Telematics and Informatics 32 (2015) 321–332

Contents lists available at ScienceDirect

Telematics and Informatics

journal homepage: www.elsevier .com/locate / te le

Social activity and structural centrality in online socialnetworks

http://dx.doi.org/10.1016/j.tele.2014.09.0080736-5853/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +49 (0) 203 379 1234; fax: +49 (0) 203 379 2195.E-mail addresses: andreas.klein@uni-due.de (A. Klein), H.Ahlf@hs-neuss.de (H. Ahlf), sharma@iup.edu (V. Sharma).

1 His primary research domain includes group buying processes on the Internet and social media issues. He also applies institutional economic thchannel management and retailing topics.

2 Tel.: +49 (0) 2131 739 8672. His primary research domain includes social network analyses, dynamic modeling and computer simulations of cbehavior, social contagion, and how information diffusion affects consumer demand and viral marketing.

3 Tel.: +1 (724) 349 4329. His primary research domain includes business-to-business marketing, and group buying in consumer markets.

Andreas Klein a,⇑,1, Henning Ahlf b,2, Varinder Sharma c,3

a University of Duisburg-Essen, Mercator School of Management, Department of Management and Marketing, Lotharstrasse 65, 47057 Duisburg, Germanyb Neuss University of International Business, School of Engineering, Markt 11-15, 41460 Neuss, Germanyc Eberly College of Business, Department of Marketing, Indiana University of Pennsylvania, 664 Pratt Drive, Indiana, PA 15705, United States

a r t i c l e i n f o

Article history:Received 8 July 2014Received in revised form 4 September 2014Accepted 12 September 2014Available online 12 October 2014

Keywords:Social networksNetwork analysisPersonal activityImpersonal activityStructural importanceOpinion leaders

a b s t r a c t

It has become well known that the knowledge about key network members is essential fordoing business successfully through online social networking sites. As of now, most studiestargeted at identifying key members have used network structural centrality measures. Lit-tle emphasis has been placed on member activities to identify key members in a network;even though gathering and utilizing such data is relatively easier than estimation of struc-tural centrality positions. Using a structural equation model on an Internet social network-ing site data, this study finds that the personal activity status of key members is highlycorrelated with their structural centrality measures. Managerial implications, limitations,and further research issues are also addressed.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Social networking has become very popular with consumers as well as businesses with the arrival of online social networksites (Yang, 2012). It appears to be a realization of their long held latent need of anytime interaction from anywhere and anyplace. For consumers, such sites enable them to communicate with friends, make new friends, and engage in expressing theirinnovativeness, and self-, and social-identities (Pagani et al., 2011). For businesses, these sites provide them with access tolarge ensembles of consumers to promote their messages and products by using key members of networks as their secondarysources of communication and consumer evangelists. Kozinets et al. (2010) call it a ‘seeding campaign’ in which firms are get-ting directly involved in word of mouth marketing with virtual community members. The underlying assumption is that net-work communication follows a two-step process of mass communication. Therefore, the practical implications for a firm areto find the key individuals, also referred as ‘influentials’ or ‘centralities’ in a network, seed them, and let them take care of therest of communication. As a result, the past three decades have seen structural centrality identification as an active focus oftheoretical and empirical research fueled by Freeman’s seminal work (Freeman, 1979).

eories to

onsumer

322 A. Klein et al. / Telematics and Informatics 32 (2015) 321–332

Over time, there have been modifications and extensions of Freeman’s three key indices of centrality, ‘degree centrality,closeness centrality, and betweenness centrality’ based upon the specific type of network flows. For a comprehensive list ofcentrality indices, see Borgatti (2006). Regardless of the modifications, Freeman’s key indices have remained robust for mostof the applications and are still being used extensively. Concomitantly, some scholars were also exploring the developmentof alternatives to the structural centrality measures. One such area that became promising is the ‘network activity’ of mem-bers. Two recent studies that of Trusov et al. (2010) and Pagani et al. (2011) have attempted to connect the network activityof members with structural centrality measures. The first uses the ‘number of site logins over time of others based upon thesite login of a user’ as an indicator of influential activities of members and connects them with structural centrality. The sec-ond classifies activities on a social network site such as viewing and posting of opinions, questions, answers, photos, videos,personal information, and knowledge about an issue, and relates these activities to personality traits such as innovativenessand self- and social-identity expressiveness.

Building upon these two works, this study develops a structural equation model and tests hypotheses about the relationbetween the activities and structure measurements at the individual level in a social networking site of approximately26,000 members in the area of event planning and organization. The results show that the personal activity status of keymembers is highly correlated with the key structural measures of their importance. Managerial implications, limitations,and further research questions of this approach are also addressed. We think that this study significantly advances the socialnetwork marketing literature in many ways. First, it directly connects the network member activities with the identificationof key members in a network. Second, it provides an alternative way of finding key members of the network based upon theiractivity levels that is more practical. Finally, the study opens up avenues for further research in network member activities.

2. Online social networks

Online social networks are essentially online social communities where people socialize or exchange information andopinions through pictures, postings, blogging, and other tools to communicate with one another (e.g., Vista, in press). Socialnetwork sites enable individuals to build their public or semi-public profile within a bounded system, articulate a list ofother users with whom they share a connection, and view and navigate their list of connections and those made by otherswithin the system (Boyd and Ellison, 2008). These networks are primarily Web 2.0 based and their functioning is mainlydependent upon the ability of extant users to add contents and other users as friends; for example, a current user can inviteothers to join him/her, who may accept or reject the invitation. However, once the individuals accept to be friends, new rela-tionship ties are built. This process carried out over time by multitude of extant users, therefore, underlies the potential sizeof the network, and hence its survival (Pagani et al., 2011). These networking sites have become popular over time; they areattracting millions of users worldwide and are growing exponentially (Katona et al., 2011). Many social networks are valuedat billions of dollars because of their business potential (Hoffman and Novak, 2012). In particular, Facebook, Twitter, Google+and MySpace, are considered as market leaders. Within these networks, not only playfulness, critical mass, and trust play acrucial role (Sledgianowski and Kulviwat, 2009), but user-generated content is indispensable for maintaining old and estab-lishing new ties (Pagani et al., 2011; Heinrichs et al., 2011). Other studies have similarly highlighted sociability and usabilityas two key factors underlying the success of online social communities. While usability mainly pertains to the technologyused, sociability relates to how members of an online community interact with each other (Lin and Lee, 2006). Accordingto Panzarasa et al. (2009), the ties between the users are established and maintained by their communication, i.e., the greaterthe communication between users and their friends in a social network, the greater the strength of ties between them.

