Fisher F Test with signifi cantly low values in small samples

Abstract

Testing for differences between groups is usually done with the Fisher-F test. In practice, it has been known to return unusually small, but significant values. This situation is seldom covered by theory, usually being considered a fault of the model or of the analysis. This paper looks at a real life situation where significantly low F-values were encountered, and presents the way these values were treated.

Keywords

• Fisher F test
• signifi cantly low F-values
• additivity
• omitted factors
• normal distribution
• small samples