Making a Global Community on the Net – Global Village or Global Metropolis?: A Network Analysis of Usenet Newsgroups

Authors

  • Junho H. Choi,

    Corresponding author
    1. Doctoral candidate in the department of communication at the State University of New York at Buffalo. He will receive his Ph.D. in May 2002. After graduation he will assume a position in the Department of Language, Literature, and Communication at Rensselaer Polytechnic Institute. He is interested in the structural patterns of intergroup communication in cyberspace. His recent work on international news flow in cyberspace has been published in Communication Research.
      Address: Department of Communication, SUNY at Buffalo, 359 Baldy Hall, Buffalo, NY 14260-1020.
    Search for more papers by this author
  • James A. Danowski Ph. D.

    Corresponding author
    1. Associate Professor at the University of Illinois at Chicago. He received his Ph. D. (1975) in communication at the Michigan State University. He is interested in communication technology & social systems and Internet theories at international, national, organizational, social network, and individual levels.
      Address: University of Illinois at Chicago, Department of Communication, 1140 Behavioral Sciences Building, 1007 W. Harrison, M/C 132, Chicago, IL 60607.
    Search for more papers by this author

Address: Department of Communication, SUNY at Buffalo, 359 Baldy Hall, Buffalo, NY 14260-1020.

Address: University of Illinois at Chicago, Department of Communication, 1140 Behavioral Sciences Building, 1007 W. Harrison, M/C 132, Chicago, IL 60607.

Abstract

This study examined the global structure of intercultural communication on a computer-mediated communication network. Extracted from a total of 232,479 discussion messages, a matrix of crossposted messages among 133 online newsgroups over a year on the Usenet was analyzed to investigate structural patterns of communication flow. This research found, unlike earlier research, that a simple structure of core-periphery relations does not fit the pattern of cross-cultural postings in Usenet discussion groups. Bonacich's centrality, cluster analysis, and multidimensional scaling analysis were conducted using UCINET V software. Results identified a multi-cored structure with decentralized and diversified patterns of information distribution in cyberspace.

Introduction

For more than 20 years computer-mediated communication (CMC) has been recognized as building new social communities formed around interests, not physical proximity (Korzenny, 1978). Attuned to the significance of these processes, communication researchers have been calling for more study of the Internet (Newhagen & Rafaeli, 1996; Parks & Floyd, 1996). Some have suggested that more exploration of networked ‘cyberspace citizens’ is needed to understand more fully the relationships and effects of fast-changing information environments and more diverse cultural encounters (Ogan, 1993; Baym, 1995).

A network analysis approach appears valuable in this pursuit. Social network researchers have asserted that when a computer network connects people or organizations, it can be fruitfully conceptualized as a social network (Garton, Haythornthwaith, & Wellman, 1997; Haythornthwaite, Wellman, & Garton, 1998; Wellman, 1997). Morris and Ogan (1996) maintain that conceptualizing the Internet as a network is a useful move and they suggest network analysis to address issues of social influence. Barnett (1997) and Barnett and Salisbury (1996) have proposed that international communication network research should be extended to computer networks such as the Internet. Such research on cross-cultural CMC networks has not yet emerged. This study conceptualizes sharing messages among online discussion groups as a form of cross-cultural communication network in cyberspace. It addresses the question as to whether the network evidence is more supportive of an undifferentiated, highly interconnected “global village” structure, a sort of ‘one world’ network, or more supportive of a differentiated and hierarchical “global metropolis” structure, one with multiple subgroups.

Global Village or Global Metropolis?

When scholars conceptualize Internet-based networks in the most general terms, the concept “global village” is likely to appear. McLuhan (1964) predicted the emergence of the “global village” as a result of television networks and satellite transmission technology. As people experience events at the other side of the world as if they were in the same physical space, “we now live in a global village… a simultaneous happening” (McLuhan & Fiore, 1967, p. 63). Today some observers have transplanted the term from its original conceptual location to describe what they perceive as resulting from Internet networks' global reach. Unfortunately the term, “global village” is used glibly (e.g. Harasim, 1993) and its theoretical underpinnings are most often unarticulated and such use unquestioned by others.

One notable exception occurred prior to the take-off period of the Internet. In the context of theoretical perspectives to international communication, Fortner (1993) argued against the use of the term “global village,” citing important features of village communication systems including some significant for this research. One is dependence on oral-aural communication in an atmosphere of intimacy. Another is the sharing of geography for living, including sharing both the sacred and the profane, as well as places of authority and of exile. Another feature of villages is that people marry, give birth, and die among those with whom they share daily life experiences. Where time and space has vanished by electronic communication technologies, people from around the world can communicate instantly as if they lived in the same village (McLuhan & Fiore, 1967; McLuhan & Powers, 1989).

Fortner (1993) posits that communication does not itself provide for these aspects of village life. Furthermore, communication alone cannot create social networks of intimacy and identity having the attributes he associates with villages. Communication can, as Lasswell (1960) and Lerner (1960) articulated, enable people separated by thousands of miles to signal events, know of the activities of others whom they do not personally know, and show potentially common elements in the ways different geographically separated people conduct their lives. Yet, mediated communication cannot foster the intimacy of village life and sharing of the same experiences.

Instead, Fortner argues that a metaphor of “global metropolis” is more fitting to mediated communication systems on a global scale. In contrast to the village, in the metropolis people rely on media for much of their information about other groups residing there. Some groups will be much more visible than others, depending on the relationships of the groups to the media. Information flow is unequal and elites are better known than strata are, because hierarchical and centrally controlled communication systems tend to reflect social inequalities among groups with ‘different degrees of socioeconomic, cultural, and political status’ (Fortner 1993, p. 25).

Given that the Internet is another global communication system with a complex and differentiated structure, extending the metropolis rather than village metaphor to the global level is more defensible to Fortner. Presented in Table 1 is a summary of contrasting features between two metaphors. Based on these arguments the question arises: What better characterizes the Internet's global cultural patterns, the global village or the global metropolis concept, or perhaps something different?

