A Word-Of-Mouth Network

Peter H. Reingen, Arizona State University
ABSTRACT - Much of the research dealing with word-of-mouth behavior was found to suffer from a lack of fit between method and the nature of phenomena investigated. To overcome this shortcoming, the present study uses graph-theoretic network analysis to examine word-of-mouth behavior in a natural environment. Word-of-mouth paths were systematically traced and a social network analysis was conducted on the word-of-mouth actors. The study demonstrates the crucial roles played by influentials and multiple group memberships in the spreading of word-of-mouth communication.
[ to cite ]:
Peter H. Reingen (1987) ,"A Word-Of-Mouth Network", in NA - Advances in Consumer Research Volume 14, eds. Melanie Wallendorf and Paul Anderson, Provo, UT : Association for Consumer Research, Pages: 213-217.

Advances in Consumer Research Volume 14, 1987      Pages 213-217


Peter H. Reingen, Arizona State University

[The author appreciates the helpful comments by William Bearden, Jacqueline Johnson Brown. Brian Foster, Jerome Kernan, Stephen Seidman and Teresa Swartz . The support by Mr. and Mrs. Laurel Jump is especially noted.]


Much of the research dealing with word-of-mouth behavior was found to suffer from a lack of fit between method and the nature of phenomena investigated. To overcome this shortcoming, the present study uses graph-theoretic network analysis to examine word-of-mouth behavior in a natural environment. Word-of-mouth paths were systematically traced and a social network analysis was conducted on the word-of-mouth actors. The study demonstrates the crucial roles played by influentials and multiple group memberships in the spreading of word-of-mouth communication.


Many studies have demonstrated the importance of WOM communication (WOM) in shaping consumers' attitudes and behaviors (e.g., Arndt 1967; Engel et al. 1969; Katz and Lazarsfeld 1955; Sheth 1971; Whyte 1954). These studies have yielded important insights into WOM behavior and have identified a significant area for research. However, traditional WOM research, as well as research on related topics such as opinion leadership (e.g., Myers and Robertson 1972; Reynolds and Darden 1971), has not been a "hot topic" in scholarly journals and symposia for well over a decade. Although some recent research has appeared (e.g., Richins 1983), interest in these phenomena has waned. This is unfortunate, as several gaps in knowledge exist in this area. These range from the relatively simple, descriptive (e.g., the number of paths of WOM comprised of senders and receivers of WOM) to the more complex, dynamic or interactive (e.g., flows of WOM within social groups and how WOM is transmitted from one social group to another to form an entire network of WOM as embedded in an overall structure of social relationships among a system's actors). The underlying reason for these deficiencies in understanding is simple. The overwhelming methodological focus in previous studies examining interpersonal topics was on sample surveys with the individual as the unit of analysis. As a consequence, the social-structural context within which WOM behavior is embedded could not be examined directly. Since WOM is an interpersonal phenomenon, however, this individualistic bias in most of the WOM research extant has resulted in only a limited understanding of this important facet of consumer behavior.


To examine WOM behavior, it is beneficial to perform a network analysis. With network analysis (Burt 1980; Knoke and Kuklinski 1982), one can examine the structure of a WOM network, the social structure of a system composed of WOM actors, and how the social structure is interrelated with the WOM network. This perspective is potentially valuable for several reasons.

First, information flows along paths that may have many links can be investigated. Reynolds and Darden (1971), Richins (1984) and Sheth (1971) have called for research examining chains of communication flow to explore questions of who shares information with whom. For example, this may be important in the determination of opinion leaders, which in previous studies was done usually on the basis of self-designated data.

Second, since a consumer's WOM behavior may agglomerate into large-scale patterns through his/her network of friends, relatives and acquaintances, network analysis is capable of yielding direct insights into the relationship between micro-level and macro-level WoM phenomena. There is a lack of research directly examining within-group WOM behavior and how the flow of WOM travels across groups.

Third, since WOM behavior arises from the interacting individuals' social environments, the relationship between the information flow network and the social network composed of WOM actors can be accorded explicit recognition.


