The Impact of eBay Ratings on Prices

The literature on markets with incomplete information begins with Akerloff's (1970) classic paper on markets in which buyers are uncertain about the quality of goods being exchanged. The result is a market in which lower quality products, 'lemons,' may drive out higher quality ones. Subsequent theoretical studies on the role of reputation in markets generally find a positive relationship between seller reputation and transaction price (Shapiro, 1983). Supporting empirical studies have been limited by a lack of good data until the emergence of online auctions like those provided by eBay, which provides a unique feedback mechanism yielding observable measures of seller reputation and handles a large volume of transactions.

With the success of eBay, a number of studies since 2000 tested the impacts of various measures of reputation on the likelihood of successful sales occurring and, especially , on the final prices for goods sold in online auctions. Though the studies have assessed the impacts of various potential influences (c.f., Stafford and Stern, 2002), they mostly focus on the importance of reputation on final bid price. ^[1] Reputation, defined as the 'current assessment of an entity's desirability as established by some external person or group of persons,' according to Standifird, Weinstein, and Meyer (1999), should play an important role in e-commerce consumer behavior because of the delayed gratification in the online marketplace (Standifird, 2001). Except for products that can be digitized and delivered by download, most purchases are transacted before receipt of the product.

It is straightforward to see the impact of reputation on the winning bid in an auction. Following Houser and Wooders (2000), assume for simplicity that either the seller delivers the good - with probability r - or does not, and that v _A is the auction winner's value of the good. Then the equilibrium bid of an auction winner, b*, equals the winner's expected value of winning the auction, rv _A . If the same item - with a certain value - is available off-line, then b* = rv _A - (v _C - p _C ) where v _C is the (certain) value of the same item when purchased off-line and p _C is the price of the off-line-purchased item. For more, see Houser and Wooders (2000). The uncertainty that develops from the time gap between order and delivery can be reduced by seller reputation as it provides the buyer with a level of guarantee that the transaction will be completed as requested .

Table 7-1 summarizes the results of various studies that have examined the impact of feedback on ratings. Past studies have yielded conflicting results:

• The number of positive feedback responses. Houser and Wooders (2000), Lucking- Reiley, Bryan, Prasad, and Reeves (2000), Standifird (2001) and Ba and Pavlou (2002) find significant positive effects, while Resnick and Zeckhauser (2001) find no effect. Because these studies use absolute numbers, they typically use logs or partition the data because, for example, increasing the number of positive feedbacks by 10 is unlikely to have the same effect for sellers with total feedback numbers of 10 and 1,000. Results appear highly sensitive to the way the data is specified.

• The number of negative feedback responses. Houser and Wooders (2000), Lee, Im and Lee (2000), Lucking-Reiley, Bryan, Pasad, and Reeves (2000), Standifird (2001), Ba and Pavlou (2002), Melnik and Alm (2002), and Melnik and Alm (2003), find significantly negative impacts of negative feedback, while Bajari and Hortacsu (2000), Kauffman and Wood (2000), Eaton (2002), Jin and Kato (2002), and Resnick, Zeckhauser, Swanson and Lockwood (2003) find no effect. Again, it appears that results are largely impacted by the way in which reputation data is entered.

• Net positive feedback. Bajari and Hortacsu (2000), Kauffman and Wood (2000), Dewan and Hsu (2001), McDonald and Slawson (2002), Melnik and Alm (2002), and Melnik and Alm (2003) find positive effects, while Ba and Pavlao (2002), Jin and Kato (2002), and Resnick and Zeckhauser (2002) find no effect. A problem with using net positive feedback as an indicator of reputation, however, is the uncertainty regarding exactly what is being measured. The number is affected by both the overall experience of the seller and the likelihood of having a good buying experience. In addition, over 99% of comments left at eBay are positive in nature leaving less than 1% as either negative or neutral (which is often viewed as negative by sellers) (Dellarocas, 2003). Thus, buyers may not perceive the variation in this statistic. Finally, while eBay provides this number, it is referred to as positive feedback, and so may be confusing to buyers.

Table 7-1: Prior research on impact of eBay feedback on winning bid price

Kauffman and Wood (2000) and Melnik and Alm (2002) also use the number of positives divided by net feedback as a measure of reputation, referring to this as a relative frequency of negative responses, and McDonald and Slaw (2002) use the number of negatives divided by the number of positives , though none find a significant impact on price. A problem is that in neither case is the denominator the total number of feedback responses, making it difficult to tell exactly what is being measured.

The Role of Product Type

One possible reason for the discrepant findings may lie in the product types studied by researchers. Most of these studies have examined standardized products (such as computers and disk drives ) or collectibles (such as coins and stamps). In recent years , businesses of all sizes have recognized the potential of using eBay as a channel for products of various types. As the focus of eBay has expanded beyond a collectibles' trading site, it becomes an even more valuable source for e-business researchers.

In past research, the product type(s) selected for study usually focused on the expense of the product (computers, PDAs, and gold coins), the volume of transactions (coins, Beanie Babies), and the likelihood for fraud (laptops) or manipulation (baseball card grade). While these criteria have helped researchers isolate factors related to reputation, the characteristics of the products chosen may influence the impact of reputation on the final bid price.

When analyzing potential purchases, consumers integrate information from both their visual and touch (haptic) senses in order to interpret a product's size and texture (Ernst and Banks, 2002). A distinctive limit for e-commerce transactions is the inability for consumers to gain information about a potential purchase by using touch. For commodity products such as books, CDs, or DVDs, haptic information is substantially less important than for heterogeneous products like clothing, and jewelry . In the latter case, the lack of haptic information leads to increased uncertainty in the evaluation of heterogeneous products (Peck and Childers, 2003).

Consider the impact of this increased uncertainty in the case of heterogeneous goods on the impact of reputation on the final bid price. Following Melnik and Alm (2003), assume a binomial probability distribution in which r is the probability that the delivered good is high quality - with value to the buyer of v _A ; (1-r) is the probability of a low quality good being delivered. For simplicity, assume the low quality value to the buyer is 0. Then b* = rv _A r. The impact of reputation on the winning bid price is clearly greater for heterogeneous goods.

^[1] Some studies (Cabral and Hortascu 2002, Eaton 2002, Jin and Kato 2002, Livingston 2002, Resnick and Zeckhauser 2002, and Bajari and Hortascu 2003) have shown that the more positive (or less negative) a feedback profile the greater the probability of sale.