Sheizaf Rafaeli is a faculty member at the School of Business Administration of the Hebrew University of Jerusalem. He is Co-Editor of the Journal of Computer-Mediated Communication, and Co-Editor of the forthcoming Network and Netplay: Virtual Groups on the Internet (AAAI/MIT Press). He is Principal Investigator for Online Textbooks and Online Welfare Counseling and Support research projects.
Fay Sudweeks BA (Psychology), MCogSc, is a Research Associate at the University of Sydney and a doctoral candidate (Business Systems) at the University of Wollongong. Her research interests are sociolinguistic aspects of computer-supported collaborative work, group development, and the application of Web-based technology to education and collaborative work. She is currently involved in the development of an interactive multimedia international journal and a MOO-based distance learning program.
Address: School of Business Administration, Hebrew University of Jerusalem, Mt. Scopus, Jerusalem, 91905, Israel.
Address: Key Centre of Design Computing, University of Sydney NSW 2006, Australia.
What makes computer-mediated groups tick and/or stick? To what degree are computer-mediated discussants really sustained “groups”? Does the grouping quality reflect anything beyond technical structure? Are technical structure and grouping related? How do threads define groups, or vice-versa? Does any of this change between academic and commercial networks?
We propose that one useful perspective for studying group computer-mediated communication (CMC) is interactivity. Interactivity is a theoretical construct that grapples with the origins of captivation, fascination, and allure that can be inherent in computer-mediated groups. In the coded data from the sample of messages collected by ProjectH, we have a representative snapshot of communication among the very large groups populating the networks. The central unit of interest in studying computer mediated groups is, in this case, the thread of messages. A message thread is a chain of interrelated messages. Rather than individuals' self-report, linguistic and sociolinguistic analyses of content, or observational data of larger units, we examine interactivity, the dependency among messages in threads.
Results indicate that the content on the net is less confrontational than is popularly believed: conversations are more helpful and social than competitive. Interactive messages seem to be more humorous, contain more self-disclosure, display a higher preference for agreement and contain many more first-person plural pronouns. This indicates that interactivity plays a role in the social dynamics of group CMC, and sheds a light on comparing interactive messages with conversation. The focus, we propose, should be on the glue: that which keeps message threads and their authors together, and what makes the groups and their interaction tick.
A network is a net. In sports, a “net” is a barrier. Elsewhere, “net” is that which remains after deductions are made. “Net” is also a means of entrapment, surrounding and captivation. Among all interpretations of the “net” in “network” this third meaning is, in our view, the most appropriate. This chapter is about the captivation (or engagement) of participatory communication on the networks. How are we trapped by group computer mediated communication (CMC)?
We begin with the observation that group CMC is an experiment in social integration and democratic participation. By some measures at least, the experiment works. People want to join, and when they do, many stay. We propose to examine how captivation occurs through interactivity. Interactivity is a variable quality of communication settings. It expresses the degree to which communication transcends reaction (Rafaeli, 1988). Following classical scholars such as Goffman (1967), and more contemporary work such as Bretz (1982), Steinfield (1987), Rafaeli (1986, 1988, 1989) and others, we believe that interactivity is a pivotal measure. It is a cause in the social dynamics of group communication. Such interactivity is made possible, but not always exercised, by CMC. It is especially interesting, but hardly studied as-of-yet, in the context of large groups.
We relate interactivity to several dimensions of interest in Project-H data: activity itself, type of list (voluntary/academic vs commercial), list climate, the emergence of leadership (the “guru” phenomenon), and others. We propose that interactivity is associated with those message qualities which invite people and make people gravitate to groups on the net. Interactivity may be a mechanism through which netting occurs on the net.
Many people congregate on the nets. Networks are centralized distribution mechanisms that are both democratic and anarchic. Computer mediated groups can be viewed as an enigma in traditional, rational and economic terms. After all, the medium is owned by no one. The process is unmanaged in any traditional sense of motivation, profit, control or censorship. Joining and departing participants do so without so much as a required introduction or an agreed upon etiquette. The groups are of an undetermined size or constitution. The situation is neither the classical written nor traditionally spoken communication. CMC groups are neither mass nor interpersonal. If we count participants and symmetry, we find that CMC is the largest form of conversation, or the smallest form of mass communication.