In line with the global village concept is an assumption that there is significant homogenization of world cultures. With the advent of the Internet, some commentators argue that today's CMC networks not only provide tools but offer venues for global community, a matrix where the world can meet and form Internet communities (Harasim, 1993). As these communities evolve over time, cultural uniformity would increase on a global scale. Such a perspective is consistent with the common social science conceptualization of ‘globalization’ (Waters, 1995). It is assumed that information as well as material goods are freely and frequently exchanged between different groups across national and cultural boundaries. It is thought that this makes different systems more consolidated, providing a basis for a new higher-order system that is more uniform.

Some critics have argued for another cause of homogenization of global culture: hierarchical and unbalanced information flow between dominant and peripheral cultures. One side of the debate about the New World Information Order (NWIO) (Gonzalez-Manet, 1992; McPhail, 1987, 1989; UNESCO, 1980) has asserted that the structure of international communications, studied using the nation-state as a unit of analysis, is unequal and unbalanced. There is interaction along the spokes, from the periphery to the center hub; but not along the rim, from one periphery member to another (Galtung, 1971). This has been alleged to results primarily from inequalities in economic power and the extent of a country's own media development.

Without a doubt, the Internet is a communication technology that is at this time distributed unequally across countries, as can be seen in the distributions of hosts by top-level domains (Internet Software Consortium; OECD, 1998a). Much like the diffusion of earlier technologies such as telephone, radio, and television, the spread of the Internet among nations has been unequal mainly depending on each country's economic and social development level (Hargattai, 1998, 1999) and has arguably contributed to a new geography of centrality and marginality (Sassen, 1997). More of the Internet connectivity, volume of data traffic, and hosts and domain names have been concentrated in the developed countries than in the underdeveloped ones (Brunn & Dodge, 2001; Danowski & Szumilas, 1999; OECD, 1998a, 1998b; Zook, 2001). The geographic distribution of Internet users and languages also shows high centralization: 35 % of on-line users are in North America and 30 % in Europe; 43 % of on-line users use English and 32 % use European languages in cyberspace, as of December 2001.1

Nevertheless, at least conceptually, the Internet provides the possibility of both balanced and horizontal flows of information between cultures. Tools and “places” are available for various groups with different cultural identities, even minorities and/or marginalized groups with fewer resources, to meet and discuss their interests. For example, there are hundreds of discussion groups with ethnic, racial, national, or cultural identity in the Usenet, listserver or bulletin board systems, and commercial on-line services such as America Online. Since this new communication technology allows people in the world to interact with each other with fewer constraints, the patterns of communication flow may potentially be different from the hierarchical and centralized flow model argued by world systems and imperialism critical theories.

This new communication medium has been expected to enhance intergroup relationships as well as interpersonal ones. Spears and Lea (1994) argue that CMC tends to equalize status, decentralize and democratize communication flow, and thus empower and liberate the individual user or group. This process facilitates establishing a more equal relationship between and among online groups. Morris and Ogan (1996), in their experiment on group interaction by e-mail, found that minority groups preserve diversity of opinion by clustering with regional neighbors. They argue that CMC can facilitate democratic structures with overlapping groups, minimizing pressures toward uniformity. Latane and Bourgeois (1996) acknowledge that the proliferation of e-mail networks has the potential to alter the dynamics of cultural change. This suggests the possibility of a substantial enhancement to multiculturalism in global communication. Minority cultures can have more choices than ever to redefine or develop their cultural heritage with more diverse alliances without losing local identity. In this sense, CMC may be a useful tool for democratizing intergroup relationships in the global social system.2

Nevertheless, with the expansion of social bandwidth for alternative voices there is also the possibility for a net reduction in the reach and effectiveness of diverse groups' messages. One potential problem is overload of receivers of Internet messages to the point that they curtail their browsing and increase selectivity, in effect tuning out on minority messages. As well, expansion of minority group voices may at some point lead the traditional mass media to label minority or marginalized groups with active Internet voices as deviant and not credible, as those media have been seen to do prior to development of the Internet (Shoemaker, 1982).

In the context of the post-industrial economy, globalization brings about rather complex layers of global culture. Waters (1995) proposes the differentiating impact of globalization on ethnicity and nationhood: globalization pluralizes the world by recognizing the value of cultural niches and local abilities. He asserts that globalization brings the periphery to the center by introducing possibilities for new ethnic identities to cultures on the periphery. Ang (1996) also argues that globalization is not leading to an easily comprehensible totality or a central value system, but to an increasing diversity of connections among all global areas. She sees that the contemporary world is “a thoroughly paradoxical place, unified yet multiple, totalized yet deeply unstable, closed and open-ended at the same time” (p.163). In her postmodern view of globalization, the capitalist world-system today is not an undifferentiated all-encompassing whole, but a fractured one, in which forces of the dominant order of Westernization are undercut by forces of localization, diversification, indigenization, and regional ethnic conflicts (Ang, 1996).

In short, these perspectives reflect different takes on the Internet's influence on world cultures. On one hand, the global village and associated world cultural homogenization may be expected. On the other hand, a picture closer to a global metropolis with multiple cultural foci may be expected. Moreover, there is a lurking murkiness lying between the lines of scholarly conceptualization of the Internet' cultural constructions. This raises the possibilities for yet other structures to emerge from empirical analysis.

Patterns of Cross-Cultural Communication

Theories of Changing Cross-Cultural Communication

Convergence theory (Kincaid, 1988; Rogers & Kincaid, 1981) argues that increased cultural uniformity occurs through increased communication between host and minority cultures. Here cross-cultural communication is defined as a process of creating, maintaining, or bridging racial, ethnic, or national boundaries (Collier & Thomas, 1988; Yum, 1984). When information is shared by two or more cultures, information processing leads to mutual understanding, mutual agreement, and collective action.

Nevertheless, it can be argued that convergence theory does not acknowledge that as communication across cultures increases, there may be greater recognition of cultural differences and similarities, as similar cultural groups form differentiated subgroups within different world regions. Thus, the net result of increased cultural communication among various cultures may be the emergence of multiple hubs of cross-cultural networks with increased convergence within these regions but increased divergence across them. Absent consideration of such processes, convergence theory taken to its logical conclusion would argue for an increasingly homogenous world culture as Internet communication among cultures increases.

World Systems and Imperialism Theory

World system theory and Galtung's imperialism theory (Galtung, 1971; Wallerstein, 1979, 1982) have argued that the world system is so unequally structured that the central countries have benefits over peripheral countries by dominating them economically, politically, militarily, communicatively, and culturally. The structure of communication flow is thought to reflect unequal power relationships in worldwide information exchange; vertical relations dominated by core countries characterize interaction between core and periphery. Peripheral countries tend to communicate only with a few core countries and rarely interact with countries in a similar peripheral position.

