How to Determine the Minimum Number of Animals for Statistical Power

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-28

Quick Answer

• Calculate the minimum animal number using a formal power analysis that combines expected effect size, variability, significance level, and desired power before you submit an IACUC protocol.
• Use free, validated software such as G*Power or R packages to run the calculation, and report the input values and assumptions in your protocol so reviewers can verify the logic.
• A power calculation is only as reliable as the effect size and variance estimates you enter, so a poorly informed guess can still produce an animal number that is too low or too high.

Why Animal Number Justification Matters in Research

Animal research protocols face scrutiny from multiple directions. Institutional Animal Care and Use Committees (IACUCs) review proposed studies to ensure that the number of animals requested is the minimum needed to achieve scientifically valid results. Funding agencies, including the National Institutes of Health, expect grant applications to include rigorous experimental design and statistical justification for sample sizes. Journals increasingly require authors to report how they determined animal numbers and to follow reporting guidelines that make those decisions transparent.

The ethical foundation for this scrutiny is the principle of reduction, one of the three Rs of animal research. Reduction means using the fewest animals possible while still obtaining reliable data. Using too few animals wastes the lives of those animals because the study may fail to detect a real effect, producing inconclusive results that cannot be published or used. Using too many animals wastes resources and exposes more animals to procedures than necessary. A formal power analysis is the accepted method for finding the balance between these two failure modes.

Beyond ethics, there is a scientific argument for careful sample size planning. Studies with insufficient statistical power are more likely to produce false negatives, meaning the researcher concludes there is no effect when one actually exists. They are also more likely to produce inflated estimates of effect size when a result does reach statistical significance, a phenomenon sometimes called the winner's curse. These problems undermine the reproducibility of animal research and waste the investment of public and private funding.

The National Library of Medicine hosts a collection of research methods resources that cover experimental design and statistical analysis in biomedical research. These resources provide background on the principles that underlie sample size estimation and can help researchers understand the statistical concepts involved. The EQUATOR Network maintains a library of reporting guidelines that help researchers document their methods clearly, including the sample size justification, so that readers and reviewers can assess the validity of the study.

Core Principles of Statistical Power

Statistical power is the probability that a study will detect an effect of a given size when that effect truly exists in the population being studied. Power is expressed as a value between 0 and 1, with 0.8 being the most common target in biomedical research. A study with 80 percent power has an 80 percent chance of producing a statistically significant result if the assumed effect is real, and a 20 percent chance of missing it.

Power is determined by four interrelated factors. The first is the significance level, usually set at 0.05, which is the probability of declaring a result significant when no true effect exists. The second is the effect size, which is the magnitude of the difference or association the researcher expects to observe. The third is the variability of the measurements, often expressed as the standard deviation. The fourth is the sample size. If you know any three of these values, you can calculate the fourth.

The relationship between these factors is intuitive. Larger effects are easier to detect, so they require fewer animals. Less variable measurements are easier to interpret, so they also require fewer animals. A more stringent significance level, such as 0.01 instead of 0.05, requires more animals to maintain the same power. A higher power target, such as 0.9 instead of 0.8, also requires more animals.

Effect size is often the most difficult value to specify because it requires prior knowledge about the system being studied. Researchers can draw effect size estimates from published studies in the same field, from pilot experiments, or from a minimum clinically or biologically meaningful difference that the study is designed to detect. When no prior data exist, researchers sometimes use standardized effect size conventions, but these should be treated as rough starting points instead of precise values.

The variability estimate is equally important. If the researcher underestimates the standard deviation, the power analysis will produce a sample size that is too small. If the researcher overestimates it, the sample size will be larger than necessary. Pilot data, published values, and historical records from the same laboratory are all legitimate sources for variance estimates.

At a Glance

Choosing the Right Statistical Test and Design

The power analysis must match the statistical test that will be used to analyze the data. A study comparing two groups with a continuous outcome, such as body weight or blood pressure, typically uses a t test or analysis of variance. A study examining survival over time uses a log-rank test. A study looking at a binary outcome, such as whether an animal develops a disease, uses a chi-square test or logistic regression. Each of these tests has its own power calculation method.

