Options
Perception-Specific Average Causal Effects of Marketing Treatments
Type
fundamental research project
Start Date
01 December 2008
End Date
31 July 2009
Status
completed
Keywords
Rubin's Causal Model
treatment perceptions
finite mixture model
Cognitive Evaluation Theory
loyalty rewards
Description
In behavioral marketing research, it seems quite natural to assume that subjects' treatment perceptions moderate treatment effects. For example, Cognitive Evaluation Theory predicts that a loyalty reward increases the loyalty of consumers interpreting the reward as a sign of appreciation, and the reward decreases the loyalty of consumers interpreting it as a manipulative trick of the firm. At first glance, these hypotheses call for a trivial interaction analysis; treatment assignment and treatment perception interact in producing effects. However, two problems occur on closer inspection: First, treatment perception is inaccessible und thus completely missing in the control condition because the treatment is obviously absent. Second, treatment perceptions are influenced by a multitude of background variables. For example, Cognitive Evaluation Theory predicts that certain prior experiences with the firm shape consumers' reward perceptions, thereby influencing the reward's expected effect on loyalty. That is, the background variable's influence on the treatment effect (moderation) is mediated by treatment perceptions, which are unobserved in the control condition ("mediated moderation hypothesis" with partially unobserved mediator). Although the importance of treament perceptions is beyond question in applied research, a suitable statistical model has not yet been developed to meet the mentioned challenges. Based on "Rubin's Causal Model" and recent extensions of it, we propose a method to estimate "perception-specific average causal effects" (PACE) of a treatment, that is, a causal effect for subjects with positive treatment interpretation and a causal effect for subjects with negative treatment interpretation. The likelihood of positive versus negative treatment perception and thus the expected treatment effect are dependent on the level of background variables. Furthermore, the mentioned mediated moderation hypothesis can be tested. As we illustrate by means of a study on loyalty rewards, the method can reveal novel insights in applied consumer research with controlled experiments.
Leader contributor(s)
Herzog, Walter
Funder(s)
Topic(s)
Rubin's Causal Model
treatment perceptions
finite mixture model
Cognitive Evaluation Theory
loyalty rewards
Method(s)
Rubin's Causal Model
finite mixture model
Range
Institute/School
Range (De)
Institut/School
Division(s)
Eprints ID
56808