Computer mediated groups beg the questions: why do people make this investment and why does this social phenomenon happen? Is the allure of group CMC in its emulation of face-to-face interaction? We argue that interactivity is the mediating phenomenon. Interactivity may occur in face-to-face contexts, but is not mandatory. It may be present in CMC contexts as well. Here, too, not always. In any case, it is not the direct similarity to face-to-face that matters.
Interactivity is a process-related, variable characteristic of communication settings. Like face-to-face communication, computer-mediated communication has the capacity of enabling high interactivity. One postulated outcome of interactivity is engagement. Interactivity can lead to sociability. We therefore propose that the concept of interactivity is a likely candidate to help in explaining how groups, especially CMC groups, stick together.
Interactivity is not a characteristic of the medium. It is a process-related construct about communication. It is the extent to which messages in a sequence relate to each other, and especially the extent to which later messages recount the relatedness of earlier messages. Following Goffman (1967, 1981), Bretz (1983), McLaughlin (1984), Rogers (1986), Tannen (1989), Schegloff (1987, 1992)Walther (1992) and others, we note that communication is mostly about and for the purpose of interaction. Interactivity places shared interpretive contexts in the primary role. Interactivity describes and prescribes the manner in which conversational interaction as an iterative process leads to jointly produced meaning. Interactivity merges speaking with listening. And it is a general enough concept to encompass both intimate, person-to-person, face-to-face communication and other forums and forms.
Logically, interactivity is indicated as a useful concept for mapping group CMC because it is (like group CMC itself) a hybrid construct. The concept of interactivity directs our focus to the intersection of the psychological and the sociological, the bridge between mass and interpersonal communication, the meeting of mediated and direct communication, and the paradox of written vs. spoken. Interactivity varies along a continuum (Rafaeli, 1988). At one end is declarative (one-way) communication (e.g. most radio and television). Reactive (two-way) communication is further down the road. In reactive communication, one side responds to the other side. Fully interactive communication requires that later messages in any sequence take into account not just messages that preceded them, but also the manner in which previous messages were reactive. In this manner interactivity forms a social reality (see Figure 1).
Interactivity is the condition of communication in which simultaneous and continuous exchanges occur, and these exchanges carry a social, binding force. Brown and Yule (1983) and Zack (1993) summarize additional qualities of interactive communication: allowing for multiple types of cues, potential spontaneity, emergent progression of the content, the ability to interrupt or preempt, mutuality, and patterns of turn-taking. But our definition of interactivity goes beyond Schegloff's simultaneous exchange and Goffman's continuous feedback. We support Schudson's (1978) contention that face-to-face conversation cannot be used as the standard of comparison for group CMC. We argue that interactivity is a continuum, a variable, not just a condition. And we insist that most communication, face-to-face or not, falls short of full interactivity. The case of group CMC differs from conversation, if only for sheer size, and because interruptability and turn-taking take on different meanings.
What does interactivity do? Interactivity in communication settings is associated with the attitudinal dimensions of acceptance and satisfaction. But it is also related to performance quality, motivation, sense of fun, cognition, learning, openness, frankness and sociability (Rafaeli, 1988). Interactivity operates in a supplementation mode. Quasi-interactive (reactive) media can allow people to use the media as a substitute for sociability. The human need for interaction (Beniger, 1987), when satisfied, allows people to use interactive media to bolster their favorable disposition toward interacting with others. Interactivity can be shown to lead to more cooperation.
We turn now to an examination of issues that arise from research on group CMC.
The picture of group CMC painted by summing laboratory studies may be somewhat incomplete. The external validity of laboratory studies of group CMC is problematic for three reasons. (1) Subjects are an atypically captive audience. (2) Groups studied in experiments tend to be unrealistically small. (3) An almost natural inclination of experimental design is to contrast computer mediated communication with a face-to-face standard of comparison. As discussed earlier, this contrast may be misleading.
