Creating Social Contagion Through Viral Product Social Contagion Through Viral Product ... social networks; viral marketing; viral product design ... Creating Social Contagion Through Viral Product ...

  • Published on

  • View

  • Download


  • MANAGEMENT SCIENCEVol. 57, No. 9, September 2011, pp. 16231639issn 0025-1909 eissn 1526-5501 11 5709 1623

    2011 INFORMS

    Creating Social Contagion ThroughViral Product Design: A RandomizedTrial of Peer Influence in Networks

    Sinan Aral, Dylan WalkerInformation, Operations, and Management Sciences, Stern School of Business, New York University,

    New York, New York 10012 {,}

    We examine how firms can create word-of-mouth peer influence and social contagion by designing viralfeatures into their products and marketing campaigns. To econometrically identify the effectiveness ofdifferent viral features in creating social contagion, we designed and conducted a randomized field experimentinvolving the 1.4 million friends of 9,687 experimental users on We find that viral featuresgenerate econometrically identifiable peer influence and social contagion effects. More surprisingly, we find thatpassive-broadcast viral features generate a 246% increase in peer influence and social contagion, whereas addingactive-personalized viral features generate only an additional 98% increase. Although active-personalized viralmessages are more effective in encouraging adoption per message and are correlated with more user engagementand sustained product use, passive-broadcast messaging is used more often, generating more total peer adoptionin the network. Our work provides a model for how randomized trials can identify peer influence in socialnetworks.

    Key words : peer influence; social contagion; social networks; viral marketing; viral product design; informationsystems; randomized experiment

    History : Received February 28, 2010; accepted March 31, 2011, by Pradeep Chintagunta and Preyas Desai,special issue editors. Published online in Articles in Advance August 4, 2011.

    1. IntroductionIt is widely believed that social contagion and word-of-mouth (WOM) buzz about products drive prod-uct adoption and sales, and firms increasingly rely onnetwork and viral marketing strategies (Hill et al.2006, Manchanda et al. 2008, Nam et al. 2010). Yet,whereas most current work has focused on viral mar-keting campaigns for existing products, less attentionhas been paid to whether (and how) firms can designproducts that are themselves more likely to go viral.The effectiveness of such viral product design strategieshave yet to be examined or causally estimated. Wetherefore conducted a large-scale randomized fieldexperiment to test the effectiveness of different viralproduct design features in creating peer influence andsocial contagion in new product diffusion.

    Viral product designthe process of explicitly engi-neering products so they are more likely to be sharedamong peershas existed at least since the firstchain letter was sent in 1888. Today, products regu-larly use information technology (IT)-enabled featureslike automated broadcast notifications and person-alized invitations to spread product awareness. Yet,although viral features have become more sophisti-cated and a central part of the design of products

    and marketing campaigns, there is almost no empir-ical evidence on the effectiveness of such features ingenerating social contagion and product adoption. Wetherefore investigate two basic questions: Can firmsadd viral features to products so they are more likelyto be shared among peers? If so, which viral featuresare most effective in inducing WOM and peer-to-peerinfluence in product adoption?

    Unfortunately, evaluating the effects of viral prod-uct design features is difficult because peer effectsand WOM are typically endogenous (Manski 1993;Godes and Mayzlin 2004, 2009; Hartmann et al. 2008;Aral et al. 2009; Aral 2011). We therefore designedand conducted a randomized field experiment test-ing the effectiveness of two of the most widely usedviral product featuresactive-personalized referralsand passive-broadcast notificationsin creating peerinfluence and social contagion among the 1.4 millionfriends of 9,687 experimental users of experiment uses a customized commercialFacebook application to observe user behavior, com-munications traffic, and the peer influence effectsof randomly enabled viral messaging features onapplication diffusion and use in the local networksof experimental and control population users. By


  • Aral and Walker: Creating Social Contagion Through Viral Product Design1624 Management Science 57(9), pp. 16231639, 2011 INFORMS

    enabling and disabling viral features among ran-domly selected users, we were able to obtain rela-tively unbiased causal estimates of the impact of viralfeatures on the adoption rates of peers in the localnetworks of adopters. Using detailed clickstream dataon users online behaviors, we also explored whetherpositive network externalities generated by additionalpeer adopters inspired further product adoption andsustained product use.