A network study was conducted focusing on pre-decision information diffusion related to an existing service offered by one marketer. Previous research has dealt primarily with the role of WOM in the diffusion of new products or services, but there is little reason to believe that it is not important for established products or services as well (Richins 1983). Emphasis was placed on the "who-told who" WOM network and how this network was embedded in the social structure composed of the WOM actors. Negative WOM or feedback loops to the recommender were not investigated. The study, therefore, likely captured only a portion of the total WOM behavior, albeit an important one.

A problem encountered with the network analysis of WOM behavior is the matter of boundary specification of the system under study. Previous network analyses of interpersonal communication have primarily been conducted on closed or semi-closed systems (e.g., Arndt 1967; Rogers and Kinkaid 1981; Weimann 1983). In perhaps most consumer behavior contexts, however, WOM systems are open in the sense that set membership is initially unknown. Thus, to be able to relate the WOM network to the social network, it was necessary to conduct the study in two phases. Briefly, in the first phase more recent users of the service were asked to reveal how they learned about the existence of the service. When a respondent mentioned another person (a user or nonuser), that person was contacted and asked the same question. The process was repeated until a WOM path was traced back to the marketer. The first phase, then, was employed to produce data on just the WOM network. Consistent with the criterion of mutual relevance for network boundary delimitation (Knoke and Kuklinski 1982), the network included only those social actors who were members of "who-told-whom" paths of information flow. System "closure" was obtained through complete backward tracing of WOM paths. Once the system was identified, the second phase was conducted which generated data on the social structure of the interpersonal relations among the system's WOM actors.

The study on which this paper is based was primarily exploratory in nature. The complete study is presented in Reingen and Kernan (1986). Here the overriding goals are to illustrate the network analysis of WOM behavior and to describe in detail the features of one WOM network as it arose from the network of social relationships among the WOM actors who were directly or indirectly tied to a marketer on "who-told-whom" paths. Understanding the details of a particular case may provide a foundation for more ambitious general studies. The marketer in question is a piano tuner (hereafter referred to as J), who had moved to a major Southwestern metropolitan area several years ago. Upon arrival, he developed a formal association with a music store. When the store sold a piano, J would tune it and would obtain a reimbursement from the store for his service. With the passage of time he developed his own clientele whom he billed directly. J does not engage in formal marketing communication. He considers himself "retired," but he tunes pianos to "keep myself busy."

Phase One

Data in the first phase were collected by telephone interview. Individuals who had been customers of J during the two years prior to the start of the study were sent first a preliminary notification and then were contacted by one of four interviewers with as many callbacks as were necessary to reach them. This resulted in a completion rate of 89 percent of the potential subject pool. The respondents were asked how they found out about J. When a subject mentioned another person, that person was contacted. This backward tracing was repeated until a path reach J. Data cross-checks on 18 sender/receiver dyads revealed that in only one instance was the receiver's determination of the source unconfirmed by the sender. However, three individuals mentioned as a source (one of whom had died) could not be reached. The first phase yielded 44 actors who were either directly or indirectly linked to J via WOM paths. Of these actors, 60 percent were female, 65 percent were college graduates and 53 percent were over 40 years of age.

The interview also yielded data on how long a receiver had known the source (less than several months, several months, half a year, a year, couple of years or many years specify), how frequently they had communicated (daily, several times a week, once a week, several times a month, once a month or less than once a month), the importance s/he attached to the social relation (10-point scale), the type of social relation between them (e.g., acquaintance, co-worker, neighbor), the information received, the circumstances surrounding information transfer, experiences with the service after using it, and some demographics.

Phase Two

Toward the end of the telephone interview, respondents' cooperation with the second phase of the study was requested. Each respondent received a thank-you note for his/her participation in the first phase. The second phase was conducted by mail. A list of the names of respondents to the first phase was prepared, and the subjects were asked to identify whom they knew on the list. For each person they knew, the subject was asked to identify the duration, frequency and importance of interaction,the type of social relationship, and the types and number of various social organizations (e.g., church, music, etc.) to which they jointly belonged. They were also asked to report on when they first used the tuning service and some additional demographics.

Before mailing the questionnaire each respondent received a preliminary notification. Respondents were promised a one-dollar reward (donated to a charity of their choice) and a non-technical summary of the research. Non-respondents to the initial mailing of the questionnaire received a maximum of two follow-ups. Of the 45 actors in the WOM network (J included), responses were eventually obtained from 38 individuals, yielding a response rate of 84 percent.