The experimental design also affects the calculation. A completely randomized design assigns animals to groups independently. A paired design, such as comparing measurements before and after treatment in the same animal, reduces variability and therefore requires fewer animals. A factorial design that examines two or more factors simultaneously has different power considerations than a simple two-group comparison. Repeated measures designs, where the same animals are measured at multiple time points, require methods that account for the correlation between measurements from the same animal.

Researchers should decide on the primary outcome and the primary analysis before running the power calculation. The primary outcome is the one measurement that the study is designed to detect a difference in. Secondary outcomes may be of interest, but they should not drive the sample size calculation because adjusting for multiple outcomes complicates the analysis and inflates the required sample size.

The choice of a one-sided or two-sided test also matters. A two-sided test, which detects differences in either direction, is the standard choice in most biomedical research. A one-sided test, which detects differences in only one direction, requires fewer animals but is only appropriate when the researcher can justify that a difference in the opposite direction is impossible or irrelevant. Most reviewers expect a two-sided test unless there is a strong justification for a one-sided approach.

Step-by-Step Workflow for Power Analysis

Step 1: Define the Primary Outcome and Effect Size

Write a clear statement of the primary outcome and the smallest effect that would be scientifically meaningful. This effect size should come from published literature in the same field, from a pilot experiment, or from a clinical judgment about what difference matters. Record the source of the effect size estimate and the reasoning behind it.

Step 2: Estimate the Variability

Identify the standard deviation of the primary outcome in the population of interest. Published papers often report standard deviations or standard errors, and the standard error can be converted to a standard deviation if the group size is known. Pilot data from your own laboratory are valuable because they reflect your specific animal strain, housing conditions, and measurement procedures.

Step 3: Select the Significance Level and Power

Set the significance level at 0.05 unless there is a specific reason to use a different value, such as a correction for multiple comparisons. Set the power at 0.8 as a minimum, and consider 0.9 for studies that are expensive, difficult to repeat, or intended to inform a critical decision.

Step 4: Run the Calculation

Enter the values into a power analysis program. G*Power is a free, widely used desktop application that supports many common tests. R has packages such as pwr that perform similar calculations. Commercial statistical software also includes power analysis modules. The program will output the required sample size per group.

Step 5: Adjust for Expected Losses

Animals may be lost during a study due to unexpected death, illness, or technical failure. Add a margin to the calculated sample size to account for these anticipated losses. The size of the margin depends on the species, the procedures involved, and historical experience in your laboratory. A common approach is to increase the sample size by 10 to 20 percent, but the exact figure should be based on your own records.

Step 6: Document the Assumptions

Write a clear paragraph for the IACUC protocol that states the primary outcome, the expected effect size and its source, the variability estimate and its source, the significance level, the power, the statistical test, and the adjustment for losses. This documentation allows reviewers to understand the logic and to identify any unreasonable assumptions.

Software Options and Their Tradeoffs

G*Power is the most commonly used free tool for power analysis in animal research. It offers a graphical interface, supports a wide range of statistical tests, and produces clear output that can be copied into a protocol or grant application. It runs on Windows and macOS and does not require an internet connection once installed.

R is a free, open-source statistical programming environment with several packages for power analysis. The pwr package covers basic tests, while more specialized packages handle complex designs such as mixed models and survival analysis. R requires some programming knowledge, but it offers greater flexibility for unusual designs and produces reproducible analysis scripts that can be shared with collaborators or reviewers.

Commercial statistical packages such as SPSS, Stata, and SAS include power analysis procedures. These are appropriate for researchers who already use these packages for their main analysis, but they may require additional modules or licenses. The choice of software is less important than the correctness of the inputs and the appropriateness of the statistical test.

