7.5 Using ANOVA: a summary

Use ANOVA for null hypothesis significance testing whenever: you have categorical variables with more than two levels in your model; you have interaction terms in your model and also want to be able to interpret the main effects as the effect of that predictor on average across all levels of the other predictor; or both. You can also use it in some situations where neither of these conditions applies. Traditionally, ANOVA was the go-to mode of hypothesis testing for factorial experiments: that is, experiments with multiple IVs where every level of one IV appears in combination with every level of the other IV, and the questions are about main effects (whether each IV affects the outcome overall), and also interactions (whether each IV modifies the effect of the other IVs). There is no reason you can’t use ANOVA in data analysis for observational studies, but this is much less often done.