Communication and Commitment in an Online ... - Semantic Scholar

23 downloads 34440 Views 817KB Size Report
CHI'12, May 5–10, 2012, Austin, Texas, USA. Copyright 2012 ..... commitment, communication, and group development, we expected ..... software development.
Communication and Commitment in an Online Game Team 1

Laura Dabbish12, Robert Kraut1, and Jordan Patton1 Human-Computer Interaction Institute and 2H. John Heinz III College Carnegie Mellon University 5000 Forbes Ave., Pittsburgh, PA 15213 {dabbish, robert.kraut}@cs.cmu.edu, [email protected]

ABSTRACT

Theories about commitment in online settings and empirical evidence from offline environments suggest that greater communication in online groups should lead members to become more committed and participate longer. However, experimental evidence is sparse, in part because of difficulties inducing communication online. Moreover, previous work has not identified the route by which communication leads to increased commitment. In this paper, we investigated whether task versus social communication modeled by a leader versus a peer influenced the amount that group members talked and their willingness to continue participating in the group. We conducted an experiment within ad hoc groups in the online game World of Warcraft. Results suggest that communication early in a group’s history causes members to talk more later on and that the early communication increases their commitment through its influence on group atmosphere rather than through increased member participation. Social communication by a peer is especially valuable in increasing commitment. Author Keywords

Virtual teams; commitment; conversation; communication; status; online games. ACM Classification Keywords

H5.3. Group and Organizational Interfaces. General Terms

Human Factors INTRODUCTION

An online group is more likely to be successful if it is composed of members who are committed to it. Commitment is “a force that binds an individual to an organization and thereby reduces the likelihood of turnover” (p. 993, [30]). By commitment, we mean members’ affective connection to and caring for the group in which they become involved [1]. Members who are Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CHI’12, May 5–10, 2012, Austin, Texas, USA. Copyright 2012 ACM 978-1-4503-1015-4/12/05...$10.00.

committed to an online group are crucial to its success. These are the members most likely to provide the content that others value—answers to others’ questions in technical and health support groups (e.g., [43]), or code in open source projects (e.g., [32]). Committed members also help enforce norms of appropriate behavior [47], police the community and sanction deviant behaviors [8], and perform behind the scenes work to help maintain the community [6]. Extensive research in social psychology and organizational behavior suggests that there are two distinct routes to commitment to a group. People can identify with the group as an entity (common identity) or they can feel close to individual members of the group (common bonds; [40]). These processes also seem to apply in online groups [45]. CSCW researchers have proposed design interventions to increase identity-based commitment to online groups by emphasizing the group as an entity and bond-based commitment by emphasizing individuals and supporting communication among them [42]. Experimental research shows that it is possible to substantially increase identitybased commitment by, for example, partitioning a larger group into subgroups and exposing subgroup members to its logo, slogan or accomplishments. However, it is much more difficult to increase bond-based commitment [13, 41], in large part because of the difficulty inducing a sufficient number of group members to communicate with each other. For example, only two percent of 12,000 unique members who had visited a money recommender site in the year after it launched discussion forums ever posted a message [18]. Failures in communication are more general, even when communication is essential to the success of the online group. For example, only approximately 14% of Wikipedia editor discuss the articles they work on in the article’s talk page [23]. A large percent of subscribers to email-based discussion groups are lurkers, never participating in discussions, with a majority of lists having no communication over a three-month period [6]. One of the goals of the current research is to investigate ways of increasing communication in online groups where little otherwise occurs. Research Question 1: How can communication in online groups?

we

increase

There is good reason to believe that conversation among members is important to generating bond-based commitment in an online group. Communication is the

basis for interpersonal relationships, on which bond-based commitment depends. Interpersonal communication drives the development of interpersonal attraction [14]. As people’s interactions increase in frequency, their liking for one another also increases [34]. In online communities, especially, frequency of interpersonal communication is a major determinant of the extent to which people can build relationships with one another [28]. Interpersonal relationships are especially likely to arise from exchanges of personal information and self-disclosure [10, 38]. Meta-analyses of the research on commitment suggest that communication among members of a group or organization is strongly related to the level of commitment to that social entity [26, 38]. However, much of this work simply correlates self-reports about the frequency of various kinds of communication with feelings of commitment. These cross-sectional, correlational analyses do not demonstrate that communication among group members actually increases commitment. If interpersonal communication in online groups actually increases commitment, the mechanisms involved remain unclear. That is, there is still an open question about what that communication should look like. Specifically, does communication lead to commitment as a function of production of communication (i.e., group members feel involved when they talk to others) or receipt of communication (i.e., they feel wanted when others talk to them)? Does the content of the communication matter? For example, if members exchange personal information through task-oriented oriented talk or if they bond through task-oriented talk? Finally, does the conversational partner matter (i.e., is communication from peers versus leaders equally as important to commitment)? A second aim of this research is to better understand whether communication in online groups increases commitment and the causal pathways that might produce this effect. Research Question 2: Does communication in online groups influence commitment? In order to investigate these research questions, we conducted the research described here in the context of ad hoc groups that form to accomplish difficult goals within World of Warcraft, a popular multiplayer, online game. These massively multiplayer games are important businesses in their right. For example, over 11 million players pay about $15 per month to subscribe to the game. Moreover, the ad hoc groups in these game share many features of virtual teams used both online (e.g., the editors creating a Wikipedia article) or in the business world (e.g., a task force designing new product). They consistent of multiple people, often strangers, coming together for a defined time to perform a meaningful (at least to the participants) interdependent task. Virtual teams are increasingly present in the business world as well. In 2002 Gartner Group suggested that more than 60

