American Sociological Review2014, Vol. 79(1) 65 85 American SociologicalAssociation 2013DOI: 10.1177/0003122413512316http://asr.sagepub.comSocial exchange has always been an important part of human sociality and arguably a driving force in the evolution of the human mind (Cos-midesandTooby1992).Inarchaicsocieties, socialexchangespannedanetworkofmulti-plex relations and was accompanied by elabo-rate ceremonies. A long tradition of research in anthropology and sociology on social exchange beganwiththeworkofMalinowskiand Mauss.Ceremonialexchanges,suchasthe Kula(Malinowski1922)orthepotlatch (Mauss [1950] 1990), were formal rituals char-acterized by solemnity, decorum, and disinter-estedgenerosity;theyoccurredsidebyside witheconomicexchanges,which,incontrast, wereguidedbyutilitarianmotives(Heath 1976).Socialexchangewasgovernedby normsofreciprocity(Gouldner1960; Malinowski 1926; Mauss [1950] 1990), and a 512316ASRXXX10.1177/0003122413512316American Sociological ReviewDiekmann et al.2013aETH ZurichbUniversity of BerncNuffield College and University of OxfordCorresponding Author:Wojtek Przepiorka, University of Oxford, Department of Sociology, Manor Road, Oxford, OX1 3UQ, United Kingdom E-mail: wojtek.przepiorka@sociology.ox.ac.ukReputation Formation and the Evolution of Cooperation in Anonymous Online MarketsAndreas Diekmann,a Ben Jann,b

Wojtek Przepiorka,c and Stefan WehrliaAbstractTheoreticalpropositionsstressingtheimportanceoftrust,reciprocity,andreputationfor cooperation in social exchange relations are deeply rooted in classical sociological thought. Todays online markets provide a unique opportunity to test these theories using unobtrusive data.Ourstudyinvestigatesthemechanismspromotingcooperationinanonline-auction market where most transactions can be conceived as one-time-only exchanges. We first give asystematicaccountofthetheoreticalargumentsexplainingtheprocessofcooperative transactions.Then,usingalargedatasetcomprising14,627mobilephoneauctionsand 339,517DVDauctions,wetestkeyhypothesesabouttheeffectsoftradersreputationson auction outcomes and traders motives for leaving feedback. Our statistical analyses show that sellers with better reputations have higher sales and obtain higher prices. Furthermore, we observe a high rate of participation in the feedback system, which is largely consistent with strong reciprocitya predisposition to unconditionally reward (or punish) ones interaction partnerscooperation(ordefection)andaltruismapredispositiontoincreaseonesown utilitybyelevatinganinteractionpartnersutility.Ourstudydemonstrateshowstrong reciprocity and altruism can mitigate the free-rider problem in the feedback system to create reputational incentives for mutually beneficial online trade.Keywordscooperation, trust, reciprocity, reputation, online markets66American Sociological Review 79(1)failure to reciprocate was punished by the loss ofreputationandstatus(Blau1964;Mauss [1950] 1990).Withtradetakingplaceovereverlonger distances, markets transcended the borders of smallcommunities,andhistoricalevidence showsthatsocialandeconomicinstitutions evolvedasasubstituteforinformalsocial control (Greif 1989; Zucker 1986). For exam-ple, trade in the early Middle Ages in Europe wascharacterizedbygeographicspecializa-tion,bookkeeping,andcashlesspayment. At that time, the Champagne fairs in France were ameetingpointfortradersfromallover Europe. Milgrom, North, and Weingast (1990) discusstheemergenceofaprivateadjudica-tionsystem(theLawMerchant)thathelped overcometrustproblemsamonganonymous traders.Administeredbyprivatejudges,it providedaplatformfortraderstosettledis-putes and document trading partners dishon-est behavior. Along with a system of notaries, suchaninstitutioncouldtrackanddissemi-nate information about traders past behavior sothatdishonesttraderscouldbedenied access to the market.More recently, similar information-collecting institutionshaveemergedtoovercometrust problemsbetweenmoneylendersandbor-rowers. Credit bureaus, for example, function as information brokers that collect and collate informationaboutborrowersliabilities, credithistories(inparticular,arrearsand defaults),andothercharacteristics(Japelli and Pagano 2002). Credit bureaus are private institutions,oftenoperatedbyagroupof lenders,andreportingofborrowerdatais voluntary.Toovercomethefree-riderprob-lem in data reporting, credit bureaus services arebasedonreciprocity;onlylenderswho submitaccurateinformationabouttheircus-tomers are granted access to a comprehensive customer database. This reciprocal information-sharingsystemcreatesfinancialincentives for lenders to contribute to the common good of customer information and for borrowers to maintainagoodreputationbytimelydebt repayment.JapelliandPagano(2002)show, moreover,thatsuchinformationsharingis indeedassociatedwithhigherlendingand lower default rates across 46 countries.Inasimilarvein,cooperationinpeer-to-peeronlinemarketsissustainedbyanelec-tronicreputationsystemthatcollectsand disseminatesinformationabouttraderspast behavior (Kollock 1999; Resnick et al. 2000). In online markets, tens of thousands of anony-mous buyers and sellers trade with each other everyday.Thesetradersarenotconnected throughasocialnetwork.Inmostcasesthey interactonlyonce(ResnickandZeckhauser 2002), andwhen cheatingthey are unlikely tobecaughtandfinedbyacentralauthority because many of these transactions transcend nationalborders(Przepiorka2013).Traders willingnesstosubmitfeedbackaftertransac-tions are finished is crucial for the functioning ofthereputationsystem.However,unlikein credit markets, information about traders rep-utations is a collective good freely available to anyone (Bolton, Katok, and Ockenfels 2004). Submittingfeedbackhasacostintermsof time and effort, so traders have no real incen-tivetosubmitfeedbackandasecond-order free-riderproblemmayexist(Heckathorn 1989).Assumingstrictrationality,actors would not rate their trading partners and, con-sequently, the whole market would degenerate or not evolve at all. How is it, then, that online markets are so successful and enjoy increasing popularity?Here we argue that other-regarding prefer-ences,suchasaltruism(Becker1976)and strongreciprocity(Gintis2000),playan important role in maintaining the provision of feedback in anonymous online markets. This feedback then generates the information nec-essary to create financial incentives for mutu-ally beneficial trade. Based on a large sample ofprocess-producedauctiondataobtained fromeBay(N350,000),wefirstcorrobo-rate previous studies findings that reputation has a market value, thus making it rational for traderstobuildandmaintainagoodreputa-tion.Weinvestigatetheeffectofreputation on the probability of sale and the selling price inahigh-andalow-costproductmarket (mobilephonesversusDVDs,respectively) Diekmann et al.67andunderhighandlowbuyeruncertainty (usedversusnewmobilephones,respec-tively). Second, we investigate traders moti-vationtoleavefeedback,withoutwhich reputationalincentiveswouldnotaccrue. Basedonalargesampleoftimedrating events,wetestpartlycompetinghypotheses withregardtothepossiblemotivesbehind tradersvoluntarycontributionstothefeed-back system. Our analyses take advantage of therepeatedmeasuresobtainedforthesame traders,tocontrolforunobservedconfound-ers (i.e., unobserved heterogeneity).Ouranalysisbringstogetherdifferent strands of research and describes theoretically andempiricallythemechanismsthatguide traderscooperativebehaviorinanonline market. In particular, we show how the insti-tutionalset-upofanonlinemarketengages tradersmoralsentimentsandmaterialinter-ests to create opportunities for mutually ben-eficial trade.THEORY AND HYPOTHESESAn interaction between a buyer and a seller in atypicalonline-auctionmarketcanbecon-ceivedasaprocesscomprisingfourstages. First(theinitiationstage),asellerdecidesto initiateanauctionandsetssomerelevant parameters. Second (the bidding stage), poten-tialbuyersplacebidsandthehighestbidder getstheoffereditem.Third(thetransaction stage),thebuyerpaysthehighestbid(i.e., sellingprice)andthesellershipstheitem. Finally(thefeedbackstage),thebuyercan rate the seller and, depending on implementa-tion of the feedback mechanism, the seller can ratethebuyer.Ratingscanbepositiveor negative, are typically accompanied by a short writtencomment,andareimmediatelyavail-able to all market participants free of charge.Transaction StageThe sequence of moves at the transaction stage isnotpredefined.However,becausesellers cannotchoosetheirbuyers,mostsellers demandadvancepaymentandshiptheitem onlyafterpaymentisreceived(Diekmann, Jann,andWyder2009).Inthiscase,buyers cannot inspect the product before payment and atrustproblemarises:sellerscoulddeliver poor quality goods or keep the money without deliveringatall(Akerlof1970;Dasgupta 1988;GthandOckenfels2003).1Howdo traders overcome this trust problem?Evidencefromlaboratoryexperiments suggests there may be a significant proportion of trustworthy sellers who care about the out-comeoftheirtradingpartnersduetoother-regarding motives such as positive reciprocity, fairness considerations, and altruism (Camerer 2003).However,thesetrustworthysellers will be driven out of the market because they cannotkeepupwithdishonestsellerscheap offers. Bad experiences with dishonest sellers will decrease buyers willingness to pay for a product, and thereby reduce trustworthy sell-ers incentives to stay in the market (Akerlof 