Mediation Analysis: Inferring Causal Processes in Marketing from Experiments

Abstract

Mediation analysis is applied to make causal inferences about the process that accounts for the effect that an intervention has on an outcome. Such an intervention can range from short-term tactics such as the content of a new on-line campaign or the size of a temporary price-cut, to far-ranging strategic interventions about customer loyalty programs, product introductions, brand extensions, retail chain mergers and so forth. Outcomes may involve any self-reported or observed state, trait, belief or action of consumers, managers, and firms. Mediation analysis is academically important because it enables tests of theories about causal processes, and it is policy relevant because improved insight into these causal processes might lead to more effective and efficient interventions. It has become an indispensable tool in the marketing researcher’s toolbox because of this hope for insight into the causal process, and because of foundational publications on mediation analysis, the development of statistical procedures to test for mediation, and because of the availability of these procedures in common statistical software (e.g., Baron and Kenny 1986; Hayes 2012, 2013; MacKinnon 2008; Preacher et al. 2007; Rucker et al. 2011; Shrout and Bolger 2002; Zhao et al. 2010). Mediation analysis is central in many academic disciplines. It is considered: