The Influence of Synchrony and Sensory Modality on the Person Perception Process in Computer-Mediated Groups
Kristine L. Nowak,
(Ph.D., Michigan State University, 2000) is an Assistant Professor in the Communication Science department, and director of the human computer interaction lab, at the University of Connecticut. Her research focuses on the person perception process and user satisfaction in computer-mediated interactions. She is also interested in design and usability issues involving computer media. See http://www.coms.uconn.edu/hcilab/ for more information.
(Ph.D., University of Wisconsin) is Director of the Rensselaer Social and Behavioral Research Laboratory and Chair of the Department of Language, Literature, and Communication at Rensselaer Polytechnic Institute. His research interests include online marketing communication and distance collaboration technologies.
(Ph.D., University of Arizona) is a professor of communication at Cornell University. His research focuses on the use of communication cues in the management of relationships and their effects, with special emphasis on computer-mediated communication in personal, social, and collaborative work settings. http://www.cmcresearch.org/jw/
Address: Department of Communication Sciences, University of Connecticut, 850 Bolton Road, U-1085, Storrs, CT 06269 USA
Address: James Watt, LL&C, Rensselaer Polytechnic Institute, 110 8th Street, Troy, NY 12180 USA
Address: Dept. of Communication, Cornell University, 336 Kennedy Hall, Ithaca, NY 14853-4203 USA
This study examined the effects of synchrony and the number of cues on the person perception process in computer-mediated communication. One hundred and forty-two participants in groups of three or four engaged in collaboration over five weeks to develop oral reports, using alternate versions of communication systems or meeting face-to-face. Consistent with the hyperpersonal model, those using low cue media felt their partners were more credible, and reported more social attraction, less uncertainty, and more involvement in the interaction than those using high cue media. People interacting with synchronous media felt increased social attraction, self-reported involvement, and certainty. They also felt that their conversations were more effective, although this effect appeared mainly in low cue groups. Results of an exploratory path analysis suggest that future research should focus on causal chains rather than direct effects, and that intervening variables (such as involvement) may be central to our understanding of the effects of communication technology systems.
Several theories of computer-mediated communication have argued that computer media that are able to engage more senses facilitate more satisfying interactions. These technologically-deterministic theoretical perspectives include social presence theory, media richness, and the social context cues hypothesis (Daft & Lengel, 1986; Kiesler, Siegel, & McGuire, 1984; Short, Williams, & Christie, 1976). However, this latter position, which states that lean media systems are inherently not well suited for social interactions, has been widely challenged (for a discussion see Walther, 1996).1 The most persuasive evidence that cue lean media can be used for social interactions comes from actual media use. Cue lean media, including text-based systems, have been used to fulfill a variety of interpersonal goals including forming and maintaining meaningful interpersonal relationships (Becker & Mark, 2002; O'Sullivan, 2000; Parks & Floyd, 1996). The Pew Internet and American Life Project (Madden, 2004) recently reported that email is the communication tool of choice and that it enhances people's sense of connection to key family and friends.
Media are used for a variety of purposes, and people adapt their communication behaviors to make the most of whatever medium they are using during an interaction (Postmes, Spears, & Lea, 2000; Walther, 1996). People are able to utilize almost any media system to fulfill interaction goals, and few differences are found in performance outcomes between face-to-face and mediated groups (van den Berg & Watt, 1991), although users generally report a preference for multimedia and full cue interfaces (e.g., Gale, 1991; Rice, 1993; Tang & Isaacs, 1993). However, this does not mean that the types of cues or the medium itself do not impact the person perception process. It may be that cue lean media (such as text systems) provide functional connections to other humans that are less satisfactory than interfaces with more sensory channels (such as videoconferencing). This apparent contradiction would support the efficiency framework, which seeks to explain the ironic relationship among media preferences and media-enabled performances (Walther, in press; Watt, Walther, & Nowak, 2002). The question examined here is whether using different systems during group interactions influences the person perception process.
Walther (1992, 1996) argued that in some situations, relationships mediated by cue lean media could be more engaging or “hyperpersonal” when compared to parallel face-to-face relationships. He further argued that people are better able to control the information others gain about them in cue lean media, and that people are more likely to believe that others are like them in the absence of physical, disconfirming information. Therefore, in cue lean media where the medium of communication does not provide nonverbal information, it is more likely that people's expectations of others will be met, leading people to like their interaction partners more.
However, in the literature on the effects of computer-mediated communication, relatively few studies have examined the effects of synchrony, which may influence interaction as much as the presence or absence of cues. Walther (1996) speculated that synchrony influences the hyperpersonal process through a possible interaction effect of synchrony and cues, but this relationship has not been tested. With that in mind, this study examined the influence of number of cues and synchrony on the process of person perception during the completion of a five-week task.
