1

TO WHOM DO CONSUMERS TRANSMIT ORGANIC WORD-OF-MOUTH?

Andrew T. Stephen INSEAD

Donald R. Lehmann Columbia University

November 10, 2010

Andrew T. Stephen ([email protected]) is an Assistant Professor of Marketing at INSEAD, Boulevard de Constance, Fontainebleau 77305, France, and Donald R. Lehmann ([email protected]) is the George E. Warren Professor of Business at the Graduate School of Business, Columbia University, 3022 Broadway, Uris Hall, New York, NY 10027, USA. The authors thank Kamel Jedidi, Leonard Lee, Jeffrey Parker, Olivier Toubia, Christophe Van den Bulte, Duncan Watts, and participants in the Columbia Marketing brown bag for their comments, Shannon Edwards and Fiona Lazar for data collection assistance, and the INSEAD Alumni Fund for funding. This paper is based on Andrew Stephen’s doctoral dissertation at Columbia University.

2 TO WHOM DO CONSUMERS TRANSMIT ORGANIC WORD-OF-MOUTH?

ABSTRACT

For widespread buzz about brands or products to spread through word-of-mouth (WOM), information often must pass through consumers who are socially well connected (i.e., “hubs”). Consumers, however, may not always share information with these well-connected consumers, making it harder for information to spread rapidly and widely throughout a population. Results from five studies examining WOM transmitters’ choices of receivers (who they talk to) show that transmitters prefer to talk to receivers who are “good listeners” (e.g., people who are interested in the topic) more than “good disseminators” (e.g., people who are socially well connected), and that for many typical transmission motives (e.g., expressing opinions, giving information, seeking information) receiver connectivity either does not influence transmission or negatively impacts it such that well connected receivers are less likely to be talked to. The tendency to take into account receiver connectivity when transmitting WOM can be increased by making salient benefits of talking to well connected consumers or by creating positive externalities on the information being transmitted.

3 One reaction to the growing influence and importance of Internet-based communication has been the increased use of marketing campaigns based on word-of-mouth (WOM), such as buzz/viral marketing and, more generally, social media marketing. Unlike traditional advertising campaigns or sales promotions, these marketing campaigns rely on consumer-to-consumer social interactions to spread WOM about products to raise awareness, generate interest, and, hopefully, influence purchase decisions. “Firm-generated” WOM campaigns typically seed small groups of consumers with information and then rely on them to spread desired information broadly throughout their social networks both online and offline. Seeding can be done in many ways including through advertising (e.g., “Big Seed” marketing; Watts and Peretti 2007), though it is more commonly done through sampling or by offering discounts or coupons to consumers (e.g., Godes and Mayzlin 2009; Toubia, Stephen, and Freud 2010) or business customers such as physicians (e.g., Coleman, Katz, and Menzel 1957; Iyengar, Van den Bulte, and Valente 2010). A growing body of literature focuses on how to effectively deploy viral marketing and related WOM-based marketing techniques (e.g., Berger and Schwartz 2010; Godes and Mayzlin 2009; Toubia, Stephen, and Freud 2010; Van der Lans, van Bruggen, Eliashberg and Wierenga 2010). While such techniques can be useful, the vast majority of consumer-to-consumer WOM is voluntary, not incented by firms, and occurs outside firms’ control (cf. Libai, Bolton, Bügel, de Ruyter, Götz, Risselada, and Stephen 2010). We refer to WOM that is not explicitly solicited or firm generated as “organic” in the sense that it occurs naturally as part of consumers’ daily lives. In these cases, firms rely on consumers to spread (hopefully positive) information about their products without formally and explicitly trying to generate it themselves. Departing from recent literature on firm-generated WOM and viral marketing, this paper focuses on organic WOM transmissions by examining to whom consumers naturally tend to transmit product-related

4 WOM. While firm-generated WOM campaigns are undeniably important and potentially effective marketing tools, at best they only allow firms to exert influence over early-stage “seed” consumers who are directly targeted. What happens in subsequent generations of WOM transmissions usually is not within the firm’s control. Thus, even firm generated WOM—if it continues to spread beyond the first generation—becomes organic WOM. In this paper we focus on individual consumers’ choices of who to talk to in organic WOM transmission situations; i.e., the selection of WOM receivers. By virtue of peoples’ social networks, these micro-level organic WOM transmission behaviors have macro-level consequences for wider information diffusion. Thus, these micro decisions must be understood in order to more fully understand information diffusion over networks. An inherent challenge, however, is that the mechanisms required to achieve widespread and rapid information diffusion are complex, multiply determined, and not easily controlled. An individual (“transmitter”) has to (i) start talking about a particular product to others (“receivers”) who will (ii) pay attention, (iii) be interested and therefore listen to what the transmitter has to say, as well as (iv) be likely to pass on (“retransmit”) the information to others, which implies that (v) receivers need to be well connected in their social networks. Getting a consumer to talk about a particular product (in a positive way) and not one of the many other topics they could conceivably talk about is not trivial; getting them to talk to someone who will listen and is likely to talk to other people who will also listen, are well connected, and likely to talk to others is even more difficult. This paper shows that while organic WOM transmitters naturally tend to prefer to talk to receivers who are likely to listen to them (“good listeners”), they do not have a similar proclivity for receivers who are well connected (e.g., social hubs with many friends) and thus wellpositioned to spread information widely (“good disseminators”). This implies an interesting but

5 troublesome paradox: organic WOM transmissions are unlikely to put information into the hands of the people who are socially best positioned to spread it rapidly and widely. Although socially well-connected consumers are preferred seeds in firm-generated WOM campaigns because they facilitate faster and wider information diffusion over networks (Goldenberg, Han, Lehmann, and Hong 2009), in organic WOM transmissions they are not disproportionately favored as receivers. We show that depending on a transmitter’s underlying motives, a potential receiver’s connectivity either has no effect on their likelihood of being talked to or in fact negatively impacts their chances of receiving information.

RESEARCH OVERVIEW AND CONCEPTUAL FRAMEWORK This paper focuses on the critical initial component of the complex process of individuallevel organic WOM: to whom do consumers talk? Specifically, if a consumer has information (e.g., an opinion about a product), and is not explicitly incentivized or influenced by an external party (e.g., a firm) to spread the information, to which people they come in contact with will they convey this information? Here this question is addressed from the perspective of marketers’ desired macro-level WOM outcomes: information spreading rapidly and widely throughout a population. To have any chance at achieving this, receivers must (i) listen to transmitters (e.g., they must be interested), and (ii) be well positioned in their networks to help diffuse information broadly if they retransmit it (i.e., they must be well connected and centrally positioned). For rapid and widespread global information diffusion to occur, the second characteristic is especially critical. While getting information to transfer from transmitter to receiver within a dyad requires the receiver to be a willing listener, this is only enough for local information diffusion. Consequently we focus on the likelihood of an individual transmitter spreading

6 information to socially well-connected people; i.e., to those individuals who are in contact with a large number of other people and therefore are “good disseminators.” We hypothesize that consumers focus more on factors related to themselves than on factors associated with achieving macro diffusion. In other words, we predict that organic WOM transmitters will be less influenced by how connected receivers are and more influenced by factors associated with either themselves or the transmitter—receiver relationship. For example, when sharing one’s opinion on a topic with others, a primary goal for a transmitter may simply be to be listened to by whomever she decides to talk to. Thus, this transmitter would prefer to talk to those people who she knows are a good fit with the topic (e.g., because they have an interest in it, such as talking about movies with a movie buff or fashion with a fashionista). Whether or not a receiver is highly connected would likely be less important. The obvious implication of this is that, unless the highly receptive and interested receiver happens to also be well connected, the transmitter will not get information into the hands of the consumers who are best-positioned to spread the word to many others quickly. We explore this hypothesis and related effects across five studies. Study 1 surveys consumers to find out what characteristics they look for in receivers when they are transmitting organic WOM for a variety of product/service categories. Study 2 examines transmitters’ preferences for a variety of receiver characteristics to see whether high social connectivity is important in organic WOM transmission decisions. Study 3 documents reasons motivating actual WOM transmissions, and examines how they relate to receiver connectivity. Overall, studies 1-3 show, consistent with our hypothesis, that receiver connectivity has a relatively minor impact on transmitters’ choices of who to talk to. Studies 4 and 5 examine why this is the case and test ways to overcome the apparent tendency of transmitters to not be inclined to share information

7 with high-connectivity receivers, illustrating some approaches managers could use to encourage consumers to transmit WOM to receivers with high connectivity so that information gets into the right hands for achieving rapid, widespread diffusion. In sum, the first three studies examine natural proclivities for transmitting WOM to different types of receivers, and the final two studies offer process insights and some approaches for overcoming the tendency to not transmit WOM to well-connected receivers.

