On this page
Statistical power is the probability that a study will detect a real effect if one truly exists. It determines whether a study has an adequate sample size and design to find what it's looking for. Low power means you might miss a real effect; high power means you're likely to find it if it's there.
- Power is the probability of detecting a real effect, not the probability that a finding is true.
- Power depends on four factors: effect size, sample size, significance level, and variability in the data.
- Larger sample sizes increase power, but the relationship is not linear, doubling the sample size does not double the power.
- A Type II error occurs when a study fails to detect an effect that actually exists, and power is the complement of this error rate.
- Low-power studies can produce published findings that fail to replicate, because they are more likely to miss real effects and more prone to noise.
When you read a research study, you want to know whether its findings are reliable. One critical measure of reliability is statistical power: the probability that the study will detect a real effect if one exists. Power answers a fundamental question about research design: does this study have enough participants, enough time, or enough precision to actually find what it's looking for?
Without adequate power, a study can fail to detect a real effect and conclude there is none. This failure, called a Type II error, is invisible in the published results. The study reports no difference, no effect, no signal. But the absence of a finding does not mean the effect is absent; it may mean the study was not sensitive enough to find it. Understanding power helps you judge whether a study's negative findings are trustworthy or whether they might simply reflect an underpowered design.
What statistical power actually is
Statistical power is the probability that a study will detect a real effect if one exists. It is a property of the study's design, not of its results. Before a study runs, power tells you the odds that your test will find something real if there is something real to find. After a study is complete, power tells you what those odds were, given the choices you made about sample size, measurement precision, and significance threshold.
Power is often expressed as a percentage or a decimal. A study with 80% power has an 80% chance of detecting a real effect. Most researchers aim for power of at least 80% when planning a study, though the target varies by field and by the cost of different types of error.
How statistical power works
Power is determined by the interplay of four factors. Changing any one of them shifts the probability that your test will detect an effect. Understanding these levers helps you see why some studies are more likely to produce reliable findings than others.
The four factors that determine power
Effect size is the magnitude of the difference or relationship you are trying to detect. A large effect, a dramatic difference between two treatments, or a strong correlation between two variables, is easier to find than a small one. If you are studying whether a new drug cures a disease, a 50% improvement is a large effect and easier to detect than a 5% improvement. Power is higher when you are looking for large effects.
Sample size is the number of observations or participants in your study. More data means more precision and more power. A study with more participants generally has more power than a study with fewer participants, all else equal. This is why large trials are more likely to detect real effects.
Significance level, often written as alpha, is the threshold at which you declare a finding statistically significant. A stricter threshold reduces power because it requires stronger evidence to reject the null hypothesis. A more lenient threshold increases power.
Variability in the data affects how clearly a signal stands out from the noise. If measurements are noisy or inconsistent, power drops because the effect is harder to distinguish. If measurements are precise and stable, power rises. Variability depends on the phenomenon you are studying and on your measurement method.
A worked example with invented numbers
Suppose you are running a study to compare two treatments for a condition. You measure a continuous outcome, like blood pressure or test score, and you want to detect a difference of 10 units between the two groups. You set your significance level at 0.05 and assume the standard deviation of the outcome is 15 units.
With 30 participants per group (60 total), you have a little better than even odds of detecting the 10-unit difference if it is real.
Now increase the sample size to 64 participants per group (128 total). Power rises substantially.
Increase again to 128 per group (256 total). Power climbs further.
Notice that power did not double when you doubled the sample size from 30 to 60 per group. This is because power follows a curve, not a straight line. Early increases in sample size produce large gains in power; later increases produce smaller gains. This is why researchers calculate sample size before a study begins: you want enough participants to achieve your target power, but not so many that you waste resources.
Why statistical power matters when reading research
When you read a published study, power tells you how likely the study was to find a real effect. A high-power study that reports no significant finding is credible: the researchers had a good chance of detecting an effect, and they did not find one. A low-power study that reports no significant finding is less credible: the researchers may have missed a real effect simply because their sample was too small or their measurement too crude.
Low power is also a reason that published positive findings sometimes fail to replicate. A low-power study that does find a significant effect may have gotten lucky: noise in the data randomly aligned with the hypothesis, producing a false positive. The effect looked real in the original study but disappears when other researchers try to reproduce it. High-power studies are more robust to this kind of noise.
When you evaluate a study, look for the sample size and the effect size the researchers expected to detect. Ask whether the sample is large enough to find that effect with reasonable power. If a study claims to have found a small effect in a small sample, be skeptical. If a study found nothing in a small sample, that null result is less trustworthy than the same null result in a large sample.
Common misreadings of statistical power
Power is often confused with other statistical concepts, and this confusion leads to wrong conclusions about what a study shows.
Power does not measure effect size. A study can have high power and detect a tiny effect, or low power and fail to detect a large one. Power tells you the probability of detecting an effect of a given size; it does not tell you how big the effect is. If a study reports a significant finding, you must look at the effect size separately to judge whether the finding is practically meaningful.
Power does not guarantee that a finding is true. A significant result from a high-power study is more likely to be real than a significant result from a low-power study, but it is not certain. Other sources of error, bias in measurement, confounding variables, selective reporting, can still produce false findings even in well-powered studies.
Power does not apply after a study is complete. Once you have data and a p-value, the question of power is moot. Power is a property of the study design, something you calculate before you run the study to decide whether your design is adequate. After the study, you have the results; power tells you what the odds were, but it does not change the results themselves. Some researchers mistakenly compute "post-hoc power" from the observed effect size and sample size, but this calculation is circular and uninformative. It will always be high if the study found a significant effect and low if it did not, regardless of whether the study was well designed.
How power connects to p-values and significance
Power and p-values are related but distinct. Understanding the difference clarifies what each one means.
A p-value is the probability of observing data as extreme as yours (or more extreme) if the null hypothesis is true, that is, if there is no real effect. A p-value of 0.03 means there is a 3% chance you would see data this extreme by random chance alone if the null hypothesis were true. P-values tell you about the data you actually observed.
Power is the probability that your test will produce a significant p-value (usually p < 0.05) if a real effect of a specified size exists. Power is about the test itself and what it is capable of detecting. It answers the question: if I run this study, what is my chance of finding an effect if one is really there?
A study can have a low p-value (a significant finding) and low power (a poor design). This happens when a study gets lucky: random noise produces a result that looks significant, even though the study was not sensitive enough to reliably detect the effect. Conversely, a study can have a high p-value (no significant finding) and high power (a good design). This means the researchers had a strong chance of finding an effect if one existed, and they did not find one, so the null hypothesis is credible.
Both power and p-values matter. Power tells you whether the study was capable of answering its question. The p-value tells you what the data actually showed. Together, they help you judge whether a finding is reliable.
Do this automatically
Calculating statistical power by hand requires working through probability distributions, effect size formulas, and sample size tables for each study design, a process that multiplies in complexity when researchers need to evaluate power across multiple scenarios or adjust parameters iteratively. Software can automate these calculations, allowing researchers to explore how changes in sample size, effect size, significance level, and study design affect the probability of detecting a real effect. Academia research runs this kind of analysis end to end and cites every figure back to its original source, across as many subjects as you need, with Mimir attaching a citation to every figure. It will not write your paper for you. It gathers and cites the evidence; the argument stays yours, and deliberately so. Every plan includes it, including the free one.
Start your first report and run it on something you already know well.
This article is general information and research commentary. It is not professional advice, and it does not account for your particular circumstances.