1970;Yamagishietal.2009).Alternatively, sellersmaybetrustworthybecausetheyare embeddedindyadicrelationsorsocialnet-works (Buskens and Raub 2002; Granovetter 1992; Gulati and Gargiulo 1999; Jones, Hes-terly,andBorgatti1997)inwhichcoopera-tionisenforcedthroughdirectandindirect reciprocity(NowakandSigmund2005;see also Blau 1964), respectively.Onlineconsumermarkets,however, merely coordinate the supply and demand of end-productsellersandultimatebuyers. Thesemarketsarenotcharacterizedbycus-tomizedandcomplexexchanges,and,aswe willshow,repeatedinteractionsbetweenthe same two traders are infrequent. Under these conditions, it is unlikely that network govern-ance structures will emerge to solve potential exchange problems (Jones et al. 1997). More-over, because online trade takes place across largegeographicdistances,offlinecommit-ment relations (Brown, Falk, and Fehr 2004; Kollock 1994; Yamagishi, Cook, and Watabe 1998)ortrustnetworks(Cook,Rice,and Gerbasi2004;DiMaggioandLouch1998), which would allow traders to obtain and dis-seminateinformationabouttheirinteraction partners,areunlikelytoform.Finally, 68American Sociological Review 79(1)althoughonlinemarketssuchaseBaytryto establish norms of good conduct by providing discussionforums,chatrooms,andclubs wheretradershavetheopportunityto exchangeinformationonselectedtopics,to our knowledge, there is no evidence that eBay tradershavedevelopedsharedsocialnorms of cooperation. Even if they have, such norms willnotpreventfraudulenttradersfrom behaving opportunistically.2Inanonymousonlinemarkets,thelackof socialembeddednessiscompensatedfor through an electronic rating system that effi-ciently and systematically disseminates infor-mationabouttradersreputations(Resnick et al. 2000). From an organizational behavior perspective,anelectronicreputationsystem isanexogenouslyestablished,fullycon-nectednetworkthatembedsanonymous onlinetradersinweakties,throughwhich informationabouttheirreputationsistrans-mitted,withoutstructurallylimitingtheir choiceofpotentialinteractionpartnersand the pursuit of their goals (Gulati and Gargiulo 1999;Uzzi1997).Inotherwords,inan anonymousonlinemarketwithareputation system, traders can be conceived as structur-allyhomogeneousandtheirinteractionsas isolated dyads (Granovetter 1992).Theresultingsocialmechanismthat resolves exchange problems is rather simple. Withtheirreputationsatstake,rationaland self-regardingsellerswhosufficientlycare abouttheirfuturebusinesshaveastrong incentivetobehavecooperatively.Evenin interactionswithbuyerswhomtheyare unlikely to meet again, these sellers will forgo theshort-termtemptationtoabuseabuyers trusttoavoidanegativeratingthatwould hamperfuturebusiness.Marketentrants, however, have no reputation and are therefore likelytobedistrustedbybuyers.Conse-quently,toenterthemarket,sellerswithno feedback record have to lower prices to com-pensatebuyersfortherisktheytakebuying fromanewseller.Ifnewsellerssufficiently care about future business, their good reputa-tionswillreimbursethemforthisinitial investmentinthefuture(Przepiorka2013; ResnickandZeckhauser2002;Shapiro 1983).3Ifbuyersdiscriminateamongsellers accordingtotheirreputations,sellingprices will be positively (negatively) correlated with asellersnumberofpositive(negative)rat-ings.Moreover,ifthemarketdoesnotclear becausesupplyexceedsdemand,ananalo-goushypothesiscanbederivedwithrespect tosellerspositive(negative)ratingsandthe probability of sale.Previousstudieshavefoundastronger absoluteeffectofnegativeratingsthanposi-tiveratingsonoutcomes(BaandPavlou 2002),possiblybecausenegativeratings carry more informationthey express unmet expectationsinrelationtoproperbusiness practicewhereaspositiveratingsmerely confirmexpectations(Lucking-Reileyetal. 2007). Finally, with both new and used prod-ucts,apotentialbuyermayfearthataseller will not deliver after payment or that the qual-ity of the item will be lower than advertised. Inthecaseofusedproducts,uncertaintyis higherforthebuyerbecausedescriptionof theproductconditionislessstandardized, leaving more room for dishonest behavior by sellers.Forusedproducts,asellersreputa-tion may thus be more important and exhibit alargereffectthaninthecaseofnewprod-ucts (Cook et al. 2004; Kollock 1994).In summary, our hypotheses for the trans-action stage are as follows:Hypothesis1:Thenumberofpositive(nega-tive) ratings increases (decreases) the proba-bility of a sale (1a) and the selling price (1b).Hypothesis2: Theeffectsofnegativeratingson theprobabilityofasaleandthesellingprice are stronger than the effects of positive ratings.Hypothesis 3: The effects of positive and nega-tiveratingsarestrongerforusedproducts than for new products.Feedback StageWearguedthatthemarketvalueofagood reputationcreatesfinancialincentivesfor traderstobehavecooperatively.However, rationalandself-regardingtraderswillnot Diekmann et al.69providefeedbackafteratransactionisfin-ished and thus reputational incentives will not accrue.Participationinthefeedbacksystem canbecomparedtovoting,inthat,justas withdemocraticelections,thereisacollec-tivegoodproblem(Downs 1957).Ifcitizens eschew the costs of voting, as strict rationality principles predict, the democratic system col-lapses, as would the online market should no buyerprovidefeedback.Fortunately,how-ever,asubstantialproportionofcitizensdo vote,andalargeproportionoftradersdo leavefeedback.Similarexplanationscanbe given for both behaviors.Consideraone-sidedreputationsystemin which only buyers can rate sellers after a trans-action. There are several possible explanations forbuyersparticipationinthefeedbacksys-tem.First,assuggestedbysignalingtheory (e.g.,Gambetta2009),abuyermightleave feedbacktogainareputationforbeingan active rater. Such a reputation signals to sellers that the buyer will publicly comment on how a transactionplaysout,whichmightenforce cooperativebehaviorbytheseller.Second, buyers might submit positive feedback because they are altruistic and the increase in the sell-ersutilityelevatestheirownutility(Becker 1976).4 Because we assume a decreasing mar-ginalutilityofpositiveratings,thelikelihood of an altruistic buyer submitting feedback will decrease with an increasing seller score. Third, somebuyersmightleavefeedbackbecause they gain satisfaction from contributing to the public good provided by the feedback system. This idea corresponds to the notion of warm glow altruism (Andreoni 1990; see also Del-larocas, Fan, and Wood 2004). Finally, there is goodreasontobelievethatsomebuyersare willing to punish defection or reward coopera-tionwithoutdirectbenefit.Thatis,giving feedback can be motivated by strong reciproc-ity (Fehr, Fischbacher, and Gchter 2002; Gin-tis 2000).5 Returning a favor with a favor and respondingtomisdeedswithsanctionsis observedinexperimentsandseemstobe rootedinhumannature(deQuervainetal. 2004).UntilMay2008,eBay.dehadafeedback systeminwhichbothtransactionpartners could leave feedback within 90 days. In such atwo-sidedratingsystem,inadditiontothe motiveslistedabove,thefeedbackstageis open to strategic thinking. Although both buy-ersandsellerscanbeledbystrategic(and nonstrategic)motives,itisprimarilysellers who have an interest in acting strategically at thefeedbackstage(seealsoBolton,Greiner, andOckenfels2013).Forexample,thebest way for a seller to establish a good reputation istoengageingoodconduct,butthereis always a small probability that a buyer is dis-satisfiedforreasonsbeyondthesellerscon-trol. Tosafeguardagainstabuyers(possibly unjustified)negativeresponse,asellerwill prefertoholdupthethreatofretaliation. Because a negative rating is more detrimental for sellers who have not yet established a good reputation than for sellers with many positive ratings, we expect sellers with fewer positive ratings to be more likely to strive for the second-moverpositionatthefeedbackstage.Onthe other hand, a seller could try to elicit positive feedbackfromabuyerbygivingpositive feedback first. Such a strategy makes sense if the seller believes the buyer adheres to a norm ofreciprocity(Gouldner1960;Resnickand Zeckhauser2002).Ifsellersmakesuchan assertion,thenthebuyersscorewillhavea positive effect on the probability of the seller giving feedback first. A further variable likely toinfluenceasellersratingdecisionis whetherthesellerhasalreadyreceivedarat-ing from the same buyer in a previous interac-tion.Atthetimeofourdatacollection, repeatedratingsfromthesameuserdidnot affect the reputation score. Hence, a seller had noreasontoelicitfurtherpositivefeedback from a buyer by giving positive feedback first (see also Dellarocas et al. 2004).For buyers, the nonstrategic motives men-tionedearlierarestillvalid.Inaddition,if buyersaremotivatedbystrongreciprocity, positivefeedbackfromthesellerwillfurther increase their likelihood of submitting a posi-tive rating. In fact, it is hard to think of reasons why a self-regarding buyer would reciprocate asellerspositivefeedback.Suchbehavior wouldbeanindicationofstrongreciprocity atthefeedbackstage.Furthermore,indirect 70American Sociological Review 79(1)reciprocity(NowakandSigmund2005),that is,rewardingsellersfortheirparticipationin thefeedbacksystem,mightbeanothernon-strategicmotiveinatwo-sidedsystem. Accordingtoindirectreciprocity,buyerswill bemorelikelytogive(positive)feedbackto sellers with a good reputation.6 Finally, due to eBaysreputationsystemignoringrepeated ratings, buyers should be less likely to submit positive feedback if they provided a rating to the same seller in a previous interaction. If no such effect exists, altruistic motives based on thebuyersbeliefthatthesellergainsutility from receiving feedback (Becker 1976) can be ruled out. For our empirical analysis, we con-fineourselvestothefollowinghypotheses (see also Dellarocas et al. 2004):Hypothesis 4: Reciprocity:Ifbuyersandsellersaremotivatedby strong reciprocity at the feedback stage, theirinclinationtosubmitapositive (negative)ratingincreasesafterreceiv-ingpositive(negative)feedbackfrom the transaction partner.Hypothesis 5: Effect of sellers score on buyers behavior:7(a)IfbuyersaremotivatedbyBeckers altruism,thesellersscoreshouldhave anegativeeffectontheprobabilityof feedback submission.