Media Affordances: Synchrony and Geography
An essential element required for fulfilling communication goals is the ability of communicators to coordinate content and realize some level of common ground (Clark & Brennan, 1991). It is important to examine how features of various media could be utilized to fulfill, if not augment, people's ability to realize this common ground and fulfill their interaction goals.
When a medium lacks one feature, people modify their communication strategies and use alternative techniques to reach common ground (Clark & Brennan, 1991). The case of asynchronous systems provides an example. Some researchers have argued that delayed feedback dampens communicators' ability to understand complex ideas and interact successfully (Daft, Lengel, & Trevino, 1987). Asynchronous systems are also suspected to cause difficulty with coherence and turn-taking (Clark & Brennan, 1991; McGrath, 1990). However, evidence suggests that users have adapted strategies to facilitate coherent interactions using asynchronous systems (Cornelius & Boos, 2003; Herring, 1999). This section will consider the ability of mediated systems to facilitate communicators' ability to reach common ground by examining some of the affordances mediated systems can provide that face-to-face communication cannot. These affordances include allowing the communicator to revise the message before it is sent, facilitating sorting and archiving the content of the interaction, and releasing users from geographic and temporal constraints.
People have displayed an ability to utilize the structural features of media to facilitate conversational coherence among temporally-independent messages. A variety of media provide features designed to allow users not only to recreate flow and structure feedback in mediated interactions (see Rapport, 1991; Turoff, 1991), but also to structure their expressions mindful of the potential separations in flow. In asynchronous systems, users are free to review, revise, and edit their message privately prior to sending it, allowing them to ensure that the phrasing and other message characteristics are consistent with the message they desire to send, without “costs” associated with delay that would be present in a synchronous interaction (Clark & Brennan, 1991). Further, asynchronous systems facilitate the user's ability to store and retrieve the content of conversations as archives, facilitating what Clark and Brennan (1991) call reviewability. These archives can be useful conversational resources for those who join an interaction late, or for those who wish to sort, organize, or re-read previous statements (Whittaker, 2003). These features allow users to take advantage of the benefits of temporal independence and help to prevent the interaction from becoming disorganized and indecipherable.
In terms of geographic constraints, when group members do not have to be in the same place in order to communicate, organizations can disperse their members geographically, even across multiple time zones (Johansen & O'Hara-Devereaux, 1994). Similarly, some asynchronous media systems liberate their users from temporal constraints by allowing groups to overcome the problems associated with competing demands for attention and time that plague synchronous meetings, allowing participants to write and respond when it is most convenient for them (Hesse, Werner, & Altman, 1988; Walther, 1996). These systems also facilitate savings in travel time and costs, and allow members to remain embedded in important sites and involved in multiple projects simultaneously.
Based on the above rationale, asynchronous communication can facilitate groups' coordination when it has certain features and they are used properly. What is less clear is the effect that asynchronous videoconferencing may have on person perception. If the issue of predicting satisfaction with an interaction were truly as simple as a count of the number of cues, videoconferencing would be similar to face-to-face interactions. On the other hand, any mediation may be significant, and as Hollan and Stornetta (1992) argued, the closer media come to replicating face-to-face, the higher people's expectations become, and the more clearly we may see that mediated interactions of any kind will never be “as good” as face-to-face.
This study examines the influence of both asynchrony and number of cues on person perception in a task-based interaction. This issue has been addressed to some extent in other studies. For example, synchronous versus asynchronous text communication has been contrasted (Arbaugh, 2000), and asynchronous text collaboration (e.g., threaded discussions or computer conferences) has been compared to face-to-face meetings (see for review Olson, Teasley, Covi, & Olson, 2002; Warkentin, Sayeed, & Hightower, 1997). However, the interaction between synchrony and number of cues has not been addressed experimentally because most investigations of asynchronous communication have been limited to text (e.g., Goodyear, 1996; Weisband, 2002; Whittaker, Nardi, & Bradner, 2000), although a small amount of work with voice mail has been conducted (e.g., Danowski & Rice, 1989; Trevino & Webster, 1992). It is not clear whether the influence of text systems would generalize to multiple-cue (audio/visual) interaction settings.
Some examinations have confounded attributes of the technologies. For example, comparisons of asynchronous text conferences with synchronous face-to-face meetings confound the mode of delivery (asynchronous vs. synchronous) with the fidelity and multiplicity of a full set of communication cues (cf. Storck & Sproull, 1995; Warkentin et al., 1997). “Video e-mail” has also been investigated in at least one study (Hopper, 1994) in an attempt to fill in the “missing cell” of a synchronous medium with multiple cues (see Watt et al., 2002), but this is essentially point-to-point communication rather than the type of collaboration that is considered here. Although as Walther (1996) points out, “[a]ny asynchronous interaction involves some mediation, some reduction in cues” (p. 31), the asynchronous audio video condition used in the present study may serve to fill the “empty cell” in order to allow a fuller investigation of these issues.