In terms of communication network structure, hierarchy can be defined as linear and vertical, rather than modular and horizontal. Information flows with one or a few agent(s) (node(s)) responsible for a large amount of traffic. More unequal distribution of power deepens the structure of the vertical communication network and may lead to cultural homogenization. This may be because there is increased uniform dependence of the peripheral nodes on the central node as hierarchicalization increases. The culture of the dominant node may become increasingly salient to the peripheral nodes; there is a combination of political, economic, and social factors that may work to make the dominant node's culture more attractive. As peripheral nodes communicate with dominant node and not much with one another they may become more culturally similar to the dominant node and indirectly to one another. Historical accounts of such cultural processes come from the body of work on imperial/colonial communication relations.

Research Seeking Core-Periphery Structures

Generally, the results of early research on international telecommunication structure have supported the contentions of world system and imperialism theories about core-periphery structures and international information flow (Barnett, 1997, 2001; Barnett & Choi, 1995; Barnett, Jacobson, Choi, & Sun, 1996; Barnett & Salisbury, 1996; Barnett, Salisbury, Kim, & Langhorne, 1999; Kim and Barnett, 1996; Sun & Barnett, 1994).

Most of these studies used data on the telecommunications traffic between nations and used network analysis to map and measure the structural properties of the global network. In studying data from the 1980s through the 1990s, it was found that the international telephone network was composed of one group with the United States and the Western economic powers at the center, and the less developed countries (LDCs) at the periphery. Networks slowly had become denser, more centralized, and more highly integrated into the core.

Emergence of Regionalism

There is increasing evidence that countries are regionally grouped in the international news exchange (Boyd-Barrett & Thussu, 1992) and telecommunications network according to their joint participation in regional organizations (Kim & Barnett, 1996b). Kim and Barnett's (1996a) research also revealed a recent trend of regionalism in patterns of media exchanges between peripheral countries. Also finding evidence supporting a trend of regionalism, Danowski (2000) observed the emergence of an Arab regional unit in international telephone traffic data.

New Units of Analysis for Cross-Cultural Communication Research?

It is important to note a change in the appropriateness of units of analysis as information societies have evolved. The state cannot be seen as the sole actor controlling and monopolizing global activities, because multiple channels connect world societies. Various actors, including individuals, groups, organizations and institutions as well as governments, are engaging in multiple layers of exchange or interaction, on a variety of issues. As a result, global information flow may be better understood in terms of a series of networks of communication among various entities at different levels of analysis.

The Global Discussion Network: The Usenet Newsgroups

The Internet has relatively fewer constraints and its use is less dependent on resources each agent owns than is the case with traditional media. Some of the proponents of free and balanced flow potential expect a different structure of communication network than core-periphery. The structures of communication networks in cyberspace have been empirically investigated in a few areas of organizational communication (Danowski & Edison-Swift, 1985; Danowski, 1988) and of Web topology (Adamic & Huberman, 1999; Albert, Jeong, & Barabasi, 1999; Broder et al., 2000; Kumar et al., 1999). Therefore, in order to expand the academic interests and scrutinize the structural role and position of each cultural group, which represents ethnic, racial, national, or regional identity, it may be useful to examine the ways in which messages about cultural issues are distributed across groups in cyberspace.

There are a number of important human communication channels in cyberspace: electronic mail (E-mail), Listserve, on-line chatting, on-line multi-user games (MUD), World Wide Web, and the Usenet, among others. Usenet, started as the UNIX users' network in 1979, is now a worldwide public discussion forum on the Internet with more than 25,000 newsgroups, organized into several categories. The mode of communication in Usenet can be characterized as voluntary, interactional, public, and global. Usenet enables managing multiple public conversations about specific topics that are not controlled in central sites but spread throughout the system. People read and respond to conversations about specific topics, similar to the way in which they read and respond to E-mail but the postings are public rather than private. In this way, Usenet participants use Internet tools as many-to-many group communication devices. It is like a ‘giant coffeehouse with thousands of rooms’ (Rheingold, 1993).

Usenet has a huge volume of messages, along with global reach. Smith (1999) investigated the entire corpus of message postings in Usenet newsgroups in 150 days in 1997. In this period Usenet contained 14,347 newsgroups carrying 6 gigabytes of messages per day. On an average day, 20,000 people posted 300,000 messages. More than one million people posted at least one message each for a total of more than 14 million unique posts. Two hundred and three nations were recognized in Usenet based on email domain.

While some might consider Usenet a mass medium based merely on volume and reach, it has different features than conventional mass media in several respects. First, Usenet depends on ordinary people's voluntary participation to create and distribute information. Each participant has the ability to create his/her own message and distribute it to many, yet targeted, audiences everywhere in the world. Traditional mass media, such as television, newspapers, magazines, film, and radio, depend on only a small number of people who have the power to determine which information should be made available to the mass audience. In this sense, Usenet is more fully a mass medium in terms of sources and receivers of messages. Each user is potentially a source, not just a receiver.

Historically in Europe and North America, until the advent of newspapers, the pamphlet was the principal means of stimulating public debate on a wide range of political, religious and cultural issues. Pamphleteering, begun with the Protestant Reformation, was employed in the American Revolution as a tool of political agitation. In terms of their role in providing a forum of debate, Usenet and the pamphlet are quite similar. Also, the content of both Usenet and the pamphlet are not institutionalized, usually containing marginal opinions. Usenet as an electronic network, however, provides distinctive features that print media cannot: openness to authorship, instant interactivity, and wide reach of global audiences. It is a place where ‘birds-of-a-feather,’ even if separated by continental or even global distances, can congregate (Pfaffenberger, 1995).

Usenet participants have been observed to share resources, exchange knowledge, develop a sense of cultural identity and sometimes argue about cultural issues within and across groups. In fact, there is a culture category of groups. The <soc.culture> division in the Usenet categories has more than 150 newsgroups that include most of the racial, ethnic, national, and regional cultural groups in the world. Each participant can post messages and also crosspost messages to other groups based on the poster's assumptions about the relationship among groups to the message content.

