MerQuR+Statistics Glossary

Statistics Glossary → Analysis of variance (ANOVA) table

Analysis of variance (ANOVA) tableconcept

A summary table that partitions the total variability into its sources (model, factors, residual) and gives, for each source, the sum of squares, degrees of freedom, mean square, F and p value.

The total variability in the dependent variable is measured by the sum of squared deviations from the mean (SS) and is split into an explained part (the model or the factors) and an unexplained part (residual, error). In each row the SS is divided by its degrees of freedom to give the mean square (MS); the F ratio is a source's MS divided by the residual MS. A large F means the variability explained by that source is greater than chance alone would produce, and the p value is taken from the F distribution.

The same layout is used not only for comparing groups but also in regression: the regression ANOVA tests the significance of the whole model, and R² = SS_model / SS_total can be read straight from the table. With several factors, the type of sum of squares (Type I sequential, Type III partial) changes the results in unbalanced designs. The F tests in the table rest on the assumptions that the residuals are normally distributed, independent and of equal variance.

An everyday example

Why do exam scores differ in a school? Suppose the total variation is 1000 units. Of these, 400 units are explained by which class a student is in, and 600 by individual differences within classes. The ANOVA table shows this split row by row and uses the F ratio to judge whether the between-class share is too large, compared with the within-class noise, to be put down to chance.

In MerQur

It appears as the 'ANOVA table + SEE' extra output in Multiple Linear Regression, as the 'Scale ANOVA' in Cronbach's Alpha analysis, as the 'ANOVA / variance components table' in the Intraclass Correlation and as the 'ANOVA + correlations + iterations' extra output in Nonlinear Regression. It is the main output of Two-Way ANOVA; the Nested Mixed Model gives the EMS table in every run, and Variance Components Analysis gives the expected mean squares table with the ANOVA (Type I) method.

Analyses where this concept appears

Two-Way ANOVAMultiple Linear RegressionNested Mixed Model (Nested LMM, Multi-Level Hierarchical Mixed Model)Cronbach's Alpha and Reliability AnalysisIntraclass Correlation Coefficient (ICC)Variance Components Analysis (VARCOMP)Nonlinear Regression

Related concepts

F testTypes of sums of squares (Type I, II, III)Coefficient of determination (R²) and adjusted R²