Survey evidence about group CMC
Many field studies of group CMC focus on the narrow bandwidth and cue deficiencies that typify CMC. Short, Williams and Christie (1976), Rice (1984), and Culnan and Markus (1987) use the term “social presence”. Sproull and Kiesler (1991) refer to the “lack of social context cues”. Trevino, Daft and Lengel (1990) mention “media richness”. Taken together, these related terms suggest that CMC is best thought of as a task-related and/or problem-solving environment. In other words, much of the survey work on CMC hints at it not being a suitable context for social interaction. Others (e.g. Rice and Love, 1987) have uncovered actual emotional use of CMC, albeit often negative emotions. Self reports of CMC users are often in contradiction to the notion of media poverty; experienced CMC users rate CMC as richer than even face-to-face (Steinfield, 1986). Other field-based studies, mostly in organizational contexts, search for reasons people adopt or “use” CMC, frequently arriving at social influence or cultural construction explanations (e.g. Fulk, 1993; Steinfield, 1987; Schmitz and Fulk, 1991). Common to these studies is a focus on constructs such as social influence and critical mass. These are all external qualities, not internal to the communication setting. A focus on content has led some to study creative ways in which members of CMC groups seek to break the bandwidth barrier. The topic of nonverbal behavior on the net, for example, has been the subject of much study (e.g. Carey, 1980; Hellerstein, 1989; Blackman and Clevenger, 1990; Reid, 1991).
Case studies of individual computer-mediated groups seem to be more upbeat and optimistic in their description of group CMC (c.f. Danowski and Edison-Swift, 1985; McCarty, 1990). Finholt and Sproull (1990) observed CMC groups within an organization behaving like real social groups, despite the fact that their members shared no physical space, were invisible, and their interaction was asynchronous. Hahm and Bikson (1989) report on a field study among retired and employed individuals, in which group CMC resulted in increased interaction among members of the group. From the perspective of searching for the “social glue”, a description of the way in which computer-mediated groups come and hold together, it seems unlikely that either social presence or emoticons hold the answer. However, because they are case studies of single groups and often intraorganizational, these studies, too, do not offer a convincing driving force that would explain the cohesiveness, “netting” force of group CMC. Group CMC is a phenomenon which deserves a social-level explanation.
Group CMC is modeled on an extension of interpersonal conversations. As such, group CMC should contain a higher proportion of “agreeing” content than disagreeing content. McLaughlin (1984) summarizes how the conversational system has a built-in preference for accord among the conversants.
The content of group CMC will demonstrate a preference for agreement
We expect interactive messages, within group CMC, to be even more agreeable than the norm for face-to-face messages. And while interactive messages may be even more agreeable than average, they will also be more opinionated, humorous, self-disclosing, and community oriented. We interpret more opinionated and self-disclosing content as indications that groups have gotten beyond preliminary introductions (Duck, 1976). Sherblom (1990) has shown how the use of personal pronouns reflects the degree of involvement in organizations. Thus, we expect interactive messages to contain more first-person plural pronouns.
If agreement is conversational, interactive messages are more conversational, but also more involving more than average
Consider group CMC on the different networks. These differ in at least two ways: the manner in which messages are stored and disseminated, and whether group members pay to use the network. There should be differences in interactivity between voluntary nets and commercial nets. These differences are expected because there is a different implied social contract when one pays to participate. Compuserve special interest groups (SIGs) are managed by sysops who may exercise varying degrees of editorial intervention. And there will be a difference between groups that appear to the reader as a newsgroup (Usenet) and those that appear (to the reader/subscriber/member) as individual messages (Bitnet Listserv groups).
Structural characteristics of networks affect their interactivity
Consider the difference between those who write often and those who make only infrequent contributions. Messages by frequent (active) authors differ in the amount and nature of interactivity of their messagthan infrequent authors. We expect frequent authors to produce more interactive messages.
Interactivity is related to individual activity and communication salience of participants
The data reported here are the result of a content analysis effort. A large international and interdisciplinary group of researchers carried out a content analysis of a representative, random sample of publicly available communication content in computer mediated discussion groups. The project had its inception and entire existence online. A complete description of the methodology is available in Rafaeli, Sudweeks, Konstan and Mabry (1994). Additionally, full copies of the data and message corpora, coding instruments, sampling and ethics policy statements are available online. See also Sudweeks and Rafaeli (1996) for a broader treatment of the group process and ethics issues. In this section we will provide a short account of sampling, coding and reliability practices.