percent of professional employees work in teams with some level of virtuality and this number has only increased since then [16, 22]. Developing commitment is particularly a challenge for a virtual team [39]. Members of these teams have difficulty forming a shared identity [46], are more likely to make negative attributions about their fellow members’ behavior [11] and feel they are not committed to the team [12], resulting in less willingness to help teammates and higher turnover [19, 52]. We were interested in whether we could induce communication in ad hoc virtual teams and whether this communication would influence team members’ subsequent commitment to the team. In an experiment, we manipulated the presence and type of communication an experimental confederate in a leadership or non-leadership role produced early in the team’s history. We examined how this influenced how much the other team members talked amongst themselves and their subsequent commitment (measured by whether they wanted to stay with the team or not). Our results suggest that injecting communication early in the team’s history significantly increased the amount of talk overall and commitment to the team. Talk by nonleaders was especially effective when it was socially oriented, not task oriented. In addition we find evidence that being talked to leads to commitment irrespective of the amount an individual talks, suggesting commitment is a more a function of the social environment rather than direct engagement with the group. BACKGROUND AND HYPOTHESES Modeling and Communication

As discussed previously, communication within an online team is not automatic. How can one foster conversation within an online group? Research on communication structure and conversation in online settings suggests a series of factors that could influence the amount of talking with a group, including the effort required to produce an utterance [18], requests and other linguistic obligations [27, 44], and behavioral mimicry or modeling [37]. In this research we focus on modeling, because it is under a manager or group leader’s control to some degree. They can serve as role models by performing the behavior they desire from their team or by recruiting a subset of team members with appropriate qualities (e.g., extraversion). Previous work suggests that in both social and nonsocial situations the behavior of others is a powerful cue for one’s own behavior [9], and is effective in influencing both private actions (e.g., towel recycling in hotel rooms, where information about the behavior of previous occupants seems to encourage individuals to conserve and reuse towels; [17]) and communication. In the domain of communication, response to models seems to occur subconsciously, for example, when individuals seem to automatically mimic a conversational partner’s nonverbal gestures, body language or speech [2, 37].

Modeling done early in a group’s history may set norms for appropriate behavior that have long lasting effects. In groups and teams the appropriate or accepted ways of behaving (or norms) serve as a model for how other members should act. Theory about group development suggests that these norms are set very early in a group’s history together and established by observable patterns of interaction [33]. These models of group development suggest that communication in the early stage of a group’s history should set the tone for the group and suggest the nature of their behavior or interaction during later stages. These considerations motivate our first hypothesis: Hypothesis 1: When a role model communicates early in a group’s history, this will cause more communication by other members later in its history. Participant roles may also influence the effectiveness of communication made early in a group’s history on subsequent communication within the group. Research on conformity and influence in groups and teams suggests that leaders, because of their high status, may have greater influence on group norms [24]. Similarly, in teams, leaders can often have more influence on a group’s conversation and decisions [48]. Social identity theory suggests this is because leaders serve as the prototypical member of the group, in that they embody the values of the team, and represent how an ‘ideal’ group member should behave [20]. Social identity theory, then, suggests that conversation by leaders should more strongly influence the behavior of other members than conversation by peers. Hypothesis 2: Conversation delivered early in a group’s history by an individual in a leadership role should have a stronger influence on subsequent group communication than conversation delivered by an individual in a peer role. Communication and Commitment

As discussed above, research on organizational commitment suggests that more communication between members should strengthen commitment to a group [26, 29, 38]. However, previous work on this relationship has been predominantly correlational in nature showing, for example, that long time members of an online community are more talkative and vice versa [36]. This makes it difficult to ascertain the route by which conversation increases commitment to the community. We suggest that there are two possible routes by which conversation increases commitment to a group: (1) by creating a friendly atmosphere or tone, or (2) through participants own communication and investment in the group. Conversational Environment

The mere presence of conversation may create an inviting atmosphere, resulting in increased feelings of commitment to a group. Theories of commitment suggest that socialization and value congruence are important antecedents of commitment to a group [30, 31]. A

communicative environment is inviting and could suggest a norm of openness in a group. For groups that have recently formed, this environment could lead to feelings of cohesion and connection among members, even if they themselves do not talk. If this were the case, the presence of higher levels of communication among some members of a group would be sufficient to increase commitment, regardless of whether an individual themselves is talking. The large number of repeat lurkers on many sites suggests that it is possible to be committed to a site without directly interacting with other members [35]. Hypothesis 3: Role model communication in the early period of a group’s history should foster higher levels of commitment to the group. Talking Increases Commitment