In social networks, some users are likely to be more valuable than others to marketers regarding diffusion of informationfor their products or for other business opportunities (DiMaggio and Garip, 2012; Karnik et al., 2013). They can impact theirvirtual friends more effectively (Goldenberg et al., 2009) by acting as word-of-mouth channels (Brown et al., 2007; Kozinetset al., 2010; Meuter et al., 2013) as well as role models to inspire others connected to them to imitate their behavior. And,they have also been referred as ‘‘key members’’, ‘‘influentials’’, or ‘‘central figures.’’ The received literature recognizes that thedesired communication (e.g., launching a new product) in social networks takes place in two steps; first, the communicationis directed at key members, who subsequently, spread the communication to their followers (Haenlein, 2013; Kozinets et al.,2010). That is why the identification of key network members has assumed such a high importance that it has become astream of active research for the past few decades. In the next section we discuss the concept of critical structural positionsin a network to lay the foundation for our work.

3. Identifying critical structural positions in a network

Centrality is an important structural attribute of a social network because it is highly related to other group propertiesand processes such as enhancement of desired network communication or controlling the disruptive cells. According toFreeman (1979) a person located at the center of a network has maximum possible degrees (number of contacts with others),falls in the geodesics (shortest possible paths) between the largest possible numbers of members, and is closest to all othermembers than any other member. These three characteristics have come to be recognized as the three key measures of cen-trality in a social network. While, the first one is viewed as an index of potential communication activity of the memberoccupying this position, i.e., it enables a member to engage in maximum communication with others in a network, the

A. Klein et al. / Telematics and Informatics 32 (2015) 321–332 323

second one is seen as the extent of control of the central position on the communication of others; because of its location onthe shortest connections between others this person can exercise greater control on the network communication than anyother member. Finally, the third characteristic pertains to the central member’s degree of avoidance of control by others onits communication in the network, i.e., degree of independence of the central position vis-à-vis others. The subsequentresearch on the identification of key structural actors in a social network has been primarily based upon Freeman (1979)centrality measures. The Appendix provides these measures as items (S1–S3).

Extending the work on network centrality, Faust and Wasserman (1992) argued that both centrality and prestige indicesare measures of the most prominent individual in a social network. While in non-directional networks, these two indicesmay be same, they are different for directional networks. Faust (1997) later on extended the network centrality measuresto affiliated networks where participants jointly participate in social occasions. Marsden (2002) introduced the concept ofegocentric measure of centrality and argued that it is as good as Freeman’s centrality measures with the advantage thatit needs relatively smaller subset of ties to develop indicators of network centrality.

Over time, suggestions have been made to customize the centrality measures for specific type of network flows ratherthan using the traditional key measures. In this context, Borgatti (2006) offered a typology of most appropriate centralitymeasures for specific kind of network flows. Building upon Borgatti’s work, Kiss and Bichler (2008) found that outdegree(number of links) and PageRank (sum of contributions from incoming hubs) measures of network centrality outperformedother centrality measures in viral marketing. Heidemann et al. (2010) also found that that the PageRank based approach toidentify key users is superior to the degree-, closeness-, and betweenness-centrality measures in a network. While it is argu-ably better to construct customized centrality measures for each network for each type of flow, however, the standard cen-trality measures were found to be robust even when the data was somewhat flawed (Borgatti, 2006). Therefore, one cansurmise that for all intents and purposes, it may be cost-effective to use the standard centrality measures for identificationof key network members; and that is why, this study used the Freeman’s centrality measures.

On a few parallel fronts, researchers were also trying to estimate attitudinal and behavior correlates of centrality mea-sures. For example, Gibbons and Olk (2003) used salient personal attributes such as gender, work experience, ethnicity,and education degree on network members’ friendship ties and their development over time. They found that attribute sim-ilarity increased friendship development among people with similar ethnic identification and also lead to structural similar-ity. More recently, using a combination of variables such as gender, age, membership period, and total items acquired from asocial network site, Goldenberg et al. (2009) were able to identify network hubs (people with exceptionally large ties). Essen-tially, these results indicate that in a social network, people seek homophily; they are more comfortable in dealing with sim-ilar people than dealing with diversity. Other scholars started to relate network activity of members to structural centrality(Harvey et al., 2011; Heinonen, 2011). The moorings of such works can be traced back to Butler (2001), who asserted that inthe absence of some form of communication activity, influence, social support, coordination, or information sharing cannotoccur in a social structure and it will fail to provide valued benefits to members. Next, we discuss the usefulness of studyingmember activity as a key to identifying important network positions.

4. Network centrality and network activity

It is well established in the marketing literature that communication between a few influential members with others whoare influenced by them is essential for the diffusion of an innovation phenomena to work in a social network (Goldenberget al., 2009; Katona et al., 2011; Zitek and Tiedens, 2012). According to Merwe and Heerden (2009), there is a strong corre-lation between the centrality of a structural position and the role of opinion leadership in a network. They also found a strongcorrelation between the domain-specificity and non-domain specific leadership, i.e., the opinion leadership may not neces-sarily be domain-specific. Similar results were found by Heidemann et al. (2010), who observed that the key users in a socialnetwork were significantly related to their communication activities. Likewise, Lee et al. (2010) observed that the centralposition in a network is significantly related to opinion leadership and also to the degree of susceptibility to interpersonalinfluence. In essence, anyone in a social network, once perceived as a leader in an area through his/her communication activ-ities can also be perceived as a leader in several other areas—a concept akin to market maven (Eirinaki et al., 2012). There-fore, it is plausible to surmise that if the activities of network members can be measured, one should be able to identify thekey network members based upon their activity level.