Usenet Crossposting as Cross-Cultural Communication

Given this sketch of the pragmatics of Usenet as a communication medium, consider more abstract communication features. Patterns of cross-cultural communication are enacted when a poster of a message (a source) sends it to multiple cultural groups (intended receivers). 3 The source is enacting an assumption about the relationship of different cultures based on their anticipated interest in the content of the message posted. The act of crossposting embodies this assumption and the intent to use the message as an enactment of necessary conditions for cross-cultural communication in cyberspace. Although this research does not examine receivers' perceptions of intercultural associations implied by the sender's posting, the networks enacted by the sources are an important part of the process. The resulting network is an aggregation of sources' perceived similarity among cultures based on message content. When members of receiver groups may in turn crosspost messages to the same groups there is evidence for mutuality in the communication process. This mutuality is not necessarily based on the same thread of message content as would be expected in convergence theory, yet the multi-content mutuality revealed by the overlapping of source-initiated crosspostings may be taken as evidence of multiplexed communication relationships. For positional analysis of this kind of social network, structural equivalence is the most widely used construct that describes the pattern of associations among actors in multirelational networks (Wasserman & Faust, 1994). To the extent that two or more actors maintain the same pattern of relations (e.g. crossposting) with other actors in a network, they occupy the same position in the social structure by mutual contact and lead to the same sources of information. (Burt, 1987, 1992). The model of structural equivalence is represented in the hierarchical cluster analysis and multidimensional scaling which are adopted for this research.

As is the case with Internet communication generally, individuals can comment on the appropriateness of particular postings. As a critical mass of criticism would mount this could have effects on changing posters' likelihood of maintaining the same patterns of crossposting. In this way the whole system of cross-cultural communication is a dynamic, self-regulated system. Accordingly, the network structure observed over a sufficient period of time is likely to reflect a relatively shared view of cross-cultural communication patterns (Rice & Barnett, 1996). This study, however, measures only the enactment of cross-cultural linkages though crossposting patterns and does not study whether feedback emerges that serves to correct the cross-cultural structure. Validation of these conceptual assumptions is a matter left to future research. As a sample message in the Appendix shows, the activity of crossposting is an intercultural act that may make local issues more global. The sample message, about Kosovo, is crossposted to ten newsgroups in the soc.culture category, indicating that the source perceives those ten groups to be related through a common interest in the issue.

Research Questions

Given the conceptual paradoxes of the literature reviewed about the cross-cultural effects of Internet communication, with some scholars arguing for effects of increased global cultural homogenization, and others emphasizing the opposite, we examine several critical research questions in this study. More broadly, which model of the Internet and global cultures best fits the data: global village, global metropolis, or something else?

Method

Sample and Procedures

The basic unit of observation of intercultural communication flow is the message crossposted among Usenet newsgroups in the soc.culture category. A Usenet message is composed of the header and the body (see an example in the Appendix). Although it has a similar format to e-mail, some of the headers are distinctive because of the individual-to-group communication mode. Each header includes its destination. <Newsgroups:> displays the newsgroup or newsgroups to which the message is posted. When a message is posted to more than two newsgroups, the link between newsgroups is established. These links were the unit of analysis, operationalized by the number of times a pair of cultures was a receiver of a crossposted message.

The crossposting data were drawn directly from each newsgroup's archives. All of the 155 cultural newsgroups in the social.culture category were included. Some newsgroups, however, were identical to each other (e.g., soc.culture.african.american and soc.culture.african-american) or divided by subgroups that represented one cultural identity (e.g., soc.culture.pakistan, soc.culture.pakistan.education, soc.culture.pakistan.history, soc.culture.pakistan.politics, etc.). For several newsgroups, a moderated group was separated from the original group (e.g., soc.culture.japan and soc.culture.japan.moderated). The above cases were recategorized into one group and their crossposting with each other (e.g., crossposting between social.culture.jewish and social.culture.jewish.holocaust) was not counted. As a result of recategorizing, 133 newsgroups from soc.culture.afghanistan to soc.culture.zimbabwe were identified in this research.

Longitudinal data gathering is useful to determine the extent to which the network has a stable structure over time and whether pooling of links across time is warranted to simplify the analysis, or if a time-series analysis is justified instead. In the study reported here, three waves of data were gathered and analyzed at approximately six-month intervals. The first included nine days of archives during three months, March, April, and May 1997. Three consecutive days were selected per month and the messages posted during the dates were downloaded. The second set of data came from ten days of archives during two months, five days per month, October and November 1997. The third data gathering was done during May and June 1998 for five days each month. The time intervals help to prevent bias that can result from the arousing of specific intercultural issues that may magnify the interactions between relevant newsgroups for some rather narrow period. Messages containing greater than two hundred lines were screened out, saving not only time and cost of downloading, but also eliminating improbable messages for discussion. Qualitative assessment showed that lengthy messages were usually junk mail, or SPAM, such as commercial announcements or pictures irrelevant to discussion. Sometimes real discussion messages are lengthy but they are not appropriate to interactive conversation and have few replies. A total of 232,479 messages were captured in the data set.

If a source sends one's message to more than one group, the name of newsgroups is found in three headers; <Newsgroups:>, <Followup-To:>, and <Xref:>. 4 <Followup-To:> is a optional header which has the same format as <Newsgroups:>. It is displayed only when the message is crossposted. If a source crossposts a message to more than one newsgroup, this poster can also specify to which newsgroups follow-up messages should be sent. <Xref:>, also an optional header, contains information about where the host Usenet site stores the message. The first author created computer software to count the links in the data archives. The program identifies the name of each newsgroup in the <Newsgroups:> header in a message and adds the name to a total count. By calculating the number of links among each pair of newsgroups, a square matrix was constructed. As a result of this procedure, a 133 by 133 matrix of newsgroups with counts of crossposted messages as cell entries was made for each of the three time periods.

Data Analysis

Network analysis is a set of research procedures for identifying structures in systems, based on the relations among components (Knoke & Kuklinski, 1982; Rogers & Kincaid, 1981, Wasserman & Faust, 1994). In order to describe the underlying structure, network analysis examines systems indicators such as centrality, connectedness, integrativeness, and system density, and also the higher-order attributes of possible clustering of the network into subgroups.