Simply seeding a group with communication may not be sufficient to foster commitment to that group. It may be that one must actively participate in the conversation to feel a higher level of commitment. Theories of commitment also suggest that involvement with a team is an important antecedent to commitment [30]. This corresponds with research on communication in interpersonal relationship development which suggests that sharing information with someone can foster bonds and make you feel closer to that person. For example, research on disclosure and liking suggests that sharing personal information about yourself with another person increases liking for that person [4]. In groups, being highly engaged with the group and involved may lead you to feel closer to the other members and increase your desire to stay with the group. Parks & Floyd’s research on membership in online text-based discussion forums showed a relationship between the amount of posts an individual had made in an online forum and their tenure in the group [36]. If communication leads to commitment through a feeling of involvement, then directly interacting with other members should increase feelings of commitment to that group. This perspective suggests that the effectiveness of a communication intervention will be dependent on the participant’s own amount of communication in response to initial seeded comments: Hypothesis 4: The influence of a role model’s communication on member’s commitment to the group will be mediated by the amount they communicate. Member role and conversation type

The relationship between communication and commitment may depend largely on who does the talking and what they say. The previous work on communication and commitment has suggested that leader communication is an important influence on commitment [50]. Research on leader communication has often focused on task-related communication in the form of feedback. However, the research on leader-member exchange has suggested that fostering a close relationship with subordinates through socio-emotional (socially-oriented) communication can

increase subordinate motivation, commitment and satisfaction [15]. Other work on communication and commitment has suggested that leaders should instead focus on task-related communication, which leads to higher organizational commitment. In a series of surveys within two organizations, Postmes et al. [38] found that leader communication about task-related topics was more strongly associated with organizational commitment than peer communication about socio-emotional topics. At the same time, peer communication about socio-emotional topics was an important determinant of commitment to the immediate group (versus the larger organization). Unfortunately, the previous work on communication and commitment has largely conflated communicator role and conversation topic. It is unclear whether socio-emotional communication from leaders would be equally effective as socio-emotional communication from peers in generating commitment to the team. Similarly, we do not know whether task-related communication from peers would increase or decrease commitment relative to socioemotional communication. The relationship between communication topic and role could further illuminate the mechanism by which communication leads to commitment. Postmes et al.’s [38] results suggest that when the leader is talking, communication should lead to commitment when that communication is related to the group’s task. This is because the communication will serve to reinforce the identity of the group as an entity [20]. Hypothesis 5a: The relationship between communication and commitment will be strongest when that communication is focused on the group’s task (rather than socio-emotional) and delivered by an individual in a leadership role (rather than a peer). However, if commitment is ultimately more a function of interpersonal relationships among members of the team [29], socially-oriented social communication delivered by peers should result in the strongest connection to the group. Hypothesis 5b: The relationship between communication and commitment will be strongest when that communication is socio-emotional (rather than focused on the group’s task) and delivered by a peer (rather than an individual in a leadership role).

the content of their communication, and measure the group’s response in terms of subsequent communication and commitment. METHOD Setting and Task

We conducted our experiment in the World of massively multiplayer online role-playing game, accessed by players over the Internet using computers. Over 11.1 million people have a subscription to World of Warcraft [51].

Warcraft which is personal monthly

We applied our experimental manipulations within groups completing dungeon raids. “Raiding” is a common activity in World of Warcraft, in which groups of players work together to accomplish a common goal which none of them could accomplish on their own. When players participate in a raid, they form a group using in-game tools and journey into an area of the game known as a “dungeon.” Players almost never venture into dungeons on their own, as their characters would quickly die without help. In a raid, however, a very well coordinated group may successfully defeat all of monsters in a dungeon. When these monsters are defeated, they release special weapons, armor, and other “loot” that players desire. To increase the generalizability of the research, we selected three dungeons that span a range of difficulties, time requirements, and participant motivations: Pit of Saron, Forge of Souls, and the Culling of Stratholme. In organizing for a dungeon raid, players can select to work with an existing team or can be randomly assigned to a team using an in-game tool called the Dungeon Finder. After each raid, they have the option to continue as a team beyond the immediate task. During the raid players have available text-based channels of communication seen by all participants (called the party chat). Players can also chat privately with selected other players. Any player can record in game activity and all chat visible to them using game tools and add-ons. Figure 1 shows a player’s screen during a dungeon raid, with chat in the lower left corner.

Summary

In summary, based on the previous research on commitment, communication, and group development, we expected that injecting communication early in a team’s history would lead to more communication among the rest of the team in a subsequent time period and ultimately higher commitment to the team. We also expected that the communicator’s role would interact with the nature of the communication. We evaluated our stated hypotheses in an experiment within real groups in an online game. The setting allowed us to manipulate the communicator’s role,

Figure 1. World of Warcraft Dungeon Image, players cooperate to defeat a dungeon monster.