Fortuitously, the measurement of network member activity is easier because the network sites are built around memberactivities such as message postings, blogging, etc. Moreover, such activities are useful to marketers as they are closer to themarketing processes such as value communication. Therefore, analyzing the relationship between the different degrees ofnetwork member activity and structural importance should be of great interest to the marketers and in particular to the mar-keting practitioners. If membership activity shows the same explanatory power as structural importance indicators, the mar-keting world can greatly benefit through improved behavioral targeting. This insight will also enhance network analysis as awhole and represent a stronger consideration of individual member properties for one-on-one targeting.

So far, only two works have tried to empirically connect network activity with structural centrality. Trusov et al. (2010)argued that two types of activities are performed by the social network participants. One, members create new contents byediting their profiles (e.g., adding pictures, uploading music, writing blogs and messages) and two, they also consume con-tents that others create (e.g., looking at pictures, downloading music, and reading blogs and messages). The primary goal of

324 A. Klein et al. / Telematics and Informatics 32 (2015) 321–332

this study was to develop a methodology to identify influential users on the basis of the activity level. The study used the‘number of site-logins of other members (followers) over time in response to the site-login of specific users’ as the surrogateof influence in the network. Their model could correctly classify influentials with an accuracy of 90% against a threshold of50% for randomly identified influential network members. They also found that on the average about 20% of the networkusers, act as influential users (perhaps, a validation of the well-known 80/20 rule). In the other study, Pagani et al. (2011)contend that most activities on a social network site entail the form of viewing and posting of opinions, questions, answers,photos, videos, personal information, and knowledge about an issue. They regarded posting as active and viewing as passiveelements of life on a social network site, and both work in a virtuous cycle to keep the site alive and be perceived valuable tocommercial websites who want to use this site to accomplish their goals. They found that individuals high in innovativenesswere both active and passive users of network information where as those who were high in self-identity expressiveness andsocial identity expressiveness were also active users of social networks.

Building upon these two works, this study focuses on individual activity levels and their relationship with key structuralpositions. In this regard, the study develops a structural equation model and tests hypotheses about the relation betweenactivity and structural measurements at the individual level. In this context, the study analyzes a social networking siteof approximately 26,000 members in the area of event planning and organization. The virtual network site originates in Ger-many and connects people all over Europe for private social events. For example, parties, exhibition visits and so on. Activityand structural information about number of friends, sent messages, network invitations, picture uploads etc. was gatheredfor this community.

5. Research model and hypotheses

5.1. Research model

Fig. 1 shows the structural equation model deployed in this study to ascertain the extent of relationship between themembership activity and their structural position in the network. In all, the model consists of four concepts: two pertainingto the structural position (Structural Importance and Neighborhood Connectedness) of the network members and twobelonging to their network activities (Personal activities and Impersonal activities).

5.2. Structural Importance

Structural centrality: The focus of structural measurements lies in identifying specific members who have structurallyadvantageous positions within a network. According to Freeman (1979), the centrality measurements are formalized as(1) the degree (regarding direct friendship connections), (2) the closeness (distance between network members), and (3)the betweenness (relevant for the control/mediation of information flows). As described before, for all practical purposes,these three structural measurements have been used individually or in combination to identify the ‘influentials’ in social net-works in most of the studies. In line with literature, this study assumes that the important members in a network are highlyconnected and have more ties and relations to other members (Lee et al., 2010). Furthermore, these individuals are viewed ascentral if they have the potential to mediate information between members who are not connected in a direct relationship.Therefore, we use all these three measures of centrality to comprehensively ascertain the structural importance of members.

5.2.1. Neighborhood connectednessFriends and social circles, which are highly connected to each other around a network member, comprise neighborhood

connectedness. Knoke and Yang (2008) discuss the clustering coefficient as an important relationship measurement for a

Personalactivity

Impersonalactivity

Structuralimportance

Neighborhood connectedness

H1

H2

H3H4

Fig. 1. Research model.

A. Klein et al. / Telematics and Informatics 32 (2015) 321–332 325

personal (sub) network’s density. This clustering of neighbors represents the probability that the connected friends of anindividual have direct friendship ties among each other as well. According to Liu-Thompkins and Rogerson (2012), the neigh-borhood connectedness can lower a member’s egocentric structural importance. The underlying assumption is that thehigher the density of a member’s friendship network, the lower the structural importance of that member in terms of thecapabilities to build bridges between other sub-networks like groups, cliques etc. But if the neighborhood connectednessof that member is low, his structural importance will probably be higher due to the member’s stronger central position.Therefore the clustering coefficient points to a highly connected social environment and represents the connectedness ofa network member’s friends among each other. We utilize the clustering coefficient (given in the Appendix as item S4) asa measurement for neighborhood connectedness in our model. The primary purpose is to see its impact on the structuralimportance. In this study, we use a single item measurement of neighborhood connectedness. That leads to our firsthypothesis:

H1. Neighborhood connectedness has a negative relationship with a network member’s structural importance.

5.3. Network member activities

According to Butler (2001) ‘‘Without some form of communication activity, influence, social support, coordination, orinformation sharing can’t occur. In fact, in the absence of communication activity, the structure will fail to provide valueto its members’’. As is widely known that members perform a variety of activities such as forum postings, picture uploadsetc. on social networking sites. Although the list of such activities is usually large because of the various communication toolsprovided by a social network site, scholars usually classify such activities into fewer but meaningful categories to developdeeper insights into their impact on the goal at hand. For example, Schoberth et al. (2006) distinguished member activitiesinto synchronous interactions (as in chats or instant messaging) and asynchronous interactions (as in forums and bulletinboards or newsgroups). The goal of their study was to conduct a longitudinal analysis of network member communication.Others have classified these activities as personal recommendation such as friends and family and impersonal recommen-dation such as discussion forums (Goyette et al., 2010; Sun et al., 2006). Given the primary goal of this study, we use thepersonal-impersonal dichotomy to classify the member activities on a social network. Our perspective is justified given thatthis study takes a cross-sectional view of the structural importance and the activities at one point in time to study the rela-tion between the two. In this work, we observed seven activities performed by the network members. These were: messagessent, status updates, picture uploads, forum activity, network invitations, friendship requests sent, and friendship requestsreceived. Table 1 depicts these activities and how they were measured.