PRIOR RESEARCH AND BACKGROUND THEORY Past Research Despite decades of research on WOM in marketing (e.g., Arndt 1967; Brooks 1957; Brown and Reingen 1987; Czepiel 1974; Frenzen and Nakamoto 1993; Godes and Mayzlin 2004; Goldenberg, Libai, and Muller 2001; Liu 2006), and related work on social contagion in other fields (e.g., Coleman, Katz, and Menzel 1957; Van den Bulte and Lilien 2001; Watts 2002), a great deal is still unknown about individual-level organic WOM transmission behavior. Exceptions that focus on individual-level psychological aspects of WOM transmission include Berger and Schwartz (2010), Cheema and Kaikati (2010), Gershoff, Mukherjee, and Mukhopadhyay (2007), and Wojnicki and Godes (2009). Nonetheless, the dominant focus in extant literature has been on aggregate consequences of WOM. Early research established the importance of WOM as a driver of new product diffusion (e.g., Arndt 1967; Brooks 1957; Czepiel 1974), and examined the influence of referrals and recommendations on consumption (e.g., Brown and Reingen 1987; Reingen and Kernan 1986). WOM has been an important component in sociologists’ models of social contagion (Coleman, Katz, and Menzel 1957; Katz and Lazarsfeld 1955; Watts 2002), which facilitates the flow of

8 information in social networks (Frenzen and Nakamoto 1993). More recent work has examined online WOM in the form of consumer-generated content such as ratings and reviews in the contexts of, e.g., book sales (Chevalier and Mayzlin 2006), television show ratings (Godes and Mayzlin 2004), retailer profitability (Mayzlin and Moe 2009), and movie revenues (Liu 2006). Our focus is on the antecedents—as opposed to the consequences—of organic WOM transmissions as they relate to macro aggregate-level diffusion outcomes. This aspect of WOM largely has been neglected and deserves attention (cf. Godes et al. 2005; Libai et al. 2010). The probability a person transmits WOM to someone else is determined by factors related to the transmitter (e.g., “opinion leaders” might be more likely to transmit), the receiver (e.g., someone with an interest in a product category might be more likely to receive WOM about products in that category), and the product itself (e.g., people are more likely to talk about topical or massappeal products, and less about socially offensive ones). WOM behaviors are somewhat deliberate, considered, and not completely spontaneous or serendipitous (Banerjee and Fudenberg 2004; Ellison and Fudenberg 1995). We therefore assume that transmitters behave selectively (“strategically”) in terms of whether or not they transmit a given piece of information to a potential receiver. Note that we do not consider situations where WOM transmissions are more spontaneous or serendipitous (e.g., small talk at a party), or, as discussed above, firm generated/incented (e.g., in viral marketing campaigns). Rather, we focus entirely on situations where a consumer has some information and has to decide with whom to share that information. Essential Elements of Word-of-Mouth Transmission for Macro Diffusion To achieve widespread information diffusion, consumers need to (i) be motivated to transmit WOM (Frenzen and Nakamoto 1993; Granovetter 1973), and when they do, transmit to

9 receivers who are (ii) receptive, i.e., listen and pay attention, and (iii) socially well connected (e.g., many friends) so they have multiple opportunities to pass information on to others. The first two requirements ensure that information is transferred from the transmitter to the receiver (i.e., merely talking to a receiver is not enough; for the message to get across the receiver has to listen and pay attention). However, they do not guarantee further information spread. The likelihood of that lies in the third requirement, which is the main focus of this paper. People with more connections have opportunities to pass information along to more people (Sahlins 1972), and products diffuse faster when well-connected people (hubs) adopt (Goldenberg et al. 2009). Widespread information diffusion is more likely in networks that have lots of hubs, and tipping points are more likely to be crossed when more hubs have received the information (Watts 2002). While some consumers are more likely to transmit WOM about certain products than others (e.g., “market mavens” or “opinion leaders”), who the receivers are matters a great deal. If a transmitter (person A) talks to a friend (person B) who has no friends other than the transmitter, then information flow will stop. Conversely, if this transmitter talks to another friend (person C), who has other friends, then information flow has a chance of continuing. If person A wants their information to spread, then presumably they will prefer transmitting to C over B. From a macro diffusion perspective, this is also the desired choice. However, what if A thought that B would be more receptive, attentive, and likely to listen? Which receiver characteristic—likelihood of listening or social connectivity—will drive A’s transmission decision? To understand when organic WOM transmission leads to desirable macro outcomes, we examine transmitters’ receiver choices (i.e., who they talk to) as a function of receiver characteristics. We also distinguish between whether the transmitted information is first-hand

10 (based on direct experience) or second-hand (being passed on), which we term initial transmission and retransmission, respectively. Initial transmission occurs when a transmitter talks about something that they have experienced personally. Retransmission, on the other hand, occurs when a transmitter talks about something that they have learned about through WOM from someone else. Initial transmission is important for starting diffusion or injecting fresh information into a network, while retransmission accelerates information spread.

STUDY 1: INITIAL EVIDENCE ON WHAT TRANSMITTERS LOOK FOR IN RECEIVERS We first used a survey to ascertain to whom potential information transmitters would tend to transmit information. A total of 448 consumers participated in a two-part online survey. First, participants were asked to imagine themselves in a situation where they could transmit WOM about a product/service to someone who they knew (e.g., friend, co-worker, relative, classmate). They were then asked to write down the single most important characteristic that they would look for in a receiver, and also rated on a seven-point scale how familiar this kind of WOM transmission situation was to them (1 = extremely unfamiliar to 7 = extremely familiar). The mean was 5.66 (SD = 1.12), indicating the situation was familiar to them. The open-ended responses were divided into three categories: (i) listening (e.g., “I want a receiver who will listen to me”), (ii) disseminating (e.g., “I want a receiver who can get this information out to others”), and (iii) “other.” Of the 448 receiver characteristics provided, 57.1% were explicitly associated with listening and about one-quarter (25.4%) of these responses included the term “good listener.” Only 3.8% of the receiver characteristics mentioned anything related to spreading, transmitting, or disseminating information to others. The remaining 39.1%

11 fell into the “other” category and mentioned receiver characteristics such as intelligence that were not directly related to either listening or disseminating.1 Second, participants were given characteristics of “good listener” and “good disseminator” receivers and asked to consider themselves as transmitters in six different scenarios that varied according to the product category that they could potentially transmit WOM about.2 All scenarios put participants in the role of initial transmitter where they had recently experienced a new product from the given category (e.g., just seen a new movie, just been to a new doctor, or just bought a new pair of casual shoes). For each scenario participants indicated whether they would prefer to share a story of their consumption experience with a receiver who was a good listener, a good disseminator, or neither of these (i.e., something else; no choice option). The order of scenario presentation was randomized, as was the order of receiver choices presented for each scenario. A repeated measures logistic regression was used to estimate the following within-subject effects: the main effect for category type (hedonic vs. utilitarian), the main effect for product type, and the category x product interaction. The no choice option was rarely selected (3-5% of responses across the six products), so we excluded those observations to simplify the analysis (the results do not change with these cases included). On average, across the six product categories, 61% of the time participants chose good listener receivers over good disseminator receivers (!2(1) = 84.50, p < .001). The category main effect was significant (!2(1) = 6.56, p = .01), with good listeners preferred even more for hedonic products. The other effects were nonsignificant (product: !2(2) = 3.38, p = .19; category x product: !2(2) = 2.36, p = .31). 1

Some of the characteristics categorized as “other” could be construed as “good listener” characteristics (e.g., “I want to talk to someone who knows me well” is a tie strength characteristic). We did not, however, do this to be conservative. Only those responses that explicitly mentioned listening (e.g., listens, pays attention) were counted. 2 The categories were: (i) movie, (ii) TV show, (iii) casual shoes, (iv) doctor, (v) computer for work, and (vi) banking services. Note that the first three are hedonic and the remainder are utilitarian.