The Impact of Synchrony and Cues on Person Perception in Group Work
The above discussion raises a number of questions about the effects of media characteristics and the extent to which people's preferences for cue rich media influence their perceptions of one another. It is possible, as posited by the efficiency framework, that even though people are able to adapt their communication behaviors to meet the features of cue lean media, they maintain a preference for cue rich media. This use of adapted behaviors may require the communicators to expend more effort, for example spending more time typing than speaking (Clark & Brennan, 1991). However, this does not necessarily mean that there will be a negative influence on the communicators' ability to reach common ground, nor would it necessarily negatively affect the person perception process. The need to adapt their communication process and the greater effort users expend in doing so may lead to the hyperpersonal experience described by Walther (1996). This section considers how number of cues and synchrony may influence the person perception process.
Uncertainty reduction theory (Berger & Calabrese, 1975) argues that a person's primary goal in an initial interaction is to reduce uncertainty about others. Uncertainty is reduced not by a sum total of information but by the quality of the information. How do people perceive the quality of the information they receive about their partners in different modalities? The efficiency framework argues that, even though people are able to fulfill interaction goals in asynchronous cue lean media, people like and perhaps trust these systems less. This preference could potentially cause people to feel less confidence in their ability to get to know their partners over these systems (Watt et al., 2002). The hyperpersonal model predicts that people like their partners more and perceive them to be similar to themselves because asynchronous cue lean media provide less information that could disconfirm that prediction, leading to more extreme attributions (see Hancock & Dunham, 2001). These contrasting possibilities lead us to ask:
RQ1: What is the effect of number of cues on uncertainty?
RQ2: What is the effect of synchrony on uncertainty?
Social attraction is an important part of person perception (McCroskey & McCain, 1974). Research has shown that people are more attracted to others when uncertainty has been reduced (Clatterbuck, 1979; Infante, Rancer, & Womack, 1997). It is possible that people will like their interaction partners less when they interact using a cue lean interface, as could occur if the predictions of the efficiency framework affect the person perception process. If increased uncertainty leads to less liking, then asynchronous cue lean media would reduce social attraction. On the other hand, the hyperpersonal model predicted that people would like their partners more following cue lean interactions. Neither theory makes direct predictions about synchrony, but it is likely that synchrony will influence social attraction in the same direction. Therefore, we ask:
RQ3: What is the effect of number of cues on social attraction?
RQ4: What is the effect of synchrony on social attraction?
The present study also examines how involved in the interaction people perceive their partners to be, which is an important part of people's perception of the quality of their interaction (Burgoon & Hale, 1987). People's own involvement is likely to affect how involved they feel their partners are as well. The hyperpersonal model argues that people will need to be more involved in the interaction when they adapt their communication behaviors to meet the features of the medium than when they are able to use more intuitive methods that come with richer media, or face-to-face interactions. The efficiency framework does not make a prediction about involvement, but the predictions of the hyperpersonal model requiring increased involvement in interactions using lean media are consistent with this position. It is likely that people's perceptions of their teammates' involvement will be highly correlated with their self-reported involvement. Neither model makes direct predictions about synchrony, although the hyperpersonal model mentions that it is likely to be influential. Therefore, the following questions are asked:
RQ5: What is the effect of number of cues on people's self-reported involvement?
RQ6: What is the effect of synchrony on self-reported involvement?
RQ7: What is the effect of number of cues on people's perceptions of their partners' involvement?
RQ8: What is the effect of synchrony on perceived involvement?
Another important consideration is the extent to which people are perceived to be credible. Credibility refers to the judgments made by the perceiver about the knowledge of their partner (McCroskey, 1971; McCroskey, Hamilton, & Weiner, 1974). The lack of social context cues hypothesis predicts that leaner media induce people to focus more on the task. Whether this enhanced task focus leads to enhanced perceptions of credibility, or amorphous partner perceptions, is unclear. However, the increased involvement predicted by the hyperpersonal model should lead to increased perceptions of credibility in a context where credibility is desirable. It is also possible that people who respond immediately are likely to be considered more credible. This is one of synchronous media's affordances that asynchronous media do not allow. Although people are likely to realize on a conscious level that immediate response is not possible in asynchronous channels, this still may influence perceptions of credibility. Alternatively, the efficiency framework predicts a reduction in social attraction following interactions in asynchronous or cue lean media. If this is the case, then it is likely that people will perceive their partners to be less credible following interactions in either asynchronous or cue lean media. Therefore, the following questions are tendered:
RQ9: What is the effect of number of cues on credibility following a task-based interaction?