5.3.1. Personal activitiesWe regard personal activities as 1:1 communication from one network member to another. Such activities entail personal

messages sent, network invitations sent to others, friendship requests sent and friendship requests received from others.Personal activities enable network members to build, cultivate, and maintain friendship ties, send personal messages aboutmeetings, events, and share their interests with them. Obviously, such activities generate personal bonding between the par-ticipating network members and add to the cohesiveness of the group. Therefore, those network members that spend greateramount of time and effort in such activities to communicate with their friends are likely to be perceived as valuable

Table 1Activity and structure measurements.

Measurements Description

Activity measurementsA1: Status updates Sum of status updates by a network member in relation to all status updates within the networkA2: Picture uploads Sum of picture uploads to an individual profile in relation to the sum of overall uploaded picturesA3: Forum activity Sum of created forum-threads and forum-posts in relation to the sum of all forum-activities in the networkA4: Messages sent Sum of messages sent by a network member to others in relation to all messages sent within the networkA5: Network invitations Sum of actively triggered network invitations by a network member (per message to non-network members) in relation

to all invitations sent by network membersA6: Friendship requests

sentSum of actively triggered friendship connections by a network member in relation to the sum of all friendship requestssent within the network

A7: Friendship requestsreceived

Sum of passively triggered friendship connections from other network members in relation to the sum of all friendshiprequests received within the network

Structure measurementsS1: Degree centrality Sum of direct contacts of a network member in relation to the sum of all network membersS2: Closeness centrality Inverse sum of distances (shortest path) between a network member and all other network membersS3: Betweenness

centralityRelation of standardized and aggregated number of shortest paths between two indirectly connected network memberswhere a third member is part of, and the number of shortest paths between those two network members

S4: Clustering coefficient Number of connections of neighbors (direct surroundings) of a network member among each other in relation to themaximum possible number of connections of these neighbors among each other

326 A. Klein et al. / Telematics and Informatics 32 (2015) 321–332

members because they add valued resources to their friends’ repertoire, and hence are likely to be perceived by their friendsas leaders who can influence them. That leads to our second hypothesis:

H2. Personal activities have a positive relationship with a network member’s structural importance.

5.3.2. Impersonal activitiesIn comparison with personal activities, impersonal activities are characterized by 1:n communication between one mem-

ber and many others. Examples of impersonal activities entail status updates, picture uploads, and forum activities. Throughonly one such single impersonal activity, network members can contact or inform a large number of friends and other net-work members about themselves, their views, or about a current affair. In line with Granovetter (1973), the efficiency ofimpersonal activities would be higher than those of the personal activities, because one single message might reach morethan one other individual within the network. Nevertheless, the effectiveness of personal activities is higher in terms ofbuilding up friendship ties and maintaining structural importance because of the 1:1 communication character. On the con-trary, impersonal communication does not generate friendship ties in the first place. For this reason, impersonal activities areless likely to have any meaningful effect on structural importance. That leads to our third hypothesis:

H3. Impersonal activities have no relationship with a network member’s structural importance.

According to Pagani et al. (2011), much of the value creation in a social network comes from consumer-to-consumerinteractions that take the form of posting and viewing of opinions, questions, answers, photos, videos, personal information,and knowledge. They further mention that while viewing is a passive network use and posting is an active network use; it isa very important cycle of consumption and content creation that keeps the network humming with life.

They found that those who are high in self-identity expressiveness and social identity expressiveness are engaged in net-work postings. And, those who post are also both active and passive users of the network. This establishes a link betweenimpersonal and personal activities of network members. For example, when such a posting is viewed by a close friend,he/she sends a feedback to the author of the posting and the author replies by a personal e-mail to that close friend. Essen-tially, impersonal activities can start a cycle of personal-impersonal activities.

Furthermore, a large amount of impersonal activity—depending on a network member’s talkativeness—is also able tostimulate the personal activity of other network members, since this activity makes the network member who posts forumthreads or status updates interesting to other inactive network members. Impersonal activity can also initiate friendshiprequests, which lead to personal activities and in the end to a higher structural importance of the network member whostarted the impersonal activity. This leads to our fourth hypothesis H4:

H4. Impersonal activities have a positive relationship with a network member’s personal activities.

6. Methodology

The data set for this study stems from an online social network for event planning and organization in local areas world-wide, and has a community size of approximately 26,000 members. This large virtual social network, compared to otherstudies in that area (e.g., Lee et al., 2010), which are sometimes build on convenience samples, contains of real activityand structural data, and includes a mixed group of members with a balanced ratio of men and women of all ages. It has aclosed network structure and does not have sub-networks or unconnected network parts. The network members are linkedthrough directed friendship connections with each other. Whether a friendship connection is actively triggered or only pas-sively accepted can be traced by the initiated friendship request or network invitation. The analyzed data set provides infor-mation whether a friendship was affiliated and from which network member this activity was initiated. In line with Gibbonsand Olk (2003) suggestion, this data set includes the relational aspects of the social network, for example, information abouteach tie between the members.

6.1. Measuring individual activity and structural importance

Some members in this virtual community actively participated in the network by submitting information about theiractivities (e.g., uploading pictures or posting information in forums) or sending messages to friends in different forms. Thosemembers were interacting with various circles of friends by supplying user-generated content (Lee et al., 2010; Ridings et al.,2006; Trusov et al., 2010). Others only visited the social networking site and acted passively by observing the activity ofothers. It appears that they used the network community to get information, but did not post any information or sendany messages (Ridings et al., 2006). These observations reveal the different behavior of network members in terms of usage,information sending and gathering as well as a disparity regarding degree of interaction. Using the items (see A1-A7 inTable 1), we gathered information about sent messages, status updates, picture updates, forum activity, and networkinvitations as well as sent or received friendship requests of social network members.

