Daniel Pyburn, Samuel Pazicni (University of New Hampshire, USA)
Interactions are important in determining whether the relationship between a predictor variable and a dependent variable differs across levels of a second predictor variable. However, testing for interactions often proves to be difficult. For example, researchers may be tempted to dichotomize continuous variables and perform an analysis of variance in search of an interaction. Problems exist in interpreting interactions using this method because dichotomizing variables may increase error variance or yield spurious results. Unfortunately, there are also issues with determining the presence of an interaction with multiple regression. The levels of predictors cannot be controlled in a quasi-experimental research situation, and any error in measurement of the independent variables will be compounded in the product term entered into the model. These matters will be discussed using data collected on the roles that comprehension ability and course performance have at varying levels of prior knowledge in a large lecture chemistry course.