Seeing Stars: How the Binary Bias Distorts the Interpretation of Customer Ratings

Abstract

Across many different contexts, individuals consult customer ratings to inform their purchase decisions. The present studies document a novel phenomenon, dubbed "the binary bias," which plays an important role in how individuals evaluate customer reviews. Our main proposal is that people tend to make a categorical distinction between positive ratings (e.g., 4s and 5s) and negative ratings (e.g., 1s and 2s). However, within those bins, people do not sufficiently distinguish between more extreme values (5s and 1s) and less extreme values (4s and 2s). As a result, people's subjective representations of distributions are heavily impacted by the extent to which those distributions are imbalanced (having more 4s and 5s vs. more 1s and 2s). Ten studies demonstrate that this effect has important consequences for people's product evaluations and purchase decisions. Additionally, we show this effect is not driven by the salience of particular bars, unrealistic distributions, certain statistical properties of a distribution, or diminishing subjective utility. Furthermore, we demonstrate this phenomenon's relevance to other domains besides product reviews, and discuss the implications for existing research on how people integrate conflicting evidence.

How online word‐of‐mouth impacts receivers

“…When volume is high, higher dispersion attenuates the impact of positive and negative WOM on receivers’ product evaluations (Khare et al, ). Further, dispersion metrics, such as the histograms displayed on Amazon, can change how receivers interpret valence (Fisher, Newman, & Dhar, ). Receivers tend to categorize 5‐star histograms into binary positive or negative buckets, resulting in less sensitivity to valence extremity (e.g., 4 vs. 5‐star ratings) and more sensitivity to distributions that are imbalanced (i.e., contain more positive vs. negative ratings).…”

“…Receivers tend to categorize 5‐star histograms into binary positive or negative buckets, resulting in less sensitivity to valence extremity (e.g., 4 vs. 5‐star ratings) and more sensitivity to distributions that are imbalanced (i.e., contain more positive vs. negative ratings). Ultimately, this (biased) product signal affects receivers’ purchase intentions (Fisher et al, ).…”

“…Although receivers rely on other consumers’ ratings, these may not truly reflect review quality, product quality, or even sender traits (Brandes, Godes, & Mayzlin, ; De Langhe et al, ). Helpfulness ratings can be biased by prior ratings (Yin et al, ), and their interpretation can be influenced by receivers’ expectations (Quaschning et al, ; Yin et al, ) and processing biases (Fisher et al, ).…”

“…Yet, we conjecture that a platform's impact may be even broader than indicated by prior work. For example, visual design elements of WOM and other website content (e.g., product information), such as prominence (Babić Rosario et al, ), display type or order (e.g., Chen et al, ; Deng, Kahn, Unnava, & Lee, ), graphical presentations (Fisher et al, ), or fonts (Huang, Li, Wu, & Lin, ), might affect receivers’ processing or attention, thereby altering the impact of WOM.…”

Spillover effects from unintended trials on attitude and behavior: Promoting new products through access‐based services

“…It has been used for a variety of marketing and consumer behavior topics; for example, the evaluation of product bundles (Janiszewski & Cunha, 2004), the formation of and changes to brand attitudes as well as spillover effects that might be caused by brand alliances (Simonin & Ruth, 1998), and coalition loyalty programs (Schumann, Wünderlich, & Evanschitzky, 2014). Moreover, the impact of country‐of‐origin information and price on perceived product quality (Hastak and Hong (1991), the impact of online ratings on product evaluations and willingness to pay (Fisher, Newman, & Dhar, 2018), and the impact of online celebrity endorsement on purchase intention (Fink, Koller, Gartner, Floh, & Harms, 2018) have been assessed applying IIT. IIT has also served as a foundation for evaluating the effects of product trials on brand and product attitudes (Smith, 1993).…”