    WOM is generally considered to be more effec-tive at promoting product contagion when it ispersonalized and active. Surprisingly, we find thatdesigning products with passive-broadcast viral mes-saging capabilities generates a 246% increase inlocal peer influence and social contagion, whereasadding active-personalized viral messaging capa-bilities generates only an additional 98% increase.Although active-personalized messaging is moreeffective in encouraging adoption per message andis correlated with more user engagement and sus-tained product use, it is used less often and there-fore generates less total peer adoption in the network.Overall, we find that viral product design featuresgenerate econometrically identifiable peer influenceand social contagion effects and provide a model forhow randomized trials can identify peer influence innetworks.

    2. Viral Product DesignSince the early work of Katz and Lazersfled (1955)there has been great interest in how WOM drives con-sumer demand, public opinion, and product diffusion(Brown and Reingen 1987, Godes and Mayzlin 2004,Aral et al. 2009) and how firms can create broad, sys-tematic propagation of WOM through consumer pop-ulations (Phelps et al. 2004, Mayzlin 2006, Dellarocas2006, Godes and Mayzlin 2009). Many campaignstarget influential individuals who are likely topropagate organic WOM most broadly (Katz andLazersfeld 1955, Watts and Dodds 2007, Goldenberget al. 2009), using referral programs to create incen-tives for them to spread the word (Biyalogorskyet al. 2001). Others use observational evidence onviral campaigns to inform viral branching models ofWOM diffusion (Van der Lans et al. 2010). However,to this point, studies of viral product design haveremained conspicuously absent from the literature onviral marketing.

    Viral product design involves incorporating specificcharacteristics and features into a products design togenerate peer-to-peer influence that encourages adop-tion. A products viral characteristics are fundamentallyabout its content and the psychological effects contentcan have on a users desire to share the product withpeers (Stephen and Berger 2009, Berger and Heath

    2005, Heath et al. 2001). A products viral features, onthe other hand, concern how the product is sharedhow features enable and constrain a products use inrelation to other consumers. Viral features may enablecommunication, generate automated notifications ofusers activities, facilitate personalized invitations, orenable hypertext embedding of the product on pub-licly available websites and weblogs. Two of the mostwidely used viral product features are personalizedreferrals and automated broadcast notifications:

    Personalized Referrals. Personalized referral featuresallow users to select their friends or contacts froma list and invite them to adopt the product or ser-vice, with the option of attaching a personalized mes-sage to the invitation. Social networking websitesenable users to invite their friends to join the servicethrough personalized referrals. When users send Web-based e-mail messages, for example, from Gmail, anautomated, pop-up hyperlink enables them to inviterecipients to join the service.

    Automated Broadcast Notifications. Automated broad-cast notifications are passively triggered by normaluser activity. When a user engages the product in acertain way (e.g., sends a message, updates his orher status), those actions are broadcast as notifica-tions to the users list of contacts. Notifications buildawareness among friends of new activities or prod-ucts a user is adopting or engaging with, and canencourage those friends to eventually adopt the prod-uct themselves. For example, social networking web-sites typically notify friends automatically when auser adopts a new application or achieves some appli-cation milestone.

    Referrals are more personalized and targeted thanbroadcast notifications. Users actively select a sub-set of their social network to receive them (target-ing) and can include personal messages in the referral(personalization). WOM is generally considered moreeffective at promoting product contagion when it ispersonalized and active. When individuals chooseto share information about products and serviceswith their friends, they tend to activate their strong-tie relationships (Frenzen and Nakamoto 1993, Araland Van Alstyne 2011). Strong ties exhibit greaterhomophily (Jackson 2008), greater pressure for con-formity (Coleman 1988), and deeper knowledge aboutone another. We tend to trust information from closetrusted sources more and to respond more oftento them because of reciprocity (Emerson 1962). Inaddition, the personalization of messages makes themmore effective, especially in online environments inwhich we are bombarded with irrelevant information(Tam and Ho 2005, Tucker 2010).