A. Klein et al. / Telematics and Informatics 32 (2015) 321–332 327

The primary focus of structural measurements lies in the identification of specific members that are in structurally advan-tageous positions within the network. The centrality measurements of degree, closeness, and betweenness were measuredby items S1–S3 and the clustering was measured by the item S4. These items are also provided in Table 1.

6.2. Correlation of structure and activity measurements

Table 2 depicts a correlational matrix analyzing the differentiation between the personal and impersonal activities andthe structural importance and neighborhood connectedness. All variables are significantly correlated with each other atthe .01 level. Two important findings about the structure measurements are worth noticing. One, the clustering coefficientis not meaningfully correlated with the activities as well as other structure measurements. Except for a correlation of .03with closeness centrality, the range of correlation of clustering coefficient with all other variables varies between �.07and �.03. This is in line with our hypothesis that the centrality as an egocentric measurement of the structural importanceof a network member is not being seriously impacted by the connectedness of a network member’s social surrounding. Two,we see a higher correlation between each of the three other structure measurements (all correlations range between .30 and.80). These findings are in line with the assumption of our model that these three centrality measures are highly correlated.

Table 2 also shows that activity measurements like messages sent, network invitations, friendship requests sent, andfriendship requests received are correlated between .27 and .67. They seem to be linked closer together. The correlationsof the remaining activity measurements, i.e. status updates, picture uploads and forum activity, range between .10 and.35. Though both activity measurements are also correlated (between .20 and .36) there is another important finding thatindicates the relation in our hypotheses. The correlations between the centrality measurements and personal activity factorsrange between .24 and .94, whereas most values are above a threshold of .50. This finding points to a close connectionbetween personal activity and structural importance.

7. Model estimation and results

The existing dataset provided the information about 26,000 members. We calculated the structural network measure-ments for each network member based upon the formulae given in the Appendix (Faust and Wasserman, 1992; Freeman,1979). The activity measurements at the individual level were directly obtained from the original dataset. We utilized astructural equation modeling approach because this is an established powerful multivariate analysis for testing and estimat-ing causal relations using a combination of statistical data and causal assumptions. Also, in comparison with the multipleregression analysis, this approach provides a simultaneous estimation of all determining factors.

7.1. Overview of estimations

Following the developed hypotheses, this research aims to check the basic argument that (1) personal and impersonalactivities are interrelated, (2) personal activities are strongly correlated with structural importance, (3) impersonal activitiesare not related with the structural importance, and (4) neighborhood connectedness is negatively related with structuralimportance. The assumption made here is that all such relationships are linear. Therefore, as a pre-condition for testingthe linearity of both activity and structural measurements, all cases were plotted against each measurement variable pre-sented in Table 1. Applying the best-fit curve method, we observed no significant outliers. Subsequently, a cluster analysisto eliminate heterogeneity within the dataset was conducted. All activity and structure measurements were included. A twocluster scenario revealed 28 cases in cluster 1 that either had activity levels much higher than the average network member(such network members could be identified as the initiators of the network since our data revealed that their high activitytook place in the initial stage of the network) or had no relevant structural position within the network (i.e. fringe member-ship account with disconnected friendship relations that classify as inactive). Both types of network members can be handledas outliers and were eliminated from the dataset.

The model was tested in SmartPLS 2.0 with a sample size of 25,823 network members (see Fig. 2 for an overview andcorresponding regression weights). The goal of our study is to test the relationship between the member activity leveland structural importance of that member. Hair et al. (2014) suggest the use of the partial least squares (PLS) path modelingmethod when such challenges are addressed. Due to the fact that indicators like messages sent, network invitations, friend-ship requests sent, and friendship requests received or status updates, picture uploads and forum activity constitute personalor impersonal activity of a network member, we had to consider the latent constructs personal and impersonal activity asformatively measured constructs (Ringle et al., 2012). We are also cognizant that the structural importance of a networkmember causes degree centrality, closeness centrality and betweenness centrality, and therefore, demands a reflective mea-surement approach. We used the PLS path modeling approach since formative and reflective measurements were simulta-neously applied. Moreover, no multivariate normal distribution of the data was observed (Becker et al., 2012). In addition,PLS path modeling can be recognized as a method without ‘factor envy’ because it is based on weighted composites (Rigdon,2012).

Table 2Bivariate correlations of activity and structure measurements.

Structural importance Neighborhoodconnectedness

Impersonal activity Personal activity

Degreecentrality

Closenesscentrality

Betweennesscentrality

Clusteringcoefficient

Statusupdates

Pictureuploads

Forumactivity

Messagessent

Networkinvitations

Friendshiprequests sent

Friendshiprequests received

Degree centrality Pearson correlation Sig. (2-tailed) .57* .80** �.06** .38** .29** .28** .62** .58** .89** .94**

.00 .00 .00 .00 .00 .00 .00 .00 .00 .00Closeness centrality Pearson correlation Sig. (2-tailed) .30** .03** .23** .23** .17** .33** .24** .54** .50*

.00 .00 .00 .00 .00 .00 .00 .00 .00Betweenness centrality Pearson correlation Sig.(2-tailed) �.06** .33** .22** .20** .52** .80** .64** .81**

.00 .00 .00 .00 .00 .00 .00 .00Clustering coefficient Pearson correlation Sig. (2-tailed) �.04** �.03** �.03** �.07** �.05** �.05** �.06**

.00 .00 .00 .00 .00 .00 .00Status updates Pearson correlation Sig. (2-tailed) .24** .35** .36** .25** .32** .36**

.00 .00 .00 .00 .00 .00Picture uploads Pearson correlation Sig. (2-tailed) .10** .20** .20** .24** .28**

.00 .00 .00 .00 .00Forum activity Pearson correlation Sig. (2-tailed) .33** .14** .10** .23**

.00 .00 .00 .00Messages sent Pearson correlation Sig. (2-tailed) .27** .64** .51**

.00 .00 .00Network invitations Pearson correlation Sig. (2-tailed) .36** .66**

.00 .00Friendship requests sent Pearson correlation Sig. (2-tailed) .67**

.00Friendship requests received Pearson correlation Sig. (2-tailed)

** Correlation is significant at the .01 level (2-tailed) n = 25,823.* Correlation is significant at the .05 level (2-tailed).