(b)If buyers are motivated by indirect reci-procity, the sellers score should have a positive effect on the buyers decision to provide feedback.Hypothesis 6: Sellers strategic behavior:(a)Sellersinclinationtomakethefirst moveatthefeedbackstageincreases with their own score; and(b)Sellersinclinationtomakethefirst moveatthefeedbackstageincreases with the buyers score.Hypothesis 7: Previous interaction:(a)Sellers are less likely to submit positive feedback if they have already received a rating from the same buyer in a previous interaction.(b)Buyers are less likely to submit positive feedback if they provided a rating to the same seller in a previous interaction.PREVIOUS FINDINGSMoststudiesinvestigatingreputationeffects in online markets analyze data collected from eBay, and most of these studies find evidence of a positive relation between sellers reputa-tions and prices or sales. Due to space limita-tions, we provide a review of these studies in theonlinesupplement(http://asr.sagepub .com/supplemental)(forearlierreviews,see BajariandHortacsu2004;Resnicketal. 2006).Here,weinsteadfocusonimportant aspects in the ongoing discussion about reputa-tion formation in anonymous online markets.In their review of 15 empirical eBay stud-ies,Resnickandcolleagues(2006)pointout that these studies quasi-experimental designs cannot account for unobserved heterogeneity inasatisfactoryway.Ifunobservedfactors, such as pre-purchase communication between buyers and sellers or the quality of the prod-uctdescription,arecorrelatedwithsellers reputations,thenestimatesoftheregression coefficients can be biased. In our dataset, we haverepeatedmeasuresforaconsiderable proportionofbothbuyersandsellers.This allowsustodealwithproblemsofunob-servedheterogeneitybyusingfixed-effects regressions (Allison 2009).SomeresearchersarguethatmosteBay studiesclaimtoestimatebuyerswillingness topayforasellersreputationbut,infact, only estimate what buyers do pay, which is a poor estimate of their willingness to pay (Sni-jders and Weesie 2009). The argument is that thehighestbidsmaydependonmanyother factorsnotobservableindataobtainedfrom eBay. For instance, bidding is a dynamic pro-cessandpotentialbuyersbiddingdecisions also depend on their beliefs about other buy-ersreservationprices.Moreover,potential buyers could consider several offers simulta-neously, but their initial choice sets are unob-served or difficult to define (but see Livingston 2003).Inourstudy,wedonotclaimtobe estimatingbuyerswillingnesstopayfora Diekmann et al.71sellersreputation.Forthesakeofourargu-ment,itisenoughtoshowthatbuyers,on average,paymoreforsellerswithhigher reputationsandthuscreatefinancialincen-tives for sellers to behave cooperatively in the market.TheAchillesheelofareputationsystem remainsmarketparticipantswillingnessto rate each other. If participants do not provide truthfulratings,themarketlosesitscapacity toidentifyfraudulenttradersandreputation loses its value. For buyers and sellers on eBay, Resnick and Zeckhauser (2002) report a rating frequency of 52 and 61 percent, respectively; Dellarocas and colleagues (2004) find 68 and 78percent;andJian,MacKie-Mason,and Resnick(2010)reportratesof33and55 percent.Moreover,averagingacrossseven different countries eBay sites, Bolton and col-leagues (2013) report a rating frequency of 71 percent.Theyattributethishighratingfre-quency to eBays two-sided rating system. On theonehand,atwo-sidedsystemprovides incentivestoleavefeedbackduetopositive reciprocityexpectations.Ontheotherhand, the threat of retaliation against a negative rat-ing leads to a positive evaluation bias. Based onadatasetofratingsobtainedfromeBay, DellarocasandWood(2008)alsoidentify reciprocityasanimportantfactordriving behavioratthefeedbackstage.Moreover, theysuggestthatnotleavingfeedbackindi-cates dissatisfaction and, based on information obtainedfromtheobservedratingpatterns, they estimate the proportion of good, bad, and mediocre transactions. They show that the 99 percentpositiveratingusuallyobservedon eBay is a biased estimate of the proportion of good transactions. Accounting for the sound ofsilence(i.e.,transactionswhereoneor bothtradersdidnotleave feedback) givesan estimateof79and86percentforsatisfied buyersandsellers,respectively.Inasimilar vein,Jianandcolleagues(2010)developeda modeltoestimatetheproportionofdifferent ratingstrategiesfromeBayfeedbackdata. Theyassumethreeratingstrategies(donot give feedback, give feedback unconditionally, andreciprocatefeedback)andestimatebuy-ersandsellerspropensitytochooseoneof the strategies. They find that the median buyer andsellerchoosethereciprocalfeedback strategy in 23 and 20 percent of cases, respec-tively, whereas they choose the unconditional feedbackstrategyin38and47percentof cases.Our study adds to these results by analyz-ing time-specific feedback decisions. Whether trading partners provided feedback first is an explanatory variable in our analysis. We esti-matediscrete-timeevent-historymodels,an approachthatdoesnotrelyonrestrictive assumptionsaboutthetimingdistributionof traders feedback. We also account for unob-servedheterogeneitybyincludingbuyerand seller fixed effects in our models. Note, how-ever,thatweobservebehaviorinourdata; therefore, any inference we make about trad-ersunderlyingmotivesisbasedonthecon-sistencyofthebehavioraldatawithour theoretical propositions.DATAOurdata,collectedfromtheGermaneBay platform,containinformationonauctionsof mobile phones and DVDs that ended between December1,2004andJanuary7,2005(for details on data collection, see the online supple-ment).Werestrictouranalysistooffersfrom sellersinGermanyandexcludeprivateauc-tions,wherethebuyerisunknown,aswellas auctionsinwhichmultiplepiecesareoffered that can be purchased by multiple buyers. Some auctionsendedearlybecausethesellerwith-drew the auction before the first bid, the seller stopped the auction with the current bid as the winning bid, or the buyer bought the item at the BuyItNow(i.e.,fixed)price.Weexclude thesecasesfromouranalysis.Finally,minor inconsistenciesintheregularexpressions caused missing or ambiguous data on key vari-ables for a small fraction of auctions, which we also exclude. In total, we excluded 14.5 percent ofmobilephoneauctionsand10.4percentof DVD auctions for these reasons.To estimate reputation effects on the high-estbidsinauctions,itissensibletoselecta homogeneous good to avoid bias due to unob-served heterogeneity (Diekmann et al. 2009). 72American Sociological Review 79(1)Table 1. Means and Standard Deviations of Key VariablesNew MobilePhonesUsed MobilePhones DVDsSellerSellers positive ratings (log) 3.77 (1.59) 3.74 (1.61) 7.00 (2.65)Sellers negative ratings (log) .42 (.70) .47 (.71) 2.00 (2.01)Buyer (if sold)Buyers positive ratings (log) 3.06 (1.70) 3.16 (1.83) 3.92 (1.69)Buyers negative ratings (log) .27 (.53) .32 (.62) .28 (.54)Feedback (if sold)Time until seller feedback (censored) 32.6 (35.3) 36.5 (36.7) 24.9 (30.1)Positive seller feedback .736 .685 .843Neutral seller feedback .002 .005 .001Negative seller feedback .012 .016 .003Seller has previous rating from buyer .032 .031 .049Time until buyer feedback (censored) 33.9 (35.7) 39.0 (36.8) 25.2 (30.2)Positive buyer feedback .706 .656 .835Neutral buyer feedback .007 .011 .004Negative buyer feedback .019 .020 .004Buyer has previous rating from seller .030 .037 .049AuctionSuccessful auction .958 .956 .533Selling price in Euros (if sold) 216.7 (37.4) 150.3 (47.1) 7.4 (10.2) Number of observations 5,499 9,128 339,517Number of sellers 4,341 7,687 33,166Number of buyers 4,881 7,454 100,046Note: Table S1 in the online supplement lists descriptive statistics of further control variables used in our model estimations.Toensurehomogeneityinourmobilephone data, we manually sorted out offers that con-tainedcomplementarygoodsormultiple items(23.1percent)anddividedthesample intonew(unusedandinoriginalpackaging) andusedproducts.Thetotalsamplethen included 14,627 mobile phones of seven dif-ferentmodels,37.6percentofwhichwere newand95.7percentofwhichreceivedat leastonebid(i.e.,theirauctionsweresuc-cessful).8Inourdata,themeansellingprice was217Eurosfornewmobilephonesand 150 Euros for used mobile phones (see Table 1 for descriptive statistics of our key variables; additional variables are documented in Table S1intheonlinesupplement).Toevaluate whetherreputationeffectscanalsobe observed for low-cost products, we analyze a sampleof339,517DVDauctions,53.3per-centofwhichweresuccessful.Wecollected dataonallDVDauctionsinthementioned timespan,coveringhundredsofdifferent movies.Thus,producthomogeneitycannot be assumed. Also, due to the lack of a reliable automaticcategorizationmethod,wedonot distinguishbetweennewandusedDVDs. Moreover,unlikethemobilephonemarket, with 