Yonathan A. Arbel, law professor at University of Alabama, wrote a paper for Wake Forest Law Review titled Reputation Failure: The Limits of Market Discipline in Consumer Markets.
My favorite part of this paper is the concept of “Regression to the Extreme” in product reviews.
« If the sample of consumers who produce reputation is randomly selected, then we would expect the outliers to experience what statisticians call “regression to the mean,” i.e., the eventual balancing of outliers towards the mean of the group. Indeed, the regression to the mean will be impeded by sluggishness, but there is the possibility of self-correction over time with a randomly selected sample. Unfortunately, the selection of consumers is all but random. »
« Regression to the extreme is the propensity of reputational data to emphasize, rather than eliminate, outlier experiences over time. Internal motivations select against middling reviews because those reviews are based on experiences that are too ‘boring’ to generate the requisite sense of spite or gratitude that will overcome the costs of producing reputational information. Additionally, reciprocity norms would lead consumers to overly-represent positive experience, in hopes of receiving reciprocal reviews from sellers, and herding would tend to silence non-popular reviews that might betray the consumer’s lack of sophistication. If a bottle of French wine receives paeans, an individual consumer may be embarrassed to reveal that she did not like it… Lastly, financial incentives select against middling reviews because shilling and cherry-picking foster creation of extreme opinions. All these tendencies lead to ‘regression to the extreme’: the propensity of reputational data to emphasize, rather than eliminate, outlier experiences over time. »
« Product reviews consistently provide strong evidence of regression to the extreme. One might expect that most products sold on the market would follow some generalized bell-shaped (Gaussian) distribution: after all, very few products are really outstanding or truly atrocious. Instead, most reviews on a large variety of online platforms form a so-called “J-shaped distribution,” with most reviews amassed in the extremes. On Amazon, more than 72% of the products have an average rating of at least four stars. In Airbnb listings, the average rating is 4.5 stars. Studies repeatedly find that middling reviews are rare and that even products with an average rating of two or three stars have few middling reviews. Further evidence suggests this pattern is not unique to online settings but carries over to offline settings. »