Choosing the right sample size in medical research is one of the most important decisions a student can make when designing a study. A sample that is too small may fail to detect a real effect, while a sample that is too large wastes time, money, and exposes more participants to unnecessary risk. This article explains what sample size means, why it matters, and how to calculate it using power, inputs, and practical examples you can apply directly to your own research projects.
In medical research, the sample size is the number of participants, patients, or observations included in a study. It directly influences the precision of your estimates and the likelihood that your results reflect the true population effect, not random chance.
Getting the sample size right is not a formality. Ethical review boards and grant reviewers expect a clear justification, and a poor choice can undermine the entire study before it begins.
Every sample size calculation depends on a few core inputs. Understanding these inputs helps you plan a study that is both scientifically valid and feasible.
You need all of these BEFORE you collect data. If you change the input values after seeing results, you are no longer planning a study — you are rationalizing it.
Power is the probability that your study will detect a true effect if it exists. It is one of the most misunderstood concepts in medical research, but it is central to sample size calculation.
Think of power as your ability to catch a real signal. If an intervention truly improves blood pressure by a certain amount, power tells you the chance your study will show that effect as statistically significant.
“Power is not about proving that your hypothesis is right. It is about giving your study a realistic chance to observe the truth.”
Higher power means you need a larger sample. For example, moving from 80% to 95% power will increase the required sample size substantially, even if every other input stays the same.
In practice, many researchers accept 80% power because it balances scientific rigor with resource constraints. However, for serious medical decisions, 90% or higher is often preferred.
The effect size is the single most important input, and also the hardest to choose. It is not the effect you hope to see — it is the smallest effect that would be clinically meaningful for patients.
One common mistake is to pick a very large effect size to keep the sample small. This makes the study underpowered for realistic effects, and if the result is not significant, you still cannot conclude that the treatment does not work.
“A study with low power is not just inefficient — it is misleading, because a null result can be interpreted as ‘no effect’ when the study was never able to detect one.”
You do not need to compute everything by hand. Several software tools and online calculators are widely taught in medical research courses.
| Tool | Best For | Examples |
|---|---|---|
| G*Power | Free desktop software for t-tests, ANOVA, regression | Comparing means between two groups |
| R (pwr package) | Flexible scripting and reproducible calculations | Mixed models and custom designs |
| OpenEpi | Online calculators for proportions and cohort studies | Estimating prevalence |
| PASS | Commercial software with extensive options | Clinical trial designs |
Regardless of the tool, you should understand the formula behind the calculation. The most common formula for comparing two means includes alpha, power, the standard deviation, and the difference between means. For two proportions, the formula uses the event rates in each group.
Suppose you want to test whether a mobile health app reduces systolic blood pressure compared with usual care. You expect the standard deviation of systolic blood pressure in your population is about 15 mmHg. You want to detect a meaningful difference of 5 mmHg between groups, with 80% power and a two-sided alpha of 0.05.
Using a standard two-sample t-test calculation, each group would need approximately 71 participants. Because you expect 15% dropout, you must recruit about 84 participants per group.
If you instead decided that a 10 mmHg difference is the smallest meaningful effect, the required sample size drops to about 18 per group. This shows how directly effect size influences sample size in medical research.
Always include your calculation assumptions in the methods section of your paper. This lets others judge whether your study was adequately powered.
Now consider a study where you are testing a new educational intervention to reduce medication errors. You expect that the error rate in the control group will be 20%, and you want to detect a reduction to 10% in the intervention group.
With 80% power and alpha 0.05, a two-proportion test would require roughly 199 participants per group. This is much larger than the mean comparison example because the effect is expressed as a difference in percentages, and variability is built into the proportions themselves.
If you lower the expected control event rate to 10% and still want to detect a reduction to 5%, the sample size per group becomes about 435. Rare events require very large samples to detect meaningful differences.
This is why many medical studies with rare outcomes use designs like case-control studies or meta-analyses to gain statistical power without huge recruitment costs.
All studies lose participants. People move away, withdraw consent, or have missing outcome measurements. Ignoring this in sample size planning leads to a final sample that is smaller than intended, which reduces power.
When you report your study, use a CONSORT diagram or similar flowchart to show how many participants were randomized, how many dropped out, and how many were included in the analysis.
A sample size justification is a required part of most research protocols. It should be specific enough that another researcher could reproduce your calculation exactly.
Write your primary outcome clearly, then state the value of alpha, power, the expected effect size, and the source for any standard deviation or event rate. Also mention the software and version used for the calculation.
If this is a preliminary or pilot study, you may not need a formal sample size calculation. Instead, describe how the study will produce estimates that inform a future larger trial. That is a completely valid approach, but it must be stated honestly.
For confirmatory studies, an inadequate sample size is a common reason for rejection by ethics committees or journals. Treat the sample size calculation as a serious scientific activity, not a box to check.
Sample size in medical research is not an arbitrary number. It is a balance between statistical power, available resources, and ethical responsibility to participants. By learning the key inputs — alpha, power, effect size, variance, and dropout rate — you can plan a study that is both rigorous and feasible. Work through your own examples using the formulas or software described here, and always write down your assumptions. A well-justified sample size is a sign of thoughtful study design, and it is one of the highest-impact skills you can develop as a medical research student.
Sample size is the number of participants in your study. Power is the probability that your study will detect a true effect if it exists. Larger samples generally give higher power, but power also depends on the effect size and the variability of the outcome.
In medical research, a p-value below 0.05 is commonly accepted as evidence against the null hypothesis. This threshold is a convention, not a law, but it is widely understood and used as a default in most sample size calculations.
It is the smallest difference in the outcome that would change clinical practice or make a difference for patients. It is not simply the effect you expect to see, but the effect you would not want to miss. It should be defined before starting the study.
Yes, pilot studies are useful for obtaining estimates of standard deviation or event rates. However, pilot studies often have small samples, so the estimates can be imprecise. Combine them with values from the published literature when possible.
Your study will have less power than originally planned. You could miss a true effect, or your confidence intervals will be wider than expected. Do not simply adjust your analysis after the data are collected; instead, plan carefully to avoid this situation.
No. Beyond a certain point, adding participants adds little statistical benefit but increases cost and time. Larger samples also increase the chance of finding statistically significant results that are too small to be clinically relevant. The goal is to recruit the minimum number needed to answer your research question reliably.
A two-sided test checks for an effect in either direction, meaning the intervention could be better or worse than the control. A one-sided test only looks for an effect in one direction. In most medical research, two-sided tests are standard because unexpected harms are possible.
For designs with multiple groups, such as analysis of variance (ANOVA), you need to specify the effect size as a measure of difference across all groups, often Cohen’s f. Software like G*Power includes options for one-way ANOVA and more complex designs.
In cluster randomized trials, groups of participants (like clinics) are randomized instead of individuals. This requires a larger total sample size because people within a cluster are more similar to each other, an effect measured by the intracluster correlation coefficient (ICC).
Yes. You should plan for a certain amount of missing data by inflating your sample size. If you expect 10% missing, multiply your calculated sample size by 1.11. For 20%, multiply by 1.25. Report this adjustment clearly in your protocol.
