Empirical versus theoretical power and type I error (false-positive) rates estimated from real murine aging research data.

We assess the degree of phenotypic variation in a cohort of 24-month-old male C57BL/6 mice. Because murine studies often use small sample sizes, if the commonly relied upon assumption of a normal distribution of residuals is not met, it may inflate type I error rates. In this study, 3-20 mice are resampled from the empirical distributions of 376 mice to create plasmodes, an approach for computing type I error rates and power for commonly used statistical tests without assuming a normal distribution of residuals. While all of the phenotypic and metabolic variables studied show considerable variability, the number of animals required to achieve adequate power is markedly different depending on the statistical test being performed. Overall, this work provides an analysis with which researchers can make informed decisions about the sample size required to achieve statistical power from specific measurements without a priori assumptions of a theoretical distribution.