Participants

Participants in the study were World of Warcraft players who used the Random Dungeon Finder tool to enter the three dungeons of interest. This tool allows players to sign up for random dungeons that require five participants. The Random Dungeon Finder matches players into teams so that all the necessary roles are filled. Additionally, it limits access only to players with characters that have attained an appropriate level, armor, and weapons to complete the dungeon. We conducted trials with 108 different groups (5 person groups made up of 3-4 naïve participants per group plus one confederate) for a total of 457 participants. Experimental Manipulations

In order to examine our hypotheses of interest, we modified the role and behavior of a confederate embedded in dungeon raid groups. Our experiment had two main manipulations: the type of chat the confederate used early in the groups existence (silent, socially-oriented, and taskoriented), and the role of the confederate within the group (leader vs. peer). These variables were manipulated across dungeon raid groups, for a 3 (chat type) by 2 (role) between subjects design. We examined the influence of these interventions on the amount of communication by other group members and their commitment to the team, measured by their willingness to stay in the group following the immediate task. We describe our independent and dependent variables in more detail below. Conversant role

We were interested in whether the relationship between conversation and commitment would vary depending upon whether the role model was a leader of the group or just a peer member. We manipulated leadership by having the same confederate play different functional roles in the fiveman dungeon raid groups as they battled monsters using two different level 80 characters. Three unique roles must be fulfilled in a five-man dungeon. The first role, called “Tank,” is a character with great defensive capabilities who will soak up damage from attacks by dungeon monsters. The Tank is commonly viewed as the natural leader of the group, as his actions determine a course through the dungeon and when and where fights will take place. The second role is called “Healer.” A Healer keeps all group members alive (and resurrects the dead) by repairing damage to the other group members as it occurs. The third and final role is called “Damage Dealer,” and may also be referred to as the “DPS” or “Damage Per Second (DPS).” Damage Dealers attack designated targets to quickly kill them in the order prescribed by the Tank. A five-man dungeon group includes one Tank, one Healer, and three Damage Dealers.

During the course of this study, the same research assistant played either the Tank role for the “leader” condition or the Damage Dealer role for the “peer” condition. Members of a group can vary in status as a function of their instrumentality to the group’s task, i.e. the skills and resources they posses that directly impact successful task performance [7]. Leadership in random dungeon groups in WoW play out differently than in other small, task-oriented groups because they are based on character’s in-game functions rather than any special leadership ability. In the dungeon raid task, the Tank role was highly instrumental because participants with Tank characters are less common in the game than Healer or Damage Dealer characters, and so more difficult to replace. The Damage Dealer role, on the other hand, was less instrumental because each team contained three Damage Dealer players that were easy to replace. Our hypotheses suggest that role modeling performed by different characters (Tank vs. DPS) should have different effects because the characters are differentially important in the game. Communication type

The five-man dungeon raids in this experiment took between 15 and 30 minutes to complete. In order to manipulate the presence and nature of conversation and view its effects on the group’s own conversation, we segmented the dungeon runs into three time periods. The confederate administered the conversation intervention during the first time period, which was approximately one third of the dungeon run (e.g. the first 5 minutes during a 15 minute run) and observed with minimal communication during the next two-thirds of the dungeon run (e.g. 10 minutes during a fifteen minute run). The confederate manipulated the presence of conversation in the first time period by initiating chat based on a script or remaining silent unless spoken to. When conversation was present, the confederate varied the type of the conversation by selecting conversational turns from a task-oriented or socially-oriented script. The scripts were created from utterances observed in use by WoW players during pilot dungeon raids. Thus we had three levels in our manipulation of communication type: silent, task-oriented, and socially-oriented. During the silent condition, the confederate did not initiate any conversation and did not respond to general questions asked to the group as a whole, but offered short, polite replies if asked a direct question. During the task-oriented and socially-oriented conditions, the confederate made utterances in the group text chat at three points during the first five minutes of the dungeon raid: (a) at the very beginning of the dungeon run, (b) after the second monster in the dungeon was killed, (c) after the fourth group of monsters were defeated. These points were selected because they are effectively subtask breakpoints within the dungeon raid task and group members would be more likely to attend to and have time to respond to the communication.

At each of these time points, the confederate selected randomly among conversational prompts. In the taskoriented condition, conversational prompts were questions directly related to the current dungeon raid task. Questions were utilized because the prior literature shows that questions are more effective at initiating conversation than statements [44]. In the socially-oriented condition, the conversational prompts were questions related to aspects of the game outside of the immediate task or related to the players’ real lives outside of the game. Table 1 shows examples of task and socially-oriented prompts. Conversational prompt

Task-oriented

Socially-oriented

(a)

At the beginning of the dungeon

“Is everybody ready?”

“What did you do today?”

(b)

After second monster

“What rotation do you use?”

“I’m thinking of creating a new alt, any recommendations?”

(c)

After fourth monster

“Does anyone need specific loot from here?”

“Any of you working on hard modes in icc?”