12 This study provides preliminary evidence that consumers prefer to transmit organic WOM to receivers who are good for local information flows (i.e., will likely listen) rather than receivers who are desirable for global diffusion (i.e., are socially well connected). We examine this effect in more detail in the following studies.

STUDY 2: FACTORS AFFECTING TRANSMITTERS’ CHOICES OF RECEIVERS Design and Procedure This study assesses the relative weights transmitters place on various receiver characteristics associated with reception of information and ability to spread information. Two hundred students participated in this study in a behavioral research lab. Transmission type (initial transmission or retransmission) was manipulated between-subjects and participants were randomly assigned to one of the two conditions. They were then given a task involving a hypothetical WOM transmission situation where they were potential transmitters of WOM about a movie and had to choose which potential receivers they would talk to. Participants were asked to imagine themselves in a social situation (e.g., at a social gathering, at the office) where they could talk to people who they know about a movie. Participants were presented with a scenario that described a prerelease movie. Depending on the condition, participants were asked to imagine either having seen the movie recently at a preview screening (initial transmission) or having heard about but not yet seen this movie (retransmission). They were also told that their current opinion of the movie was either positive or negative (this was also manipulated between-subjects but, as reported below, did not have an effect on choices and therefore we pooled data across the valence conditions). The movie was

13 described as prerelease to reduce the likelihood that participants would think that potential receivers had seen the movie (although conceivably they still could have heard about it).3 Nested within each between-subjects condition was a repeated choice task that required participants to choose whether or not to transmit WOM information to each of 16 possible receivers (presented in random order). The 16 receivers were described on four characteristics related to requirements for successful diffusion, each with two levels (i.e., a 24 full factorial):4 •

Tie Strength: “close friend” (strong) versus “acquaintance” (weak).

•

Receptivity: “typically listens to your opinions” (high) versus “does not typically listen to your opinions” (low).

•

Relevance: “likes movies” (high) versus “dislikes movies” (low).

•

Connectivity: “knows a lot of people” (high) versus “knows few people” (low). Tie strength, receptivity, and relevance are related to receivers being “good listeners.”

Receivers who are close to the transmitter (strong tie), have been receptive in the past (high receptivity), or are interested in the topic (high relevance) should be more likely to listen and pay attention to the transmitter’s message and hence be talked to. While receivers who are socially well connected (high connectivity) are better positioned to spread information widely, it is unclear how this will impact transmission. The choice task involved making a series of binary choices between “tell this person about the movie” (transmit) or “not tell this person about the movie” (not transmit). This is similar to a repeated choice conjoint-type task, and is also similar to that used by Wuyts et al. (2004) to study managers’ preferences for business-to-business relationship ties.

3

In a pretest, a separate group of participants found this scenario believable and easy to relate to. We did not ask participants to associate profiles with real people, although some participants wrote peoples’ names next to each profile. This suggests reasonable task involvement, and that the scenario was sufficiently realistic for participants to be able to associate potential receivers with real people. 4

14 Results The choice probabilities for each receiver profile are listed in Table 1. For initial transmission, 71% (30%) of high (low) relevance receivers, 63% (38%) of high (low) receptivity receivers, 65% (36%) of strong (weak) tie receivers, and 53% (48%) of high (low) connectivity receivers were selected. For retransmission, 64% (33%) of high (low) relevance receivers, 51% (45%) of receivers with high (low) receptivity, 58% (39%) of strong-tied (weak-tied) receivers, and 56% (40%) of high (low) connectivity receivers were selected. [INSERT TABLE 1 ABOUT HERE] To more fully examine the data, the choices were modeled with a logistic regression, including a random participant effect to control for within-subject repeated choices. The model estimated the main effects of the four receiver characteristics (effects coded: +1, -1) and their two- and three-way interactions on transmission choices (the four-way interaction was nonsignificant and removed from the model reported here). Effects for transmission type and valence were also estimated. Valence was not a significant factor but transmission type was (Wald test for differences in the drivers of initial transmission and retransmission: !2(14) = 158.32, p < .001); see Table 2). Therefore we estimated models separately for initial transmission and retransmission groups (see Table 2). [INSERT TABLE 2 ABOUT HERE] All four main effects were significant and positive in both models. The strongest main effect was for relevance (i.e., how interested the receiver is perceived to be in the topic), irrespective of transmission type. For initial transmission, the next-strongest main effects were tie strength and receptivity, while for retransmission they were tie strength and connectivity.

15 Interestingly, the relative importance of these drivers of receiver choices differs between transmission types. Of particular interest is connectivity. For initial transmission, it had a small effect compared to the other factors (all of which are characteristics of “good listeners,” which reflects an underlying human desire to be paid attention to; Festinger, Schachter, and Back 1963). A typical initial transmitter talks to receivers for whom the information is thought to be relevant (i.e., a good fit between topic and receiver), who have been receptive in the past, and who are friends; i.e., initial transmitters talk to people who are expected to be good listeners, consistent with study 1. How potentially effective a receiver is likely to be in disseminating information widely, by contrast, is much less important. The positive interactions for initial transmission indicate reinforcement effects between the good listener characteristics. For example, when the topic is highly relevant or the receiver is a friend (but not both), the average transmission probabilities are 57% and 41%, respectively. But if the topic is relevant and the receiver is a friend, the probability rises to 96%. Adding high receptivity to this combination makes transmission almost certain (99%). Connectivity plays a greater role in driving WOM retransmissions, mostly through interactions with the good listener factors. For example, consider two receivers who have low relevance, low receptivity and are weak ties, but differ only on their connectivity. The high connectivity receiver was selected by 75.4% of retransmitters, while the low connectivity one was selected only 11.8% of the time (for initial transmission, neither were popular: 11% for high connectivity, and 5.5% for low connectivity). Similarly, consider two receivers who have high relevance, high receptivity, and are weak ties but differ on connectivity. Here the high connectivity one was selected 80.3% of the time, versus 11.1% for the low connectivity one (for initial transmission, both were popular: 71.2% for high connectivity, and 69.9% for low

16 connectivity). Having high (low) connectivity makes an otherwise unappealing (appealing) receiver desirable (undesirable). This difference between initial transmission and retransmission has important implications for macro diffusion. If widespread and rapid diffusion depends on information getting into the hands of well-connected receivers, it is less likely that this will happen when people are transmitting WOM based on their own experiences. Initial transmitters do not seek high connectivity receivers; rather, they seek people who they think will listen to them. If such people by chance are also well connected, then widespread diffusion would be a fortunate but unlikely side effect since highly connected people in social networks are rare (Barabási and Albert 1999). Thus, information about one’s own experiences—which may be more credible, fresher, and more influential than second-hand (retransmitted) information—may not get to the people who can best help spread it widely. Fortunately, retransmitters are more likely to talk to well-connected receivers, which implies that while WOM from an initial transmitter might not get to someone who is well connected, second-hand information might.