Our goal was to obtain as broad a scope of group CMC as possible. We needed to consider varying units of analysis: the single message, a thread, the group, and the network within which the group resided. The sampling method chosen was of fixed-length message threads, at a given starting date, within randomly selected groups stratified to equal numbers by network type. Messages were selected over a restricted domain, in a single temporal sequence of 100 messages per each group. Equal numbers of groups were randomly selected from Bitnet, Usenet and CompuServe groups. Sampling was completely random, with specific non-discussion groups filtered. Corpora of messages were downloaded from three networks of group CMC: Internet's Usenet, Bitnet Listservs, and Compuserve SIGs, resulting in group numbers as outlined in Table 1.
Table 1. Groups and messages sampled and coded.
Groups in completed sample
Messages in completed sample
The sampling period began on Monday 15 March 1993. Coders shared the task of downloading, using several news servers in locations around the world. Articles were collected according to the date and time of arrival at each news server. Including the post-sampling screening, 77 Bitnet lists, 39 Usenet newsgroups and 23 CompuServe SIGs were selected to get samples of 20 groups for each network. Due to coder attrition, the final database contains complete data for 40 lists (including 10 duplications), and partial data for 4 additional lists, totaling 4322 messages.
A codebook containing 46 closed items was prepared and pretested by all coders. The rewriting of the codebook and pretests were repeated until we achieved average agreement percentages exceeding 90% on all items. The items included in the codebook are listed in the Appendix.
Coders of each list completed a questionnaire to gather descriptive information about the coders, the technology used, impressions of the list, and problems experienced. Data from this questionnaire are not reported here.
Approximately one-third of the lists were double coded to establish reproducibility and reliability of coding. Of the 37 lists (batches of 100 messages) distributed to the research group members, 20 were single coded, 12 were double coded, and 5 were not coded. Of the 32 coded, 4 were unfinished, giving a final tally of 20 single coded and 10 double coded complete lists. The database(s), therefore, has a total of 4322 messages from 30 fully-completed lists, and 4 partially completed lists. Of these, 3322 are unique messages. In addition, there are 1000 doubled-up codes. Coding reliability was furthered ensured through a “blind” enquiry process (see Rafaeli et al., 1994).
The first reportable result was that the project was completed and a usable database was made available. Some evidence for the success of this group is in the measures of the reliability of coding between coders. Intercoder agreement exceeded 75% for the variables used in the following analyses.
Interactivity varied among groups from a high level of 40% in some groups, to absolutely no interactive messages in some groups. On average, just under 10% of the messages were coded as directly referring to how previous messages related to others. We consider these 388 messages to be interactive messages. More than half the sample of messages (52.5%) were coded as referring to a single message that preceded them. These 2269 messages are considered “reactive”. Table 2 compares characteristics of reactive, interactive and all messages in the sample.
Table 2. Messages content and nature: All messages vs. interactive messages (percentages).
All Messages (N=4322)
Reactive messages are those coded as responding to one message.
Interactive messages are those coded as containing references to the manner in which previous messages related to those preceding them.
Significance (Mann Whitney U Test):
a Interactive messages differ significantly (p <0.001) from noninteractive.
b Reactive messages differ significantly from noninteractive but not from interactive.
c Reactive messages differ significafrom interactive AND noninteractive.
d Reactive messages differ significantly from interactive but not from noninteractive.
Contains agreement within list/group (COALIT1)
Contains disagreement within list/group (COALIT1)
Contains agreement with persons/statements external to the list (EXTCOAL)
Contains disagreement with persons/statements external to the list (EXTCOAL)
Primarily provide information (NATURE=1)
Contain a fact (FACT)
Primarily request information (NATURE=2)
Contain a question (QUESTION)
Primarily opinionated (NATURE=3)
Contain an opinion (OPINION)
Contain self-disclosure (FIRSTPER)
Use of first-person plural (COALIT2)
Almost one third of the messages in the corpora quoted other messages verbatim (QUOTE1). Interactive messages were slightly longer than reactive messages. Reactive messages, in turn, were slightly longer than all other messages.