RQ10: What is the effect of synchrony on credibility following a task-based interaction?
The efficiency framework would argue that even though cue lean media facilitate people's ability to fulfill their goals, they would feel that the conversation was less effective. In contrast, the hyperpersonal model's prediction that people will like their partners more and be more involved in the interaction suggests that these factors may increase people's perception of conversational effectiveness. The same rationale would possibly apply to asynchronous media.
RQ11: What is the effect of number of cues on conversational effectiveness?
RQ12: What is the effect of synchrony on conversational effectiveness?
Participants were 142 students enrolled in a communication course at a large public university in the eastern United States. The students were randomly assigned to 39 groups of three or four members each. These groups, in turn, were assigned to collaborate using one type of communication medium to complete a class project.
All participants did the same task, which was part of the course and worth 20% of each student's overall grade. Students were asked to research and prepare a 12-15 minute oral report, as if it were to be presented to the United States Congress, arguing how to balance privacy and national security. The groups met at assigned times once a week for five weeks and were instructed to discuss the issues and to prepare a final, smoothly flowing oral presentation. The final oral report required each group member to present a portion of the arguments, thus preventing “social loafing.” In addition to a course grade for the project, the group that gave the best presentation received a $100 prize. Students in all groups were asked not to discuss the project with their partners outside their assigned medium. The final videotaped oral presentations were evaluated by outside raters.
Asynchrony and Synchrony
All participants were assigned a time and day of the week to come to a computer lab to participate in this project. Each individual came at the same time and day each week for five weeks. With synchronous groups, all group members were participating at the same time (although they may have been in different locations). Each member of the asynchronous groups was assigned distinct days and times of the week for the duration of the project. No two members of the same asynchronous group participated either at the same time, or during adjacent times. An asynchronous participant would download and review his or her teammates' messages and respond as desired, and that would end the session. The participant would return the following week at the assigned time to repeat this procedure.
To disentangle the impact of cues and synchrony on group performance, five different group collaboration conditions were used. Seven face-to-face groups (multiple-cue, synchronous) were used as a reference to four factorial combinations of time mode and cue multiplicity as summarized in Table 1.
Table 1. The four factorial combinations of time mode and cue multiplicity
High Cue Multiplicity
Low Cue Multiplicity
Synch Video conference(CUSeeMe) 8 groups
Synchronous text(WebBoard Chat) 8 groups
Asynchronous video(TIC System) 8 groups
Threaded text discussion(WebBoard Conference) 8 groups
The asynchronous audio video combination was created to fill in the “missing cell” of the full crossing of time mode with cue richness (see Watt et al., 2002). Eight groups completed their task by using an asynchronous audio-visual group collaboration system known as the Time Independent Collaboration system (TIC), development of which is detailed in Watt et al. (2002). The TIC system allowed users to record messages by using a computer equipped with a microphone and Webcam. Messages were stored in a server database, and presented to users via an interface that allowed them to see the date of message creation, author, and a text subject identification.
Upon completion of the group project, self-administered paper and pencil measures were used to operationalize the variables described below. The full set of scale items and whether individual or group means were used are shown in Appendix A.
Uncertainty reduction was measured using seven Likert-type items on a 7-point metric from Clatterbuck's (1979) Attributional Confidence Scale, including how comfortable the participants felt about their ability to predict other group members' values, attitudes, feelings, and emotions. These items achieved a unidimensional inter-item Cronbach alpha reliability of .94.
Social attraction was measured using eight Likert-type items on a 7-point metric, based on McCroskey and McCain's (1974) scale. The items measured the extent to which participants felt their partner was pleasant or offensive and whether or not the participant desired a future interaction. These items achieved a unidimensional inter-item alpha of .91.
Self-reported involvement was measured using three Likert-type items on a 7-point metric. These items were revised from indicators of involvement (Burgoon & Hale, 1987) to ask the participants to report their level of involvement in the interaction. These items achieved a unidimensional inter-item alpha of .71.
Perceived Partner Involvement
For perceived partner involvement, all questions were asked about each group member (meaning each person responded to these items for all of his or her group members). Ten Likert-type items on a 7-point metric were selected from a combination of the indicators for involvement and immediacy (Burgoon & Hale, 1987) after conducting tests of internal consistency and reliability. These items achieved a unidimensional inter-item alpha of .91.
Credibility was measured using seven semantic differential items from McCroskey et al. (1974). These items achieved a unidimensional inter-item alpha of .91.