Online power calculators are convenient but vary widely in quality. Some are maintained by reputable institutions and provide clear documentation of their methods. Others are poorly documented and may use outdated or incorrect formulas. If you use an online calculator, verify that it provides a citation for its methods and that the results match a validated program for a simple test case.

Common Failure Patterns in Sample Size Justification

One common failure is using a standardized effect size convention without justification. Conventions such as small, medium, and large effects were developed for social science research and may not correspond to meaningful differences in a specific animal model. Reviewers increasingly reject protocols that use these conventions without supporting evidence.

Another failure is ignoring the difference between the number of animals and the number of independent observations. In some designs, multiple measurements are taken from the same animal, and these measurements are not independent. Treating them as independent inflates the effective sample size and produces an underpowered study. The power analysis should be based on the number of independent experimental units, which is usually the number of animals.

A third failure is failing to account for multiple comparisons. If the study tests several outcomes or several group comparisons, the chance of a false positive increases. The power analysis should either focus on a single primary outcome or adjust the significance level for the number of comparisons. Ignoring this issue produces a sample size that is too small to support the full set of planned analyses.

A fourth failure is using a pilot study to estimate effect size without recognizing the uncertainty in that estimate. Pilot studies are typically small, so their effect size estimates have wide confidence intervals. A sample size calculation based on a pilot effect size may be substantially wrong. Some researchers address this by using the lower bound of the confidence interval for the effect size, which produces a more conservative and larger sample size.

A fifth failure is submitting a protocol with no power analysis at all, often with a statement that the sample size is based on previous studies or on convention. IACUCs and funding agencies increasingly require a formal calculation, and the absence of one is a common reason for protocol revision or grant application criticism.

Records and Measurements to Keep

Maintain a laboratory notebook or electronic record that documents the source of every input to the power analysis. For the effect size, record the citation of the published study, the relevant data from that study, or the reasoning behind a minimum meaningful difference. For the variability estimate, record the citation, the pilot data, or the historical records that produced the standard deviation.

Keep the output from the power analysis software, including the date, the version of the software, and the input values. This output should be attached to the IACUC protocol or stored in the laboratory's electronic records. If the protocol is revised, record the reason for the revision and the new calculation.

Track actual losses during the study and compare them to the margin added in the power analysis. This record is valuable for planning future studies in the same model. If losses consistently exceed the margin, the margin should be increased. If losses are consistently lower, the margin can be reduced, which reduces the number of animals used.

Record the actual effect size and variability observed at the end of the study. These values can be compared to the assumptions used in the power analysis. Discrepancies between assumed and observed values are informative for future study planning and can be reported in the methods section of the published paper.

Welfare and Safety Context

The power analysis is a direct expression of the reduction principle in animal research ethics. A study with too few animals is ethically problematic because the animals used cannot yield a reliable answer, so their lives are wasted. A study with too many animals is ethically problematic because it exposes more animals to procedures than necessary. The power analysis is the tool that identifies the minimum number needed for a valid result.

The welfare of animals during the study is also affected by the quality of the experimental design. An underpowered study may need to be repeated, subjecting a second cohort of animals to the same procedures. A poorly designed study may produce ambiguous results that require additional experiments to resolve. Careful planning at the outset reduces the total number of animals used across the lifetime of a research program.

The National Institutes of Health requires that grant applications include a data management and sharing plan, and the agency has published a policy that describes the expectations for such plans. While this policy is not specifically about sample size, it reflects a broader expectation that research methods, including statistical design, be transparent and reproducible. The data management plan should describe how the data will be preserved and shared, which allows other researchers to use the data for their own power calculations and study planning.

Researchers should also be aware of the publication ethics context. The Committee on Publication Ethics publishes core practices that cover authorship, peer review, data handling, and misconduct. Transparent reporting of sample size calculations is part of responsible research conduct, and journals that follow these practices expect authors to describe their methods fully.

Professional Escalation Criteria

If a power analysis produces a sample size that is so large that it is not feasible within the available budget, animal supply, or housing capacity, the researcher should revisit the study design before increasing the number of animals. Options include reducing the variability through better measurement techniques, using a paired or repeated measures design, or reconsidering whether the effect size assumption is unnecessarily small.

