Revisiting Self-Selection Bias in E-Word- of-Mouth: An Integrated Model and Bayesian Estimation of Multivariate Review Behaviors

摘要

This paper studies the consumer self-selection bias in the e-word-of-mouth (eWOM) systems, e.g. consumer review websites. Under Bayesian framework, this study extends our understanding of this bias and discovers two new sources through developing a system of structural models of consumer review behaviors tested by a large data set. Our model and results provide evidences that the timing and content of a review introduce significant amount of bias into ratings in a simultaneous fashion. Specifically, we find that after controlling for various exogenous effects the two sources of bias persist: a subsequent rating is positively associated with the time interval between two consecutive reviews by the same consumer, and is negatively associated with the length of a review. Clearly, our findings confirm that modern eWOM systems have notable flaws despite of their mechanical advantages. We further discuss the possible mechanisms as well as the economic impact underlying these findings.