In statistics, a pattern of data obtained from multifactorial analysis of variance (ANOVA) or log-linear analysis in which the effect of an independent variable or factor (1) varies across levels of another independent variable, or across combinations of levels of other independent variables or factors. When this occurs, variation in the dependent variable is not the result of a simple additive combination of the independent variables or factors. For example, suppose a well-practised skill is performed better in front of an audience than alone, whereas an unpractised skill is performed better alone than in front of an audience. In this case the independent variables or factors may be the audience condition and the level of practice, and the effect of the audience factor on the dependent variable varies across levels of the practice factor, leading to improved performance at one level of the practice factor (well practised) and worse performance at the other level (unpractised). In this case, performance is not a simple additive effect of audience condition and level practice. Interactions are notoriously difficult to interpret, and an interaction graph is usually helpful. A main effect, in contrast to an interaction effect, is a significant difference between two or more means; a two-way interaction is a significant difference between two or more differences between means (as in the above example, in which one difference is positive and the other negative); a three-way interaction is a significant difference between two or more differences between two or more differences between means, and so on, but after that it begins to become too complex to grasp or to depict graphically. Compare main effect.