STUDY 3: TRANSMISSION MOTIVES AND RECEIVER CHARACTERISTICS Design and Procedure This study documents motives (reasons) consumers have for transmitting WOM and then links these motives to measured receiver characteristics in actual WOM transmissions. Fiftythree students participated for a nominal payment. A single between-subjects factor, transmitter type (initial transmitter, retransmitter) was manipulated, with random assignment to conditions. Participants in the initial transmission condition were asked to think about when they had recently talked about a movie that they had personally seen, and those in the retransmission

17 condition were asked to think about when they had recently talked about a movie they had heard about from others but had not yet seen. Participants provided details about the movie (title, actors) and described the conversation that they had (including the first name of the receiver to make the task more concrete and vivid). The instructions made it clear that they were to recall a situation where they had started the conversation to increase the likelihood that the reasons for their recalled transmission tapped into intrinsic motivations and not extrinsic factors such as complying with conversational norms. We solicited reasons for participants’ actual WOM transmissions in two ways. First, using an open-ended question asking participants why they told this person about this movie and what motivated them to do so. Second, after some filler questions we presented participants with 14 possible reasons for transmitting WOM and had them rate how important each reason was in motivating them to talk to the other person on five-point scales (1 = “not at all important” to 5 = “extremely important”). We generated the 14 items based on prior behavioral WOM research (e.g., Frenzen and Nakamoto 1993) and our own understanding of WOM transmission situations. After the movie WOM transmission was described and the motives measures taken, we asked participants to describe the receiver they transmitted the information to with respect to tie strength, receptivity, relevance, and connectivity. Tie strength was measured following Marsden and Campbell (1984), with a five-point item for relationship type increasing from stranger to spouse, and two five-point items for frequency of discussion about movies and about general topics (" = .77). Receptivity was measured with a single item (“This person usually listens to what I have to say about movies”) on a five-point Likert scale (1 = “strongly disagree” to 5 = “strongly agree”). Relevance was measured by rating receivers on five five-point Likert scales (1 = “strongly disagree” to 5 = “strongly agree”) on items such as “This person has good taste in

18 movies” and “This person is interested in movies” (" = .81). Connectivity was measured with a single five-point Likert-scaled item capturing the perceived extensiveness of the receiver’s active personal network (“This person talks to lots of other people”). Results Transmission motives. We content-analyzed and categorized the open-ended “why did you transmit?” responses. This revealed a number of general reasons for WOM transmission, which closely corresponded to our scale items (in fact, the items covered all the reasons mentioned in the open-ended data). For convenience, therefore, we used the data on the scale items in our analysis (see Table 3). One of the 14 scale items was not represented in the openended responses (“to surprise them”), and was dropped from subsequent analysis. The reasons offered did not differ between transmitter type conditions, although the reasons’ strengths did. [INSERT TABLE 3 ABOUT HERE] A factor analysis of the thirteen retained items identified five factors that accounted for 73.9% of the variance. Two of the factors heavily overlapped (both tapped into transmitting due to a desire to give information to receivers), so we combined them as one factor. This resulted in a set of four factors, reflecting four transmission motives: to (i) seek information from receivers, (ii) give information to receivers, (iii) express one’s opinion, and (iv) make small talk. The items for each factor and their reliability scores are reported in Table 3. Composite measures were created for each factor by averaging the appropriate items for each participant. The four composite items were then subjected to a multivariate analysis of variance to see if the strengths of the various transmission motives differed depending on whether the participant was an initial transmitter or a retransmitter (controlling for within-subject item correlations). Means are plotted in Figure 1. Overall, there was a significant effect of transmission type on the

19 reasons (F(4, 48) = 2.91, p = .03). This difference was driven almost entirely by the expressing opinion factor, which was a stronger motive for initial transmitters (M = 3.44) than for retransmitters (M = 2.61; F(1, 51) = 9.68, p < .01). The giving information factor was a marginally significantly stronger reason for initial transmitters (M = 2.73) than for retransmitters (M = 2.37; F(1, 51) = 3.01, p = .09). The other reasons—seeking information and making small talk—did not differ across conditions (p = .66 and .32, respectively). [INSERT FIGURE 1 ABOUT HERE] Although not necessarily exhaustive, the consistency between open-ended and rating scale data suggests that the four general transmission motives are reasonably comprehensive and representative. WOM transmission is not only about giving information to others, but also is driven by a need to get information and to use transmission as a “soap box” from which to express one’s opinions in a social forum. The main motivational difference between initial transmission and retransmission is that expressing opinions is a much stronger motive in the initial transmission condition, which helps explain why receiver connectivity was found the be a weak driver of receiver choice in study 2. The retransmission reasons (light bars in Figure 1) are similar in strength except for making small talk, which is slightly less important. Based on repeated measures ANOVAs within each transmission type, there is a significant overall difference between reason strength for initial transmission (F(3, 75) = 7.18, p < .001), but not for retransmission (F(3, 78) = 1.43, p = .24). Thus, initial transmissions are characterized by a comparatively strong need for socially expressing opinions. Linking transmission motives to receiver characteristics. Next we regressed the four receiver characteristics on the transmission motives using seemingly unrelated regression. This allowed us to control for correlated dependent variables through an error covariance structure

20 (correlations between the four dependent receiver characteristics measures ranged between -.01 to .49, suggesting that at least some of these dependent variables should not be treated as independent). Each characteristic was regressed on the four transmission motives. Given that the variables on both sides of these equations were measured (as opposed to manipulated), econometric efforts to control for potential bias (e.g., endogeneity) were employed. Of particular concern were reverse effects (i.e., characteristics influencing reasons), since they cannot be ruled out because the transmission reasons were not manipulated. We controlled for this by adding four more equations to the model, one per transmission motive, where reasons were regressed on receiver characteristics. We are interested in the main equations where transmission motives influence receiver characteristics (see Table 4 for standardized parameter estimates). The overall fit of the model was reasonable (R2 = .50). [INSERT TABLE 4 ABOUT HERE] The more driven transmitters were to seek information, socially express opinions, and avoid making small talk, the more likely they were to have talked to good listeners (those who are high on tie strength, receptivity, and/or relevance). Participants who wanted to express opinions, for instance, were more likely to transmit to people who had been receptive in the past (b = .47, t = 3.71, p < .001) and who were close friends (b = .62, t = 5.99, p < .001). Similarly, participants who engaged in WOM to seek information also had a preference for receivers for whom the topic had high relevance (b = .48, t = 3.40, p = .001), were close friends (b = .28, t = 2.61, p = .01), and to a lesser extent had high receptivity (b = .23, t = 1.72, p = .09). The positive relationship between these common transmission motives and receivers who are likely to be good listeners is encouraging, since it suggests that micro-level, self-focused transmission motives can lead to the selection of receivers who possess necessary (although not

21 sufficient) characteristics for achieving macro diffusion. Talking to people who are going to listen is a prerequisite for local information flow. However, rapid and widespread diffusion and global information flow of course also requires receivers to be well connected. Consistent with the findings of studies 1 and 2, connectivity plays a relatively minor role here. The only significant relationship between motives and receiver connectivity was for wanting to give information (a particularly important motive from a marketer’s perspective), and this effect was negative (b = -.35, t = -2.31, p = .02). In other words, the more strongly motivated a participant was to transmit WOM to inform a receiver, the less likely their chosen receiver was to have high connectivity (e.g., perhaps because they assumed they would already have the information). As a result, hubs may be less likely to get potentially useful information given to them, making it harder for such information to spread. Overall, this study shows that micro transmission behaviors sometimes do not align with what is desired from a macro diffusion perspective. Individual transmitters appear to be driven to derive self-focused benefits from organic WOM conversations. Two important transmission motives—socially expressing opinions (which can influence receivers’ attitudes), and giving information (which can help receivers make consumption decisions)—are particularly misaligned with macro diffusion. The more motivated someone is to express their opinion, the more likely they are to talk to a good listener, which is good for local information transfer. Unfortunately, they also are not more likely to want to talk to a good disseminator (someone with high connectivity), which is bad for global information transfer. Similarly, the more motivated someone is to give information to others, the more likely they are to be talking to a friend (likely a good listener) and the less likely they are to talk to someone who is well connected.