The major findings will be presented in three parts. First, the network of the WOM relations is described, followed by a structural analysis of the social network of the individuals involved. Finally, the WOM structure embedded in the social structure will be described.

Network Analysis of WOM Relations

Information flow paths. Figure A shows the digraph of the WOM flow of information among the 45 actors, 44 of whom are reachable from J. Informally, the digraph consists of points (i.e., actors) and directed lines indicating who told whom about the service. Of the 44 individuals directly or indirectly tied to J, only 4 (9%) were non-users of the service. Thus, in the vast majority of instances, the sender of information was a previous user. No individual reported employing more than one source. The digraph of WOM has 25 maximal paths originating at J and leading to a terminal point. There was one path (#22) with length one (4%), 16 with length two (64%) , 6 with length three (24%), and 2 with length four (8%), where length is defined by the number of directed lines on a path.

Influentials in the WOM network. Perhaps more important from a substantive viewpoint regarding the notion of opinion leadership is a simple analysis of the difference between outdegree (i.e., the number of arrows "coming out") and indegree (i.e., the number of arrows "coming in") of the actors directly or indirectly linked to J. Of the 44 individuals, 25 had a negative difference (were receivers of WOM but not senders), 14 had a difference of zero (were receivers once and senders once), and five had a positive difference (hat more sender than receiver relations with others). These five individuals were actor 126 (paths #5, 6, 7, 8), 142 (paths #9, 10, 11, 12), 177 (paths #11, 12), 148 (paths {13, 14, 15, 16) and 78 (paths 218, 19, 20) . They were particularly influential as transmitters of WOM information in that they were intermediate links for 15 (60%) of the paths, although they represented only 11 percent of the individuals directly or indirectly tied to J. It is noteworthy that all of these five actors were music professionals (e.g., music teachers). Although not all intermediate links who were music professionals had more sender than receiver relations, individuals possessing this attribute (n - 7) had significantly more sender relations than individuals (n - 12) not possessing the attribute (Z - 2.57 and X - 1.00, respectively; t - 4.27, p < .001).

Information flows and relational attributes. For type of social relation between senders and receivers of WOM, 44 percent of the receiver responses indicated a friend relation, 23 percent an acquaintance relation, 13 percent a colleague relation and 8 percent a relative relation. Thus, the majority of activated ties for the flow of WOM arose from stronger (friends, relatives, colleagues), rather than weaker social relations (mere acquaintances). There was no significant relationship between the time order of ties (earlier ties with distance--length of any shortest path of which two persons are members--of one or two from J and later ties with distance of three or four from J) and these types of social relation (X2 - 2.67, p > .10). Finally, a log-linear analysis was performed on the last tie (n - 25) and earlier ties (n - 19) on a path activated for WOM flow as the dependent variable and communication frequency, duration of relation and importance of interaction as the independent variables. These independent variables were employed to operationalize Granovetter's (1973) notion of tie strength which recent research suggests is related to information flows (Weimann 1983). Following the approach of Knoke and Burke (1980 , p. 38), the model postulating that none of the independent variables has a significant relationship with the dependent variable could not be rejected (L2 = 5.11, p > .10), indicating that last ties did not differ significantly in strength from earlier ties on a path.

Sociological context and WOM flows. The data indicate clearly that WOM behavior arises from and is conditioned by the interacting individuals' social environments. Some examples:

(J -> 102 -> 148 -> 84 -> 112) J and 102 go to the same church where they met. 148 gives piano lessons to 102's daughter and they were talking about getting a tuner. 84 knew that 148 is a piano teacher. They know each other well through their church. The asked 148 for advice. 84 and 112 are neighbors. 112 knows that 84 hat just bought a new piano, so she asked 84 to recommend a tuner.

(J -> 142 -> 177 -> 154) 142 is a music teacher at a local high school. J tuned pianos at the school and 142 knows him from there. 177 was a music teacher at the same high school. 154 also teaches at that school and he asked 177 to recommend a tuner ("Since she (177) is a piano teacher, I trusted her opinion").

