision is usually the input that drives the number, not prevalence, which surprises people the first time they see it.

G*Power Tutorial for Sample Size Calculation

G*Power is a free program built for power analysis and sample size calculation. It handles t-tests, ANOVA, chi-square, correlation, and regression, which is why it ends up being the tool of choice for most student and early-career projects.

The analysis itself usually moves elsewhere once data collection starts. SPSS and STATA both run the tests you planned for here, and STATA can compute power directly.

Step-by-Step G*Power Tutorial

Step 1: Pick the statistical test that matches your primary analysis. Set “Test family” and “Statistical test” before you touch anything else, because every field below them changes depending on what you choose here.

Step 2: Under “Type of power analysis,” choose “A priori: Compute required sample size.” This is the one you want. It calculates the sample you need before the study starts, which is the whole point.

Step 3: Fill in your parameters. Which ones appear depends on the test, but you’ll usually see effect size, α error probability, power (1 − β), number of groups, number of measurements, and allocation ratio.

Step 4: Hit Calculate.

Step 5: Write down the total sample size and the output parameters, along with every single assumption you typed in. You’ll need them for your Methods section, and you will not remember them in three months.

A Short G*Power Example

Take an independent-samples t-test comparing two groups:

Effect size (d) = 0.5 α = 0.05 Power = 0.80 Two-tailed, allocation ratio = 1

G*Power gives you 64 per group, so 128 in total.

Change one input and that answer moves. Switch to a one-tailed test, use an unequal allocation ratio, or choose a different statistical test, and you’ll get a different figure each time. This is why your assumptions carry more weight than the number itself.

How to Adjust Sample Size for Attrition

Here’s the part people miss. The number your calculation produced is what you need sitting in the final analysis, not what you need to recruit. Participants drop out. They withdraw consent, move away, stop answering the phone, return half-empty questionnaires, or turn out to be ineligible after enrollment.

So adjust:

Adjusted sample size = Required sample size / (1 − expected attrition rate)

Say your calculation gave you 200 and you’re expecting 10% attrition.

200 / 0.90 = 222.2

You recruit roughly 223 participants.

Notice that this is not the same as adding 10% to 200. That gives you 220, and you end up three short of what you needed. When the goal is to protect the analyzable sample, divide.

Attrition is partly a data collection problem. REDCap reduces the half completed questionnaires and missing follow ups that quietly shrink your analysable sample.

Common Sample Size Calculation Mistakes

Sample size problems usually trace back to a fairly short list:

• Choosing a sample size arbitrarily
• Using the wrong statistical test
• Confusing confidence level with statistical power
• Choosing an unrealistic effect size
• Ignoring attrition or non-response
• Using a calculator without knowing what it assumes
• Calculating sample size after the data is already collected
• Reporting the final number with none of the assumptions behind it
• Borrowing a formula built for a different study design
• Ignoring clustering or design effects in complex sampling

Almost all of these disappear if you settle your primary outcome and your analysis plan before you open any calculator.

These are the errors that cost a project rather than a mark. Research mistakes IMGs make and how to avoid them covers the wider set, from design through to authorship.

How to Report Sample Size Calculation in a Research Paper

A reviewer should be able to sit down with your Methods section and reproduce your number. That means listing the primary outcome, the statistical test, the expected effect size or prevalence, alpha, power, the calculated sample, the attrition adjustment, and the final target.

It reads something like this:

Sample size was calculated for the primary outcome of mean change in systolic blood pressure using an independent-samples t-test. Assuming an effect size of 0.5, α = 0.05, and 80% power, 128 participants were required. Allowing for 10% attrition, the target sample was 143 participants.

Three sentences. That’s all it takes.

Frequently Asked Questions

Q1. How do you calculate sample size in research?

It depends on your study design, primary outcome, expected effect size or prevalence, alpha level, desired power, and required precision. Choose the statistical test first, then apply the matching formula or software.

Q2. What is a good sample size for clinical research?

There is no universal number. A good sample size is one that is justified statistically for your specific outcome and design.

Q3. What is the usual power used for sample size calculation?

80% is the standard. 90% is used when missing a true effect would carry a high cost.

Q4. What is the difference between alpha and beta error?

Alpha is a Type I error: detecting an effect that is not there. Beta is a Type II error: missing an effect that is there.

Q5. What is Cohen’s d used for?

Cohen’s d expresses the standardized difference between two means, and it is the effect size input for t-test-based calculations.

Q6. Can G*Power calculate sample size?

Yes. G*Power performs a priori power analyses for a wide range of statistical tests and reports the required sample size directly.

Conclusion

Sample size calculation is not about finding a number. It is about making sure your study has enough participants to answer its primary research question reliably.

The workflow stays the same across designs: define the research question → identify the primary outcome → select the statistical test → determine the effect size or prevalence → set alpha and power → calculate the sample size → adjust for attrition.

Before you start recruiting, walk through your assumptions with a statistician or an experienced researcher.If you are planning your first review and want guidance through protocol development, search strategy, screening, and manuscript preparation, the American Academy of Research and Academics offers mentorship and training for students and early career researchers. Get in touch to talk through your project and get support from your first research question through to submission.

Disclaimer:

Articles published by American Academy of Research & Academics are prepared by our team using information from direct experience, publicly available resources, and educational references. AI tools may be used to assist with drafting, proofreading, and formatting; however, all content undergoes review and approval before publication.
The information provided is intended for educational purposes only. Requirements, policies, and processes may change over time. Readers should consult official sources for the most current information.