Table 1. Examples of task and socially-oriented prompts

After the confederate manipulated his communication, in the remaining two-thirds of the dungeon run, he was silent, responding only to direct questions. Outcome Measures

We collected two dependent variables: the amount of talk by team members following the conversation intervention and commitment to the team. Conversation generated

In order to measure the effectiveness of conversational modeling on subsequent conversation among group members, we measured the amount of talk in the second period of the dungeon run. We did this by logging the chat from all group members using an in-game feature that captured all text communication and recorded it to a file on the confederate’s computer. Chat logging appended timestamps to all text messages. We calculated the amount of talk generated by each member of the group during the second time period of play in terms of lines of communication, with each line representing an utterance, after eliminating the confederate’s communication. It is important to note that this measure includes only chat messages sent using the party chat (hence visible to all members of the group) while private messages between pairs of players were not captured in the logs. Commitment

We were interested in whether the initial conversation directly or indirectly influenced commitment to the team. Here we defined commitment as a desire to stay with team beyond the original team goal of completing the dungeon

raid for which they formed [5]. In order to measure this, near the end of the dungeon raid but before participants left the dungeon, our confederate queried the group to ask whether they wanted to continue together on an additional dungeon raid. Before the final monster (the boss) in the dungeon was killed, the confederate posted a message to the group chat asking if they want to queue for another dungeon. After the boss was killed, he/she immediately activated the Random Dungeon Finder and recorded which (if any) team members queued for another dungeon. Thus our measure of individual commitment was behavioral -- a binary variable set to one if a team member re-queued for the next dungeon, and zero if not. Control variables

In addition to our outcome measures of interest (communication volume and behavioral commitment), we also identified control variables measured to account for other factors that may influence the dependent variables. These included the specific dungeon name, the participant’s potential performance (character level, armor quality) based on data retrieved from a public database that publishes information on characters from World of Warcraft, the participant’s actual performance during the dungeon raid task (damage or heals per second), and their role (Tank, Healer, or Damage Dealer). RESULTS

We conducted 108 trials within the dungeons in World of Warcraft, with approximately 4 participants per trial for a total of 457 participants. Data from participants across all three roles are included in our analysis. Twenty-two team members dropped out of the raid before completion and were replaced by the Random Dungeon Finder. The results reported below do not change whether or not the 22 replacements were included in the analysis. We were interested in whether our communication manipulation generated communication by the other members of the team (Hypothesis 1), whether communication role modeling was more effective when done by a group leader or peer (Hypothesis 2) whether the presence of communication increased team member commitment directly (Hypothesis 3) or in a mediated fashion (Hypothesis 4) and whether these possible effects varied by member role (Hypothesis 5a and 5b). We first examined the influence of our communication manipulation in the first period of play on the amount of talk generated by members of the group in the second period. We conducted a negative binomial regression of the number of lines of talk generated by each member in the second time period of the session to account for the skewed distribution of the count data, with individual nested within group to account for the non-independence of observations across members of the same group. We included member role, conversation type, and their interaction as independent variables in our model, along with the dungeon and player

s  

performance potential as control variables. Statistical tests are reported as Incident Report Ratios (IRR), the ratio of the amount of lines of talk as an independent variable changed from a baseline condition (e.g., the peer role model) to another (e.g., the leader role model). Figure 2 shows the pattern of results for amount of communication.

in odds ratios (OR), the ratio of the odds of staying in the group when an independent variable increases by a unit compared to a baseline condition. Because decisions within a single group are not independent of each other, the analysis nested members within groups with participant id included as a random effect in the model, to account for this non-independence of observations. Players were substantially more likely to want to continue playing in groups where the confederate was a leader than when it was a peer (OR=4.94; p=.001).

Figure 2: Effects of confederate role and communication type on players’ communication

H1 predicts that group members would talk more when a role model talked early in the groups history than when the role model was silent. This hypothesis was confirmed. We created a nominal variable to contrast talk versus silent conditions (effectively collapsing the task-oriented and socially-oriented talk manipulations). When the role model talked in the first third of a group’s life, the other members talked more in the remaining two-thirds (IRR=1.61; p=.05). We conducted exploratory analyses to determine whether the type of talk mattered, including a nominal variable representing the three levels of talk in our experiment (silent, task- and socially-oriented). Both types of talk had an influence on the amount of participant talk, with early socially-oriented talk by our confederate having the strongest influence on the lines of talk (IRR=1.75; p=.04) and task-oriented talk only marginally increasing the amount of talk among the other members of the group (IRR=1.49; p=.15). H2 predicted that the effects of role model talk would increase when the role model was a group leader. However, this hypothesis was not confirmed. The effects of role modeling were not different when the model was a leader or peer (for the Role by Talk interaction, IRR=0.79; p=.47). In addition, the leadership role did not influence the power of role modeling when we separately examine their task-oriented communication (for the Role by task-oriented communication interaction, IRR=0.82; p=.61) and their socially-oriented communication (for the Role by Sociallyoriented communication interaction, IRR= 0.76; p=.48). Conversant role, communication type, and commitment

Because the outcome measure of commitment is binary (whether a group member signed up to play with the group again or nor), we conducted a random-effects logistic regression on the decision to continue. Results are reported

Figure 3: Effects of confederate role and communication type on players’ commitment

