Statistical power is the probability that your study will detect a real effect when one genuinely exists in the population. Put simply, it is your chance of correctly rejecting a false null hypothesis. Power is reported as a value between 0 and 1, and a study with low power can miss a true finding entirely, which is why it matters so much for the credibility of your dissertation.
What power protects you from in your dissertation
Every hypothesis test can go wrong in two ways. A Type I error is finding an effect that is not really there, controlled by your significance level. A Type II error is missing an effect that is genuinely present, and power is exactly the protection against it: power equals one minus the probability of a Type II error. If your thesis study is underpowered, a non-significant result is ambiguous, because you cannot tell whether the effect is truly absent or whether your study was simply too small to see it. High power lets you interpret a null result with confidence, which your committee values. You can work out the power of a planned sample directly, or solve the calculation in reverse for the n it would take.
How big should power be, and what the numbers mean
By long-standing convention, 0.80 is treated as the minimum acceptable power for most research, meaning an 80 percent chance of detecting a real effect of the size you expect. A power of 0.90 is stronger still, giving a 90 percent chance of detecting that effect and only a 10 percent risk of a Type II error. The right target depends on the cost of missing a true effect in your field, but for a typical dissertation aiming for 0.80 is the standard floor and 0.90 is a comfortable goal. Power is never a single fixed property of a test; it rises and falls with the inputs described below.
The three levers that move statistical power
Three things drive power, and you can reason about each before you collect data. The first is sample size: larger samples give more power because they pin down your estimates more precisely. The second is effect size, the true magnitude of the difference or relationship you are studying, explained in effect size explained; bigger effects are easier to detect. The third is your significance level, the threshold for your p-value, covered in what a p-value really tells you; a stricter threshold lowers power for a fixed sample. Because these levers are linked, you can solve for any one if you fix the others, which is the engine behind a formal power analysis and sample size calculation.
Statistical power and Type II error are two sides of one coin
Power and the Type II error rate are locked together: power equals one minus beta (β), where β is the probability of missing a real effect. A study with power of 0.80 carries a 20 percent chance of a Type II error, so β is 0.20. This is the mirror image of the Type I error rate, alpha (α), which is the chance of a false positive. The two error rates pull in opposite directions, since tightening α to guard against false positives lowers power and raises β unless you compensate elsewhere.
That trade-off is why you decide both rates before collecting data rather than discovering them afterwards. A common, defensible stance treats a missed true effect as four times less costly than a false alarm, which is how the conventional pairing of α = .05 and β = .20 arises. Framing your design around both errors, not just the familiar significance threshold from the p-value explainer, is what lets you read a non-significant result as informative rather than merely inconclusive.
How to increase statistical power in a study
When a planned study comes out underpowered, you have several practical moves before you abandon the question. The most direct is to increase the sample size, which sharpens every estimate and is the lever you control most cleanly; the sample size calculator shows how many more cases a target power demands. You can also reducemeasurement error with more reliable instruments and repeated measures, since cleaner data behaves like a larger sample by cutting the noise your effect must rise above.
Design choices help too. A within-subjects or repeated measures design removes between-person variation and often reaches the same power with far fewer participants, and a well-chosen one-tailed test gains power where the direction of the effect is genuinely known in advance. Finally, recruiting a sample where the effect is expected to be larger, or using a more sensitive outcome, raises the true effect size the test has to detect. Each route trades something away, so weigh them against your research question rather than reaching reflexively for more participants.
Reporting power the right way in your thesis
Power belongs in your methods chapter, not as an afterthought once results are in. The strongest approach is an a priori calculation: before collecting data, you state the effect size you expect, the significance level you will use, and the power you are targeting, then justify the sample size that follows. Avoid running a so-called post hoc power calculation based on the effect you happened to observe, because it merely restates your p-value and adds nothing. Choosing the matching test for your design also feeds the calculation, as set out in the guide to picking a statistical test. Framed and reported this way, power turns from a vague worry into a clear, defensible part of your dissertation design.