328A

.Klein

etal./Telem

aticsand

Informatics

32(2015)

321–332

Network invitations

Picture uploads

Closeness centrality

Friendship requests sent

Forum activity

Betweennesscentrality

Friendship requests received

Messages sent

Status updates

Degree centrality

Clusteringcoefficient

Personalactivity(R2 = .22)

Impersonalactivity

Structuralimportance

(R2 = .94)

Neighborhoodconnectedness

.98***

.01**.47*** .00

H2 √

H3 √H4 √ H1 --

.12

.18

.38***

.50***

.60**

.45***

.36

.88***

.66***

.96***

1**

Significance level: *** p <= 0.001 ** p <= 0.01 * p <= 0.05

Fig. 2. Model estimation.

A. Klein et al. / Telematics and Informatics 32 (2015) 321–332 329

7.2. Essential findings

Results reveal that the degree centrality has the highest influence (0.96) on structural importance, followed by the between-ness centrality (0.88) and closeness centrality (.66). All factor loadings regarding structural importance are above 0.50(Hulland, 1999). The composite reliability analysis also shows values of 0.88 and a Cronbach’s alpha of 0.79, which is abovethe recommended threshold of .70 (Cronbach, 1951). Besides, an average variance extracted of 0.71 outperforms the suggestedvalue of .50 (Fornell and Larcker, 1981), and therefore, indicates the high discriminant validity of the construct. Concerning theeffect of neighborhood connectedness on structural importance, the result shows that the assumed negative effect cannot beconfirmed, since the estimated value is .01 (p 6 .01). Thus, hypothesis H1 is rejected. The connectedness degree of the virtualcircle of friends and the egocentric perspective underlying structural importance seem to be independent. Friendship ties andcommunication structures among friends do not affect a network member’s structural importance. An additionally performedStone–Geisser-Test (Q-square) reveals that all path coefficients are significant with values above zero (Geisser, 1998).

The standardized regression weights show that a network member’s personal activities—consisting of messages sent, net-work invitations, friendship requests sent and friendship requests received—are strongly positively related to his/her struc-tural importance—measured by degree centrality, closeness centrality and betweenness centrality. The correlation value (of0.98 at p 6 .001), almost close to 1.0, is very high, which strongly supports hypotheses H2. A corresponding R-square value of.94 for structural importance proves substantial explanatory power of the latent construct (Chin, 1998). In contrast, imper-sonal activities like status updates, picture uploads and forum activities have no positive effect on the structural importanceof a network member (.00). This result supports hypothesis H3. Taken together, these results of hypotheses H2 and H3strongly corroborate the fundamental premise of our study that network member activities (in particular, the personal activ-ities) can be taken as a close substitute of the structural importance of a member.

The results also show that the influence of impersonal on personal activity is positive with a value of 0.47 (p 6 .001),which proves both a direct and an indirect effect (.46 * .98 = .46) on structural importance. The direct effect of impersonalactivity on personal activity confirms hypothesis H4. Unlike structural importance, the R-square for personal activity ranksat .22 and therefore has only weak informative value (Chin, 1998), which likewise reduces the significance of the indirecteffect. Hence, impersonal activity can only partly explain personal activity. This fact supports the assumption that bothare separate constructs in terms of discriminant validity.

Summing up, the structural importance of a network member can be very well explained through that member’s personalactivity. Therefore, the measurement of personal activity can simplify the identification process of important network mem-bers. Accepted friendship requests, either sent or received, have the highest influence in this context. In contrast, the struc-tural importance of a social network member cannot be explained by either impersonal activity or by a network member’ssocial surrounding, since regression weights in this study show no effect.

8. Discussion and conclusions

The fundamental premise underlying this research was that the individual activities of a network member (personal activ-ities) are highly correlated with the egocentric structural position of that member (structural importance). And, the results ofthis study corroborate our position. Structural measurements and indices of influential people though well established, their

330 A. Klein et al. / Telematics and Informatics 32 (2015) 321–332

estimation are difficult for larger networks without the availability of greater amount of adequate resources. On the contrary,activity data can be simply observed and gathered, for example, by analyzing message counts or status updates on the individ-ual level. Furthermore, this kind of data is readily available for social network site operators as well as for third parties who canuse their marketing strategies without having to apply complex statistical analyses to identify influential users. Therefore,results of this study show that by and large, the network influential users can be relatively easily identified using their personalactivity levels. And, this shows the potential of providing an easier way for (usually cash-strapped) small and medium sizedfirms to also identify key network people, seed their marketing campaigns, and compete with larger firms.

However, a caution is in order; that the sole dependence on personal activities of network members in the identificationof key members may have some limitations. For example, a lot of personal communication may not necessarily indicate astructurally advantageous position of a network member as such a communication may rest upon only a few friendship ties,which will rank a network member as active, but from a structural point of view as not important. Also, besides the signif-icance of network activity, structural holes are also important descriptors of a social network (Zaheer and Soda, 2009),because lightly connected bridges between denser sub-networks might exist (Ganley and Lampe, 2009; Geys andMurdoch, 2010; Katona et al., 2011). In his pioneering research about social structure of competition Burt (1995) also dis-cusses the efficiency and effectiveness of network structures and network holes. Therefore, in our view, in the overall schemeof accurate identification of key members, we see the analysis of personal activity as the first step that can be used to sep-arate ‘more important’ members from ‘less important’ or ‘unimportant’ social network members. The subsequent steps canentail the use of appropriate centrality indices on those ‘more important’ members to identify the key members, therebyenhancing the accuracy of the identification of key members.