H3, which predicted that more communication early in a group’s life cycle would increase group members’ commitment, was confirmed. Figure 3 shows the pattern of results for commitment across conditions. The presence of conversation doubled the odds of staying with the group (OR=2.10; p=.01). The effect of the leadership role was large. However, the communication had different effects when the confederate performed different roles in the group (for the role by communication interaction, OR=.28, p=.04). In particular, communication increased players’ commitment when the role model was a peer (OR=2.75, p=.12), but depressed it when they were leaders (OR=.61, p=.10). Distinguishing among the types of communication shows that socially-oriented chat tripled the odds of staying with the group (IRR=3.10; p=.02), while task-oriented communication did not reliably increase the odds of staying (OR=1.35, p=.54). However, as Figure 2 shows, the effects of different types of communication varied with the confederate’s role. Socially oriented communication increased the probability of players’ staying when the confederate was a peer (OR=4.65, p=.04), but had no effect when the confederate was a leader (OR=.79, p=.50; for role by socially-oriented communication interaction, p=.03). In contrast, task-oriented communication depressed a player’s probability of staying when the confederate was a leader (OR=.48, p=.03), but had no effect when the confederate was a peer (OR=1.64, p=.49; for the role by task-oriented communication interaction, p=.08). These results suggest that players wanted to continue in groups when the confederate was a leader. He made the

group attractive by dint of his status and irreplaceability. Communication didn’t enhance his position in the group, and even hurt it when he engaged in task-based communication. In contrast, players were less likely to want to continue playing in groups when the confederate was their peer. His position, gameplay and replaceability made the group less attractive than if he had been a leader. However, he could compensate for these problems of status and performance through social communication. We suspect that other players grew to like the group more because of his social rather than task competence. Finally we examined whether the influence of early communication on commitment occurred as a function of communication environment, or was mediated by member communication, suggesting involvement as a pathway. Having established the direct influence of chat presence and type on the amount of member communication, and the influence of communication type on commitment, we next added member communication to our model of commitment (as lines of talk). We found that the presence of chat continued to have a positive influence on commitment (likelihood of staying with the group), even with the amount of member communication included in the model (OR=2.05; p=.11), particularly for the peer conversant’s (role by talk interaction: OR=1.75; p=.03). This result suggests that member talk does not mediate the influence of leader or peer communication on member commitment to the group. We examined this pathway using a binomial regression in order to compare the coefficients for the direct effect of commitment with and without member communication in the model using mediation tests. We found the mediation was not significant according to this test [3]. Thus, we find support for Hypothesis 3, that conversation influences commitment by creating a positive climate in the group. Support for our hypotheses is summarized in table 2. DISCUSSION

The results from our study indicate that it is possible to induce communication in a temporary online team through the use of a role model. We found that communication early in a group’s history led to increased levels of communication by group members. In addition, the role model’s communication was effective in increasing commitment to the group in the form of member retention. Our results also illuminate the causal pathway between communication and commitment to a group. We found that communication influences commitment to a group by creating a supportive environment within which members choose to remain. This is evidenced by the fact that early communication by either a peer or a leader led to higher levels of member communication (which could be construed as evidence of affective or behavioral commitment), but this member communication did not explain the increase in likelihood to stay with a group (continuance commitment). Thus it is through an

environmental effect, rather than direct involvement, that communication can increase commitment to a group. Additionally, our results showed this occurred regardless of the type of communication (task or socially-oriented talk). Hypothesis

Supported?

1: When a role model communicates early in a group’s history, this will cause more communication by other members later in its history.

Yes. Presence of chat increased the amount of member chat across all role conditions.

2: Conversation delivered early in a group’s history by an individual in a leadership role should have a stronger influence on subsequent group communication than conversation delivered by an individual in a peer role.

No. Early conversation by both peers and leaders increased member conversation volume.

3: Role model communication in the early period of a group’s history should foster higher levels of commitment to the group.

Yes. Chat manipulation significantly increased likelihood of staying in the group.

4: The influence of a role model’s communication on member’s commitment to the group will be mediated by the amount they communicate.

No. Early conversation had a direct influence on the likelihood of staying with a group. In particular, socially-oriented communication had the strongest influence on retention.

5a: The relationship between communication and commitment will be strongest when that communication is delivered by an individual in a leadership role (rather than a peer) focused on the group’s task (rather than socioemotional).

No. Communication type did not influence commitment when delivered by an individual in a leadership role.

5b: The relationship between communication and commitment will be strongest when that communication is socio-emotional and delivered by a peer.

Yes. Significant interaction between role and conversation type, with talk type having the strongest influence on commitment to peer conversant (and no influence of talk type for leaders).

Table 2 – Summary of Hypotheses and Experimental Support

This suggests that the nature of the communicative environment may matter less than the fact that members are interacting with one another. Future work is needed to ascertain exactly how and why the communicative environment leads to commitment to a group. Previous work has suggested that social support is particularly important antecedent to commitment to a social group, and it may be that presence of communication induces a feeling of this type of support. It may also be that communication changes member perceptions of the group as an entity. For example, communication may increase the general attractiveness of a group, by suggesting members are more involved or committed to the task. Particularly in

a distributed environment, this evidence of commitment may reduce uncertainty about whether or not members are engaged and committed to the group. We also found that the communicator role had a significant influence on the effect of communication content, with socially-oriented communication substantially increasing commitment only when the communicator was a peer (rather than the leader of the group). This relates to work on liking and relationship formation in work groups, where personality can increase attractiveness of less skilled members but may not influence the perceived utility of clearly instrumental members [7]. It may be that communication is attended to more when it is delivered by members whose utility is uncertain to the group.