22 STUDY 4: INCREASING TRANSMISSIONS TO WELL-CONNECTED RECEIVERS Study 3 showed that individuals’ motives for transmitting WOM sometimes do not lead them to share their information with those receivers who are best-able to spread the information widely among a population. This augments studies 1 and 2 showing receiver connectivity is a relatively minor consideration for transmitters, particularly for the initial transmitters who inject new information into networks and are therefore vital for starting diffusion processes. This study (and study 5) sheds light on why receiver connectivity does not have a larger positive influence on transmitters’ choices of whom to talk to, and illustrates how the tendency to not consider connectivity can be overcome. Design and Procedure One hundred and forty-six members of an online panel participated in this study using the same scenario in study 2. The study had a 2 (transmission type: initial transmission, retransmission) x 3 (transmission motive: seek information, give information, express opinion) x 2 (connectivity salience: no, high) between-subjects design. Transmission type was not close to significant in any of the analyses so we pooled initial transmission and retransmission, resulting in a reduced 3 (transmission motive) x 2 (connectivity salience) design. We manipulated the reason that participants had to transmit WOM, using the same reasons uncovered in study 3 except for “small talk” since it is not of theoretical importance here. Participants were told there are three main reasons that people have when transmitting WOM, and to focus on one when considering the scenario and subsequent task (see appendix A). Connectivity salience was manipulated by emphasizing how important it is to transmit WOM to well-connected receivers. Participants read newspaper articles describing recent research findings from “a team of scientists at a Harvard University social research center” (see

23 appendix B). In the no-connectivity-salience condition, participants read about the health benefits of eating cabbage, which is unrelated to social connectivity. In the high-connectivitysalience condition, participants read an article that described some benefits of talking to wellconnected people. The benefits were matched with each participant’s transmission motive. For example, participants with the “give information” motive under high salience read an article about how if you want to get information out to other people the best way to do it is to talk to a well-connected person. Connectivity salience is the key factor in this study since it allows us to see whether the tendency to pay little attention to receiver connectivity is at least partly due to transmitters simply not thinking about it. Since the connectivity salience manipulation emphasizes benefits to the transmitter, finding an increased preference for high connectivity receivers under high connectivity salience will suggest that the tendency to not choose wellconnected receivers can be overcome by associating connectivity with benefits to transmitters. The procedure was as follows. Participants were told that they would be completing three unrelated studies. In the “first” study, the connectivity salience manipulation was administered as a “reading comprehension task.” The “second” study was an unrelated filler task. The “third” study (i) manipulated the transmission motive, (ii) described the scenario and the decision task, and (iii) measured the dependent variables (characteristics of preferred/intended receivers) employing receiver characteristics measures similar to those in study 2 for tie strength, receptivity, relevance, and connectivity. The items were modified to capture how important each receiver characteristic was to a participant (measured on 7-point scales from 1 = “not at all important” to 7 = “extremely important”). Results

24 We screened-out participants who did not comprehend their assigned news article based on a multiple-choice question that asked, “According to the newspaper report, well connected people are good to talk to because…” or “According to the newspaper report, eating cabbage is good because…” and then gave a series of possible answers, only one of which was correct. We also included an open-ended item asking for a few sentences summarizing the main point of the news article; the same people who got the multiple-choice question wrong also did not accurately summarize the article in their own words. Of the original sample size of 190, 44 were dropped, leaving 146 valid participants (dropped participants were distributed fairly evenly across cells). We analyzed the data using a 3 (transmission motivation) x 2 (connectivity salience) multivariate analysis of covariance (MANCOVA), with the four receiver characteristics as the dependent variables, and how useful participants found the information contained in the news article (a single-item 7-point Likert-scale, measured at the same time as the comprehensioncheck items) as a covariate. We controlled for perceived usefulness of the news article information because participants in the no-connectivity-salience condition gave a higher usefulness rating to their “eat more cabbage for health benefits” article (M = 5.14) than participants in the high-connectivity-salience condition gave to their connectivity-related articles (M = 4.52; F(1, 144) = 8.77, p < .01), which was not surprising since the cabbage article was about personal health benefits. The focal dependent variable was the importance of talking to a well-connected receiver (connectivity). The results are shown in Figure 2. Across transmission motives, strength of preference for having a well-connected receiver increased from 4.77 in the no-connectivity-salience condition to 5.17 in the high- connectivity-salience condition (F(1, 139) = 4.13, p < .05). This main effect was not moderated by the specific transmission objective (transmission motivation x

25 connectivity salience F(2, 139) < 1, p = .95), and there was no main effect of transmission motivation (F(2, 139) = 1.56, p = .21). As expected, connectivity salience did not affect preferences for the other receiver characteristics (main effect of connectivity salience on relevance p = .70, receptivity p = .78, tie strength p = .13). [INSERT FIGURE 2 ABOUT HERE] In summary, making salient self-focused benefits of transmitting WOM to wellconnected receivers increased the importance that motivated transmitters placed on connectivity. This demonstrates that transmitters may not naturally associate receiver connectivity with personal benefits they can derive from a WOM transmission. We increased the importance of receiver connectivity by connecting it to transmitter benefits aligned with transmission motives.5 Note that this study’s purpose was not to ascertain the best approach for making connectivity more important to transmitters; rather, it was to demonstrate that preferences for well-connected receivers can be strengthened with a simple manipulation.

STUDY 5: UNDERSTANDING WHY RECEIVER CONNECTIVITY IS LESS IMPORTANT Studies 1-3 demonstrated that WOM transmitters place relatively less weight on how well connected potential receivers are compared to the weight they put on transmitting to receivers who are likely to be “good listeners.” Study 4 then showed that it is possible to increase the importance that transmitters place in talking to well-connected receivers by making salient some benefits of receiver connectivity that aligned with transmitters’ motives. This study builds on study 4 and tests an explanation for the main result in studies 1-3 by manipulating whether the

5

As a follow-up we tested connectivity salience manipulations describing benefits not aligned with transmitters’ motives. There was no effect of connectivity salience on preference for high connectivity receivers.

26 information to be transmitted has a positive or negative externality; i.e., whether the value of the information to the transmitter increases or decreases as more people become aware of it. Transmitters may be less inclined to transmit to receivers who are “good disseminators” due to a number of related reasons. For example, transmitters may think that well-connected people would have already received the information from someone else, which discourages transmitters from sharing the information for fear of telling a receiver something that, at least in the receiver’s eyes, is not novel or “fresh.” This may reduce the likelihood of the receiver reciprocally providing the original transmitter with useful, novel information in the future (per social capital theory; Lin 2001). Another possibility is that transmitters perceive it to be riskier telling something to a receiver with high connectivity since, if the transmitter’s information is in some way flawed, more people could end up finding out. Another explanation is that transmitters simply do not want their information to spread widely (e.g., the location of a good place to fish or trail to ski) and hence do not want it in the hands of people with many social connections. Each of these reasons relates to a transmitter-perceived information externality. Under social exchange theory, people expect to benefit from their social interactions (Homans 1961; Sahlins 1972). If the information has a negative externality, then the more people who receive it, the lower a transmitter’s utility and the worse off they will be. In this case a transmitter will be less inclined to give information to a well-connected receiver. This study tests this supposition. Design and Procedure Seventy-five students participated in this experiment. We used a 3 between-subjects (information externality: positive, negative, neutral/control) x 3 within-subject (receiver connectivity: high, medium, low) mixed design, with random assignment of participants across the three between-subjects conditions. Participants were asked to imagine they had received an