Conversational effectiveness was measured using nine Likert-type items on a 7-point metric from Canary and Spitzberg (1987). These items achieved a unidimensional inter-item alpha reliability of .83.
The research questions were tested with 2-way ANOVAs, examining both main effects of number of cues and of synchrony, and their interaction. The analyses were first done with only the mediated groups, excluding the face-to-face groups. The analyses were then repeated with the face-to-face groups included in the synchronous, high cue cell. The results of the latter analysis added to the power of the tests, but the results did not meaningfully differ from those obtained by examining only the four conditions that were mediated. Even the cell means differed by very small amounts. The results reported below exclude the face-to-face groups and include only the four mediated conditions.
RQ1 and RQ2: What are the effects of number of cues and synchrony on uncertainty?
There was a significant main effect for number of cues, F(1,116)=5.84, p<.02, partial eta sq.=.05, and for synchrony, F(1,116)=4.09, p<.05, partial eta sq.=.03, on uncertainty. However, these main effects must be qualified, as there was a significant interaction between cues and synchronicity, F(1,116)=15.32, p<.001, partial eta sq.=.12.
Those using asynchronous low cue media (M=26.24) felt more certain than those using asynchronous high cue technology (M=17.27). This difference was significant (p<.001) by post hoc Scheffe test. The difference in certainty between low and high cue medium users of synchronous technology was much smaller, with those using low cue media (M=23.56) reporting less certainty than those using high cue media, (M=25.81). A Scheffe test showed that this difference was not significant.
RQ3 and RQ4: What are the effects of number of cues and synchrony on social attraction?
There was a significant effect of the number of cues on social attraction, F(1,116)=7.54, p<.01, partial eta sq.=.06. High cue media (M=35.71) provided significantly less social attraction than low cue media (M=39.80). There was also a significant effect of synchrony on social attraction. Synchronous media (M=39.43) provided significantly higher social attraction than asynchronous media (M=36.08) F(1,116)=5.04, p<.03, partial eta sq.=.04. There was no significant interaction between synchronicity and cues.
RQ5 and RQ6: What are the effects of number of cues and synchrony on people's self- reported involvement?
The effect of number of cues on self-reported involvement was not significant F<1 (M=13.53 for high cue and M=13.86 for low cue media). However, the effect of synchronicity on self-reported involvement was significant, F(1,116)=13.30, p<.01, partial eta sq.=.10. Those using synchronous media reported higher self-involvement (M=14.69) than those using asynchronous media (M=12.70). There was no significant interaction effect on self-reported involvement, F<1.
RQ7 and RQ8: What are the effects of number of cues and synchrony on people's perceptions of their partners' involvement?
There was a significant effect of number of cues on perceived involvement F(1,116)=9.48, p=.003, partial eta sq.=.08. Those using high cue media perceived less involvement (M=51.22) than those using low cue media (M=55.47). There was no significant effect of synchrony on perceived involvement F(1,116)=3.16, p=.08, between those using synchronous (M=54.57) and asynchronous media (M=52.12). There was no significant interaction of cues and synchrony for perceived involvement, F<1.
RQ9 and RQ10: What are the effects of number of cues and synchrony on credibility following a task-based interaction?
There was a significant effect of number of cues on credibility, F(1,116)=8.77, p<.005, partial eta sq.=.07. Those using a lean cue medium rated their partners as more credible (M=40.49) than those in a rich cue medium (M=37.38).
There was no significant effect of synchrony on credibility, asynchronous media M=38.73, and synchronous media M=39.13, F<1. There was no significant interaction between credibility and synchrony F<1.
RQ11 and RQ12: What are the effects of number of cues and synchrony on conversational effectiveness?
There was no significant difference on conversational effectiveness between those using high cue and those using low cue media F(1,109)=3.93, p>.05. Synchrony had a significant effect on conversational effectiveness F (1,109)=11.50, p=.001, partial eta sq.=.10, However, there was a significant interaction between cues and synchrony F(1,109)=9.37, p<.01, partial eta sq.=.08. Subjects using synchronous low cue media rated conversational effectiveness significantly higher (p<.05 by post hoc Scheffe test, M=53.17) than did subjects in all other experimental groups (asynchronous low cue M=43.82; synchronous high cue M=45.86; asynchronous high cue M=45.38), among which the means did not differ.
The present design and results suggest several implications for conventional theories of CMC, and suggest the importance of examining the relationship between communication technology and behavior in terms of its impacts on fundamental processes, rather than examining simple direct effects.