9. Limitations and guidelines for future research

The activity data used in this study did not contain adequate consumer information in terms of their attitudes, preferences,or motivations for individual decision making. This limits our ability in suggesting marketing strategies to firms in carryingout product specific marketing activities among network members in order to gain competitive advantage. Additional dataabout the message-specific content of network members has to be taken into account. Conducting field experiments andinterviews can help to reach that goal (see Trusov et al., 2010). Another limitation of our study is that our analysis was cen-tered on one social network. Consequently, our results can only be described as exploratory. However, they offer an insightinto the development of a more accurate methodology for the identification of key network members.

Further research in social network analysis should ascertain more information about the influence of important memberson the individual decision making processes of friends and contacts in business situations. Passively observed data aboutsocial networking sites cannot provide this type of information. Therefore, the specific analysis of purchase behavior triggeredby social networking sites was not part of this research, but should be included in future studies. Based on such information,effective economic value of important network members can be estimated. Hence, social network analysis can also contributeto concepts and methods of customer lifetime value estimation, if information about purchase behavior exists. Moreover, thestatic approach of this study has to be transferred to a dynamic or at least a comparative static view to observe individualbehavior over time. Since friendship structures and network usage are likely to change, formerly unimportant memberscan become critical for future network success. The research approach so far does not incorporate potential dynamics in termsof fluctuating user influences over time and network evolution. Overall, the analysis of networks should be part of the stra-tegic planning of social network operators, capture the behavior of their network members and analyze the interdependencebetween those network members as well as dynamic developments (Stephen and Toubia, 2010).

Appendix A

Selected structural measurements

S1. Degree centrality

CDðniÞ ¼PK

k¼1friendsi

N�1 :

CD ¼ Degree centrality:ni ¼ Current=considered network member:N ¼ Total amount of network members:

S2. Closeness centrality

CCðniÞ ¼ N�1PN

j¼1dðni ;njÞ

:

CC ¼ Closeness centrality:ni ¼ Current=considered network member:d ¼ Geodesic distance between network members:N ¼ Total amount of network members:

A. Klein et al. / Telematics and Informatics 32 (2015) 321–332 331

S3. Betweenness centrality

CBðniÞ ¼

Pj–k

i–j;k

gjkðniÞ=gjk

ðN�1Þ�ðN�2Þ :

CB ¼ Betweeness centrality:ni ¼ Current=considered network member:gjk ¼ Number of shortest paths between network members j and k:

gjk ðniÞ ¼ Number of shortest paths where network member ni is part of :N ¼ Total amount of network members:

S4. Clustering coefficient

CCCðniÞ ¼2�ctjk

ki �ðki�1Þ :

CCC ¼ Clustering coefficient:ni ¼ Current=considered network member:ctjk ¼ Number of connection ties ðeÞ between neighbors ðj; kÞfor network member ni:

ki ¼ Number of connection ties for network member ni:

References

Becker, J.-M., Klein, K., Wetzels, M., 2012. Hierarchical latent variable models in PLS-SEM: guidelines for using reflective–formative type models. Long RangePlan. 45 (5–6), 359–394.

Borgatti, S.P., 2006. Identifying set of key players in a social network. Comput. Math. Organ. Theor. 12 (1), 21–34.Boyd, D.M., Ellison, N.B., 2008. Social network sites: definition, history, and scholarship. J. Comput. Mediat. Commun. 13 (1), 210–230.Brown, J., Broderick, A.J., Lee, N., 2007. Word of mouth communication with online communities: conceptualizing the online social network. J. Interact.

Mark. 21 (3), 2–20.Burt, R.S., 1995. Structural Holes: The Social Structure of Competition. Harvard University Press, Cambridge, MA.Butler, B.S., 2001. Membership size, communication activity and sustainability: a resource-based model of online social structures. Inf. Syst. Res. 12 (4),

346–362.Chin, W.W., 1998. The partial least squares approach to structural equation modelling. In: Marcoulides, George A. (Ed.), Modern Methods for Business

Research. Lawrence Erlbaum Associates Inc., Mahwah, NJ, pp. 295–336.Cronbach, L.J., 1951. Coefficient alpha and the internal structure of tests. Psychometrika 16 (3), 297–334.DiMaggio, P., Garip, F., 2012. Network effects and social inequality. Ann. Rev. Sociol. 38 (1), 93–118.Eirinaki, M., Monga, S.P.S., Sundaram, S., 2012. Identification of influential social networkers. Int. J. Web Based Communities 8 (2), 136–158.Faust, K., 1997. Centrality in affiliation networks. Social Networks 19 (2), 157–191.Faust, K., Wasserman, S., 1992. Centrality and prestige: a review and synthesis. J. Quant. Anthropol. 4 (1), 23–78.Fornell, C., Larcker, D.F., 1981. Evaluating structural equation models with unobservable variables and measurement error. J. Mark. Res. 18 (1), 39–50.Freeman, L.C., 1979. Centrality in social networks: conceptual clarification. Social Networks 1 (3), 215–239.Ganley, D., Lampe, C., 2009. The ties that bind: social network principals in online communities. Decis. Support Syst. 47 (3), 266–274.Geisser, S., 1998. The predictive sample reuse method with applications. J. Am. Stat. Assoc. 70 (350), 320–328.Geys, B., Murdoch, Z., 2010. Measuring the bridging versus bonding nature of social networks: a proposal for integrating existing measures. Sociology 44 (3),

523–540.Gibbons, D., Olk, P.M., 2003. Individual and structural origins of friendship and social position among professionals. J. Pers. Soc. Psychol. 84 (2), 340–351.Goldenberg, J., Han, S., Lehmann, D.R., Hong, J.W., 2009. The role of hubs in the adoption process. J. Mark. 73 (2), 1–13.Goyette, I., Ricard, L., Bergeron, J., Marticotte, F., 2010. E-WOM scale: word-of-mouth measurement scale for e-services context. Can. J. Admin. Sci. 27 (1), 5–

23.Granovetter, M.S., 1973. The strength of weak ties. Am. J. Sociol. 78 (6), 1360–1380.Haenlein, M., 2013. Social interactions in customer churn decisions: the impact of relationship directionality. Int. J. Res. Mark. 30 (3), 236–248.Hair, J.F., Hult, G.T.M., Ringle, C.M., Sarstedt, M., 2014. A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM). Sage Publications Ltd.,

Thousand Oaks.Harvey, C.G., Stewart, D.B., Ewing, M.T., 2011. Forward or delete: what drives peer-to-peer message propagation across social networks? J. Consum. Behav.