    For these reasons, one might suspect that personal-ized referrals are more effective than broadcast notifi-cations. But, although each personalized referral may

  • Aral and Walker: Creating Social Contagion Through Viral Product DesignManagement Science 57(9), pp. 16231639, 2011 INFORMS 1625

    be more persuasive (more effective per message), thepervasiveness of automated broadcast messages thatdo not require additional time and energy on the partof the user may lead to greater overall peer adop-tion. The effort required by the user to actively selectand invite peers to adopt the product may curtailwidespread use of the personalized referral and solimit its effectiveness in encouraging broad adoption.The relative overall effectiveness of these viral fea-tures is therefore ultimately an empirical question.

    3. Experimental Design andProcedures

    Evaluating the effects of viral product design fea-tures on social contagion is difficult because peereffects and WOM are typically endogenous (Manski1993, Hartmann et al. 2008, Van den Bulte and Lilien2001, Godes and Mayzlin 2004, Van den Bulte andIyengar 2011). Several approaches for identifying peereffects have been proposed, including peer effectsmodels and extended spatial autoregressive mod-els (e.g., Oestreicher-Singer and Sundararajan 2008,Trusov et al. 2009, Bramoulle et al. 2009), actor-oriented models (e.g., Snijders et al. 2006), instrumen-tal variable methods based on natural experiments(e.g., Sacredote 2001, Tucker 2008), dynamic matchedsample estimation (Aral et al. 2009), structural models(e.g., Ghose and Han 2010), and ad hoc approaches(Christakis and Fowler 2007). However, randomizedtrials are considered to be one of the most effec-tive ways to obtain unbiased estimates of causal peereffects (Duflo et al. 2008, Hartmann et al. 2008).

    We therefore partnered with a firm that developscommercial applications hosted on the popular socialnetworking website and collectedexperimental data on the peer influence effects ofenabling viral features on the diffusion one of theirapplications. This application is free to adopt and pro-vides users the opportunity to share information andopinions about movies, actors, directors, and the filmindustry in general. We designed multiple experimen-tal versions of the application in which personalizedinvitations and broadcast notifications were enabled ordisabled, and randomly assigned adopting users tovarious experimental and control conditions. As usersadopted the application, each was randomly assignedto one of the two treatment conditions or the baselinecontrol condition. The application collected personalattributes and preferences from users Facebook pro-files, as well as data on their social networks and thepersonal attributes and preferences of their networkneighbors.1

    1 Facebook allows users to specify privacy settings that may restrictan applications access to some or part of their profile. This is

    The experiment enabled experimental group usersto use passive-broadcast and active-personalized viralmessaging capabilities to exchange messages withtheir network neighbors, while disabling those fea-tures for the baseline control group. The applicationthen recorded data on the use of these viral fea-tures by experimental group users, as well as click-stream data on recipient responses to viral messagesand their subsequent adoption and use of the appli-cation. When an individual adopted the applicationas a result of peer influence, their treatment sta-tus was also randomized to ensure that the stableunit treatment value assumption held. This facilitatedanalysis of the relative effectiveness of different viralmessaging channels in generating peer adoption andnetwork propagation. Randomization also enabledexploration of the mechanisms by which a particularviral channel influenced recipient behavior. Two pri-mary viral features were examined:

    Automated Broadcast Notifications (Notifications).When enabled, notifications were generated automat-ically when an application user performed certainactions within the application, such as declaring afavorite movie or writing a movie review. Whennotifications were generated, they were distributedto a random subset of an application users peersand displayed in a status bar at the bottom of thepeers Facebook environment. When a peer clickedon the notification, they were taken to an applicationcanvas page where they were given the option toinstall the application. These notifications required noeffort beyond normal application use, making theirengagement relatively costless to the user. Becausethey were randomly distributed to a Facebook userspeers and were not accompanied by a personalizedmessage, they also exhibited low personalization.

    Personalized Referrals or Invitations (Invites). Whenenabled, invites allowed application users to sendtheir Facebook peers personalized invitations toinstall the application. Peers received the invitation intheir Facebook inbox and could click on a referral linkcontained within the invitation. If they did so, theywere taken to the application canvas page where theywere given the opportunity to install the application.Each invite required a conscious and deliberate actionfrom the user be...


View more >