Analyzing the Relationship between Dependent and Independent Variables in Marketing: A Comparison of Multiple Regression with Path Analysis

Abstract

Multiple regression models continue to be widely used in marketing. Within the regression framework, researchers have to grapple with and resolve several contentious issues. For example, multicollinearity, non simultaneous estimation of parameters, inherent measurement error in independent variables, absence of overall goodness of fit indices, and lack of compelling guidelines for adding and deleting model variables are some common estimation problems associated with this method. In the absence of universally acceptable guidelines, researchers often use judgment calls to deal with these issues. Employing ad-hoc approaches, in turn, compromises the potential usefulness of multiple regression models. In this paper, we position path analysis as a competing technique that can address in a relatively unambiguous way, many of the above mentioned limitations of multiple regression. We illustrate the superiority of path analysis by reanalyzing data from selected marketing studies that have used multiple regression models. To enable researchers use path analysis more frequently, we provide a technical appendix depicting use of the EQS software for re-estimating multiple regression models. We discuss several implications of our results and outline avenues for future research.