1.25 auctions per seller, on average, the DVDmarket,withanaverageof10.24auc-tionsperseller,isdominatedbyarelatively small number of professional sellers.We operationalize buyers and sellers rep-utationswiththeirnumberofpositiveand negativeratings.Becausedistributionsof Diekmann et al.73thesevariablesarehighlyright-skewed,we takethelogarithmoftheirvalues(afteradd-ing 1). Note that in terms of ratings, too, sell-ers in the DVD market are considerably larger than sellers in the mobile phone market.At the feedback stage, we have the time (in days) until feedback for both the seller and the buyer of a successful auction. Recall that these variablesarecensoredat90days.Themean durationseemstobelowerintheDVDmar-ket,andpositivefeedbackisverycommon, whereas neutral and negative ratings are rare. Negative ratings are more frequent in auctions formobilephonesthanforDVDs,indicating thatproblematictransactionsaremorelikely when they involve more complex and expen-sive products (although this could also be due tothelowerdegreeofprofessionalisminthe mobile phone market). Another important var-iableatthefeedbackstageiswhetherthe sellerorbuyerhasalreadyreceivedarating fromthesamepersoninapreviousinterac-tion. Such repeated encounters are more likely intheDVDmarket(5percent)thaninthe mobile phone market (3 percent).9RESULTSEffect of Reputation on Auction SuccessWe first turn to reputation effects on sales and sellingprice.Table2givesanoverviewof resultsforourkeyvariables.Wecontrolfor manyadditionalvariablesinourmodelesti-mations,butduetospacelimitations,we reportanddiscusstheireffectsintheonline supplement only.Models 1 through 3 in Table 2 show results fromlogisticregressionsofauctionsuccess onsellerspositiveandnegativeratings.In line with Hypothesis 1, we find a significantly positive effect of positive ratings and a nega-tive effect of negative ratings on the probabil-ity of sale for new mobile phones (Model 1) andforDVDs(Model3).Contrarytoour expectation, we find no significant effects for used mobile phones (Model 2). Recall, how-ever, that over 95 percent of all mobile phones were sold, which leaves relatively little infor-mation for the estimation of reputation effects on the probability of sale.Becausewearedealingwithauctions and not fixed-price offers, a large part of the reputationeffectwillbeexertedonprice. Table 2 also reports results from linear regres-sions of selling price on sellers positive and negativeratings(Models4through6).As expected,weobserveastrongrelation betweenasellersreputationandthehighest bidinourdata.Forbothmobilephonesand DVDs,thenumberofpositiveratingshasa significantly positive effect on price, and neg-ative ratings reduce price.10 Because both the numberofratingsandthesellingpriceare log-transformed,theestimatedcoefficients can be interpreted as a percent increase in the selling price due to a one percent increase in thenumberofratings(theelasticity).Thus, doublingthenumberofpositiveratings increasesthesellingpriceofnewmobile phones (Model 4) by about .005 ln(2) 100 = .35 percent. For used mobile phones (Model 5)andDVDs(Model6),theeffectofa100 percentincreaseinpositiveratingsisabout .55and3.7percent,respectively.Evaluating theserelativeeffectsatthemeanpricesfor new and used mobile phones and DVDs (see Table 1) yields absolute effects of 61, 97, and 27cents,respectively.Althoughthese amounts may appear small at first sight, they arenot,infact,ifonetakesintoaccountthe rangeofsellersreputationsinoursample, wheredifferencesof1,000percent(e.g.,40 versus 440 positive ratings) and more are the rule rather than the exception. As Hypothesis 2 predicts, the absolute effect of negative rat-ings on price and sales is larger than the effect ofpositiveratingsinallmodels(Models1 through 6), although the difference is statisti-cally significant at the 5 percent level only in Model4( p=.003foratwo-sidedtest).In terms of absolute effects at mean prices, dou-bling the number of negative ratings decreases thepriceby1.3Euros,2.2Euros,and52 cents in Models 4, 5, and 6, respectively.In line with our expectation that used prod-uctswillexhibitmoreuncertaintythannew 74 Table 2. Effect of Reputation on Sales and PricesProduct Sold (0/1)Selling PriceSelling Price (with Fixed Effects) 12345678910 New Mobile PhonesUsed Mobile PhonesDVDsNew Mobile PhonesUsed Mobile PhonesDVDsNew Mobile PhonesUsed Mobile PhonesDVDsDVDsSellers positive ratings (log).344*(.136).019(.064).117**(.040).005***(.001).008*(.003).053***(.015).027**(.010).002(.045).016***(.002).005(.007)Sellers negative ratings (log).670*(.301).207(.132).145*(.058).011***(.002).018**(.006).101**(.035).055(.034).025(.033).012**(.005).036**(.014)McFadden R-squared.858.678.133 R-Squared.844.513.111 Number of observations5,4999,128339,5175,2698,727180,8811,6121,944113,276103,030Number of sellers4,3417,68733,1664,2427,47430,0185856916,901Number of titles18,05415,964Note: Models 1, 2, and 3: Coefcient estimates of logistic regressions are shown; the binary dependent variable is equal to one for successful auctions. Models 4, 5, and 6: Coefcient estimates of OLS regressions are shown; the dependent variable is the logarithm of the selling price (in Euros). Models 7, 8, 9, and 10: Coefcient estimates of xed-effects regressions are shown (seller xed effects in Models 7 and 8, title xed effects in Model 9, title and seller xed effects in Model 10); the dependent variable is the logarithm of the selling price (in Euros). Numbers in parentheses are robust standard errors (adjusted for seller-clusters). Models contain various control variables (starting price, length of product description, number of competing offers, and dummies for private prole, veried identity, Me-Page, PowerSeller, auction picture, thumbnail listing, bold listing, payment modes, auction duration, time and date of auction ending, and product subcategory); for detailed results see the online supplement.*p < .05; **p < .01; ***p < .001 (two-tailed tests).Diekmann et al.75products,andbuyerswillthuspayahigher reputationpremiumtosellersofusedprod-ucts(Hypothesis3),theeffectsofpositive andnegativeratingsonthesellingpriceare strongerforusedmobilephones(Model5) thanfornewmobilephones(Model6).11 Recall, however, that reputation has a signifi-canteffectonsalesonlyfornewmobile phones.Ifwetakeintoaccountbothout-comes (the selling price and the probability of sale),thetotaleffectofreputationiseven largerfornewthanforusedmobilephones, althoughthedifferenceisstatisticallyinsig-nificant (not shown).12Asmentionedearlier,wecannotassume producthomogeneityforourDVDdataset. Wethereforereplicateouranalysisusinga modelwithfixedeffectsformovietitles (Model9).13Althoughsmallerinabsolute sizethanintheordinaryDVDmodel,the effectsofpositiveandnegativeratingsare still highly significant.One could object that our effects are spuri-ousduetounobservedsellerheterogeneity. For example, some sellers might be more suc-cessfulforreasonsunrelatedtoreputation and,therefore,morelikelytohaveagood reputation, resulting in a spurious correlation betweenreputationandsellingprices.We checkthisargumentbyanalyzingtheeffects ofchangesinreputationforsellerswho appear repeatedly in our dataset. We estimate modelswithsellerfixedeffectsfornewand usedmobilephones(Models7and8)anda modelwithcrossedfixedeffectsforsellers andtitlesforDVDs(Model10).14Inthe modelfornewmobilephones,effectsof positiveandnegativeratingspersist,asdoes the effect of negative ratings in the model for DVDs,buttheeffectsdisappearforused mobile phones. However, the fact that we still find reputation effects in most of our models providesstrongsupportforthereputation hypothesis (Hypothesis 1). Note that account-ingforsellerfixedeffectsresultsinfairly conservativeestimatesofreputationeffects. These estimates are based only on changes in sellersreputationsduringtheobservation window of about one month, and all between-sellervariabilityisdiscarded.Consequently, modelestimationsarebasedonsellerswith hightradingfrequency,whoaremorelikely to have a good reputation, whereas newcom-ers, for whom an increase in reputation might be most valuable, tend to be excluded.Insummary,ouranalysesreplicateprevi-ous findings of a positive (negative) effect of positive (negative) ratings on sales and prices (Hypothesis 1); we also find a larger absolute effectofnegativeratingsascomparedto positiveratings(Hypothesis2).Ouraddi-tionalmodelestimationsalsomakeitrather unlikely that the effects we find are spurious. Evidencefortheuncertaintyhypothesis (Hypothesis 3) is mixed, however; the reputa-tion effect does not seem to be larger for used mobile phones than for new ones.Participation in the Feedback SystemAsarguedearlier,iftoofewtraderspartici-pateinthefeedbackforum,thereputation system will be ineffective in detecting cheat-ers,andreputationalincentivesfor(first-order)cooperationatthetransactionstage willnotaccrue.However,wedofindhigh levelsof(second-order)cooperationatthe feedbackstage.Buyerssubmittedpositive feedback in over 65 percent of all transactions inourmobilephonesample,whereasonly3 percent of transactions were rated neutrally or negatively. In the DVD sample, the difference isevenlarger.Buyersrated83percentof transactionspositively;neutralornegative feedback was submitted in less than 1 percent ofcases(seeTable1).Furthermore,Table3 shows that over 80 percent of transactions in themobilephonemarketand90percentof