MerQuR+Statistics Glossary

Statistics Glossary → F test

F testconcept

A test based on the ratio of two variance estimates (mean squares) that follows an F distribution under the null hypothesis.

The F statistic is the variance explained by an effect (the effect mean square) divided by the unexplained variance (the error mean square): F = MS_effect / MS_error. If the null hypothesis is true, both estimate the same variance and the ratio is close to 1; if the effect is real, the numerator grows. The statistic is compared with the F(df₁, df₂) distribution defined by the numerator and denominator degrees of freedom.

In ANOVA it tests whether group means are equal; in regression, whether all slopes are zero or whether a block of terms adds anything. With one numerator degree of freedom, F equals the square of the corresponding t statistic. The classical F test assumes independent, normally distributed errors with equal variances; when the variances differ, corrections such as Welch's ANOVA are used. The F test that compares two variances directly is very sensitive to non-normality.

An everyday example

The weight of loaves from three bakeries is compared. If the differences between the bakeries' means are small compared with the natural variation among loaves from the same bakery, F is close to 1 and no difference can be claimed. If the bakery means are spread out about five times as much as the within-bakery variation, F is about 5 and the difference may be significant.

In MerQur

In the One-Way ANOVA (classical or Welch), Two-Way ANOVA, ANCOVA and Mixed-Design ANOVA tables; in the model F of Multiple Linear Regression; and in the F tests based on expected mean squares in the Nested and Crossed Mixed Models.

Analyses where this concept appears

One-Way ANOVATwo-Way ANOVAMixed-Design ANOVA, Split-PlotANCOVAEffect Size CalculatorMultiple Linear RegressionNested Mixed Model (Nested LMM, Multi-Level Hierarchical Mixed Model)Crossed Mixed Model (Crossed LMM, Crossed Random Effects Mixed Model)Difference-in-Differences (DiD)Discriminant Analysis (LDA / QDA)Intraclass Correlation Coefficient (ICC)Weighted Linear Regression (Survey Regression)Weighted Logistic Regression (Survey Logistic Regression)

Related concepts

Analysis of variance (ANOVA) tableTypes of sums of squares (Type I, II, III)