Elaboration likelihood model, endogenous quality indicators, and online review helpfulness

Abstract

Given strong influences of online customer reviews on consumer purchase decisions, identifying helpful reviews has received broad attention from practitioners and researchers. The elaboration likelihood model (ELM) has been adopted to explain the review feature–helpfulness link. However, when analyzing reviews from websites, existing studies tend to ignore that quality indicators such as length and readability are merely cues and have not circumvented endogeneity induced by unseen argument quality. Hence, we propose an extended ELM application to observational data on review helpfulness. We develop a research model that integrates relevant quality indicators and sentiment features based on a circumplex model of affect. To test our hypotheses, we use publically available review datasets from three platforms (Amazon.com, Drugs.com, and Yelp.com) and adopt an instrument-free method that allows for arbitrary correlations between unseen argument quality and multiple endogenous indicators. Our analysis shows that ignoring endogeneity would result in invalid effect size and hypothesis-testing. In addition to identifying effects of endogenous quality indicators on review helpfulness, we find asymmetric effects of positive and negative valence contingent on low or high arousal. By articulating conceptual pitfalls and illustrating empirical remedies, our study aims to be a prototypical example of performing ELM-grounded analyses of online customer reviews.