Investigating the measures of relative importance in marketing research

Harvir S. Bansal

University of Waterloo

Philippe Duverger

Towson University


Empirical results in marketing research derived from multiple linear regression models are often susceptible to issues of dimensionality and multicollinearity. The current business environment is data rich and managers need the most parsimonious models to yield actionable results for the firm. Model parsimony is particularly important in customer satisfaction, customer loyalty and marketing research in general relying on tracking studies. For example, the question of which predictor is the main driver of retail loyalty or switching behaviour is crucial, given the firm’s limited resources. At the same time, researchers want to retain all variables in the models on the basis of theory, and study the evolutions of these parameters throughout repeated tracking studies, in order to inform managerial decisions over time. We first briefly motivate the need for a measure of relative importance in the context of marketing research, with an example taken from the customer retention/loyalty literature.