The basic network data set is an n×n matrix S, where n equals the number of nodes in the network. A node might be an individual or higher-level component such as an organization or a nation from which the system is composed. Each cell, Sij, indicates the strength of the relationship among nodes i and j. In communication research, this relationship is generally the frequency of communication among the nodes (Barnett, Danowski, & Richards, 1993).

In the current research, cultural newsgroups were the nodes, and the relationship was operationalized as the number of messages crossposted. For example, if node i is <soc.culture.usa> and node j is <soc.culture.korean>, Sij would be the number of messages that are posted in both newsgroups. S is symmetrical (Sij=Sji) when one is not concerned with directionality. In those instances when the source and receiver of the communication would be differentiated, S is asymmetrical (SijSji). In this analysis, S is symmetrical because the crossposting messages have no directionality. To analyze the structure of intercultural communications in Usenet, three methods of network analysis were used: Bonacich centrality, cluster analysis, and metric multidimensional scaling.

Bonacich's centrality. Freeman (1979) characterized three facets of actor centrality based on linkage itself: degree, betweenness, and closeness. Degree centrality refers to the number of direct links and is an index of an actor's connectedness (Freeman, 1979). In other words, a degree centrality equals to the number of direct links a node has (e.g., the number of crossposted groups). Betweenness refers to the frequency with which a node falls between pairs of other nodes in the network and closeness shows minimum steps to all other nodes (Freeman, 1979; Freeman, Roeder, & Mulholland, 1980).

However, Freeman's three centrality measures considered neither the weighting of nodes nor the paths other than geodesics in the network. An alternative is Bonacich's (1972) measure of centrality. It considers the strength of the relationships among the nodes by taking the eigenvector of the largest eigenvalue of matrix S, standardized so that its length is equal to the eigenvalue. The loadings on this vector indicate a node's centrality. Reading the eigenvector centrality index, however, requires some caution because of an inherent circularity involved in the calculation of the centrality measure. A node, connected to central nodes, has its own centrality boosted, and, in turn, boosts the centrality of other nodes to which it is connected (Scott, 2000). The algorithm from UCINET V (Borgatti, Everett, & Freeman, 1999) was applied to determine the centrality of each newsgroup.

Cluster analysis. Cluster analysis was also used to identify subgroups within the intercultural communication flow network. It is a method for finding groups of similar entities in data (Aldenderfer & Blashfield, 1984). It identifies those groupings or clusters of nodes that best represent their measured relations. From a similarity matrix of n nodes, the pair of nodes with the greatest similarity is combined to form a cluster, C1. Then a new n-1×n-1 matrix is generated, with the pair of nodes combined as a single node. The process is repeated, with a third node added to C1 or a new pair of nodes combined to form C2. This procedure continues until all nodes are included to form a Cn cluster. In this case, Johnson's hierarchical cluster analysis from UCINET V (Borgatti, Everett, & Freeman, 1999) was used. This form of cluster analysis has been employed in many studies of global communication systems (Kang and Choi, 1999; Kim and Barnett, 1996a, 1996b). Moreover, it is the default method in most statistical analysis programs, attesting to its popularity. If cohesive clusters (subgroups) are found, this would be evidence of regionalized networking of local culture. Using the same data reduction procedures as used in prior research adds validity to our findings.

Two different clustering modes were used, agglomerative (average method) and diameter (complete link method). In complete-link clustering (also called the diameter or maximum method), the distance between one cluster and another cluster is equal to the longest distance from any member of one cluster to any member of the other cluster. In average-link clustering, the distance between one cluster and another cluster is equal to the average distance from any member of one cluster to any member of the other cluster (Borgatti, 1994). Both methods were used and compared because they are thought to sometimes yield different results. To the extent they both reveal similar clusters validity of the clusters is enhanced.

Multidimensional scaling. Multidimensional scaling (MDS) examines the spatial structure of the network. This analysis was conducted to map distances among newsgroups. This helps to capture the combined picture of centrality and cluster analyses for spatial models of social and communication networks. In this spatial model, centrality is defined as the position in the space closest to the origin of the coordinate system created from the patterns of interaction (Barnett, 1993; Bonacich, 1972; Kincaid, 1988). Along with dimensional interpretation, a neighborhoods approach can reveal patterns in the data based on how cluster members are represented in the plot (Borg & Groenen, 1997; Kruscal & Wish, 1978; Rice & Richards, 1985).

In multidimensional scaling, matrix S may be converted to a space of many dimensions, with each node located on a series of projections on orthogonal dimensions. The stronger their relationship (the more frequently two nodes communicate) and the more similar their interaction patterns, the closer they are in the network space (Barnett, 1988). Mathematically, the process is equivalent to converting a matrix of intercity distances to Cartesian coordinates, where latitude, longitude, and altitude are the dimensions and the cities' location given on each dimension. From these coordinates, a graphic representation, a map, may be drawn. This process is known as metric multidimensional scaling (Torgerson, 1958). In this analysis, the metric MDS software from UCINET V (Borgatti, Everett, & Freeman, 1999) was employed.

Results

Comparison of Time Periods

To address the question as to whether the three time periods were significantly different, a Quadratic Assignment Procedure (QAP) was run, correlating each pair of time period matrices. A QAP method is a nonparametric procedure to test the spuriousness of correlations between two matrices (Krackhardt, 1987). The QAP algorithm from UCINET V (Borgatti, Everett & Freeman, 1999) was used to test the relationship between three matrices.

The correlations between each of the pairs of the three time period matrices were significant (p > < .000). For period 1 and 2 the correlation was .67, for period 1 and 3 it was .58, and for period 2 and 3 it was .69. This provides empirical justification for combining the data from the three time periods into an aggregate matrix. This will simplify the presentation of results and add reliability without compromising validity.

Degree of Crossposting

Descriptive statistics were computed on the extent of crossposting for each of the 133 Usenet cultural groups. Forty-six percent of posts to newsgroups in the soc.culture category are crossposted, indicating that it is an important feature of communication. The amount of crossposting provides evidence as well about the likelihood of identifying emergent network structures that are potentially robust enough to reflect a valid representation of the crossposting structure. In contrast, when there is a low volume of link data network solutions may default to a single large group structure with a core-periphery pattern simply because there are not enough links to reveal a more differentiated structure. Having a large amount of link data does not, however, introduce the opposite bias in network analysis, constraining the identified network solution to a non core-periphery structure.