5.

6.

7.

8.

Limitations

It is important to note the potential limitations of our study, which was conducted in a virtual environment primarily used for entertainment by its players, with small groups. It is unclear to what extent these results would generalize to a traditional organizational setting where members are being paid for task completion, and rewards depend on successful task completion. World of Warcraft is a setting where individuals engaged in game tasks like dungeon raids typically converse only when needed to support task performance. The way communication is viewed and used in this environment, and the level of communication typically experienced in these types of groups, may have influenced our results. In addition, groups used in this study were small in size. Thus it will be important to examine whether these results hold in other settings with different communication norms and with larger group sizes. CONCLUSION

Overall our results suggest that online community and virtual team managers and social media system designers may have the ability to influence member commitment by introducing communication early in a group’s history. This communication seems to influence commitment by creating a supportive or friendly environment within a group (versus by inducing member involvement).

9.

10.

11.

12.

13.

14. 15.

16.

REFERENCES

1.

2.

3.

4.

Allen, N. J., and Meyer, J. P. The measurement and antecedents of affective, continuance, and normative commitment to the organization. J. of Occupational and Organizational Psychology 63 (1990), 1-8. Bargh, J., and Williams, E. The automaticity of social life. Current Directions in Psychological Science 15, 1 (2006), 1-4. Baron, R.M., and Kenny, D.A. The moderatormediator variable distinction in social psychological research. J. of Personality and Social Psychology 51, 6 (1986), 1173-1182. Berg, J.H., and Archer, R.L. The disclosure-liking relationship: Effects of self-perception, order of

17.

18.

19.

disclosure, and topical similarity. Human Communication Research 10, 2 (1983), 269-281. Bettenhausen, K. L. Five years of groups research: What we’ve learned and what needs to be addressed. J. of Management 17 (1991), 345-381. Butler, B. Membership size, communication activity, and sustainability: A resource-based model of online social structures. Information Systems Research 12, 4 (2001), 346–362. Casciaro, T., and Lobo, M.S. When task competence is irrelevant: The role of positive and negative interpersonal affect in instrumental ties, Administrative Science Quarterly 53 (2008), 655-684. Chua, C., Wareham, J., and Robey, D. The role of online trading communities in managing internet auction fraud, MIS Quarterly 31, 4 (2004), 759-781. Cialdini, R.B., and Goldstein N.J. Social Influence: Conformity and Compliance. Annual Review of Psychology 55 (2004), 591–621. Collins, M., and Miller, L.C. Self-disclosure and liking: A meta-analytic review. Psychological Bulletin 116, 3(1994), 457-475. Cramton, C.D., Orvis, K.L., and Wilson, J.M. Situation invisibility and attribution in distributed collaborations. J. of Management 33, 4 (2007), 525-546. Crossman, A., and Lee-Kelly, L. Trust, commitment, and team working: The paradox of virtual organizations. Global Networks 4 (2004), 375-390. Farzan, R., Dabbish, L., Kraut, R. E., and Postmes, T. (2011). Increasing Commitment in Online Communities via Building Social Attachment, In Proc. CSCW 2011, ACM Press (2011), 321-330. Festinger, L. Informal social communication. Psychological Review 57 (1950), 271-282. Gerstner, C.R., and Day, D.V. Meta-analytic review of leader-member exchange theory: Correlates and construct issues. J. of Applied Psychology 82, 6 (1997), 827-844. Gibson, C., and Gibbs, J.L. Unpacking the concept of virtuality. Administrative Science Quarterly 51 (2006), 451-495. Goldstein, N. J., Cialdini, R. B., and Griskevicius, V. A room with a viewpoint: Using social norms to motivate environmental conservation in hotels. J. of Consumer Research 35(2008). Harper, F., Frankowski, D., Drenner, S., Yuqing Ren, Kiesler, S., Terveen, L., et al. Talk Amongst Yourselves. In Proc. IUI 2007, ACM Press (2007), 6271. Herbsleb, J. D., and Mockus, A. An empirical study of speed and communication in globally-distributed software development. IEEE Transactions on Software Engineering 29, 3 (2003), 1-14.

