Would you agree if N is three? On statistical inference for small N

Area of Science:

• Primate Research
• Statistical Inference in Animal Studies

Background:

• Traditional non-human primate studies often use two or three animals.
• Previous work suggested using one animal for sample inference or five+ for population inference.
• A recent framework proposed using three animals and majority outcome for population representation.

Purpose of the Study:

• To evaluate the error rate of a proposed framework for non-human primate studies using three animals.
• To determine the framework's validity across various typicality (true probability) assumptions.
• To assess the inferential value of using two or three animals for population insights.

Main Methods:

• Analysis of a recently proposed framework for non-human primate study sample sizes.
• Testing the framework under diverse assumptions of the representative outcome's typicality.
• Utilizing conjunction analysis to assess inferential bounds from small sample sizes.

Main Results:

• The framework's error rate is highly dependent on the typicality of the outcome.
• Acceptable error rates necessitate very high typicality (87%), making larger sample sizes redundant.
• Reducing error rates by increasing sample size from one to three animals is most effective only for typicality values between 70-90%.

Conclusions:

• The use of two or three animals in non-human primate studies offers limited useful inference about the population.
• The proposed framework's validity is constrained by the typicality of the observed outcome.
• If small sample sizes (2-3 animals) are used, reporting the inferred lower bound of typicality is recommended.