As was predicted by the hyperpersonal model, those in low cue media felt the best about their partners. They felt that their partners were more credible, felt more social attraction for them, felt less uncertainty, and perceived their partners as more involved in the interactions. Further, in a test of synchrony, the results showed little support for the simple crossing of effects that the TIC system enabled: asynchrony plus more visual cues. People interacting with synchronous media felt increased social attraction and self-reported involvement, though they felt more uncertainty. There were no significant influences of synchrony on perceived involvement or credibility. This suggests that the reported preference for “rich” media (multiple cue, synchronous interaction) does not affect the person perception process.
It should be recognized that the conditions under which asynchronous communication took place were not typical of asynchronous communication technologies in common use. Subjects could not log on whenever or wherever they wished. One might have expected that this removal of some of the primary advantages of asynchronous communication would disadvantage this mode over synchronous communication, and thus bias the results toward the prediction that synchronous mode subjects would find their partners more credible. Given that this prediction was not supported even under these special conditions, there is little reason to believe this would be true under “real” asynchronous interactions. However, this should still be further explored in future research. Whereas the research questions do not predict the direction of effect of asynchrony on involvement, it is unlikely that the testing situation decreased the effect of asynchrony from having a positive impact on involvement to the null difference found. One would expect the opposite.
In terms of the influence of the high/low cue dichotomy there were several significant differences, which have important implications for our understanding of the influence of the medium on the person perception process. Consistent with predictions based on the hyperpersonal model, those using low cue media (text only) felt more certain, perceived more social attraction and more involvement from their partners, and felt their partners were more credible than those in high cue interactions.
Although neither the hyperpersonal nor the efficiency model made directional pr edictions about the effect of synchrony, both suggest that this would be an impo rtant variable, and it was. People interacting with synchronous media felt more certain, more social attraction, conversational involvement and self-reported in volvement. These results indicate that being able to respond in real time has a positive influence on person perception and makes people feel more involved in the interaction.
The significant interaction effects on uncertainty reduction and conversational effectiveness showed that those using asynchronous low cue media felt significan tly more certainty and conversational effectiveness than those using asynchronou s high cue media. It is possible that these interactions are at least partially explained by a novelty effect. People have limited experience using high cue asy nchronous media, compared to their experience using cue lean asynchronous media (such as email). The increased uncertainty and perceived conversational effectiv eness may have been due to the novelty of the system. However, the fact that thi s project took place over five weeks should have diminished a novelty effect. An other possible explanation is that high cue media introduce ambiguous cues, whic h are not as readily available in low cue media. Also, synchrony may introduce s ome ambiguity by temporal context, which may have reduced the difference between low and high cue media. Future research should further examine these possibilit ies.
In the current experiment, visual information in videoconferencing presented participants' faces as they worked on a task. Questions about the focus of visual cues have been raised in other streams of research. For instance, Kraut, Fussell, and Siegel (2003) examined the relative advantage of video cues depicting objects that are the focus of distributed collaboration over video cues depicting the participants themselves. Such research has found greater efficiency in language use and better productivity when object-focused video cues accompany discussions about the manipulation of such objects. While a simple conclusion might be that the present treatment was not successful because the focus of video was misdirected, this might be overly simplistic. Recent thinking about different types of conversational foci-object-oriented versus person- or attitude-oriented-suggests that there may be a time and place for sharing images of things or images of people, depending on which set of cues complements the attentional and conversational needs of the discussion (Walther, 2003). When working on a project or preparing a presentation, being able to see referential objects or an outline may be more beneficial than seeing each other (Whittaker, 2003). This potential benefit was not tested in this experiment, although the videotaped interactions showed students trying to show one another images and papers in the TIC system. Future research should examine these possibilities.
The findings also have implications for media richness theory (Daft et al., 1987), a popular if not time-tested approach to mediated task communication. Media richness predicts that full cue and synchronous media (in addition to media capable of individual personalization and natural language) are superior for ambiguous communication tasks. Lean media such as text-based, asynchronous CMC are preferable for simple tasks. One of the theory's weaknesses, however, is its binary classification of sense engagement and its consequent inability to specify the fit of media that may saturate one cue but barely engage another. Although it is not certain whether the present task was equivocal or simple in media richness terms, no previous theorizing would suggest that the combination of minimal-cue but synchronous media should have such positive implications for the person perception process. The implications of these results as further impetus for the reconceptualization of media richness should be considered.
As with any experiment, there are factors that limit the generalizability of these results. For example, the design required people to work on this project at assigned times. As discussed above, this eliminated a potential advantage of asynchronous media that would have come with being able to use the medium whenever and wherever participants felt like discussing the project.
Also, to examine the potential unique experience of face-to-face interactions (as compared to any mediated condition), the ANOVA analyses were run twice. In the second analyses, face-to-face groups were included in the high cue-synchronous experimental condition. It might be predicted that these unmediated groups would be fundamentally different from mediated groups. However, these analyses, which had both synchronous audio/visual mediation and face-to-face meeting in the high cue-synchronous cell, produced identical conclusions to the analyses reported above. These results provide no support for the presumption of a fundamental difference between face-to-face and mediated synchronous high cue communication.