27 email with a special 50% off discount coupon for any purchase at Amazon.com as part of a three-day promotion, and that they could, if they wanted to, share this coupon via email with anyone. The stimuli presented participants with the coupon and the “fine print,” which they were instructed to read carefully. In the positive externality condition, the fine print stated that the code would not work unless a minimum of 2,500 redemptions per day (over a three day sale) were achieved6 (this is similar to the discount offers provided through group buying websites like Groupon.com). In the negative externality condition, the code would not work after a maximum of 2,500 redemptions per day (over three days); i.e., a scarcity restriction, similar to the manipulation used by Frenzen and Nakamoto (1993) (and to “private sales” websites like Gilt.com). In the neutral/control condition there were no such restrictions on the coupon’s use. After examining the coupon and its fine print, participants were asked to consider how likely they would be to forward this email coupon to a friend. Similar to the task in study 1, we presented each participant with three different friend profiles (in random order), manipulating connectivity across the potential receivers (high/above average, medium/average, low/below average).7 For each profile, participants indicated on a 0-100% scale how likely they would be to forward (transmit) the coupon to that friend. We also measured, on a 0-100% scale, how likely participants would be to redeem the coupon themselves. Results We expected externality would matter only for high connectivity receivers. Specifically, transmission intentions for high connectivity receivers should be higher (lower) in the positive (negative) externality condition. We analyzed the data with a mixed ANCOVA with 6

To make this believable, the coupon’s fine print also explained details of how the discount, if activated, would be applied after the sale period and before purchases shipped, and that if the discount was not activated then customers could cancel their order before shipping without penalty. 7 All friend profiles said that they were Amazon shoppers to ensure that there was some fit between the information (coupon) and the potential receivers, and thus any WOM would be relevant.

28 transmission intention as the dependent variable, receiver connectivity as the within-subject factor, externality as the between-subjects factor, and redemption intention as a covariate. We control for redemption intention since the externality of the coupon should matter more to the transmitter as their own redemption probability increases. The results are plotted in Figure 3. [INSERT FIGURE 3 ABOUT HERE] Redemption intention was positive and significant (F(1, 71) = 14.14, p < .001), as was the main effect of externality (F(2, 71) = 4.83, p = .01). The within-subject main effect of connectivity was not significant (F(2, 142) = 2.11, p = .13), but, as predicted, the externality x connectivity interaction was (F(4, 142) = 2.86, p = .03). The simple effects of externality on transmission intention were significant for medium and high connectivity (p = .03 and p < .001, respectively), but not low connectivity (p = .28). There were no differences between negative and neutral conditions of externality (all contrasts p > .44), which may reflect a tendency for consumers to believe that the information they possess has a negative externality. These results show that transmitters are influenced by whether the information they possess has a negative or positive externality. When transmitters have higher utility if the information spreads widely, we see higher transmission rates, especially when receiver connectivity is high. Like study 4, this study demonstrates that the weight transmitters place on high connectivity receivers can be altered in a relatively simple way—in study 4 by making salient some motive-aligned benefits of transmitting information to high connectivity people, and here by highlighting positive externality if the information is shared. Importantly, this study sheds some light on the mechanism driving the effects found in studies 1-3. As we speculated, transmitters may not want to transmit to high connectivity receivers if they anticipate negative consequences for themselves—in this case a negative externality.

29 DISCUSSION This paper’s objective was to better understand WOM transmission at the individual level with respect to the types of receivers that transmitters tend to share information with. It focused on whether transmitters’ natural, organic WOM behaviors (i.e., WOM not incentivized by firms) lead them to pass-on information about products to receivers who possess desirable characteristics for achieving widespread and rapid diffusion, i.e., are both good listeners and good disseminators. Ideally, receivers both have an interest in the topic (so that they pay attention and listen) and are socially well connected (so they can reach many others if they retransmit information they receive via WOM). If consumers act in a mostly self-focused manner—as suggested in recent work on WOM as a means for self-enhancement (Wojnicki and Godes 2009)—then it is unlikely that they will target receivers who would be the best disseminators. Rather, they are more likely to seek-out people who will listen to them and pay attention to what they have to say. If this is the case, the micro transmission behaviors are misaligned with desired macro diffusion outcomes. This misalignment or “micro–macro disconnect” implies that organic WOM (as opposed to explicitly incentivized WOM such as in viral marketing campaigns) will have a tendency to spread locally but not rapidly and more widely on a global scale. Evidence of this disconnect was found in studies 1-3, where we found that (i) transmitters place more weight on receiver characteristics that make them good listeners and less weight on characteristics that make receivers good disseminators, and (ii) this occurs across a range of product/service categories, common transmission motives and transmitter types. In particular, study 3 found that while “good listener” characteristics are positively associated with various self-focused transmission motives, these motives are not associated with connectivity. In fact,

30 there was a slightly negative relationship between wanting to give others information and receiver connectivity, suggesting that transmitters are less likely to share information with wellconnected people when their motive is to give information (e.g., advice, recommendations, suggestions) to others. Thus, at best receiver connectivity did not matter, and at worst it discouraged transmitters from giving potentially useful information to others who are highly connected. Overall, studies 1-3 demonstrated that the self-focused nature of WOM transmission behaviors has the consequence of making it less likely that information will get into the hands of people who are structurally positioned in the social network to spread the information widely. Given that most WOM between consumers is organic and not firm incented, these findings suggest that, if left to their own devices, consumers are unlikely to deliberately put information into the hands of the socially well-connected people who are best positioned to help widely and rapidly diffuse that information. Instead, transmitters appear likely to share information with receivers who they think will listen to them. This fosters local information diffusion, which is a necessary but not sufficient condition for the global diffusion that marketers care most about. Hence, while organic WOM transmission behaviors are not antithetical to marketers’ macro diffusion objectives, they are nonetheless suboptimal. This also provides another explanation of why some products do not “cross the chasm.” If innovators only talk to other innovators (whom are most likely to be receptive to their information), the communication with the main market will be limited and information about the innovation will tend to be confined to the relatively small set of innovators. Studies 4 and 5 attempted to find ways to increase the likelihood of a transmitter sharing information with high connectivity receivers. Specifically, we demonstrated that if some benefits of talking to well-connected receivers are made salient to transmitters or if the information itself

31 has a positive externality to the transmitter, then transmitters are more inclined to transmit WOM to receivers with high social connectivity. These studies suggest that well-connected receivers may not be transmitted to because of transmitters’ concerns that they may be worse off from doing so. Further investigation of this would be a fruitful direction for future research. In practice, a number of approaches can be taken to encourage consumers to put information into the hands of well-connected receivers. Providing explicit incentives for transmitting to certain types of receivers is one possibility. For example, in viral marketing campaigns seeds (initial transmitters) are given free products, discounts, or other economic incentives for spreading WOM as widely as possible. However, incentives need not be economic and can instead rely on transmitters’ self-interest. For instance, study 4 demonstrated that simply making salient personal benefits of transmitting WOM to well-connected receivers can increase preference for well-connected receivers. Along similar lines, study 5 showed that if the transmitter’s information has a positive externality this serves as a natural economic incentive to get them to talk to well-connected receivers. The current work is, of course, not without its limitations. Some of our studies used hypothetical scenarios. While scenario-based experiments are often used in the marketing literature, it is important to also have participants report on actual WOM transmissions, as we did in study 3 (where the results were consistent with the scenario-based studies). In future work it would be interesting to unobtrusively examine organic WOM in the field. Also, work is needed to more fully understand how information externalities interact with organic WOM transmissions. Although WOM is not a new phenomenon of interest to marketers, much more needs to be understood at an individual level. This paper represents a modest step toward this goal. Hopefully the current research encourages further work on this and related topics.

