Questions concerning the numbers of animals to be used in certain studies frequently arise during the review of protocols. The number of animals to be used in the study must be adequately justified and should be the minimum needed.
Not infrequently, the number of animals is dictated by factors other than that predicted necessary to attain statistical significance, e.g., the amount or yield of tissue which can be extracted from each animal, the duration for which the tissue is viable, and the amount needed for each experiment. In such cases, this information should be detailed in the application.
Statistical methods have been developed for estimating sample sizes in order to assure that one will study enough subjects to (1) identify a difference if it does exist, and (2) determine that the difference will be statistically significant at the chosen level of certainty. Basically, one must estimate ("guess") the results of the study before one can determine how many subjects will be required. Sometimes a pilot study will be needed to obtain a reasonable estimate of the parameters. The statistical method to be used, the expected variance, and the chosen level of certainty must be included in the proposal where appropriate.
Many research protocols are simply testing a null hypothesis (i.e., that something occurs or does not occur). In such instances, the 95% confidence limit often can be determined by a non-parametric test using smaller group sizes (e.g., 10 control and 10 experimental animals). Again, in such cases, the relevant statistical data should be included in the proposal.
An alternative approach is to evaluate the data as they are being collected (e.g., Wald's Sequential Probability ratio test). This allows one to terminate the experiment as soon as a significant result is obtained.
In some experiments, a threshold for an "all or none" response is determined (e.g., studies calculating the ED50 or LD50 for a drug or radiation treatment). In such cases, sequential or staircase designs may provide equivalent information with fewer experimental animals than traditional multiple group designs. When feasible, the ARC encourages such designs. Additionally, the ARC encourages the use of endpoints other than lethality when possible (see ARC Policy on Death as an Endpoint).
As a starting point in determining animal numbers, investigators may find the following resources helpful:
- Sample Size Determination, Appendix A in Guidelines for the Care and Use of Mammals in Neuroscience and Behavioral Research, National Academies Press, Washington DC, 2003
- Estimating Animal Numbers, Appendix B in Guidelines for the Care and Use of Mammals in Neuroscience and Behavioral Research, National Academies Press, Washington DC, 2003
- Statistical Calculators, available on the Statistics Online Computational Resource website
- UCLA Biomathematics Department
- Statistical Considerations for Clinical Trials and Scientific Experiments, Harvard University
Approved 9/89; Revised 8/91, 1/26/04; Updated 10/11/04