# A Researcher's Guide to Power Analysis > Learn how to conduct a power analysis to determine the right sample size for your research. This practical guide covers effect size, alpha, and common pitfalls. Source: https://www.alfredscholar.com/blog/a-researchers-guide-to-power-analysis/ Published: 2026-10-30 Author: Alfred Scholar Team ## What Is Statistical Power, Really? Imagine you’re looking for a specific rare bird in a vast forest. Statistical power is the probability that your search plan (your study) will actually find the bird, assuming it’s really there. A low-power study is like searching for a tiny sparrow with blurry binoculars at dusk; even if the bird is perched right in front of you, you’re likely to miss it. An adequately powered study is like having clear binoculars, a good map, and searching at dawn. More formally, statistical power is the probability of correctly rejecting the null hypothesis when it is false. In simpler terms, it’s your study’s ability to detect a real effect. The standard convention is to aim for at least 80% power, which means you have an 80% chance of finding a statistically significant result if a real effect exists. This also means you accept a 20% chance of making a Type II error (a false negative), which is failing to detect an effect that is actually there. Conducting a power analysis *before* you collect data is a critical step in research design. It forces you to think through your study's logic and prevents you from wasting time and resources on a study that was doomed from the start. ## Why You Absolutely Need to Conduct a Power Analysis "How many participants do I need?" is one of the first questions a grant committee, ethics board, or supervisor will ask. A power analysis is the only defensible answer. Guessing or using an arbitrary number is a major red flag that suggests a lack of methodological rigor. Here are the two main reasons to conduct an *a priori* (before the fact) power analysis: 1. **To Avoid Underpowered Studies:** This is the most common statistical mistake. An underpowered study has a low probability of detecting a real effect. If you run an experiment with too few participants and get a non-significant result (p > 0.05), you’re left in an ambiguous position. Is there truly no effect, or was your study too small to detect it? An underpowered study that yields a null result is largely uninformative. 2. **To Avoid Overpowered Studies:** While less common, using a sample size that is too large is also problematic. It wastes resources, time, and money. More importantly, it can be unethical to expose more participants than necessary to an experimental intervention. Excessively large samples can also make tiny, practically meaningless effects statistically significant. A power analysis helps you find the "Goldilocks" sample size: just right. It’s the smallest sample you need to have a good chance of detecting an effect of a scientifically meaningful size. For a deeper dive into common statistical mistakes, check out our [guide to avoiding common statistical traps](/blog/a-researchers-guide-to-avoiding-common-statistical-traps/). ## The Four Ingredients of a Power Analysis A power analysis is an equation with four interconnected variables. If you specify any three, you can calculate the fourth. When planning a study, you typically set the first three to calculate the fourth: the sample size (N). ### 1. Effect Size (d, r, f, etc.) The effect size is the magnitude of the effect you expect to find. Are you looking for a huge, obvious effect (like the impact of a parachute on survival) or a small, subtle one (like a minor difference in reaction times)? This is the most important, and often most difficult, input to determine. You can't just guess. The best sources are: * **Previous Research:** Look for meta-analyses or similar studies in your field to get a realistic estimate. You can often calculate it from reported means and standard deviations in a published paper. * **Pilot Studies:** A small pilot study can give you a preliminary estimate of the effect size. However, be cautious: effect sizes from small pilots are often noisy and inflated. * **Minimum Practical Importance:** Determine the smallest effect size that would be practically or clinically meaningful. For example, what is the smallest improvement in test scores that would justify implementing a new teaching method? ### 2. Significance Level (α) This is your threshold for statistical significance, representing your tolerance for a Type I error (a false positive). It’s the probability of rejecting the null hypothesis when it's actually true. By convention, alpha (α) is almost always set to 0.05 (or 5%). ### 3. Statistical Power (1 - β) This is the variable we defined earlier: the probability of detecting a true effect. The conventional target is 0.80 (or 80%). This means you are accepting a 20% chance of a Type II error (β). For studies where a false negative would be particularly damaging (e.g., a clinical trial for a life-saving drug), researchers may aim for a higher power, such as 0.90 or 0.95. ### 4. Sample Size (N) This is the number of participants or observations in your study. In an a priori power analysis, this is the variable you are solving for. It’s the primary output you need to justify your research plan. ## How to Conduct an A Priori Power Analysis: A 5-Step Workflow Let's walk through the process of determining your sample size before you start your research. 