Paul Resnick, Richard Zeckhauser,

Eric Friedman, and Ko Kuwabara

For buyers and

sellers alike, there’s
no better way to
earn one another’s
trust in online
interactions.

Reputation
Systems
T
he Internet offers vast new opportu- provided references, Better Business Bureaus tallied
nities to interact with total complaints, and past personal experience and per-
strangers. These interactions can be son-to-person gossip told you on whom you could
fun, informative, even profitable. rely and on whom you could not. Participants’
But they also involve risk. Is the standing in their communities, including their roles
advice of a self-proclaimed expert at in church and civic organizations, served as a valu-
expertcentral.com reliable? Will an unknown dot- able hostage.
com site or eBay seller ship items promptly with Internet services operate on a vastly larger scale
appropriate packaging? Will the product be the
same one described online?
Prior to the Internet, such questions were
answered, in part, through personal and
corporate reputations. Vendors
MARC MONGEAU

COMMUNICATIONS OF THE ACM December 2000/Vol. 43, No. 12 45

than Main Street and permit virtually anonymous reviewer receives). iExchange.com tallies and displays
interactions. Nevertheless, reputations still play a major reputations for stock market analysts based on the per-
role. Systems are emerging that respect anonymity and formance of their picks.
operate on the Internet’s scale. A reputation system col- Why are these explicit reputation systems so
lects, distributes, and aggregates feedback about partic- important for fostering trust among strangers? To
ipants’ past behavior. Though few producers or answer, it helps to first examine how trust builds nat-
consumers of the ratings know one another, these urally in long-term relationships.
systems help people decide whom to trust, First, when people interact with
encourage trustworthy behavior, and deter one another over time, the
participation by those who are unskilled history of past interactions
or dishonest. auction sites informs them about their
For example, consider eBay, the abilities and dispositions.
largest person-to-person online auc- enable trash to be Second, the expectation
tion site, with more than four mil- shuttled across the of reciprocity or retalia-
lion auctions active at a time: it tion in future interac-
provides limited insurance, and country and in the tions creates an incentive
buyers and sellers both accept for good behavior. (Polit-
significant risks. There are prob- process transmuted ical scientist Robert
lematic transactions to be sure. Axelrod calls this
Nevertheless, the overall rate of into treasures. the “shadow of the future”
successful transactions remains [2].) An expectation that
astonishingly high for a market as people will consider
“ripe with the possibility of large- one another’s pasts in future
scale fraud and deceit” as eBay [5]. interactions constrains
The high rate of successful trans- behavior in the present.
actions is attributerd by eBay to its Among strangers, trust is
reputation system, called the Feed- understandably much more diffi-
back Forum. After a transaction is cult to build. Strangers lack known
complete, the buyer and seller have past histories or the prospect of future
the opportunity to rate each other interaction, and they are not subject to a net-
(1, 0, or 1) and leave com- work of informed individuals who would
ments (such as “good transac- punish bad and reward good behavior. In
tion,” “nice person to do business some sense, a stranger’s good name is not at
with,” “would highly recom- stake. Given these factors, the temptation to
mend”). Participants have running “hit and run” outweighs the incentive to coop-
totals of feedback points attached erate, since the future casts no shadow.
(visibly) to their screen names, which Reputation systems seek to establish the
might be pseudonyms. Yahoo! Auc- shadow of the future to each transaction by creat-
tion, Amazon, and other auction sites ing an expectation that other people will look back
feature reputation systems like eBay’s, on it. The connections among such people may be sig-
with variations, including a rating scale of nificantly weaker than in transactions on a town’s
15, several measures (such as friendliness, prompt Main Street, but their numbers are vast in comparison.
response, quality product), and averaging instead of At eBay, for example, a stream of buyers interacts with
total feedback score. the same seller. They may never buy an item from the
Reputation systems have also spread beyond auction seller again, but by sharing their opinions about the
sites. For example, Bizrate.com rates registered retailers seller via the Feedback Forum, they construct a mean-
by asking consumers to complete a survey form after ingful history of the seller. Future buyers, lacking per-
each purchase. So-called “expert sites” (www.expertcen- sonal histories with particular sellers, may still base
tral.com and www.askme.com) provide Q&A forums their buying decisions on a sufficiently extensive pub-
in which self-proclaimed experts provide answers for lic history. If buyers do behave this way, the sellers’ rep-
MARC MONGEAU

questions posted by other users in exchange for reputa- utations will affect their future sales. Hence, they seek
tion points and comments. Product review sites (such to accumulate as many positive points and comments
as www.epinions.com) offer rating services for product as possible and avoid negative feedback. Through the
reviewers (the better the review, the more points the mediation of a reputation system—assuming buyers