32 REFERENCES Arndt, Johan (1967), “Role of Product-Related Conversations in Diffusion of a New Product,” Journal of Marketing Research, 4 (3), 291-295. Banerjee, Abhijit and Drew Fudenberg (2004), “Word-of-mouth Learning,” Games and Economic Behavior, 46 (1), 1-22. Barabási, Albert-László and Réka Albert (1999), “Emergence of Scaling in Random Networks,” Science, 286 (5439), 509-512. Berger, Jonah and Eric Schwartz (2010), “What Do People Talk About and Why?” working paper, University of Pennsylvania. Brooks, R. C. (1957), “Word-of-Mouth Advertising in Selling New Products,” Journal of Marketing, 22 (2), 154-61. Brown, Jacqueline Johnson and Peter H. Reingen (1987), “Social Ties and Word-of-Mouth Referral Behavior,” Journal of Consumer Research, 14 (3), 350-62. Cheema, Amar and Andrew M. Kaikati (2010), “The Effect of Need for Uniqueness on Word of Mouth,” Journal of Marketing Research, 47 (3), 553-563. Chevalier, Judith A. and Dina Mayzlin (2006), “The Effect of Word of Mouth on Sales: Online Book Reviews,” Journal of Marketing Research, 43 (3), 345-354. Coleman, James S., Elihu Katz, and Herbert Menzel (1957), “The Diffusion of an Innovation among Physicians,” Sociometry, 20 (4), 253-70. Czepiel, J. A. (1974), “Word-of-Mouth Processes in Diffusion of a Major Technological Innovation,” Journal of Marketing Research, 11 (2), 172-80. Ellison, Glenn and Drew Fudenberg (1995), “Word-of-Mouth Communication and Social Learning,” Quarterly Journal of Economics, 110 (1), 93-125.

33 Festinger, Leon, Stanley Schachter, and Kurt Back (1963), Social Pressures in Informal Groups: A Study of Human Factors in Housing, Stanford, CA: Stanford University Press. Frenzen, Jonathan and Kent Nakamoto (1993), “Structure, Cooperation and the Flow of Market Information,” Journal of Consumer Research, 20 (December), 360-375. Gershoff, Andrew D., Ashesh Mukherjee, and Anirban Mukhopadhyay (2007), “Few Ways to Love, but Many Ways to Hate: Attribute Ambiguity and the Positivity Effect in Agent Evaluation,” Journal of Consumer Research, 33 (4), 499-505. Godes, David and Dina Mayzlin (2004), “Using Online Conversations to Study Word-of-Mouth Communication, Marketing Science, 23 (4), 545-560. –––––, –––––, Yubo Chen, Sanjiv Das, Chrysanthos Dellarocas, Bruce Pfeiffer, Barak Libai, Subrata Sen, Mengze Shi, and Peter Verlegh (2005), “The Firm’s Management of Social Interactions,” Marketing Letters, 16 (3-4), 415-428. ––––– and ––––– (2009), “Firm-Created Word-of-Mouth Communication: Evidence from a Field Study,” Marketing Science, 28 (4), 721-739. Goldenberg, Jacob, Barak Libai, and Eitan Muller (2001), “Talk of the Network: A Complex Systems Look at the Underlying Process of Word-of-Mouth,” Marketing Letters, 12 (3), 211-223. –––––, Sangman Han, Donald R. Lehmann, and Jae Weon Hong (2009), “The Role of Hubs in the Adoption Process,” Journal of Marketing, 73 (2), 1-13. Granovetter, Mark S. (1973), “The Strength of Weak Ties,” American Journal of Sociology, 78 (6), 1360-1380. Katz, Elihu and Paul F. Lazarsfeld (1955), Personal Influence, Glencoe, IL: Free Press.

34 Libai, Barak, Ruth Bolton, Marnix Bügel, Ko de Ruyter, Oliver Götz, Hans Risselada, and Andrew T. Stephen (2010), “Consumer-to-Consumer Interactions,” Journal of Service Research, 13 (3), 267-282. Lin, Nan (2001), Social Capital: A Theory of Social Structure and Action, Cambridge, UK: Cambridge University Press. Liu, Yong (2006), “Word of Mouth for Movies: Its Dynamics and Impact on Box Office Revenue,” Journal of Marketing, 70 (July), 74-89. Marsden, Peter V. and Karen E. Campbell (1984), “Measuring Tie Strength,” Social Forces, 63 (2), 482-501. Mayzlin, Dina and Wendy Moe (2009), “The Impact of Consumer Reviews on Firm Profits,” working paper, Yale University. Reingen, Peter. H. and J. B. Kernan (1986), “Analysis of Referral Networks in Marketing: Methods and Illustration,” Journal of Marketing Research, 23 (4), 370-78. Sahlins, Marshall (1972), Stone Age Economics, Chicago: Aldine Atherton. Watts, Duncan J. (2002), “A Simple Model of Global Cascades on Random Networks,” Proceedings of the National Academy of Sciences of the United States of America, 99 (9), 5766-5771. Wojnicki, Andrea and David Godes (2009), “Word-of-Mouth as Self-Enhancement,” working paper, University of Toronto. Wuyts, Stefan, Stefan Stremersch, Christophe Van Den Bulte, and Philip Hans Franses (2004), “Vertical Marketing Systems for Complex Products: A Triadic Perspective,” Journal of Marketing Research, 41 (November), 479-487.

35 TABLE 1 INITIAL TRANSMISSION AND RETRANSMISSION PROBABILITIES (STUDY 2) Percentage of Participants Choosing Receiver

Receiver Characteristics Relevance

Receptivity

Tie Strength

Connectivity

Initial Transmission

Good Good Good Good Good Good Good Good Poor Poor Poor Poor Poor Poor Poor Poor

High High High High Low Low Low Low High High High High Low Low Low Low

Strong Strong Weak Weak Strong Strong Weak Weak Strong Strong Weak Weak Strong Strong Weak Weak

High Low High Low High Low High Low High Low High Low High Low High Low

100.0 97.3 71.2 69.9 78.1 75.3 42.5 34.3 56.2 53.4 31.5 23.3 32.9 26.0 11.0 5.5

Retransmission

97.6 94.4 80.3 11.1 69.8 68.3 43.7 44.9 37.8 43.3 26.2 19.7 22.8 27.0 75.4 11.8

36 TABLE 2 DRIVERS OF INITIAL TRANSMISSION AND RETRANSMISSION (STUDY 2) Effects

Initial Transmission (A)

Retransmission

Comparison of (A) with (B)a

(B)

Main Effects: Relevance (high/low) Receptivity (high/low) Tie strength (strong/weak) Connectivity (high/low) Valence (positive/negative)

1.47** 1.02** 1.14** .23* -.04

(.123) (.122) (.122) (.107) (.329)

1.22** .55** .93** .81** -.05

(.087) (.087) (.089) (.082) (.250)

Not different A stronger** Not different B stronger** Not different

Interaction Effects: Relevance ! Tie strength Receptivity ! Tie strength Tie strength ! Connectivity Relevance ! Receptivity Relevance ! Connectivity Receptivity ! Connectivity Relevance ! Receptivity ! Tie strength Relevance ! Tie strength ! Connectivity Receptivity ! Tie strength ! Connectivity Relevance ! Receptivity ! Connectivity

.29 ** .15 -.03 .23 -.03 -.03 .28** .10 .09 .04

(.118) (.118) (.091) (.118) (.093) (.089) (.121) (.107) (.107) (.107)

.89** .82** -.20** .59** .35** .38** .36** .38** .26** .77**

(.087) (.086) (.074) (.083) (.082) (.078) (.089) (.078) (.080) (.084)

B stronger** B stronger ** B stronger * B stronger * B stronger ** B stronger ** Not different B stronger * B stronger * B stronger **

Other Parameters: Overall intercept Random effect variance

.21 1.53**

(.251) (.380)

.22 1.14**

(.221) (.240)

* p < .05; ** p < .01. Statistically significant estimates are in bold. Standard errors are in parentheses. a Comparison is based on estimating the model on the pooled dataset with a dummy variable (1 = initial transmission, 0 = retransmission) and estimating additional interaction terms for all of the effects crossed with this indicator. Wald test for model difference: !2(14) = 158.32, p < .001.