10 (6), 365–372.Heidemann, J., Klier, M., Probst, F., 2010. Identifying key users in online social networks: a pagerank based approach. In: Proceedings of the Thirty First

International Conference on Information Systems, ICIS, Saint Louis, Missouri, pp. 1–22.Heinonen, K., 2011. Consumer activity in social media: managerial approaches to consumers’ social media behavior. J. Consum. Behav. 10 (6), 356–364.Heinrichs, J.H., Lim, J.-S., Lim, K.-S., 2011. Influence of social networking site and user access method on social media evaluation. J. Consum. Behav. 10 (6),

347–355.Hoffman, D.L., Novak, T.P., 2012. Toward a deeper understanding of social media. J. Interact. Mark. 26 (2), 69–70.Hulland, J., 1999. Use of partial least squares (PLS) in strategic management research: a review of four recent studies. Strateg. Manag. J. 20 (2), 195–204.Karnik, A., Saroop, A., Borkar, V., 2013. On the diffusion of messages in on-line social networks. Perform. Eval. 70 (4), 271–285.Katona, Z., Sarvary, M., Zubcsek, P.P., 2011. Network effects and personal influences: the diffusion of an online social network. J. Mark. Res. 48 (3), 425–443.Kiss, C., Bichler, M., 2008. Identification of influencers – measuring influence in customer networks. Decis. Support Syst. 46 (1), 233–253.Knoke, D., Yang, S., 2008. Social Network Analysis. Sage Publications Inc., London.Kozinets, R.V., de Valck, K., Wojnicki, A.C., Wilner, S.J.S., 2010. Networked narratives: understanding word-of-mouth marketing in online communities. J.

Mark. 74 (2), 71–89.Lee, S.H., Cotte, J., Noseworthy, T.J., 2010. The role of network centrality in the flow of consumer influence. J. Consum. Psychol. 20 (1), 66–77.Lin, H.-F., Lee, G.-G., 2006. Determinants of success for online communities: an empirical study. Behav. Inf. Technol. 25 (6), 479–488.Liu-Thompkins, Y., Rogerson, M., 2012. Rising to stardom: an empirical investigation of the diffusion of user-generated content. J. Interact. Mark. 26 (2), 71–

82.Marsden, P.V., 2002. Egocentric and sociocentric measures of network centrality. Social Networks 24 (4), 407–422.Merwe, R.V.D., Heerden, G.V., 2009. Finding and utilizing opinion leaders: Social networks and the power of relationships. South Afr. J. Bus. Manage. 40 (3),

65–76.

332 A. Klein et al. / Telematics and Informatics 32 (2015) 321–332

Meuter, M.L., McCabe, D.B., Curran, J.M., 2013. Electronic word-of-mouth versus interpersonal word-of-mouth: are all forms of word-of-mouth equallyinfluential? Serv. Mark. Q. 34 (3), 240–256.

Pagani, M., Hofacker, C.F., Goldsmith, R.E., 2011. The influence of personality on active and passive use of social networking sites. Psychol. Market. 28 (5),441–456.

Panzarasa, P., Opsahl, T., Carley, K.M., 2009. Patterns and dynamics of users’ behavior and interaction: network analysis of an online community. J. Am. Soc.Inform. Sci. Technol. 60 (5), 911–932.

Ridings, C., Gefen, D., Arinze, B., 2006. Psychological barriers: Lurker and Poster motivation and behavior in online communities. Commun. Assoc. Inf. Syst.18 (1), 329–354.

Rigdon, E.E., 2012. Rethinking partial least squares path modeling. In praise of simple methods. Long Range Plan. 45 (5–6), 341–358.Ringle, C.M., Sarstedt, M., Straub, D.W., 2012. Editor’s comments: a critical look at the use of PLS-SEM in MIS quarterly. MIS Q. 36 (1), iii–xiv.Schoberth, T., Heinzl, A., Preece, J., 2006. Exploring communication activities in online communities: a longitudinal analysis in the financial services

industry. J. Organ. Comput. Electron. Commerce 16 (3/4), 247–265.Sledgianowski, D., Kulviwat, S., 2009. Using social network sites: the effects of playfulness, critical mass and trust in a hedonic context. J. Comput. Inf. Syst.

49 (4), 74–83.Stephen, A.T., Toubia, O., 2010. Deriving value from social commerce networks. J. Mark. Res. 47 (2), 215–228.Sun, T., Youn, S., Wu, G., Kuntaraporn, M., 2006. Online word-of-mouth (or mouse): an exploration of its antecedents and consequences. J. Comput. Mediat.

Commun. 11 (4), 1104–1127.Trusov, M., Bodapati, A.V., Bucklin, R.E., 2010. Determining influential users in internet social networks. J. Mark. Res. 47 (4), 643–658.Vista, A., in press. Mass media, the ‘Sensational Message’, and metamorphic truths. Telemat. Inform.Yang, T., 2012. The decision behavior of facebook users. J. Comput. Inf. Syst. 52 (3), 50–59.Zaheer, A., Soda, G., 2009. Network evolution: the origins of structural holes. Adm. Sci. Q. 54 (1), 1–31.Zitek, E.M., Tiedens, L.Z., 2012. The fluency of social hierarchy: the ease with which hierarchical relationships are seen, remembered, learned, and likes. J.

Psychol. Soc. Psychol. 102 (1), 98–115.