20. Hogg, M.A. A social identity theory of leadership. Personality and Social Psychology Review 5 (2001), 184-200. 21. Johnson, S.K., Bettenhausen, K., and Gibbons, E. Realities of working in virtual teams. Small Group Research 40, 6 (2009), 623-649. 22. Kanawattanachai, P., and Yoo, Y. Dynamic nature of trust in virtual teams. Case Western Reserve University, USA . Sprouts: Working Papers on Information Systems, 2, 10 (2002). http://sprouts.aisnet.org/2-10. 23. Kittur, A., and Kraut, R. E. Harnessing the wisdom of crowds in Wikipedia: Quality through coordination. In Proc. CSCW 2008, ACM Press (2008), 37-46. 24. Levine, J., and Moreland, D. Small Groups. Handbook of Social Psychology (2998), 415-469. 25. Luthans, F., Wahl, L.V.K., and Steinhaus, C.S. The importance of social support for employee commitment. Org. Dev. J., Winter (1992), 1-10. 26. Mathieu, J. E., and Zajac, D. M. A review and metaanalysis of the antecedents, correlates, and consequences of organizational commitment. Psychological Bulletin 108, 2 (1990), 171-194. 27. Markey, P. M. Bystander intervention in computermediated communication. Computers in Human Beh. 16, 2 (2000), 183-188. 28. McKenna, K.Y.A., Green, A.S., and Gleason, M.E.J. Relationship formation on the Internet: What's the big attraction? J. of Social Issues 58,1 (2002), 9-31. 29. Meyer, J.P., and Allen, N.J. A three-component conceptualization of organizational commitment. Human Resource Management Rev. 1, (1991), 61-89. 30. Meyer, J.P., Becker, T.E., and Vandenberghe, C. Employee commitment and motivation: A Conceptual analysis and integrative model. J. of Applied Psychology 89, 6(2004), 991-1007. 31. Meyer, J.P., and Herscovitch, L. Commitment in the workplace: Toward a general model. Human Resource Management Review 11 (2001), 299-326. 32. Mockus, A., Fielding, R., and Herbsleb, J.D. Two case studies of open source software development: Apache and Mozilla. ACM Transactions on Software Engineering and Methodology 11, 3 (2002), 309-346. 33. Moreland, R. L., and Levine, J. M. Socialization in organizations and work groups. In M. E. Turner (Ed.), Groups at Work: Theory and Research (pp. 69112). Erlbaum: Mahwah, NJ, 2001. 34. Newcomb, T. The Acquaintance Process. Holt, Rinehart & Winston: New York, 1961. 35. Nonnecke, B., and Preece, J. Lurker demographics: Counting the silent. ACM CHI Conference, CHI Letters 4, 1(2000), 73-80. 36. Parks, M.R., and Floyd, K. Making friends in cyberspace. J. of Communication 46 (1995), 80-97.

37. Pickering, M. J., and Garrod, S. Toward a mechanistic psychology of dialogue. Behavioral and Brain Sciences 27 (2004), 169-226. 38. Postmes, T., Tanis, M., and de Wit, B. Communication and committment in organizations: A social identity approach. Group Processes and Intergroup Relations 4, 3 (2001), 227-426. 39. Powell, A., Piccoli, G., and Ives, B. Virtual teams: A review of current literature and directions for future research. ACM SIGMIS Database 35, 1 (2004), 6-36. 40. Prentice, D. A., Miller, D. T., and Lightdale, J. R. Asymmetries in attachments to groups and to their members, Personality and Social Psychology Bulletin 20, 5 (1994), 484-493. 41. Ren, Y., Harper, F. M., Drenner, S., Terveen, L., Kiesler, S., Riedl, J., et al. Increasing attachment to online communities. MIS Quarterly (under review). 42. Ren, Y., Kraut, R. E., and Kiesler, S. Applying common identity and bond theory to design of online communities,” Org. Studies 28,3 (2007), 377-408. 43. Rodgers, R., and Chen, Q. Internet community group participation: Psychosocial benefits for women with breast cancer. JCMC 10, 4(2005) np. 44. Sacks, H., andJefferson, G. Lectures on conversation: Wiley-Blackwell: New York, 1995. 45. Sassenberg, K. Common bond and common identity groups on the internet: Attachment and normative behavior in on-topic and off-topic chats, Group Dynamics 6,1(2002), 27-37. 46. Shapiro, D. L., Furst, S. A., Spreitzer, G. M., and Von Glinow, M. A. Transnational teams in the electronic age: Are team identity and high performance at risk? J. of Org. Behavior 23 (2002), 455-467. 47. Smith, C.B., McLaughlin, M.L., and Osborne, K.K. Conduct control on usenet. JCMC 2, 4(1997), np. 48. Strodbeck, F.L., James, R.M., and Hawkins, C. Social status in jury deliberations. American Sociological Review 22 (1957), 713-719. 49. Tangirala, S., and Alge, B. J. Reactions to unfair events in computer-mediated groups: A test of uncertainty management theory. OBHDP 100 (2004), 1-20. 50. Vandenberghe, C., Bentein, K., and Stinglhamber, F. Affective commitment to the organization, supervisor, and work group. J. of Vocational Beh. 64 (2004), 4771. 51. Wikipedia. (2011). World of Warcarft Retrieved Sep 9, 2011, from http://en.wikipedia.org/wiki/World_of_Warcraft 52. Workman, M., Kahnweiler, W., and Bommer, W. The effects of cognitive style and media richness in commitment to telework and virtual teams. J. of Vocational Beh. 63 (2003), 199-219.