If the researcher cannot find any published data or pilot data to support an effect size or variability estimate, the appropriate response is to conduct a small pilot study to gather preliminary data. The pilot study should be designed to estimate variability instead of to test the hypothesis, and the results should be used to inform the full power analysis.

If an IACUC reviewer or a funding agency reviewer questions the power analysis, the researcher should be prepared to explain the reasoning behind each input value and to provide the source documentation. If the reviewer requests a different approach, such as a Bayesian analysis or a different effect size, the researcher should engage with the request and revise the calculation if the criticism is valid.

If a study is completed and the observed effect size is much smaller than the assumed effect size, the researcher should consider whether the study was underpowered and whether the results are reliable. This situation may require consultation with a statistician to interpret the findings and to plan any follow-up work.

Frequently Asked Questions

What is the minimum acceptable power for an animal study?

The most common target is 0.8, meaning an 80 percent chance of detecting a true effect. Some confirmatory studies use 0.9, especially when the study is expensive or difficult to repeat. The choice should be justified in the protocol, and a lower value such as 0.7 is generally considered inadequate for a definitive study.

Can I use a power analysis from a published paper for my own study?

You can use the effect size and variability estimates from a published paper as inputs to your own calculation, but you should not copy the sample size directly. Your study may differ in the animal strain, the specific procedures, the outcome measurement, or the statistical test, all of which affect the required sample size.

What if I have no pilot data and no published estimates for my outcome?

You should conduct a small pilot study to estimate the variability of your primary outcome. The pilot study does not need to be large enough to test your hypothesis, but it should be large enough to provide a reasonable estimate of the standard deviation. You can also consult a statistician for guidance on study design.

How do I account for animals that die or are removed during the study?

Add a margin to the calculated sample size to cover expected losses. The margin should be based on your own historical records for the same species, strain, and procedures. A common starting point is 10 to 20 percent, but the exact value should reflect your experience.

Is it acceptable to use a standardized effect size like small, medium, or large?

These conventions are rough starting points and are generally not sufficient justification on their own. Reviewers expect an effect size that is tied to the specific biology of your model or to a minimum meaningful difference. If you use a convention, explain why it is appropriate for your study.

What is the difference between the number of animals and the number of observations?

The number of animals is the number of independent experimental units. If you take multiple measurements from the same animal over time, those measurements are correlated and are not independent. The power analysis should be based on the number of animals, not the total number of measurements.

Should I use a one-sided or two-sided statistical test?

A two-sided test is the standard choice and is expected by most reviewers. A one-sided test requires fewer animals but is only appropriate when you can justify that a difference in the opposite direction is impossible or irrelevant. Most researchers should use a two-sided test unless there is a strong reason to do otherwise.

Where can I find reporting guidelines for animal research methods?

The EQUATOR Network maintains a searchable library of reporting guidelines for health research. These guidelines help authors describe their methods, including sample size determination, in a transparent and complete way. Following an appropriate guideline improves the quality of the methods section and helps reviewers assess the study.

Using the Evidence

Related Articles

• Statistical Power And Sample Size
• Sample Size Calculation: Formulas and Practical Considerations
• Sample Size Calculation for Diagnostic Accuracy Studies: A Practical Guide
• Statistical Power: What It Is and Why It Matters in Research
• Ethical Guidelines for Animal Research in Psychology: A Practical Overview

References and Further Reading

• Research Methods Resources. National Library of Medicine.
• EQUATOR Network. EQUATOR Network.
• Core Practices. Committee on Publication Ethics.
• NIH Grants and Funding. National Institutes of Health.
• ORCID for Researchers. ORCID.
• Data Management and Sharing Policy. National Institutes of Health.
• NCBI Literature Resources. National Center for Biotechnology Information.
• PubMed. National Library of Medicine.

This article is educational and does not replace institutional policy, professional advice, or applicable safety and regulatory requirements.