DVDtransactionswereratedbyatleastone trader.After an auction ends, traders have 90 days to submit their feedback; because traders can observetheirpartnersgivingthemfeedback, theirfeedbackdecisionsarenotindependent from each other. Therefore, it is important to taketheexacttimingoftradersfeedback decisionsintoaccountwhenanalyzingtheir rating behavior. Figure 1 gives an overview of thedistributionoffeedbackdecisionsover the90days.Thesurvivalfunctions(upper 76American Sociological Review 79(1)panel)showhowtheproportionsofbuyers andsellerswhohavenotyetgivenfeedback decrease over time; the hazard rates for posi-tiveandnegativeratings(themiddleand lowerpanel,respectively)show,roughly speaking, the feedback probability on a given dayconditionalonnofeedbackbeinggiven before; dotted lines indicate 95 percent confi-dence intervals (see the online supplement for computational details).Survivalcurvesforsellersandbuyersare synchronizedinbothmarkets,butratingsare submittedfasterintheDVDmarket. After10 days,almost50percentofsellersandbuyers had submitted a rating in the DVD market, but only about 40 percent had done so in the mobile phone market. Most notably, sellers are some-whatmorelikelytoquicklysubmitapositive rating. That is, the hazard rate of a positive rat-ing is higher for sellers than for buyers over the first couple of days in both markets. This indi-catesthatsomesellersmightactstrategically when trying to elicit a positive rating from the buyer by taking the first step. For negative rat-ings, the picture is different (note that the scale ofthehazardrateismagnitudessmallerthan for positive ratings). The curves are very simi-larforsellersandbuyers,althoughinthe mobile phone market the hazard rate for buyers is higher than for sellers. Furthermore, negative ratings are submitted more slowly than positive ratings. For positive ratings, hazard rates peak betweendays5and10;fornegativeratings, they are highest between days 15 and 35. Inter-estingly, unlike for positive ratings, we observe an increase in the hazard rates for negative rat-ingsbetweendays80and90(althoughconfi-denceintervalstendtobelargeasdatamass getssmallinthisregion).Thisindicatesthat some actors may delay their negative rating to escape retaliation.Thesurvivalcurvesandhazardratesin Figure 1 only provide descriptive information onthedistributionofratingsovertime.To analyze the effects of covariates, we divide the processtimeintodaysandapproximatethe hazard ratesusing discrete-timelinear proba-bilitymodels(Long1997;Mood2010)con-tainingindicatorvariablesfordifferent segmentsoftheprocesstime(i.e.,wemodel the baseline hazard rates shown in Figure 1 as stepfunctions).15Wealsoincludetwo-way buyer and seller fixed effects in the models to control for unobserved buyer and seller char-acteristicsthatmightotherwisebiasthe results.Forinstance,apositiverelation betweentradersreputationsandtheirrating probabilitycouldbeduetothefactthatfre-quent raters accumulate more ratings because their ratings are reciprocated. Likewise, trans-action quality or speed might be higher among traders with better reputations, thus increasing thelikelihoodofreceivingpositivefeedback ordecreasingthetime-to-feedbackreaction. Suchpotentialbiasesarekeptataminimum byincludingsellerandbuyerfixedeffectsin the models.Table 4 provides estimates from the discrete-timelinearprobabilitymodelsforpositive andnegativefeedbackfrombuyersandsell-ersintheDVDmarket(wedonotpresent resultsformobilephonesherebecausethe Table 3. Feedback Patterns (%)Mobile Phones DVDsAuctions with mutual feedbackBuyer rated first 32.3 48.6Seller rated first 28.1 30.2Simultaneous .1 .3Auctions with buyer feedback only 10.1 5.1Auctions with seller feedback only 11.8 5.7Auctions without feedback 17.7 10.1Total 100.0 100.0Observations 13,996 180,881Diekmann et al.77sampleistoosmallforthetwo-wayfixed-effects estimation). The displayed coefficients arescaledbyafactorof100;theycanbe interpreted as percentage-point effects on the conditional probability of submitting a rating on a specific day given that no rating had yet been submitted.Our results are consistent with the hypoth-esis that ratings are driven by strong reciproc-ity(Hypothesis4).AllmodelsinTable4 includetime-dependentvariablesindicating whether the trading partner submitted a posi-tive,neutral,ornegativeratingfirst.Oncea sellersubmitsapositiverating,thebuyers conditionalprobabilityofsubmittingaposi-tiveratingsharplyincreasesbyaboutfive percentage points (Model 11); for sellers, this effect is even stronger (Model 12). Likewise, theprobabilitythatanegativeratingwillbe reciprocatedwithanegativeratingisvery Figure 1. Survival Functions and Hazard Rates of Sellers and Buyers Rating Decisions78American Sociological Review 79(1)Table 4. Hazards of Positive and Negative Feedback in the DVD MarketPositive Feedback Negative Feedback11 12 13 14Buyer Seller Buyer SellerPositive first move by partner (time-dependent)5.106***(.112)13.754***(1.013).003(.004).006**(.002)Neutral first move by partner (time-dependent).410(.732)2.176***(.461).011**(.004).440***(.109)Negative first move by partner (time-dependent).544(.327)1.859***(.461).535**(.166)1.871***(.311)Positive ratings (log) .880***(.183).421(.234).025*(.012).012**(.004)Negative ratings (log) .616(.508).225(.258).102*(.051).008(.006)Partners positive ratings (log) .840***(.137).259(.269).017***(.005).021***(.006)Partners negative ratings (log) 1.023***(.259).894(.755).043**(.014).059(.035)Previous interaction ratingReceived only .067(.459).943(.596) Provided only 2.051***(.428)1.448*(.676) Received and provided 1.147***(.219).158(.286) Received or provided .002(.005).006*(.003) Number of observations 96,055 96,055 96,055 96,055Number of events 80,601 80,343 310 223Number of sellers 9,309 9,309 9,309 9,309Number of buyers 26,188 26,188 26,188 26,188Note: The table shows coefcient estimates for effects on the conditional probability of submitting a rating on a specic day given that no rating had been submitted yet (scaled by a factor of 100) for discrete-time linear probability models (LPMs) with seller and buyer xed effects (standard errors in parentheses, adjusted for buyer-clusters in Models 11 and 13 and for seller-clusters in Models 12 and 14). Time indicators parameterizing the baseline hazard are not displayed.*p < .05; **p < .01; ***p < .001 (two-tailed tests).high(Models13and14).Notethatthese effectsaremuchsmallerbecauseofthelow baselineprobability.16 Asomewhatinconclu-sive finding is the positive effect of a buyers negativeratingonthesellersprobabilityof giving a positive rating.Moreover,ourresultssupportHypothesis 5a, not 5b. The more positive ratings a seller has, the less likely the buyer is to submit feed-back. That is, in addition to strong reciprocity, buyersratingbehaviorisconsistentwith Beckers altruism rather than with warm glow altruism,signaling,orindirectreciprocity. The remaining evidence in Table 4 is consist-entwiththeviewthatbuyers,inparticular, taketheirinteractionpartnersutilityinto account when leaving feedback. First, in line with Hypothesis 7b, buyers (and sellers) seem toacknowledgethattraderstowhomthey providedaratingbeforedonotbenefitfrom another rating. Second, the more negative rat-ings a seller has, the more likely a buyer is to Diekmann et al.79leaveapositiverating.Thismightreflect buyerswillingnesstosupportsellerswho want to redeem their good reputation. Finally, we find that buyers (and sellers) are reluctant togiveanegativeratingifatradingpartner alreadyhasmanynegativeratings,butthey are more likely to give a negative rating to a tradingpartnerwithmanypositiveratings. This might reflect the fact that, after a failed transaction,tradersfeelgreaterdisappoint-ment toward trading partners with good repu-tations than toward those with bad reputations.We do not find support for our hypotheses regardingsellersstrategicbehavioratthe feedbackstage.Neithersellersownnumber ofpositiveratings(Hypothesis6a)norbuy-ersnumberofpositiveratings(Hypothesis 6b)exhibitapositiveeffectonsellersprob-abilityofleavingpositivefeedback.Also, sellers are not less likely to give feedback to a buyer from whom they have received a rat-inginapreviousinteraction,whichcontra-dicts Hypothesis 7a.Insummary,weobservehighlevelsof participationinthefeedbacksystem,which seem to be driven mainly by strong reciproc-ityandaltruism.Bothsellersandbuyers reciprocatereceivedratingsfromtheirtrans-actionpartners(Hypothesis4)andappearto taketheirinteractionpartnersutilityinto account in their feedback decisions (Hypoth-eses 5a and 7b). We do not find evidence for sellers strategic considerations, as our corre-spondinghypothesescannotbeconfirmed (Hypotheses6a,6b,and7a).However,this doesnotmeanthatstrategicmotivesare unimportantordonotexist.Forexample, accordingtoourdescriptiveresults,sellers are on average quicker than buyers to submit positivefeedback,whichcanbeseenasa strategic attempt to elicit reciprocation.DISCUSSION AND CONCLUSIONSIn commerce, trust problems have been ubiq-uitous ever since humans started trading com-modities.Culturalevolution,togetherwith thecreativityofsocialengineers,ledtoa greatvarietyofinstitutionsintendedtohelp mitigatethesetrustproblems.Theseinstitu-tionsincludethereputationsystemofthe Maghribi traders in the Middle Ages described byGreif(1989),thelawmerchantsystem analyzed by Milgrom and colleagues (1990), theformationofformalsocialstructuresin nineteenth-centuryNorthAmericaexplored