Why Do People Rate? Theory and Evidence on Online Ratings

The rapid growth of online retail in the last decade has led to widespread use of consumer-generated ratings. This paper theoretically and experimentally identifies in- fluences that drive consumers to rate products and examines how those factors can create distortions in product ratings. By manipulating payoffs and effectively “de- activating” either the buyer or seller side of an artificial laboratory market, raters’ behavior is decomposed into buyer-centric and seller-centric components. The cost of providing a rating also plays a major role in influencing rating behavior, with high and low quality sellers being rated more often than those of moderate quality.

Introduction. Internet commerce is a large and rapidly growing component of the economy. Internet retail accounted for $224.3 billion in sales for 2012, up 16.2% from 2011. Typical growth over the past decade has been even higher, averaging approximately 20% annually. Online retail’s share of total U.S. retail has also increased tremendously over the past decade, climbing from just over 1% in 2001 to 5.2% by the end of 2012.1 The rapid growth and popularity of internet retail is not surprising. Virtually any good can be purchased on the internet, in every model, style, or color produced. The enormous selection offered to consumers means that they must often choose between several goods with similar observable characteristics but potentially different levels of quality. Without firsthand experience, it may be difficult or impossible for consumers to tell which of several similar-looking products is of the highest quality. In an effort to alleviate this problem and to encourage sales, many internet retailers provide customer-based rating and review systems for their products.

Discussion / Conclusion. What can we learn from these findings, and how can they be applied? Understanding why people rate may help to improve the design of future rating systems. Different systems may affect ratings’ ability to accurately reflect product quality. For example, consider a product which is of acceptable quality but has some small probability of failure. In the face of even a small cost, consumers who receive a functional, though unremarkable, product would be unlikely to provide a rating. However, the small number of consumers who do have a bad experience would be likely to provide negative ratings for the product. This would lead the product to have inaccurately poor ratings. As a simple example, consider a product which is generally of moderate quality, but occasionally fails utterly. If 80% of consumers receive a product of quality 5 and 20% receive quality 0, the average rating in the presence of a cost will be 3.07, compared with 3.6 when rating is free. Designers of rating systems should pay special attention to minimizing any costs which may discourage consumers from providing ratings. Given that the very act of providing a rating may be burdensome, designers may want to provide small incentives to buyers for rating products. A small discount on future purchases, as in Avery et al.