46 December 2000/Vol. 43, No. 12 COMMUNICATIONS OF THE ACM

provide and rely on feedback—isolated interactions restaurants and Consumer Reports magazine for appli-
take on attributes of long-term relationships. In terms ance repair histories, systematically capture and dis-
of building trust, a boost in the quantity of informa- seminate feedback. For the most part, however,
tion compensates for a significant reduction in its reputations travel haphazardly through word of
quality. mouth, rumor, or the mass media.
For people trying to sell off, say, their old LP-record The Internet can vastly accelerate and add structure
collections, reputation systems might seem like a nui- to the process of capturing and distributing informa-
sance. But consider such an effort in a market with no tion. To post feedback, users need only fill out an
such system, and hence no obvious distinction online form; a mere mouse click is often enough.
between sellers in terms of, say, quality of goods and Where interactions are mediated electronically, objec-
shipping service. Buyers would be reluctant to pay full tive information about performance may be captured
prices given their uncertainty about the sellers’ quality automatically (such as delay from question to response
(such as whether they reveal scratches in the records at at an expertise site). The same technology facilitating
the time of sale). However, high-quality sellers would market-style interaction among strangers also facili-
be reluctant to accept discounted prices. Over time, tates the sharing of reputations that maintain trust.
high-quality sellers would desert the market. Eventu- Despite this promise, significant challenges remain
ally, only the lowest-quality sellers would remain, a in the operating phases of such systems: eliciting, dis-
dynamic economist George Akerlof memorialized as tributing, and aggregating feedback.
the “market for lemons” [1]. Eliciting feedback encounters three related prob-
Reputation systems can reverse this flow and lems. The first is that people may not bother to provide
“unsqueeze” a bitter lemon. With clear reputation feedback at all. For example, when a trade is completed
markers, low-quality sellers get lower prices, leaving a at eBay, there is little incentive to spend another few
healthier market with a variety of prices and quality of minutes filling out a form. That many people do so is
service. For example, sellers with stellar reputations a testament to their community spirit, or perhaps their
may enjoy a premium on their services; some users gratitude or desire to exact revenge. People could be
may be willing to pay for the security and comfort of paid for providing feedback, but more refined schemes,
high-quality services. Such premiums are observed in such as paying on the basis of concurrence with future
auctions in auctions of coins and computer chips on evaluations by others, would be required to assure that
eBay [3, 6, 7]. The benefits of informative reputation their evaluations are thorough.
systems return to buyers and to sellers, enabling the Second, it is especially difficult to elicit negative
old LPs to spin out the door. feedback. For example, at eBay, it is common practice
Ratings are not the only way to convey reputations. to negotiate first before resorting to negative feedback.
When agreeing to be rated is optional (such as when Therefore, only really bad performance is reported.
registering as a retailer at bizrate.com), doing so is Even then, fear of retaliatory negative feedback or
likely an indication of higher-quality services, even simply a desire to avoid further unpleasant interac-
before ratings are available. Other ways to indicate tions may keep a dissatisfied buyer quiet. In the end,
quality are to use one’s real name, rather than a pseu- because information about patterns of moderate dis-
donym, and to indicate on a Web site that one also content may remain invisible, buyers cannot shun the
has a physical store with its attendant overhead costs. sellers who foster such discontent.
To operate effectively, reputation systems require at Third is the difficulty of ensuring honest reports.
least three properties: One party could blackmail another, threatening to
post negative feedback unrelated to actual perfor-
• Long-lived entities that inspire an expectation of mance. At the other extreme, in order to accumulate
future interaction; positive feedback, a group of sellers might collaborate
• Capture and distribution of feedback about cur- and rate one another positively, artificially inflating
rent interactions (such information must be visi- their individual reputations.
ble in the future); and Distributing feedback, the second phase, poses its
• Use of feedback to guide trust decisions. own challenges. One is name changes. At many sites,
people choose pseudonyms when registering. If they
In the offline world, capturing and distributing register again, they might choose another pseudonym,
feedback is costly. Businesses often collect feedback effectively erasing prior feedback. Reputations can still
from consumers but tend not to publicize the com- have effects, since newcomers want to accrue positive
plaints. A few independent services, such as Zagat’s for feedback, and those with established reputations want