by Zucker (1986), and informal network gov-ernance structures as described by Jones and colleagues (1997). Factors such as the degree of traders social embeddedness (Buskens and Raub2002;DiMaggioandLouch1998; Granovetter1992),thetypeofcommodity (Kollock 1994), the complexity of the interac-tion(Jonesetal.1997),and,mostimpor-tantly,thelackoftimelyandinexpensive informationaboutatradingpartnersreputa-tion(Milgrometal.1990)havebeenimpor-tantdeterminantsintheevolutionofthese institutions.Recenttechnologicalprogresshasdrasti-callydiminishedthecostoffastinformation transmission,andauctionmarketswere among the first to establish online reputation systems for economic transactions. Based on asellersreputation,buyerscantradeoff prices against the possibility of being cheated, andthepricepremiumforagoodreputation providesafinancialincentiveforsellers cooperativebehavior.Moreover,tradersdo nothavetobeembeddedinofflinesocial networkstoaccessinformationaboutother traders as long as they have access to the mar-ket platform online. In the past decade, online reputationsystemshavespreadalloverthe web, and the Internets technical possibilities havebroughttheirfunctioningclosetoper-fection.However,theexistenceandstability of online reputation systems cannot be taken forgranted.Areputationsystemseffective-nessinpromoting(first-order)cooperation crucially depends on traders voluntary provi-sionoffeedback.Becauseatradergainsno direct benefits from leaving feedback, partici-pationinthefeedbacksystemissubjecttoa (second-order) free-rider problem.Ourstudyinvestigatestheroleofreputa-tion in promoting cooperation and the process 80American Sociological Review 79(1)ofreputationformationinthefield.Thatis, we explore the mechanisms underlying coop-eration by observing transactions and ratings amongtradersinanonymousonlinemarkets inanunobtrusiveway.Weconsideronline markets as integrated systems of institutional rulesthatgoverntradersdecisions,andour analysiscoverstheentirechainofinterac-tionsbetweenabuyerandaseller,starting with the sellers offer and ending with either or both leaving feedback. Our theoretical con-siderationsarerootedinclassicalsociology and anthropology, with their insights into the rolesoftrust,reciprocity,andreputationin socialexchange,butwealsodrawonrecent findingsfromexperimentalsocialsciences. Basedonalargesampleofdatafromthe mobilephoneandDVDmarkets,wetest variousandpartlycompetinghypotheses derived from our theoretical argument.In both markets, we find that sellers num-ber of positive ratings has a positive effect on sales and selling prices, and sellers negative ratingshavetheoppositeeffect(Hypothesis 1). Moreover, in line with Hypothesis 2, nega-tive ratings have a greater effect on sales and prices than do positive ratings. This corrobo-ratespreviousfindingsthatreputationhasa market value and that buyers take into account informationaboutsellersreputationswhen theyplacetheirbids(seetheonlinesupple-ment for a review of previous findings). How-ever, the empirical evidence for the uncertainty hypothesis(Hypothesis3)ismixed.As hypothesized, reputation has a larger effect on priceforusedmobilephonesthanfornew ones, but this is not the case for the probabil-ityofsale.Thegrosseffectofreputation, takingintoaccountboththepriceandthe probability of sale, seems even larger for new mobile phones than for used ones. A possible explanation for this finding is that used prod-uctsattractbuyerswhoarewillingtotakea riskandthereforedonottakeinformation aboutsellersreputationsintoaccountas carefully as buyers of new products do.At the feedback stage, cooperation is very high.Sixtypercentoftransactionsinthe mobilephonemarketand80percentinthe DVDmarketwereratedbybothsellersand buyers,andonlyasmallfractionoftransac-tions had no rating at all. This is comparable to what Dellarocas and colleagues (2004) and Boltonandcolleagues(2013)foundintheir eBay data. Our study corroborates, moreover, that buyers and sellers behavior at the feed-backstageislargelyconsistentwithstrong reciprocity(Hypothesis4).Althoughone receives no apparent benefits from reciprocat-ing a rating received from ones trading part-ner,thelikelihoodofprovidingsimilar feedbacksharplyincreasesforbothbuyers and sellers after receiving a rating. The effects arestrongandhighlysignificantforboth positiveandnegativeratings.Thisisinline withstatisticsreportedinBoltonandcol-leagues (2013) and with what Dellarocas and Wood (2008) and Jian and colleagues (2010) found based on their model estimations.Besidesreciprocity,altruisticpreferences seemtomotivatetraders(inparticular,buy-ers) decisions. Buyers are more likely to give a positive rating to traders with few positive or manynegativeratings(Hypothesis5a).This contradictsJianandcolleagues(2010),who foundthatbothbuyersandsellersaremore likely to give feedback to experienced traders thantoinexperiencedtraders.Wearguethat this could be an artifact resulting from Jian and colleagues(2010)disregardingtransaction quality in their model estimations. Transaction qualityislikelytobehigherforexperienced traders and could thus lead to a positive corre-lationbetweentradersexperienceandthe probability of giving feedback. We circumvent this problem by including partner fixed effects in our model estimations. If we do not include them,ourresultsaresimilartothoseofJian and colleagues (2010). Moreover, we find that buyers(andsellers)arelesslikelytogive positivefeedbacktothesametradingpartner repeatedly(Hypothesis7b).Hypotheses5a and 7b are in line with Beckers (1976) notion of altruism; our results are consistent with the idea that traders take their interaction partners utility into account when leaving feedback and are more likely to give feedback the more ben-eficial it is for their partner.Diekmann et al.81Contrarytoourexpectationsandunlike theresultsbyDellarocasandcolleagues (2004) would suggest, we find no support for ourhypothesisregardingsellersstrategic considerationsatthefeedbackstage.Inpar-ticular, we find no support for Hypotheses 6a and6b,thatsellersstrategicallytakeinto account their own or the buyers score at the feedbackstage.Dellarocasandcolleagues (2004)conclusionwasbasedonthefinding that a buyers first move at the feedback stage hadanegativeeffectonthesellersoverall probabilityofleavingfeedback.Weargue that this result could be an artifact due to their use of a probit model; if sellers have a lower a priori probability of providing a rating, buy-ers will more likely be the first movers at the feedbackstage.Theevent-historymodels usedinouranalysisdonotsufferfromthis problembecausethebuyersfirstmoveis treated as a time-varying variable. Yet, traders seemtoanticipatetheirtradingpartners reciprocalbehavior,especiallyinthecaseof negative feedback, as some try to avoid retal-iationbysubmittingtheirnegativerating close to the end of the rating period. Thus, we cannotconcludethatstrategicmotivesare unimportantintradersdecisionstoleave feedback; they are difficult to identify in our data.In Gouldners (1960:176) words, the norm ofreciprocityisastartingmechanism [that]helpstoinitiatesocialinteraction. ReciprocityisalsoacornerstoneinBlaus (1964)socialexchangetheory.Bothauthors elaborateontheprincipleofreciprocityand outline its implications for the emergence and stabilityofexchangerelations.Theirnotion of reciprocity and the reciprocity we observe in our data is similar to the notion of strong reciprocitypioneeredbyGintis(2000). Accordingtostrongreciprocity,actorswill returnafavorwithafavorandretaliate against an unfriendly act without expecting to becompensatedforthecoststheyincurin doing so. In laboratory experiments, Fehr and colleagues(2002)demonstratedthatstrong reciprocity is a key mechanism in promoting cooperation in voluntary contribution games, andtheysparkedacross-disciplinarydebate concerning the determinants of human social cooperation(Guala2012).Inourstudy,we show how the interplay of simple institutional rulesandhumanmotivationestablishesa highlevelofcooperationamonganonymous traders in online markets.Basedonourfindingswearguethat,in thesemarkets,altruism,strongreciprocity, and,mostlikely,alsostrategicmotivesplay animportantroleinmaintaining(second-order)cooperationatthefeedbackstage, which generates the information necessary to createfinancialincentivesfor(first-order) cooperationatthetransactionstage.Inother words, giving positive or negative feedbackasecond-orderissuecanbeconceivedas rewardingorpunishingatradingpartners first-ordercooperationordefection,respec-tively (Resnick and Zeckhauser 2002). Given thattheconstrainedset-upofonlinemarkets mirrors laboratory conditions relatively well, ourfindingsbolstertheexternalvalidityof experimentalevidenceofsocialcooperation andpeerpunishment. Atthesametime,our study makes apparent that unobtrusive behav-ioraldatafromonlineplatformsneedtobe complementedwithlaboratoryandfield experiments, which are better suited to disen-tanglingthevariousmotivesbehindhuman behavior.Althoughstrongreciprocityand altruismappearasthemostlikelymotives behind the behavioral patterns we observe at the feedback stage, carefully designed experi-ments should be used to verify our results and rule out alternative explanations.Electronicreputationsystemshavebeen constantly refined to better meet the require-mentsofasteadilygrowingonlinecommu-nity. In fact, the change in eBay.des reputation system in spring 2008 was guided by a thor-oughtheoreticalandempiricalanalysisthat