37 TABLE 3 MEASURES OF WORD-OF-MOUTH TRANSMISSION REASONS (STUDY 3)

Reason (Factor) Seeking Information

Items 1. I wanted to see what this person thought about this movie. 2. I wanted to see if they agreed with my opinion about this movie. 3. I hoped that they would then share their knowledge about movies with me. 4. I wanted to impress this person with my taste in movies.

Alpha .79

Expressing Opinion Socially

5. I just had to tell someone about this movie. 6. I had a strong opinion about this movie, so I talked about it. 7. I wanted to let this person know that I had seen this movie.

.74

Making Small Talk

8. I wanted an excuse to start a conversation with them. 9. To make small talk.

.90

Giving Information (to Benefit Them or Me)

10. They gave me some information previously, so I was returning the favor. 11. I wanted to tell them so that I could say that I had talked about this movie. 12. To help them choose a movie to watch/see. 13. I thought they would be interested in my opinion about this movie

.55

38 TABLE 4 EFFECTS OF TRANSMISSION MOTIVES ON RECEIVER CHARACTERISTICS (STUDY 3) Main Equations: Effects of Transmission Motives on Chosen Receiver’s Characteristics

Seeking information Expressing opinion socially Giving information Making small talk Intercept

Relevance .48 (3.40)*** .22 (1.63) -.50 (-3.56)*** -.25 (-1.82)* 1.05 (4.43)***

Receiver Characteristics Receptivity Tie Strength .23 (1.72)* .28 (2.61)*** *** .47 (3.71) .62 (5.99)*** .17 (1.23) .31 (2.87)*** *** -.36 (-2.78) -.50 (-4.85)*** .49 (2.16)** .30 (1.57)

Connectivity .07 (.49) -.07 (-.50) -.35 (-2.31)** -.22 (-1.50) 1.56 (6.22)***

Control Equations: Effects of Chosen Receiver’s Characteristics on Transmission Motives

Receiver topic relevance Receptivity of receiver Tie strength with receiver Receiver’s connectivity Intercept

Seeking Information .24 (1.66) .07 (.44) .31 (2.05)** -.18 (-1.31) .56 (2.22)**

Transmission Reasons Expressing Opinion Giving Socially Information .06 (.48) -.39 (-2.98)*** .21 (1.49) .22 (1.53) *** .50 (3.64) .36 (2.55)** -.18 (-1.52) -.30 (-2.37)** * .42 (1.82) 1.11 (4.72)***

* p < .10, ** p < .05, *** p < .01. Note: t-statistics are in parentheses; statistically significant estimates are in bold.

Making Small Talk -.10 (-.64) -.01 (-.08) -.21 (-1.34) -.27 (-1.91)* 1.59 (6.09)***

39 FIGURE 1 TRANSMISSION REASON MEANS BY TRANSMISSION TYPE (STUDY 3)

40 FIGURE 2 MEAN STRENGTHS OF PREFERENCE FOR RECEIVER CHARACTERISTICS BY CONNECTIVITY SALIENCE (STUDY 4)

7.0

No connectivity salience High connectivity salience

Strength of Preference for Receiver Characteristic

6.5

6.0

5.5

5.0

4.5

4.0

Tie Strength

Relevance

Receptivity

Receiver Characteristic

Connectivity

41 FIGURE 3 MEANS OF TRANSMISSION PROBABILITIES TO RECEIVERS WITH DIFFERENT LEVELS OF CONNECTIVITY FOR DIFFERENT INFORMATION WITH POSITIVE, NEUTRAL, AND NEGATIVE EXTERNALITIES (STUDY 5) 100 Negative externality 90

Neutral/control (no externality) Positive externality

80

Pr(transmit)

70 60 50 40 30 20 10 0 Below average (low)

Average (medium)

Receiver Connectivity

Above average (high)

42 APPENDIX A: TRANSMISSION MOTIVE MANIPULATION USED IN STUDY 4 General part People talk to each other all the time about things like new products, movies, brands, and companies. This is generally called “word of mouth” (WOM). People spread WOM for various reasons. Sometimes people talk to others just for the sake of talking (e.g., to make conversation). Other times people spread WOM for a more specific purpose. For example, you might want to give someone else some information or inform them about something. Or you might want to express or share an opinion that you have. Or you might be trying to get some information from them by starting a conversation about a particular product or topic. These are just some examples of the reasons why people spread WOM. Next we will present you with a scenario describing a very common situation in which people spread WOM. Put yourself into this scenario and consider talking about a particular product (in this case a new movie) to various people at a party. < Then specific text for condition, see below > Seeking information condition In this situation your objective is to get information about this movie. That is, your only reason for talking or spreading WOM is to seek or obtain information from others. Be sure to answer the questions that follow with this specific objective in mind. Giving information condition In this situation your objective is to give information about this movie to others. That is, your only reason for talking or spreading WOM is to give information and inform others. Be sure to answer the questions that follow with this specific objective in mind. Expressing opinion condition In this situation your objective is to express your opinion about this movie. That is, your only reason for talking or spreading WOM is to tell others what you think. Be sure to answer the questions that follow with this specific objective in mind.

43 APPENDIX B: CONNECTIVITY SALIENCE MANIPULATION USED IN STUDY 4 Matching seek information motivation Want to get good information? Talk to someone who is well connected Cambridge, MA—A new report by a team of scientists at a Harvard University social research center shows that people who have lots of active social ties (e.g. friendships, business associates, classmates) are better informed across a wide variety of topics, in particular popular culture. “People who are well connected are also very much in the know on a lot of things,” said Dr. R. Parker Brown, one of the report’s authors. The report explains that people who have many social connections—or are more “central” in social networks—are exposed to more information from more sources. This greater exposure means that they are more likely than people with fewer connections to be well informed. “These well connected people are who you would want to talk to if you want to learn about new things. Tapping their knowledge could be very helpful,” said Dr. Brown. Matching give information motivation Want to inform the world? Talk to someone who is well connected Cambridge, MA—A new report by a team of scientists at a Harvard University social research center shows that people who have lots of active social ties (e.g. friendships, business associates, classmates) play a critical role in disseminating information and knowledge around the country and even the world. “People who are well connected are amazingly good at spreading information not only across small communities such as their neighborhood or town, but also across cities, states and beyond,” said Dr. R. Parker Brown, one of the report’s authors. The report explains that by having many social connections—or being more “central” in social networks—these types of people are linked to many people who are spread over different regions. This then allows for many geographically dispersed people to be exposed to information that a well-connected person has. “These well connected people are who you would want to talk to if you want to get information to spread to lots of people over a wide area. This can be particularly helpful if the information has the potential to help others in some way,” said Dr. Brown. Matching express opinion motivation Want to be listened to and have your opinion taken seriously? Talk to someone who is well connected Cambridge, MA—A new report by a team of scientists at a Harvard University social research center shows that people who have lots of active social ties (e.g. friendships, business associates, classmates) are receptive to information and opinions from other people. “People who are well connected are extremely interested in what other people have to say. They are like sponges and absorb a wide variety of viewpoints and opinions,” said Dr. R. Parker Brown, one of the report’s authors. The report explains that by having many social connections—or being more “central” in social networks—this type of person has access to information, ideas, and opinions from a large number of people. This makes them more receptive to and interested in what other people have to say than regular with fewer social connections. “These well connected people absolutely love hearing what other people have to say about all kinds of things. They have all these connections so that they can expose themselves to as many opinions as possible on a regular basis,” said Dr. Brown.