A Fake Review Sells More Copies
Who doesn't look at online ratings before booking a table at a restaurant, reserving a room in a hotel, or buying a book? Product and service ratings generated by users play a fundamental role in the business model of online platforms such as eBay.com and Amazon.com. Consumers often value information expressed by their peers more than messages relayed over traditional channels such as advertising or expert reviews.
In order to verify how sellers benefit from online reviews, Andrea Vittadini, a recent Bocconi graduate in Economics and Social Sciences, has gathered and analyzed data on a sample of books sold on Amazon.com and Barnesandnoble.com. He found that the number of copies sold is higher, when the number of reviews is larger, the average rating is higher, and positive reviews are more detailed.
By exploiting the peculiarities of ratings posted on Amazon.com the issue can be further explored. Every consumer can rate a product from one to five stars. The average rating of reviewers, rounded to the nearest half-star, is shown next to the product. The rating ticks up a notch when either the .25 or .75 threshold is surpassed. For instance, a book with an average rating of 4.74 will be given 4-and-a-half stars; while a book with a rating of 4.76 will have 5 stars placed next to it on the site. Limiting the analysis to Amazon.com, data show that a half-star increase has a positive and significant effect on sales. It could thus appear that the system of online ratings is beneficial to all actors involved.
However, each time the Web affords an opportunity for progress, we must be careful in looking for the other side of the story. Given the positive effects on sales and the ease with which reviews can be posted, isn't there an incentive to falsify them to turn higher profits?
In order to answer this question, we have verified whether the frequency of ratings is continuous around the half-star trigger values. When we only consider books with less than 50 reviews, such statistical test shows that the likelihood of observing an average rating which is slightly higher than the critical threshold is systematically larger than the likelihood of observing an average rating which is immediately below the threshold, as the following graph shows.
In other words, it really seems that are "too many" reviews just above the threshold causing the half-star increment. These are in fact anomalous reviews posted to obtain a half-star increase and thus sell more copies. The fact that this occurs only when the number of reviews is comparatively low, but not when the number of reviews exceeds 50, can be easily explained. Intuitively, it's relatively easy to manipulate the ratings of a book with few reviews by calling on friends and relatives of the author to rate it positively. It's a lot more expensive asking someone to write dozens of original reviews that are not filtered by Amazon's anti-spam system.