1. **Choose Your Statistical Test:** The power calculation is specific to the statistical test you plan to use (e.g., a t-test, ANOVA, correlation, regression). You must decide on your analysis plan *before* you can calculate power. This choice is fundamental to your [research methodology](/blog/how-to-choose-a-research-methodology/). 2. **Determine Your Desired Power:** For most studies, 0.80 is the accepted standard. Use a higher value like 0.90 if the consequences of a false negative are severe. 3. **Set Your Alpha Level:** Stick with the convention of α = 0.05 unless you have a very strong, field-specific reason to change it. 4. **Estimate Your Expected Effect Size:** This is the hardest step. Do your homework. Dig through the literature for a justifiable effect size. Avoid using generic "small," "medium," or "large" labels from Cohen unless you have absolutely no other information to go on. 5. **Calculate the Required Sample Size:** Use software to plug in the values from the steps above. ## A Note on Tools: G*Power and Beyond You don’t need to do these calculations by hand. Several excellent tools are available. **G*Power** is a free, widely used program that covers a huge range of statistical tests (t-tests, F-tests, χ² tests, and more). To use it, you simply: 1. Select your test family and the specific statistical test. 2. Choose the type of power analysis: "A priori: Compute required sample size." 3. Enter your effect size, alpha, and power. 4. Click "Calculate." G*Power will instantly tell you the total sample size you need. It's an indispensable tool for any quantitative researcher. Other popular options include the `pwr` package in **R** and built-in functions in commercial software like **Stata** and **SPSS**. ## Common Pitfalls to Avoid Power analysis seems straightforward, but a few common mistakes can invalidate your results. ### The Myth of Post Hoc Power Never, ever calculate "post hoc" or "observed" power after your study is complete. This is a measure of the power to detect the effect size you found in your specific sample. It is mathematically redundant with your p-value and adds no new information. If your result is not significant, your post hoc power will be low. If it is significant, it will be high. Reviewers who ask for post hoc power are mistaken; the correct approach is to calculate confidence intervals for your effect. ### Using "Canned" Effect Sizes Blindly It’s tempting to plug in Cohen's d = 0.5 for a "medium" effect and call it a day. Don't do it. An effect that is "medium" in one field might be gigantic in another. The effect size you use must be justified based on the specific context of your research question and prior literature. ### Forgetting to Account for Attrition The sample size from your power analysis is the number of participants you need for your final analysis. But people drop out of studies. If you anticipate a 20% attrition rate, you need to inflate your recruitment target to compensate. To do this, divide your target sample size by the expected retention rate (e.g., if you need 100 participants and expect 80% retention, you need to recruit 100 / 0.8 = 125 participants). ## How to Report Your Power Analysis When you write your methods section, you need to describe your power analysis transparently. An independent researcher should be able to reproduce your calculation. State clearly: * That you conducted an *a priori* power analysis. * The software you used (e.g., G*Power 3.1). * The specific statistical test being powered. * The input parameters: the effect size (and how you justified it), the target power, and the alpha level. * The resulting required sample size. **Example for an independent t-test:** > "We conducted an a priori power analysis using G*Power 3.1 to determine the required sample size for an independent-samples t-test. Based on a similar study by Smith (2022) which found a medium effect size (Cohen's d = 0.55), we aimed to detect a similar effect. With an alpha of .05 and a desired power of 80%, the analysis indicated that a total sample size of 104 participants (52 per group) is required."