Centrality of Newsgroups

There were considerable differences in centrality among the newsgroups. Generally, newsgroups with higher numbers of links and crosspostings showed higher centrality. The Singapore newsgroup showed the highest centrality. Among the top newsgroups of higher centrality, many Asian groups were included: Singapore, Taiwan, Hong Kong, Asian-American, China, and the Malaysia newsgroup. Most of European and North-American newsgroups were listed in the top 30. Latin-American and Arab groups were located in the middle. African and local groups with fewer numbers of links and crosspostings were less central. Although the number of links (connectedness) did not show many differences among the top 50% of newsgroups, the centrality scores produced from UCINET V revealed large differences among newsgroups. The newsgroups of Bonacich's Centrality are in Table 2 listed by the normalized eigenvector index.

Bonacich's centrality results, however, showed that an important criterion for a valid solution was not met. The ratio of the first eigenvalue to the second should be at least 1.5 and preferably 2.0 or greater (Borgatti, Everett, & Freeman, 1999). The current ratio was 1.36. This is suggestive of a non-core/periphery structure because a single dominant eigenvector does not emerge. When such a result occurs the centrality measures are biased because of the effects of subgroups in the data. Because of the inherent circularity in the calculation of Bonacich's centrality measure (Scott, 2000), many of Asian newsgroups may have boosted the centrality of their own and of other nodes to which they are circulrarly connected. The recommended follow up course of action is to perform a full factor analysis on the data in order to identify an underlying pattern of correlations within a set of variables.

This was done using UCINET's implementation of the principal components method, using Varimax rotation of factors, and the matrix was decomposed into a product of the most dominant eigenvectors. The Scree test on the eigenvalues indicated that a seven-factor solution was meaningful (see Table 3). This can be interpreted as evidence for a multiple subgroup structure, and a more sophisticated view of a hidden pattern of grouping is presented by the hierarchical cluster analysis.

Clustering of Newsgroups

Two common types of cluster analysis were used and compared: agglomerative (average) and diameter (complete link) methods (see Table 4). A cluster analysis shows the extent to which each subgroup has similar patterns of communication flow.

Agglomerative Method (Average): One large cluster (106 members) was identified at the 2.87 level with two dyad and 10 drops. When the level went up to 5.35, the cluster started to ramify. The level of 39.0 was the optimal cut-off point, where Cluster F′ and G′ ramified. Above the cut-off level 8 clusters (94 members), 6 dyads (12 members) were detected with 27 drop-outs.

Diameter Method (complete link): A major cluster was not identified above the 0.0 level. At 40.0 as the optimal cut-off point, 15 clusters (86 members) and 9 dyads (18 members) were detected with 29 drop-outs. The results show that the diameter (complete link solution) method produced more clusters and dyads. There were more ramified clusters that consisted of one cluster in the agglomerative method (average solution). Nevertheless, the membership of corresponding groups in the agglomerative method is quite similar. Because the diameter approach emphasizes closeness as affected by similar distance from all others and is less subject to problematic results (Wasserman & Faust, 1994, p.381), the diameter method was adopted in the description of clustering and each member in a cluster was given a code with an identification number, which will be used in a multidimensional scaling plot.

The result of clustering roughly corresponds with membership in the factor analysis in Table 3. Also, it shows several underlying factors that explain the distinctions. The main factor may be geopolitical proximity. Clusters A, B, C, and D were composed of Asian newsgroups with the Australian newsgroup. European newsgroups showed more diversified patterns, clustered into 7 different subgroups. Cluster L and M are composed of Latin American newsgroups plus the Spain newsgroup. Since Spanish is widely used in discussions by Latin American groups, language may be another dimension of clustering. The Mexican-American newsgroup, in which English is used, resulted in a dyad with another U.S. ethnic minority group, the African-American, whereas the Asian-American group was clustered into the Asian cluster (B). Many messages written in French in the Laos newsgroup were crossposted to the Quebec newsgroup and caused it to be assigned to cluster G. Although language appears to be a very important key to the groupings as seen in above examples, testing this hypothesis by QAP may not be desirable because most of Asian groups do not share a common language in actuality. The situation is also similar for the European groups. Given that typical Internet users are comparatively well-educated, English is used as a global language especially when crossposting to multiple groups. Spanish and French are used as major languages for regional networking, as seen in cluster L-M, and cluster G.

Worth noting is the fact that many pairs of clusters whose members have politically confronted one another were grouped into the same cluster: Pakistan and India (A), China and Taiwan (B), British and Irish (K), Iran and Iraq (P), and Israel and Palestine (Q). These findings reflect known political situations in regional interactions. Some subgroups were clustered primarily by shared cultural similarity, heritage, and language, as evidenced in the Spain newsgroup being a member of a cluster composed of most of the Latin American newsgroups.

These patterns of clustering are an important distinction from those in other telecommunication network research, that identified only one cluster composed of a few Western countries. They point toward the increasing regionalization suggested by Kim & Barnett (1996) and the empirical evidence presented by Danowski (2000).

Multidimensional Scaling

Like cluster analysis, MDS takes a holistic approach to representing the proximities among actors in the network. For a correspondence with the cluster analysis, a total of 104 nodes, generated by the diameter method (i.e., 86 group members and 18 dyad members), were selected for analysis and display. After deleting 29 drop-outs from the original 133 by 133 matrix, a new matrix of size 104 by 104 was created to put into the metric multidimensional scaling analysis.

A standard measure of the adequacy of a MDS solution is the stress index. Solutions with stress values less than 0.20 are considered to reflect acceptable goodness-of-fit (Kruskal & Wish, 1978). Given that the large number of nodes and null values in a matrix negatively influences the stress index, the relatively large size of this matrix (104 nodes) with 34% null cells (3648 out of 10712 cells) was not a good condition to produce acceptable stress level.

An initial attempt at a two-dimensional solution produced an excessive stress level of 0.41. As indicated in the factor and cluster analyses, this matrix seemed to have multiple dimensions in the proximity structure. Theoretically, like the assumptions in the factor analysis, the number of objects minus one (i.e., n-1) is the number of possible dimensions, but a space higher than four dimensions is impractical or impossible to visualize. Setting the scaling to produce a three-dimensional solution, however, also resulted in a rather high stress level of 0.32.