(J -> 181) They go to the same church. 181 sat next to J in church and asked what he did. He "merely mentioned the fact (that he was a tuner), neither persuaded nor discouraged."





These examples indicate that co-membership in social organizations (e.g., neighborhood, work, church, family) appears to be important for the activation of WOM flow. This co-membership enables an individual to seek information directly from a person who is likely to possess the information needed (e.g., 148 b 84). In fact, of the 21 WOM instances that could be content-analyzed for direction, 14 (67%)- were solicited from satisfied customers. However, several WOM occurrences were induced by situational or environmental cues (e.g., J -> 181). Also noteworthy is that of the 10 cases involving influentials on which data for direction are available, 90 percent were solicited by receivers. Finally, many WOM flows seem to travel from one social organization to another due to a person's dual membership in them (e.g., 84 who heard about J from a church tie and who told another person through a neighbor tie). It may therefore be concluded on the basis of this brief qualitative analysis that much of the WOM behavior observed here was due to the intersection of individuals' group affiliations and the intersection of groups within the individual. More formal support for these conjectures is obtained by the following analyses.

Network Analysis of Social Structures

Overall structure. The network analysis on the 38 individuals who responded to the second phase produced 13 groups. The groups were identified with SONET-I (Foster and Seidman 1978). They consisted of at least three socially connected individuals, and any tie within a group was on a potential path for information flow. The identification of such more closely-knit "clumps" of social structure is important because it allows for a direct examination of how information travels within and across groups.

The 13 groups contained 22 (58%) of the individuals. Thus, these individuals had access to several sources of information, but (as was shown above) they employed only one of them. The remaining 16 (42%) individuals are labeled isolates (defined later). Of these isolates, 13 (81%) were in only one dyadic relation with another person, namely the sender. Their access to information was much more limited. Figure B shows the sociogram of the structure (non-directed lines). In the following more detailed presentation, emphasis is placed on those aspects of the structure of social relations which are helpful in analyzing WOM processes as embedded in the social structure.

Intersection of groups. It is noteworthy that of the 13 groups, 12 (92%) contained members who belonged to more than one group. Thus, the groups were tied together primarily by overlapping membership. Some matrix algebra is helpful in isolating individuals interconnecting groups and their joint membership in groups with other actors. This may be crucial as these individuals may spread information across groups, allowing for the emergence of larger-scale WOM patterns.

Let R be a binary matrix with k (groups) columns and i (individuals) rows, entering a 1 in the i,k th cell if and only if individual i belongs to group k. Let X - RRT, where RT, is the transpose of R. The i,> th element of X gives the number of groups to which i and j both belong, and xii is the number of groups to which individual i belongs (Wilson 1982). Multiple group membership was not uniformly distributed across individuals (Kolmogorov-Smirnov Z = 3.10, p <-.002). Of the 22 individuals, nine (41%) intersected groups, accounting for 33 (72%) of the total number of group memberships of 46 (i.e., the sum of the diagonal values of X). J belonged to 10 (77%) of the 13 groups and the number of memberships in groups jointly held by J and the j others was 25 or 54 percent of the total, indicating that much of the social structure was induced by the marketer himself.

Kinds of groups. Although the groups strongly overlap in membership several distinct kinds of groups can be identified on the basis of the commonality of relation that typifies membership in them. Of the 11 groups that could be classified, five (45%) were "church" groups, two (18%) were "teacher colleague" groups, two (18%) were "music store colleague" groups, one (9%) was a "band" group and one (9%) was a "neighbor" group. Thus, the groups were tied together by some individuals with multiple membership in several kinds of groups rather than in the same kind of group. In general, WOM behavior within and between different kinds of groups may facilitate a wider spreading of information.

Structural roles. Social-network participants were classified according to one of four roles. These roles are important in the understanding of the flow of WOM to illustrate how dyadic interaction aggregates to form larger-scale patterns.

