Regression analysis in biological research: sample size and statistical power
Sharp, D.S.; Gahlinger, P.M.
Medicine and Science in Sports and Exercise 20(6): 605-610
1988
ISSN/ISBN: 0195-9131 PMID: 3237050 Document Number: 311217
Regression analysis is often used to demonstrate associations among variables believed to be biologically related. Failure to demonstrate a "significant" relationship may be due to two factors: 1) the variables are truly unrelated, or 2) a relationship exists but goes undetected due to inadequate statistical power. Investigators must consider the second possibility since failure to detect a statistically significant relationship is often taken as evidence for no biological relationship. These issues are addressed in the context of the interrelationship between four features common to all statistical methods: the size of effect or relationship worth detecting, the Type I (.alpha.) error, the sample size, and the Type II (.beta.) error. An example derived from published data relating morphological characteristics of muscle fiber type and isokinetic strength performance illustrates the practical significance of this dilemma.