To reduce the stress level and distortion of distances, the raw matrix was transformed into a correlation matrix by using the Similarities solution in UCINET V (Borgatti, Everett, & Freeman, 1999). A neighborhood or pattern approach provides more meaning than a dimensional interpretation of the proximity matrix when the matrix is expected to have higher dimensionality (Kruskal & Wish, 1978). Also, in a neighborhood approach, correlation coefficients have been used as proximity measures (Kruskal & Wish, 1978; Wasserman & Faust, 1994). In the transformed matrix the correlation coefficient of each cell of the matrix represents the proximity between two newsgroups.

A two-dimensional solution of the correlation coefficients matrix produced a stress level of 0.31. Setting the scaling to produce a three-dimensional solution resulted in a good stress level of 0.21. The spatial distances among newsgroups were illustrated in three two-dimensional diagrams (dimension 1 and 2, dimension 1 and 3, and dimension 2 and 3). In all three diagrams most of the clustered members were located at close distances.

The distance plot diagram of dimension 2 by dimension 3 is presented in Figure 1. It gives the greatest correspondence with the result of cluster analysis and, therefore, contributes to convergent validity. The USA newsgroup (E01) was located around the center because of the highest connectedness (101 ties, that is, 98% connected to all nodes) and prestige. The USA newsgroup has strong connections to most of the central members in other cluster groups, along with members in the E cluster. However, other members in this cluster (E02: Europe, E03: Canada, E04: French) are located in the west with European groups due to strong external ties with them.

Cluster members were easily identified as neighborhoods in the diagram. The Middle East newsgroups (Group P and Q) were grouped in the north. In the east are South Asian newsgroups (Group A and Dyad X), but they distanced themselves from the center. Latin American newsgroups (Group L and M) also formed a cohesive block in the south, but the closeness of Latin American groups and Asian Groups was not easily configured. Southeast (Group C and D) and East Asian (Group B) newsgroups displayed their proximity. In the west to northwest were European groups (Group E, F, G, H, J, K, and N) displayed. Members of some clusters and dyads (F, N, S, T, W, and Y) were not located at close distance because less central nodes tend to show more distortions in the diagram.

Cluster analysis shows the cohesive subsets of actors in a dendrogram, but the inter-group relationship among those subsets or ‘possible cleavages between groups’ (Wasserman & Faust, 1994, p. 287) can be clearly displayed in a MDS diagram. One interesting example is Dyad R (R01: African-American, R02: Mexican-American). The African-American newsgroup shared more messages with the Asian-American newsgroup (B04) than with the Mexican-American newsgroup: 768 messages (17%) were crossposted to the Asian-American newsgroup and only 149 (3%) were posted to the Mexican-American newsgroup, out of 4,432 messages. The Asian-American newsgroup, however, with stronger ties with China, Japan, Hong Kong, and Taiwan newsgroups, was grouped with this East Asian cluster (B). The external relationship with the Asian-American group made dyad R locate itself closer to cluster B in the south of the diagram.

This spatial analysis again showed the multi-cored structure of the intercultural communication system with cluster patterns among culturally and geopolitically similar newsgroups. However, the USA newsgroup is positioned in the center, partially supporting the interpretation of center-peripheral structure. The spatial representation of the intercultural discussion network indicates a rather complex structure that has both a center and strong regional connections.

Discussion

The research reported here investigated the structure of intercultural communication networks among discussion groups in Usenet. The results of the centrality and cluster analyses suggest that the structure of communication flows in cyberspace is more complex than world system theory and international news flow research have described them to be. Rather than showing the simple center-periphery structure, we found that the global crosscultural community in cyberspace has several cohesive subgroups structured by cultural and geopolitical similarity. Before interpreting these findings, we discuss limitations.

Limitations

The data for the study reported here were based on natural, voluntarily created groups in Usenet. Their organizing themes show multiple levels of cultural identity. For example, Mexican-american, Mexican, and Latin-American newsgroups have different, yet interconnected, levels of cultural identity. Nevertheless, this multiplicity of levels among units of analysis is embedded in the results. Adjustments were made in the analysis to reduce some multiplicity, although some multiplicity of identity remains. Another limitation is that the study focused on only one aspect of electronic networking. International Internet traffic also includes major activity on the World Wide Web and in private e-mail, which were not studied here. A further limitation is that the study did not investigate changes over time. Longitudinal study over years is highly recommended for future research. Thus, time-series analysis for data gathered over long periods of time would reveal possible changes in the structure of cross-cultural communication within Usenet.

Message content was crucial to the study in that the network analysis was grounded in the crossposting of messages. Variations in the substantive content of messages, however, were not analyzed. In future research, semantic network analysis (Danowski, 1982; 1993) may provide a useful structural view on meaning construction in different on-line communities. For example, this method can answer questions like the following: (1) what kind of content would stimulate greater cross-cultural discussion, and (2) what kind of content would be crossposted to more diverse groups. Other methods of content analysis could also be usefully deployed. Additionally, further methodological triangulation is recommended. Here we triangulated with three data reduction methods, yet many other kinds of future triangulation are warranted.

Interpretation of Findings

In this study of the structure of intercultural communication patterns among Usenet cultural discussion groups we observed that communication flow in cyberspace appears to be more complex than the simple center-periphery structure based on hierarchical power relations. We found that the global community in cyberspace has multiple cohesive subgroups. The social distances among them appear to be determined by the members' regional geopolitical proximity, language, cultural affinity, political conflicts and other cultural factors, not solely by economic power. The center was identified in the multidimensional scaling, but the concentration of communication flows into the center was not manifestly visible.

The results are consistent with a trend of regional coalition in global communications. Asian groups and Latin American groups, located at the periphery in earlier research on international news and media flows, formed cohesive subgroups in a network identified through crossposting messages and sharing opinions and information. The clustering patterns of these multiple cohesive subgroups in global community reflect their diversity. Regional information exchange among the Third World ethnic/national groups may be indicative of colonial struggles against dominant Western culture. This communication flow pattern may reflect the form of regional cooperative activities for balanced relationships, as well as reflect regional rivalries. The slogan of NWICO (New World Information and Communication Order), “many voices, one world,” appears evidenced in this domain of cyberspace. That vision of connected global community coexisting with the retention and enhancement of local/regional identity has some support here.