Likewise, there was some concern about the synchronous audio/visual groups. Their conversations were realistic, but disrupted due to network congestion. Data were examined for the possibility that this might have caused more negative perceptions, and that this might be producing the negative relationship found between high cues and involvement. Again, whether or not these groups were included in analyses did not influence the results. Essentially, it does not appear that either the “gold-standard” of face-to-face communication or the perceived technological woes of video conferencing affected the perception process.
Other uncontrolled factors in this design may actually increase the external validity of the results. First, in this design, students enrolled in the same courses where they saw each other several times a week, so their interactions were not limited to their assigned media. As with most real-world group projects, people may have met one another before being assigned to the project together. They also may have disregarded instructions and met and/or interacted via other media over the period of weeks required to complete the project. Even if this was the case, the assigned media still had an impact on person perception.
Along with reliance on a homogeneous student sample, a serious limitation of this research is the use of a single open-ended, academically-oriented task for all the groups. While this control improves the internal validity of the study, it is possible that the results reported here are specific to this particular type of task. Future research should replicate this design with differing populations, group tasks, and goals.
Exploratory Structural Analysis of Results
The ANOVA results discussed above imply a direct link between synchrony and cues and the six outcome variables. To test this implicit assumption, a structural analysis using the AMOS package was conducted. In this analysis, synchrony and cues were the independent variables, which were tested against the six separate dependent outcome variables. As Figure 1 shows, the overall chi-square for this structure was large and significant, indicating that this structure of independent effects does not fit the data. There are clearly interdependencies among the outcome variables that are not captured by multiple independent ANOVAs, and that require explanation.
To investigate these, the observed correlations table was examined. The pattern of correlations showed fairly strong relationships of the involvement variables with other perception variables. The other perception variables were only moderately or weakly associated with the synchrony and cues variables, while self-involvement was strongly associated with synchrony. Although not perfect, this pattern of correlations suggested a model in which the involvement variables mediated the effect of the technology mode variables (synchrony and cues) on the other perception variables.
This model was investigated, and while it did not fit the observed data sufficiently well to be accepted, the overall goodness-of-fit measures were much better than those produced by the original model shown in Figure 1. Comparison of the implied correlations computed from this intermediate model (not shown) with the observed correlations indicated that several covariance paths between endogenous variables were necessary to account for larger observed correlations between these variables. The resulting model is shown in Figure 2.
The overall goodness-of-fit of this model, as measured by the Root Mean Square Error of Approximation (RMSEA), is .07, midway between the desirable value of .05 or less and the unacceptable value of .10 or greater (see Arbuckle & Wothke, 1999). The overall model chi-square is 28.7 with 17 d.f., which differs significantly from the original data at p=.04. This is very close to the desirable significance level of .05 or greater. All structural coefficients in the diagram are significant at p<.01 with the exception of the relationship between self-reported involvement and social attraction, p=.056. This path was retained because it improved the overall goodness of fit of the model. On balance, the structure of the final exploratory model appears to be a reasonably good fit with the observed data (see Table 2).
Table 2. Observed correlations compared to implied correlations from final model
High cue medium
Involvement with partners
Involvement - Self report
Social attraction to others
Credibility of others
High cue medium
Involvement with partners
Involvement - Self report
Social attraction to others
Credibility of others
Since this model was developed by empirical observation rather than theoretical deduction, it is possible that its structure is partially due to improbable chance covariation, and thus will not replicate in future tests. An independent replication of this study with new subjects would give the definitive test that might exclude this possibility, and should be explored in future research. However, a partial test of replication can be made with a split-sample test. The original data are randomly separated into two subsamples, each with N/2 observations. These subsamples are used to test the structure of the model. If the structure is a good representation of both the subsamples, it is less likely that the model structure capitalized on extreme (and thus improbable) chance variation, as this variation would likely be attenuated in at least one of the subsets.
As Figures 3 and 4 show, the model structure fit both random subsets of the data better than the full sample. Because each subsample had N=71 versus N=142 for the full sample, some of the smaller structural coefficients that were significant in the full model did not reach significance. However, the coefficients in all three models are similar. The results of the subset analyses do not indicate that the structure of the exploratory empirical model was the result of incorporating chance variation.