Intersectors are group members who belong to more than one group (nine of the 38 individuals, 245) . In general, they may represent a cut point in the graph of social relations in that deleting an intersector and his/her associated lines may result in a disconnected graph (e.g., 126 at the 3 o'clock position in Figure B is an intersector who is a cut point). Disconnected graphs of social structure inhibit the wide-spread dissemination of information through interpersonal channels

Bridge pillars are members of two different groups who are linked to each other (two individuals, 5%). If the line connecting such members is deleted and results in a disconnected graph, it represents a bridge between system actors. The line connecting 102 and 148 at the 6 o'clock position of Figure B is a bridge. Similar to intersectors, bridge pillars are important for linking micro-level interaction to macro-level WOM phenomena.

Members are individuals who belong to only one group but who are not bridge pillars (11 individuals, 29%; e.g., 60 at the 6 o'clock position).

Isolates are individuals with no group membership (16 individuals, 42%; e.g., 130 at the 1 o'clock position).

WOM Structure Embedded in Social Structure

Given this social structure among relevant WOM actors in the system, a rather clear picture emerges of how WOM interaction at the micro-level developed into larger-scale processes. Since there was strong overlap among groups, intersectors played a crucial role in the spreading of WOM information. Overall, 62 percent of the WOM dyads had an intersector as the sender although intersectors comprised only 24 percent of the system's actors. J was the intersector in 56 percent of these instances (13123), followed by 142 (13%, 2/23; 4 o'clock position, 78 (13%, 3123; 9 o'clock position) and 126 (9%, 2/23; 3 o'clock position). These four individuals represent cut points in the graph of social structure, and they alone accounted for 57 percent (21/37) of the dyadic WOM relations.

The importance of intersectors in the present study is also indicated by an analysis focusing on the four roles of social network actors specified above. WOM from intersector b member was most prominent (24%), followed by intersector 4 intersector (22%), member b isolate (22%), and intersector x isolate (13%). There was no member t intersector and member + member WOM occurrence. Since 15 (94%) of the 16 isolates were end points of WOM paths and since their social integration for the system under study was low, it is not surprising that all of these 15 individuals were receivers but not senders of WOM. Thus, there was a positive correlation between number of social choices received and sender-to-receiver WOM occurrences (r - .88, p < .001). It is also of interest to note that of the previously identified five individuals with a positive difference between outdegree and indegree of WOM relations, four were intersectors (126, 142, 177, and 78) and four were cut points (126, 142, 148, and 78). Finally, although only two individuals were bridge pillars (102 and 148), they were crucial in transmitting WOM information from one distinct group to another, thus resulting in the longest WOM paths.


It is recognized that many of the descriptive results of the foregoing case study may be unique to the service marketer analyzed and that they represent merely an initial step in theory development. A variety of WOM networks must be examined before we can generalize about the length of paths, opinion leaders, search behavior, WOM flows within and across groups, etc., in such network structures. At a minimum, however, the study demonstrates that network analysis can yield detailed insights into complex phenomena such as WOM. Traditional research methods cannot provide the requisite relational data.

With regard to the WOM network in the present study, many paths were found, with the longest involving five individuals. Typically, the sender of WOM was a previous customer. This is consistent with Arndt's (1967) findings. Individuals relied on only one source, although many had access to multiple sources. Although only a minority of WOM instances was induced by weak interpersonal ties, weak ties were crucial in extending path lengths. The prominence of stronger ties as sources of information may have been due to several factors; for example, they may have been more readily available and more motivated to help.

In contrast to some of the previous research on WOM and opinion leadership (e.g., Belk 1971; Summers 1970), transmitters and receivers of WOM information were clearly distinguishable, and WOM was induced usually by information seeking consumers. This divergence in findings may be due to differences in the products which were studied. For example, Belk (1971) investigated WOM behavior regarding freeze-dried coffee. For such a product, WOM conversations may be more dependent upon environmental or situational cues, and overt information search may be less prevalent. Similarly, Summers (1970) studied opinion leaders for women's fashion. He found that in most cases the traditional view held--opinion leaders offered unsolicited advice. It is less likely, however, that individuals offer unsolicited advice for services such as piano tuning. Instead, many individuals in the present study sought advice from music professionals. That music professionals were prominent sources is consistent with previous research suggesting that opinion leaders are more knowledgeable about the product category (e.g., Summers 1970).