The intercultural communication patterns in this cybercommunity have implications for cultural issues of globalization. Our findings can be interpreted as supportive of Waters's and Ang's views on the complexity of globalization. We found a slice of evidence of such chaotic and fragmented order in the changing global community in this interconnected, yet multi-clustered structure of the communication network in cyberspace among cultural newsgroups. Asian and Latin American groups, which have been regarded as peripheral in the hierarchy of communication networks, formed regional allegiances and moved toward the center from the periphery. Ethnic minorities actively utilize Usenet, with many near the top of the list of number of posted messages. The Irish, Jewish, Vietnamese, African-American, Cuban, and Asian-American newsgroups are in the top 20. Apparently, participants in the Vietnamese and Cuba newsgroups are diasporic peoples and many of the other newsgroups above are seemingly so.

It should be noted, however, that the technological resources that make the Internet distribution of information possible are still largely in the hands of a very limited number of powerful nations and groups. These still have dominant power in the world system by maintaining the infrastructure of communication network systems. It also should be noted that various studies on international news flow and telephone networks have shown that the world communication network has a hierarchical structure with many peripheries around a few cores. This study has focused entirely on communication flow structure in a totally different context where resources are no longer as strong as a set of constraints to information flows. Nevertheless, in this study infrastructure constraints were a possible deterrence; for example, only 12 African newsgroups were identified and only a few of them were clustered.

The fundamental question of this study was: which best accounts for the structure of global communication: global village, global metropolis, or something else? This research presents results more in line with the global metropolis model than with the global village model. Some aspects of the global village model are evidenced here: equal access to information, less hierarchy, and decentralization. However, segmentation rather than homogenization was the strongest evidence in support of the global metropolis metaphor. In a communication environment with equal control of information distribution, cultural newsgroups form multiple subgroups rather than create one encompassing community.

It seemed, to some degree, that hierarchy existed among the newsgroups. The USA newsgroup was identified as a structural center in the network, perhaps due to the United States' dominant position in contemporary international relations. In contrast, newsgroups organized around countries or peoples with fewer technological resources and fewer participants were located in the periphery and connected very weakly to other newsgroups. Compared to the control of other communication technologies, however, the Internet provides conditions of less centralized communication flow. The relatively decentralized communication flow was evidenced in many Asian and ethnic minority newsgroups listed in the higher ranks of the centrality table.

Future research may find that a hybrid of the global village and the global metropolis is the best model. Cultural homogeneity for some aspects of world culture may be increasing based on such factors as more widespread acceptance of common macro-, micro-, and consumer economic models and of democratic forms of government. At the same time there may be increasing retention and enhancement of local and regional cultures made possible by the relatively unconstrained Internet, particularly as it continues to diffuse and extend its reach into most countries and ethnic groups. Examination of such “glocalism” (Swyngedouw, 1996) in future research would be valuable in providing a more definitive answer to the basic question than our preliminary findings have been able to accomplish. Glocalism would be the “something else” model providing a synthesis of global metropolis and village models. Such a model merits future study.

Footnotes

  • 1

    The estimated population of on-line users, as of April 2002, is 544.2 million, which is approaching ten percent of the world population. In September 1999, the figures were 201 million and 3.4 per cent. More information on geographic distribution of on-line users can be found at http://www.nua.ie/surveys/how_many_online/index.html. The source of global Internet statistics by language is available at http://www.euromktg.com/globstats/. Specifically, the Internet had been and is US-centric. First, historically, the infrastructure of the Internet was created and developed mainly in the United States (i.e., ARPANET and NSFNET). Second, the number of host servers and registered domain names in the United States provide evidence of this concentration. The US accounts for 58% of all Internet hosts and only 7 of the top 100 websites are based outside the US (Telegeography, 2000; original source from Web21. http://www.web21.com). Additionally, the US controls the Internet's directory system (Telegeography, 2000). Out of 13 root servers of the domain name system (DNS), ten servers are deployed in the US. Third, the US-centric Internet presents a clearer picture of concentration with respect to backbone connectivity. The US is the hub whose backbone spokes connected the rest of the world and plays a role as Internet switching office for international data flows (Abramson, 2000; Telegeography, 2000). The largest international ISPs are giant telecommunication companies of the US (e.g., MCI-WorldCom, GTE, Sprint, and AT&T) and they control the vast majority of the Internet's backbone cables and routing tables (Cukier, 1999). It is estimated that over 70% of Internet traffic in non-US locations goes first through to the US because of bandwidth capacity (Cukier, 1999; Huffaker, 2000).

  • 2

    In global politics, various small organizations can make connections more easily on-line and make their voice heard more easily. For example, Labor Net, provided by the Institute for Global Communications (http://www.igc.org), is an international online network for progressive social movement groups. Even local conflicts, often with government and corporations, are distributed through affiliated networks and calls for collective action to resolve them are communicated (Herod, 1998).

  • 3

    Anonymity is one of the common characteristics of on-line communication. It is very hard to determine if another has identified himself/herself honestly. Hiding or misrpresenting one's biological, social, or ethnic identity appears to happen frequently when one communication about interpersonal issues is involved. In public discussion, however, there is not much motivation to misrepresent one's cultural identity.

  • 4

    <Newsgroup:> is a default header in most newsgroup reader software. In Netscape there are two header view modes: all and normal. Normal mode is set to a default that shows the newsgroup header. Microsoft Outlook also shows the newsgroup header as default. All text-based newsreaders such as Pine and Web-based Google Group similarly show the newsgroups header as default. Because all crossposted groups appear in the <Newsgroups:> header, readers will be aware of crossposting when they read a message. One exception is AOL, who changed their policy and no longer allow crossposting, citing traffic load.As a default function in most newsreader modules and software, a reply to a cross-posted message automatically get cross-posted, just like replying to e-mail with multiple receivers. Follow-up is a feature to control the range of replied messages in a thread. However, this feature is rarely used and not provided in many newsreaders (e.g., AOL and Google Groups don't have this feature). Instead, when a poster want to change the membership of crossposting, generally he/she posts a message in a new thread.Some news readers (Google, AOL) display only the initial messages in each thread. In this research WinVN Newsreader version 0.99 was used for data gathering. All messages in every thread were included in the data set and every crossposted group in the soc.culture category was counted.

Ancillary