These exploratory results indicate that both self-reported involvement and the perception of the involvement of the other group members mediate the effects of synchrony and cues on person perception. In other words, technological factors are only indirectly related to perceptions of the credibility of other participants, conversational effectiveness, uncertainty reduction, and social attractiveness of other group members. Instead, synchrony and cues directly affect the experience and perception of involvement, which then subsequently affect the other perceptions.2
In this model, synchronous media produce more positive perceptions of self-involvement in the communication, and this in turn leads to a much higher perception of conversational effectiveness, but also to reduced social attraction. As predicted by the hyperpersonal model, low cue media lead to more positive perceived involvement with others in the group (as noted in the ANOVA analyses in the Results section), and this involvement leads to increases in all other perception variables, particularly social attraction (beta=.74) and credibility of others (beta=.64).
Beyond the theoretical contribution of this research, a major issue raised by the results has to do with how best to approach the study of new media effects. Our analyses show connections between technology variations and person perception, but most importantly they show connections between those moderated by technology's effects on technology-independent processes, such as involvement, which in turn affected distal outcomes. Traditional technologically deterministic “cues-filtered-out” hypotheses of CMC such as social presence theory (Short et al., 1976) have posited rather monolithic relationships between the amount of cues and the quality of social outcomes. These positions have garnered inconsistent support and much refutation over the years (see Walther & Parks, 2002). The present results suggest that conversational processes provide the missing link in technology-behavior relations, refocusing theoretical inquiry on the communication processes affected by artifacts, rather than on their assumed characteristics.
Conversational processes appear to be essential to understanding these results. That is, it was conversational involvement variables (self-reported and perceived) that directly predicted social outcomes in the structural model. Variations in media characteristics did produce outcome differences, but involvement intervened. While it is not entirely clear at this point how media attributes prompt the involvement/effectiveness/attraction chain, it is not difficult to imagine how intervening processes might take center stage in the next steps of research on video and other forms of conferencing.
The results from this study suggest that researchers should focus on causal chains rather than direct effects, and that intervening variables from non-technical social processes may offer greater potency in revealing the ultimate and indirect effects of communication technology systems. Such suggestions have arisen from the empirical results of other CMC research. For instance, Walther (1994) demonstrated that it was the degree to which participants in CMC and face-to-face groups anticipated future interaction that predicted their relational communication, not the medium in which they operated. CMC did affect the degree to which interactants anticipated future interaction, but it was the latter condition, not the medium, that accounted for most or all of the variance on a number of relational dimensions. In another instance, recent research has found that anonymity and face concerns, which CMC may instill but which also can be varied in CMC, account for interpersonal language variations such as sarcasm and irony (Hancock, 2005).
Aspects of CMC certainly raise questions about the impact of technology on human communication. While recommendations to focus on the more abstract properties or constructs that can be linked to such system variations are not altogether unwise (see Eveland, 2003), it appears that understanding CMC requires deeper understanding of the basic interpersonal and conversational processes that may be triggered by CMC variations, rather than simply technologically deterministic predictions. Media configurations and the social arrangements of their use prompt different interpersonal processes, and while media attributes may or may not be unique and interesting, the social processes they evoke, at least thus far, appear to be more robust predictors of what happens online or off.
Overall, both synchrony and level of cues appear to be important influences on people's perceptions of one another, but examining main effects alone may not provide the most interesting information about the person perception process. These results are consistent with the hyperpersonal model (Walther, 1996), specifically that users of low bandwidth systems believe that others are more “like them” than users of high bandwidth systems with more indicators (e.g., voice, pictures) (Walther, Slovacek, & Tidwell, 2001). This is important for both media users and designers to understand as they continue to utilize such systems for group interactions.
It was suggested that this position was “a dead horse no longer to be beaten” over ten years ago (Walther, 1994, p. 476).
Significant correlations among several endogenous variables in the model were discovered during exploratory model building. These covariances, shown in the figures as correlated error terms, represent common variance among variables. For example, the self-reported involvement and perceived involvement residuals were correlated at .27 (p<.01). Likewise, the correlations between conversational effectiveness and perceptions of the credibility of other group members (r=.22, p<.01) and between uncertainty reduction and social attraction (r=.33, p<.01) were significant. These might represent two kinds of group cohesion outcomes, and are certainly fodder for further research.
Appendix A. Items for Collaboration Outcome Measures
Uncertainty Reduction (Scale: 1=unable to answer; 7=completely confident). The mean for individual evaluations of all other group members was used in these analyses.
Ability to predict behavior
Certain that he/she likes you
Ability to predict his/her values
Ability to predict his/her attitudes
Predict his/her feelings and emotions
Empathize with the way he/she feels about himself/herself
How well do you know him/her
Social Attraction (Scale: 1=strongly disagree; 7=strongly agree). The mean for individual evaluations of all other group members was used in these analyses.
Could be a friend of mine
Would like to have a friendly chat with user
Reverse: we could never establish friendly relationship with one another
Reverse: would never fit into my circle of friends