Measuring interaction: An empirical comparison of three OLS regression models

Abstract

The capacity to correctly assess the existence of interaction is a high-value modeling capability among researchers of information systems (IS), especially those focusing on behavioural paradigm studies. Interaction is a notable aspect for the major theoretical frameworks of the IS field, particularly the adoption theories. Allowing for crossover effects in the Theory of Planned Behaviour resulted in improvements in model prediction (Taylor & Todd, 1995b). This study presents the trimmed model, which does not permit crossover effect relations among variables. In complex models, as mentioned by Pedhazur (1997), one variable may affect another variable indirectly through multiple paths. According to him, it stands to reason that indirect effects, through certain paths, may be more meaningful and/or stronger than others. The findings of this quantitative study lead one to conclude that crossover effect models are more capable of showing the interaction among models’ variables, as well as explaining the highest percentage of variation for a single dependent variable, in comparison to the full and trimmed models