combined field data from online markets with evidencefromlaboratoryexperiments(Bol-tonetal.2013).Themainconcernwiththe oldtwo-sidedfeedbacksystemwasthatit might foster a positive evaluation bias. On the onehand,itencouragedpositiveevaluations duetopositivereciprocityexpectations;on 82American Sociological Review 79(1)theotherhand,itdeterredtruthfulnegative evaluations due to negative reciprocity expec-tations.Withoutabandoningtheoldsystem, buyerswerenewlygiventheopportunityto leaveamoredetailedandanonymousseller rating.Thatis,afteratransaction,buyers could rate sellers on several attributessuch asitemdescription,communication,transac-tionspeed,andcostsandsellerswerenot abletoidentifythebuyerwhoratedthem because they were only shown average feed-backscoresoneachattribute.Surprisingly, althoughthecostsforleavingfeedback increased for buyers, both buyers and sellers conventionalratingbehaviorhardlychanged 10 weeks into the system change, and 70 per-centofbuyerswhogaveconventionalfeed-backalsogavedetailedsellerfeedback (Bolton et al. 2013). This provides additional supportforourassertionthatcooperationat thefeedbackstageislargelymotivatedby strongreciprocityandaltruism,ratherthan merestrategicconsiderationsencouragedby the two-sided feedback system.FundingThis research was partly supported by the Swiss National Science Foundation (grant number 100017_124877).AcknowledgmentsWe thank Stephanie Eckman, Debra Hevenstone, Claudia Jenny, and six anonymous reviewers for their perceptive and helpful comments.Data NoteAn online supplement to this article with additional infor-mation on previous research, data collection, and results is available from ASRs website (http://asr.sagepub.com/supplemental).Thedataandanalysisfilesusedforthis articlecanbedownloadedathttp://www.socio.ethz.ch/research/datafiles.Notes1.Mosteconomicandsocialexchangeissequential. Actor A transfers resources to actor B and expects B to reciprocate at some point in the future so that both A and B earn the gains from trade. A trust problem arises when A cannot expect B to reciprocate with certainty(Coleman1990). Wecall Asexpectation regardingBspropensitytoreciprocatetrust.A trusting A is more likely to transfer resources to B. We call Bs propensity to reciprocate trustworthi-ness. B can be trustworthy because she is commit-ted to act in As interest out of self-interest (Hardin 2002),orduetoother-regardingpreferencesand internalizednormsoffairnessandreciprocity(for othernotionsoftrust,seeCook,Hardin,andLevi 2007).2.eBayGermanyhas14.5millionactiveusers.We counted 142 eBay clubs with a total of 2,686 mem-bers representing .019 percent of active users (eBay .de,March28,2013). Asomewhathighernumber ofparticipantsareindiscussionforums,butrela-tive to the millions of users and transactions, their proportion is negligible. Moreover, repeated trans-actionswiththesametraderarerare.Thedensity ofthetradernetworkwillthusbeclosetozero, evenforsubmarkets.Finally,onlinerelationsmay betoothin(Hardin2004)toestablishnormsof good conduct to the same extent as offline relations would (Granovetter 1992).3.OneBay,traderscanleavefeedbackonlyaftera transaction,andforeverytransactionthemarket platform charges a small fee. Building a good repu-tation,evenwithshamtransactions,thusrequires timeandmoney.Thismakesagoodreputation costlytofakeandthereputationsystemfairly robustagainstdeliberateattemptstoinfluenceit. Onplatformswherethepossibilitytoleavefeed-back is not restricted to traders of a particular trans-action, reputations can be faked more easily. eBays reputationsystemisfurtherstabilizedbecausethe platform bans traders with too many negative evalu-ationsandmaypreventthemfromopeninganew account under a different alias.4.NotethatBeckersaltruismmodelisagnostic regarding the emotional triggers of altruistic behav-ior. Regret and guilt, for instance, can be subsumed under the model.5.Fehrandcolleagues(2002:3)pointoutthatthe essential feature of strong reciprocity is a willing-nesstosacrificeresourcesforrewardingfairand punishing unfair behavior even if this is costly and provides neither present nor future material rewards for the reciprocator (emphasis in original). Behav-ior as defined by strong reciprocity cannot therefore be explained by reciprocal altruism (Trivers 1971) orbyreputationandindirectreciprocity(Nowak and Sigmund 2005).6.Ahighnumberofratingscanbeinterpretedasan indicationofbeinganactiveraterduetothehigh correlation between incoming and outgoing ratings.7.ThetwomotivesdescribedinHypotheses5aand 5barenotmutuallyexclusiveandmaybepres-entsimultaneouslyinaheterogeneouspopulation. Empiricalestimates,however,willprovideinfor-mationaboutwhichmotiveispredominant.Other motives,suchasthewarmgloworsignaling, Diekmann et al.83mightalsobepresent,butweseenowaytodis-entangle them with our data. These motives would be consistent with a null effect of sellers scores on buyers behavior. Furthermore, effects described in Hypothesis5couldalsobecausedbyfearofdis-agreeablee-mailsfromthesellerifonedoesnot leave feedback. However, we cannot offer any clear ideas about how such fears would relate to a sellers score.8.Wecollecteddataforfivemoremodels,butwe excluded these cases from our analysis because the modelswerereleasedpriorto2004. Theexcluded cases are composed almost entirely of used products and thus are more heterogeneous than the offers of newer models. Including these older models in our analysis does not change our main findings.9.Wecalculatedthesenumbersbasedontraders entireratinghistories.Theyarelowerbounds fortheactualproportionsofrepeatedencounters because transactions appear in the feedback history only if a rating occurred.10.Censoring at the starting price might bias effects of positiveandnegativeratingsonpricebecauseour estimation is based on successful auctions only. To evaluatetherobustnessofourresults,weapplied variousmodelsthattakecensoringintoaccount (e.g.,censorednormalregressionandHeckmans sample selection model). Results are very similar in all models (not shown).11.Abootstraptestofthejointnull-hypothesisthat theabsolutecoefficientsofpositiveandnegative ratings are not larger in the model for used mobile phonesthaninthemodelfornewmobilephones yieldsap-valueof.04. Thisp-valueisanempiri-calstrengthprobabilitycomputedasthefraction ofbootstrapsamplesinwhichthepointestimates forthecoefficientsareinaccordancewiththe null-hypothesis(seeDavison,Hinkley,and Young 2003). The p-values for the separate hypotheses are .13 for positive and .10 for negative ratings.12.Wedeterminedthetotaleffectbycomputing averagemarginaleffectsonthelogarithmofthe expected price, where the expected price was com-puted as the predicted selling price weighted by the estimated probability of sale.13.Titlematchescouldnotbefoundforallauctions, even after removing special characters. The sample size in this analysis is thus reduced to 113,276 sold items.Theseitemsweresoldby15,545different sellersandcomprised18,054differenttitleswith 6.3 auctions per title, on average (the largest group beingHarryPotterandthePrisonerofAzkaban, with 383 auctions).14.Forthetwo-wayfixed-effectsmodel,weuse Schmieders(2009)implementationofanalgo-rithm proposed by Guimares and Portugal (2010).15.We chose the segments such that a sufficient num-berofratingeventsareavailableineachinterval. For positive feedback we use one-day intervals up to day 30, followed by two-day intervals up to day 50,followedbyfive-dayintervals.Fornegative feedback,theintervalsindaysare:110,1115, 1620,2125,2630,3140,4150,5170,and 7190. The exact choice of intervals does not seem to matter much for our findings.16.In the subsample used for the two-way fixed-effects estimation,theaverageconditionalprobabilityof feedbackbeinggivenonaspecificdayisabout3 percentforpositivefeedbackand.01percentfor negative feedback.ReferencesAkerlof,GeorgeA.1970.TheMarketforLemons: QualityUncertaintyandtheMarketMechanism. Quarterly Journal of Economics 84:488500.Allison, Paul D. 2009. Fixed Effects Regression Models. Thousand Oaks, CA: Sage.Andreoni, James. 1990. Impure Altruism and Donations toPublicGoods: A Theoryof Warm-GlowGiving. Economic Journal 100:46477.Ba,SulinandPaulA.Pavlou.2002.Evidenceofthe EffectofTrustBuildingTechnologyinElectronic Markets: Price Premiums and Buyer Behavior. MIS Quarterly 26:24368.Bajari,PatrickandAliHortacsu.2004.Economic InsightsfromInternetAuctions.JournalofEco-nomic Literature 42:45786.Becker,GaryS.1976.Altruism,EgoismandGenetic Fitness:EconomicsandSociobiology.Journalof Economic Literature 14:81726.Blau, Peter M. 1964. Exchange and Power in Social Life. New York: Wiley.Bolton, Gary E., Ben Greiner, and Axel Ockenfels. 2013. EngineeringTrust:ReciprocityintheProduction ofReputationInformation.ManagementScience 59:26585.Bolton, Gary E., Elena Katok, and Axel Ockenfels. 2004. HowEffectiveAreElectronicReputationMecha-nisms? An Experimental Investigation. Management Science 50:15871602.Brown, Martin, Armin Falk, and Ernst Fehr. 2004. Rela-tionalContractsandtheNatureofMarketInterac-tions. Econometrica 72:74780.Buskens,VincentandWernerRaub.2002.Embedded Trust: Control and Learning. Pp. 167202 in Group Cohesion, Trust and Solidarity, edited by E. J. Lawler and S. R. Thye. Amsterdam: JAI Press.Camerer, Colin F. 2003. Behavioral Game Theory. Princ-eton, NJ: Princeton University Press.Coleman, James S. 1990. Foundations of Social Theory. Cambridge, MA: The Belknap Press of Harvard Uni-versity Press.Cook,KarenS.,RussellHardin,andMargaretLevi. 