Also important is an examination of how the network structure of WOM related to the social structure of WOM participants. Whereas WOM expanded in a linear or radial fashion in the present study, there was a rich degree of social structure among the WOM actors who were directly or indirectly tied to the marketer. This is indicated by most WOM actors belonging to groups of different kinds . Thus, most of the WOM dyads existed in groups. The vast majority of groups were linked together by overlapping members, and therefore individuals who intersected groups were fount to be crucial for the spreading of WOM. These individuals transmitted WOM information to other group members or to isolates. By definition, isolates were less well integrated into the social system under study. One likely reason for this is that they were relatively more recent additions to the system as is indicated by the face that the vast majority of isolates were end points of WOM paths (i.e., they were not senders of WOM). Group members who received information from an intersector either transmitted the information to isolates or they transmitted it to members of another group since they themselves intersected groups.

Most prominent of the individuals intersecting groups was J, suggesting that multiple membership in different kinds of groups by a service provider is important to generate marketing transactions brought about by interpersonal interaction and to stimulate WOM processes. Finally, the single instance of a bridge over which WOM information was transmitted should not be overlooked because it resulted in the longest WOM paths with another music professional again functioning as an influential.

In conclusion, it is suggested that network analysis provides a good fit between method and WOM phenomena. The research is structurally posed in a fashion similar to that in which the target occurs (Bonoma, Bagozzi and Zaltman, 1978) . Network analysis is not a panacea, however. It has problems which are potentially more acute than those encountered with more traditional methods. For example, it is ewident that extra effort is required in the gathering of WOM network data, as the consequences of non-response are more severe in network analysis. Boundary specification issues and inaccurate or incomplete retrieval of WOM events from episodic memory pose additional problems which need tn be addressed to capitalize on the method's potential.


Arndt, Johan (1967), "Role of Product-Related Conversations in the Diffusion of a New Product," Journal of Marketing Research, 4 (August), 291-295.

Belk, Russel W. (1971), "Occurrence of Word-of-Mouth Buyer Behavior as a Function of Situation and Advertising Stimuli," Proceedings of the American Marketing Association's Educators Conference, 419-422.

Bonoma, Thomas V., Richard Bagozzi and Gerald Zaltman (1978), "The Dyadic Paradigm with Specific Application Toward Industrial Marketing," in Organizational Buying Behavior, Thomas V. Bonoma and Gerald Zaltman, Eds., Chicago: American Marketing Association, 49-66.

Burt, Ronald S. (1980), "Models of Network Structure," in Annual Review of Sociology, Vol. 6, Palo Alto, CA: Annual Reviews, Inc., 79-141.

Engel, James E., Roger D. Blackwell and Robert J. Kegerreis (1969), "How Information is Used to Adopt an Innovation," Journal of Advertising Research, 9, 3-8.

Foster, Brian L. and Stephen B. Seidman (1978), SONET-I: Social Network Analysis and Modeling System, Vol. 1, Binghamton, NY: Center for Social Analysis, State University of New York.

Granovetter, Mark S., "The Strength of Weak Ties," American Journal of Sociology, 78, 1360-1380.

Knoke, David and Peter J. Burke (1980), Log-Linear Models, Beverly Hills, CA: Sage.

Myers, James H. and Thomas S. Robertson (1972), "Dimensions of Opinion Leadership," Journal of Marketing Research, 9 (February), 41-46.

Reingen, Peter H. and Jerome B. Kernan (1976), "Referral Networks in Marketing: Methods and Illustration," Journal of Marketing Research, 13 (November), in press.

Sheth, Jagdish N. (1971), "Word-of-Mouth in Low-Risk Innovations," Journal of Advertising Research, 11 (June), 15-18.

Summers, John O. (1970), "The Identity of Women's Fashion Opinion Leaders," Journal of Marketing Research, 7 (May), 178-185.

Weimann, Gabriel (1983), "The Strength of Weak Conversational Ties in the Flow of Information and Influence," Social Networks, 5 (September), 245-267.

Wilson, Thomas P. (1982), "Relational Networks: An Extension of Sociometric Concepts," Social Networks, 4 (June), 105-116.

Whyte, William }i. Jr. (1954), "The Web of Word of Mouth," Fortune, 50 (November), 140-143.