Fully 57% of all messages (first column) contain statements of facts. More than a quarter of the messages contain a question or request. Examined in a different way, when coders classified the overall nature of each message, they found 40% of the messages to be primarily providing information, probably in response to the 14.6% of the messages that were predominantly requests for information. The other half of the message body was classified as “mixed” (22%), opinionated (18.7%), and persuasive (3.6%).
It appears that messages exchanged by participants in group CMC are predominantly factual, conversational, agreeable, and supportive. More than one in every five messages contained at least an attempt at humour. And more than a third of the messages contain personalizing content, in the form of a verbal self-disclosure, an admission or introduction. 9.3% of the entire corpora of messages use first-person plural pronouns. Messages were more likely to contain agreement than disagreement. This finding is repeated, using both measures of the expression of opinion about statements and persons within the list ((16% over 12% agreement to disagreement) and opinions external to the list (9.5% to 5.2%).
Among interactive messages (reading across Table 2), there is more statement of opinion in general, and specifically more expression of agreement. For example, interactive messages contain statements of agreement with persons or statements on the list twice as often as the general sample. Likewise, and even more interestingly, interactive messages are twice as likely as the general sample to contain statements of agreement. Using both the internal-to-the-list and the external measures, we find interactive messages to be even more agreeable than messages overall. These differences are statistically significant. Interactive messages contain more opinions, more self disclosure, and more than twice as much use of first-person plural pronouns.
Reactive messages resemble the general sample (and differ from interactive messages) in all these respects, with two exceptions: Reactive messages are as opinionated as interactive messages, and they tend, like interactive messages, to contain fewer requests for information. Interactive messages do not differ from others in how factual they are.
Table 3 displays the distribution of reactive and interactive messages by the three networks. Listserv mediated messages were significantly less interactive than either Usenet or Compuserve SIG messages. Compuserve SIG messages were least likely to contain no reference to previous messages.
Overall, 1450 different authors contributed messages to the corpora content analyzed here. The most frequent contributor is responsible for 39 messages, however most frequent contributors had about a dozen messages. More than two thirds of the sample of messages were written by authors who appear in the sample only once or twice. Table 4 displays interactivity and reactivity of messages by frequent contributors. Messages by the most frequent contributors (10 or more messages per author), as well as those by frequent contributors (4 to 9 messages per author), are significantly more reactive than the norm. However, messages by frequent authors are not more likely to be interactive.
Table 4. Interactivity and reactivity in messages by frequent contributors (percentages)
Messages by most frequent contributors (>10 msgs) (N=1021)
Messages by frequent contributors (4-9 msgs) (N=691)
Entire sample (N=4322)
Reactive messages (responding to only one preceding message)
Reactive messages (responding to one or more preceding messages)
Significance: * connotes significantly different from rest of sample, using Mann Whitney U test
Communication theory is based on a split in levels of analysis. Individual level motivations and the building blocks for interpersonal relationships reside in the minds of participants. But the actual social actions and relations are transacted through observable behaviors, the exchanges of messages (Watzlawick, Beavin and Jackson, 1967; Palmer, 1994). When we come to the new reality of group CMC there is yet another split. That which is communicated, the messages, are the fruit of an unknown proportion of the participating audience. Only those who actively contribute. There is a silent portion of participants about whom we can only speculate. In a time when much is spoken of virtual reality, there may be a big irony here.
The most “real” part of the social phenomenon of communication is the text exchanged – more real even than the groups, people, and emotions involved. We follow this irony by focusing an empirical lens on the “real” artifacts of a new kind of communication. In a departure from previous perceptions, we believe the groups formed on the net are neither pseudo (Beniger, 1987), nor imagined (if that is what virtual means, e.g. Palmer, 1994). We believe the documented presence of interactivity in the behavior of these groups is both evidence for their reality, and a mechanism for their formation.