2007. Cooperation Without Trust? New York: Russell Sage Foundation.Cook,KarenS.,EricR.W.Rice,and AlexandraGer-basi. 2004. The Emergence of Trust Networks Under 84American Sociological Review 79(1)Uncertainty:TheCaseofTransitionalEconomiesInsightsfromSocialPsychologicalResearch.Pp. 193212inCreatingSocialTrustinPost-Socialist Transition, edited by J. Kornai, B. Rothstein, and S. Rose-Ackerman. New York: Palgrave MacMillan.Cosmides, Leda and John Tooby. 1992. Cognitive Adap-tationsforSocialExchange.Pp.163228inThe Adapted Mind, edited by J. H. Barkow, L. Cosmides, and J. Tooby. New York: Oxford University Press.Dasgupta,Partha.1988.TrustasaCommodity.Pp. 4972inTrust:MakingandBreakingCooperative Relations,editedbyD.Gambetta.Oxford:Basil Blackwell.Davison, A.C.,D.V.Hinkley,andG. A. Young.2003. RecentDevelopmentsinBootstrapMethodology. Statistical Science 18:14157.Dellarocas,Chrysanthos,MingFan,andCharlesA. Wood.2004.Self-Interest,Reciprocity,andPar-ticipationinOnlineReputationSystems.Working Paper 205, MIT Center for Digital Business.Dellarocas,ChrysanthosandCharlesA.Wood.2008. The Sound of Silence in Online Feedback: Estimat-ing Trading Risks in the Presence of Reporting Bias. Management Science 54:46076.Diekmann, Andreas, Ben Jann, and David Wyder. 2009. Trust and Reputation in Internet Auctions. Pp. 13965ineTrust:FormingRelationshipsintheOnline World, edited by K. S. Cook, C. Snijders, V. Buskens, and C. Cheshire. New York: Russell Sage.DiMaggio,PaulandHughLouch.1998.Socially EmbeddedConsumer Transactions:For WhatKinds of Purchases Do People Most Often Use Networks? American Sociological Review 63:61937.Downs, Anthony. 1957. An Economic Theory of Democ-racy. New York: Harper & Row.Fehr, Ernst, Urs Fischbacher, and Simon Gchter. 2002. StrongReciprocity,HumanCooperation,andthe Enforcement of Social Norms. Human Nature 13:125.Gambetta, Diego. 2009. Signaling. Pp. 16894 in The Oxford Handbook of Analytical Sociology, edited by P.HedstrmandP.Bearman.Oxford:OxfordUni-versity Press.Gintis,Herbert.2000.StrongReciprocityandHuman Sociality.JournalofTheoreticalBiology206:16979.Gouldner, Alvin W. 1960. The Norm of Reciprocity: A Preliminary Statement. American Journal of Sociol-ogy 25:16178.Granovetter,Mark.1992.ProblemsofExplanationin EconomicSociology.Pp.2556inNetworksand Organizations:Structure,Form,andAction,edited by N. Nohira and R. G. Eccles. Boston: Harvard Busi-ness School Press.Greif, Avner. 1989. Reputation and Coalitions in Medi-eval Trade: Evidence on the Maghribi Traders. Jour-nal of Economic History 49:85782.Guala,Francesco.2012.Reciprocity: WeakorStrong? WhatPunishmentExperimentsDo(andDoNot) Demonstrate. Behavioral and Brain Sciences 35:115.Guimares,PauloandPedroPortugal.2010.ASimple FeasibleAlternativeProceduretoEstimateModels with High-Dimensional Fixed Effects. Stata Journal 10:62849.Gulati,RanjayandMartinGargiulo.1999.WhereDo InterorganizationalNetworksComeFrom?Ameri-can Journal of Sociology 104:143993.Gth, Werner and Axel Ockenfels. 2003. The Coevolu-tion of Trust and Institutions in Anonymous and Non-anonymousCommunities.Pp.15774inJahrbuch fr Neue Politische konomie, edited by M. J. Holler, H. Kliemt, D. Schmidtchen, and M. Streit. Tbingen: Mohr Siebeck.Hardin,Russell.2002.TrustandTrustworthiness.New York: Russell Sage Foundation.Hardin, Russell. 2004. Internet Capital. Analyse & Kri-tik 26:12238.Heath,Anthony.1976.RationalChoiceandSocial Exchange:ACritiqueofExchangeTheory.Cam-bridge: Cambridge University Press.Heckathorn, Douglas D. 1989. Collective Action and the Second-OrderFree-RiderProblem.Rationalityand Society 1:78100.Jappelli,TullioandMarcoPagano.2002.Information Sharing,LendingandDefaults:Cross-CountryEvi-dence. Journal of Banking & Finance 26:201745.Jian, Lian, Jeffrey K. MacKie-Mason, and Paul Resnick. 2010.IScratched Yours: ThePrevalenceofRecip-rocationinFeedbackProvisiononeBay.TheB.E. Journal of Economic Analysis & Policy 10:Article 92.Jones,Candace,WilliamS.Hesterly,andStephenP. Borgatti. 1997. A General Theory of Network Gov-ernance:ExchangeConditionsandSocialMecha-nisms. Academy of Management Review 22:91145.Kollock,Peter.1994.TheEmergenceofExchange Structures:AnExperimentalStudyofUncertainty, Commitment, and Trust. American Journal of Soci-ology 100:31345.Kollock, Peter. 1999. The Production of Trust in Online Markets.Pp.99123inAdvancesinGroupPro-cesses,Vol.16,editedbyE.J.LawlerandM.W. Macy. Amsterdam: JAI Press.Livingston,Jeffrey A.2003.What AttractsaBidderto aParticularInternet Auction?Pp.16587inOrga-nizing the New Industrial Economy, edited by M. R. Baye. Amsterdam: Elsevier.Long, J. Scott. 1997. Regression Models for Categorical andLimitedDependentVariables.ThousandOaks, CA: Sage.Lucking-Reiley,David,DougBryan,NaghiPrasad, andDanielReeves.2007.PenniesfromeBay:The DeterminantsofPriceinOnline Auctions.Journal of Industrial Economics 55:22333.Malinowski, Bronislaw. 1922. Argonauts of the Western Pacic: An Account of Native Enterprise and Adven-ture in the Archipelagoes of Melanesian New Guinea. London: Routledge & Kegan Paul.Diekmann et al.85Malinowski, Bronislaw. 1926. Crime and Custom in Sav-age Society. London: Routledge & Kegan Paul.Mauss,Marcel.[1950]1990.TheGift:TheFormand ReasonforExchangeinArchaicSociety.London: Routledge.Milgrom, Paul R., Douglass C. North, and Barry R. Wein-gast. 1990. The Role of Institutions in the Revival of Trade:TheLawMerchant,PrivateJudges,andthe Champagne Fairs. Economics and Politics 2:123.Mood,Carina.2010.LogisticRegression:WhyWe CannotDoWhatWeThinkWeCanDo,andWhat We Can Do About It. European Sociological Review 26:6782.Nowak,Martin A.andKarlSigmund.2005.Evolution of Indirect Reciprocity. Nature 437:129198.Przepiorka,Wojtek.2013.BuyersPayForandSellers InvestinaGoodReputation:MoreEvidencefrom eBay. Journal of Socio-Economics 42:3142.deQuervain,DominiqueJ.-F.,UrsFischbacher,Valerie Treyer,MelanieSchellhammer,UlrichSchnyder, Alfred Buck, and Ernst Fehr. 2004. The Neural Basis of Altruistic Punishment. Science 305:125458.Resnick,PaulandRichardZeckhauser.2002.Trust amongStrangersinInternet Transactions:Empirical Analysis of eBays Reputation System. Pp. 12757 inTheEconomicsoftheInternetandE-Commerce, edited by M. R. Baye. Amsterdam: Elsevier.Resnick,Paul,RichardZeckhauser,EricFriedman,and Ko Kuwabara. 2000. Reputation Systems. Commu-nications of the ACM 43:4548.Resnick,Paul,RichardZeckhauser,JohnSwanson,and Kate Lockwood. 2006. The Value of Reputation on eBay: A Controlled Experiment. Experimental Eco-nomics 9:79101.Schmieder,JohannesF.2009.gpreg:StataModuleto EstimateRegressionswithTwoDimensionalFixed Effects. Boston College, Department of Economics, Statistical Software Components, S457048.Shapiro,Carl.1983.PremiumsforHighQualityProd-uctsasReturntoReputation.QuarterlyJournalof Economics 98:65980.Snijders,ChrisandJeroenWeesie.2009.OnlinePro-gramming Markets. Pp. 16685 in eTrust: Forming RelationshipsintheOnlineWorld,editedbyK.S. Cook, C. Snijders, V. Buskens, and C. Cheshire. New York: Russell Sage Foundation.Trivers,RobertL.1971.TheEvolutionofReciprocal Altruism. Quarterly Review of Biology 46:3557.Uzzi, Brian. 1997. Social Structure and Competition in Interfirm Networks: The Paradox of Embeddedness. Administrative Science Quarterly 42:3567.Yamagishi, Toshio, Karen S. Cook, and Motoki Watabe. 1998.Uncertainty,TrustandCommitmentForma-tion in the United States and Japan. American Jour-nal of Sociology 104:16594.Yamagishi,Toshio,MasafumiMatsuda,Noriaki Yoshikai,HiroyukiTakahashi,andYukihiroUsui. 2009.SolvingtheLemonsProblemwithReputa-tion. Pp. 73108 in eTrust: Forming Relationships in the Online World, edited by K. S. Cook, C. Snijders, V. Buskens, and C. Cheshire. New York: Russell Sage Foundation.Zucker,LynneG.1986.ProductionofTrust:Institu-tionalSourcesofEconomicStructure,18401920. Research in Organizational Behavior 8:53111.AndreasDiekmannisProfessorofSociologyatthe Swiss Federal Institute of Technology, ETH-Zurich. His research interests focus on theories of social cooperation, experimentalgametheory,researchmethodsandstatis-tics,andenvironmentalandpopulationsociology.His publicationshaveappearedinProceedingsoftheRoyal Society B, the Journal of Conflict Resolution, Sociological Methods&Research,andtheAmericanSociological Review.Heiscurrentlyconductingresearchonsocial norms and institutions, and on environmental protests in China.BenJannisAssociateProfessorofSociologyatthe UniversityofBern,Switzerland.Hisresearchinterests include social-science methodology, statistics, social strat-ification, and labor market sociology. Recent publications include papers in Sociological Methodology, Sociological Methods & Research, and Public Opinion Quarterly.WojtekPrzepiorkaisaresearchfellowatNuffield College and the Department of Sociology at University of Oxford. His research interests are in quantitative method-ology,inparticularexperimentalmethods,analytical sociology,behavioralgametheory,andtheevolutionof socialcooperation.Hisworkhasbeenpublishedin EuropeanSociologicalReview,EvolutionandHuman Behavior, and Proceedings of the Royal Society B. In his current research, he studies the signaling benefits of altru-istic acts and mechanisms of social norm enforcement.StefanWehrliisaPhDcandidateinsociologyatthe Swiss Federal Institute of Technology, ETH-Zurich, and manageroftheETHDecisionScienceLaboratory (DeSciL). His research interests include decision science and social network analysis.