Analysis of Variance

Objective of Analysis of Variance

Analysis of variance is a classic method for studying experimental designs. The goal is to determine how the variation of one or several influencing variables affects one dependent variable. The dependent variable is scaled metrically, while classes must be present for the influencing variables (nominal scaling). The aim of such a study is to ascertain how class membership affects the level of the dependent variable.

One question of interest is whether each influencing variable will in itself trigger a significant variation of the dependent variable. In addition, mutual interactions are also revealed, or in other words, particularly strong influences due to specific combinations of influencing variables.

Prerequisites

The dependent variable must be scaled metrically, whereas classes are required for the independent variables. Should one of these variables be present in metric form, classes must be created.

Analysis of variance looks at the global influences of all existing variations of an independent variable. Hence it is possible for two variants (e.g. two product alternatives) to lead to significantly different results in the dependent variable (for instance, to differing levels of buying intention) even if this variable has no significant main effect. This can occur if the other variations lead to relatively similar results. If individual group differences are relevant, appropriate significance tests (post-hoc tests) should be used.