We are still far from a theory of interactivity. But these data may bring us slightly closer. The findings reported here do not prove the proposed definition of interactivity, or its role in group CMC, namely that interactivity leads to engagement. A typical lament of communication studies is of the paucity of data. This chapter is no exception to that practice. We have no information about unverbalized reactions, “lurkers”, and only little dynamic, time based information in these data. However, the central claim of this chapter, that interactivity plays a role in creating the attraction of networks and in generating their growth patterns, has some support in the data. The contribution here is an unprecedented, cross-sectional, representative account of group CMC. These data indicate a certain relationship between interactivity and captivating/engaging communication parameters as well as discrimination between varying levels of interactivity.
Although more opinionated, interactive messages have a higher propensity to agree. Interactive messages are significantly more humorous, and more likely to contain self disclosure. Interactive messages are more than twice as likely to contain first-person plural pronouns in reference to members of the list, indicating that interactivity is associated with a sense of involvement and belonging (Table 2). In addition, there are significant structural correlates of interactivity. The networks differ in the interactivity and reactivity of their messages (Table 3).
Table 3. Networks and interactivity level.
messages coded as interactive
Not at all
Yes, 1 message
Yes, > 1
Yes, a sequence
The single unsupported hypothesis is the predicted association between frequency of contributions by the author and interactivity (Table 4). Frequent authors write significantly more reactive messages, but are just as likely as all others to write interactive messages.
These data contain indirect validation of the construct of interactivity itself: (1) Not all networked content is interactive. Interactivity is a variable. Messages, threads and groups can be more or less interactive. (2) These data discriminate well between interactive, reactive and other messages. Messages, networks types, and authorship type relate to interactivity. In other words: interactivity is a richer construct than a plain dichotomy. And, (3) several postulated behavioral correlates of interactivity were shown to covary with the interactivity of messages. In other words: it matters whether messages, threads or groups are interactive.
The underlying hypotheses about interactivity suggest that less interactive uses of the net are not likely to see stable memberships. Individuals may come, but they will not tarry. While less interactive groups may be or even grow, they may be doomed to a rotating-door, shifting existence. In such groups there could be many who stop to visit, but few would be “netted” to stay because the content offerings are reactive at best. Interactive groups are more likely to sustain their memberships, and yield other desired outcomes, such as symmetry in contributions, creativity and productivity, agreement, humor, and sense of belonging. Interactivity is related to longevity too. It can, therefore, turn from just a theoretical construct to practical use as well.
Future work on interactivity can try to address the theoretical as well as the practical aspects. Will the speculations just outlined prove true? Is it possible to increase interactivity through structural interventions? Baseline measures of the content of public, large scale group CMC are now available. Longitudinal follow-ups, in which the comprehensive trend is measured, are obviously indicated. Is there a negative correlation between group size and interactivity? Is there an optimal size of group in the CMC context, analogous to the case in small groups? Does interactivity act the same in synchronous contexts?
ProjectH was made possible by ProjectH members. The full list of our membership with addresses and affiliations is available as part of the ProjectH technical report. ProjectH enjoyed the support of the following: Comserve (vm.its.rpi.edu) sponsored the project and unwittingly endowed it with a name. The Key Centre of Design Computing, University of Sydney, Compuserve, and The Zaggagi Fund provided funds, computing resources and coordinating time.
Brief description and comments
Number of lines?
Is it appropriate?
Does message contain an opinion?
Does message contain a fact?
Does message contain an apology?
Does message contain a question/request?
Does message contain a call for action?
Does message contain a Challenge/dare?
Does message contain attempts at humour?
Does message contain metacommunication?
Is the message formatted appropriate?
Appropriate/excessive use capitalization?
Is there colloquial spelling?
Overall rhetorical style
Icon for emotion?
Device for emotion?
Art, other than emotion
From this list?
From other CMC?
Reference to previous message or messages?
Message number referenced?
Is there reference to how previous messages related to even earlier msgs?
Introduce new topic?
Does message contain agreement/disagreement with persons or ideas ON the list?
Does message contain use of first-person plural pronouns (“us”, “we”) about the group?
Directly address any persons on the list?
Does message contain agreement/disagreement with persons or ideas OFF the list?