Knowing When to Draw the Line: Designing More Informative Ecological Experiments

Linear regression and analysis of variance (ANOVA) are two of the most widely used statistical techniques in ecology. Regression quantitatively describes the relationship between a response variable and one or more continuous independent variables, while ANOVA determines whether a response variable differs among discrete values of the independent variable(s). Designing experiments with discrete factors is straightforward because ANOVA is the only option, but what is the best way to design experiments involving continuous factors? Should ecologists prefer experiments with few treatments and many replicates analyzed with ANOVA, or experiments with many treatments and few replicates per treatment analyzed with regression? We recommend that ecologists choose regression, especially replicated regression, over ANOVA when dealing with continuous factors for two reasons: (1) regression is generally a more powerful approach than ANOVA and (2) regression provides quantitative output that can be incorporated into ecological models more effectively than ANOVA output.