COMMUNICATIONS OF THE ACM December 2000/Vol. 43, No. 12 47

to avoid negative feedback. Game-theory analysis work in theory. Individuals shouldn’t be expected to
demonstrates that there are inherent limitations to the make the effort to provide evaluations; negative eval-
effectiveness of reputation systems when participants uations should be avoided completely; and vendors
are allowed to start over with new names [4]. In par- should be expected to develop sophisticated ways to
ticular, newcomers (those with no feedback) should manipulate and trick the system. Even if all reporting
always be distrusted until they have somehow paid were complete and honest, users would find it virtu-
their dues, either through an entry fee or by accepting ally impossible to utilize the torrents of information
more risk or worse prices while developing their repu- they receive on other participants, given the lack of
tations. Another alternative is to prevent name satisfactory summary statistics.
changes—either by using real names or preventing Despite their theoretical and practical difficulties,
people from acquiring multiple pseudonyms, a tech- it is reassuring that reputation systems appear to per-
nique called “once-in-a-lifetime pseudonyms” [4]. form reasonably well. Systems that rely on the partic-
A second difficulty in distributing feedback stems ipation of large numbers of individuals accumulate
from the lack of portability from system to system. trust simply by operating effectively over time.
Amazon.com initially allowed users to import their rat- Already, Internet-based reputation systems perform
ings from eBay. But when eBay protested vigorously, commercial alchemy. On auction sites, for example,
claiming its user ratings were proprietary, Amazon dis- they enable trash to be shuttled across the country
continued its rating-import service. Limited distribu- and in the process transmuted into treasures.
tion of feedback limits its effectiveness; the future casts We conclude with an allusion to democracy,
a shadow on only a single online arena, not on many. another theoretically flawed and practically chal-
Efforts are under way to construct a more universal lenged system that nonetheless appears to perform
framework. For example, virtualfeedback.com provides miracles. Were Winston Churchill, the World War II-
a rating service for users across different systems, but it era British prime minister, to comment on reputation
has yet to gain wide public acceptance. systems and building trust as he did on democracy
Finally, there is also potential difficulty in aggre- and government, he might say: “Reputation systems
gating and displaying feedback, so it is useful in influ- are the worst way of building trust on the Internet,
encing future decisions about whom to trust. Net except for all those other ways that have been tried
feedback (positives minus negatives) is displayed at from time-to-time.” c
eBay; other sites, including Amazon.com, display an
average. These simple numerical ratings fail to convey References
important subtleties of online interactions; for exam- 1. Akerlof, G. The market for ‘lemons’: Quality uncertainty and the market
mechanism. Quart. J. Econom. 84 (1970), 488–500.
ple, Did the feedback come from low-value transac- 2. Axelrod, R. The Evolution of Cooperation. Basic Books, New York, 1984.
tions? What were the reputations of the people 3. Bajari, P. and Hortacsu, A. Winner’s curse, reserve prices, and endogenous
entry: Empirical insights from eBay auctions; see www.stanford.edu/
providing the feedback? ~bajari/wp/auction/ebay.pdf.
As a solution to the ubiquitous problem of trust in 4. Friedman, E. and Resnick, P. The social cost of cheap pseudonyms; see
new short-term relationships on the Internet, reputa- www.si.umich.edu/~presnick/papers/identifiers/index.html.
5. Kollock, P. The production of trust in online markets. In Advances in
tion systems have immediate appeal; the participants Group Processes, vol. 16, E. Lawler, M. Macy, S. Thyne, and H. Walker,
themselves create a safe community. Unfortunately, Eds. JAI Press, Greenwich, CT, 1999; see also www.sscnet.ucla.edu/soc/f
aculty/kollock/papers/online_trust.htm.
these systems face complex challenges, many of which 6. Lucking-Reiley, D., Bryan, D., Prasad, N., and Reeves, D. Pennies from
yield no easy solutions. Efforts are under way to eBay: The determinants of price in online auctions; see
address these problems; for example, the Reputations www.vanderbilt.edu/econ/reiley/papers/PenniesFromEBay.pdf.
7. Houser, D. and Wooders, J. Reputation in auctions: Theory and evidence
Research Network (see databases.si.umich.edu/reputa- from eBay; see bpa.arizona.edu/~jwooders/ebay.pdf.
tions) represents a first step toward recognizing repu-
tation systems as a subject of study and as a vital asset Paul Resnick ([email protected]) is an associate professor at
the University of Michigan School of Information in Ann Arbor.
for the safety of online interaction environments. Richard Zeckhauser ([email protected]) is the
Internet-based reputation systems, like traditional Frank P. Ramsey Professor of Political Economy at Harvard
markets, aggregate vast amounts of information, University's John F. Kennedy School of Government.
which then significantly influences choices made by Eric Friedman ([email protected]) is an assistant
professor in the Department of Economics at Rutgers University in
businesses, as well as by individuals. The parallel may New Brunswick, NJ.
end there. The theoretical underpinnings of the effec- Ko Kuwabara ([email protected]) is a Ph.D. student at the
tive operation of markets are well understood, and the University of Michigan School of Information in Ann Arbor.
aggregation to a brief set of statistics, namely a single This work is supported in part by the National Science Foundation under grant IIS-
price for each item, proceeds automatically. 9977999.
Today’s reputation systems, by contrast, shouldn’t © 2000 ACM 0002-0782/00/1200 $5.00

48 December 2000/Vol. 43, No. 12 COMMUNICATIONS OF THE ACM