Treffer: Development of Computer-Supported Collaborative Social Networks in a Distributed Learning Community

This study examines the formation and change of collaborative learning social networks in a distributed learning community. A social network perspective is employed to understand how collaborative networks evolved over time when 31 distributed learners collaborated on a design project using a computer-mediated communication system during two semesters. Special attention was paid to how pre-existing friendship networks influenced the formation of macro-level collaborative learning networks and individual level social capital. We discovered that pre-existing friendship networks significantly influenced the formation of collaborative learning networks, but the effect was dependent on the developmental phase of community. Also, pre-existing networks generally acted as a social liability that constrained learners' ability to enhance their social networks and build social capital when they participated in a new learning environment. The results suggest that, in order to fully understand how to build effective collaborative learning and work environments, participants' social network structures need to be considered. (Contains 2 tables and 4 figures.)

Verfasser

AN0120749262;b6q01dec.05;2019Mar22.08:32;v2.2.500

Development of computer-supported collaborative social networks in a distributed learning community 

<sbt id="AN0120749262-2">1 Introduction</sbt>

This study examines the formation and change of collaborative learning social networks in a distributed learning community. A social network perspective is employed to understand how collaborative networks evolved over time when 31 distributed learners collaborated on a design project using a computer-mediated communication system during two semesters. Special attention was paid to how pre-existing friendship networks influenced the formation of macro-level collaborative learning networks and individual level social capital. We discovered that pre-existing friendship networks significantly influenced the formation of collaborative learning networks, but the effect was dependent on the developmental phase of community. Also, pre-existing networks generally acted as a social liability that constrained learners' ability to enhance their social networks and build social capital when they participated in a new learning environment. The results suggest that, in order to fully understand how to build effective collaborative learning and work environments, participants' social network structures need to be considered.

A perspective of community-based learning and knowing is receiving growing attention by both researchers and practitioners. Proliferation of popular terms such as communities of practice (Brown and Duguid [11], Lave and Wenger [35]), knowledge-building communities (Scardamalia and Bereiter [47]), knowledge communities (Erickson and Kellogg [19]), communities of knowing (Boland and Tenkasi [8]), and online collaborative learning communities (Alavi [2]) reflect the growing theoretical concerns on, as well as practical interests in, such new ways of learning and knowing. Increasingly, these communities are moving beyond face-to-face exchanges, to interact in Computer-Mediated Environments (CMEs) such as email lists, online forums, and shared web spaces, requiring us to reconfigure our conceptual as well as physical boundaries of community forms and community-based learning and knowing (Preece [42], [43], Woodruff [61]).

The purpose of this study is to gain a better understanding of this new form of social system—distributed learning community (DLC)—by examining the building process of community social infrastructure, i.e., Computer-Supported Collaborative Social Network (CSCSN). A social network perspective is employed to understand how collaborative learning networks evolved over time when 31 distributed learners had collaborated on design projects using computer-mediated communication (CMC) systems. Special attention was paid to how pre-existing face-to-face networks influenced the over-time change and formation of CSCSN, and whether they constrained or facilitated the way the individual learners developed their social capital in this emergent social structure. In other words, we asked:

1. how does a community social infrastructure emerge in a computer-mediated learning environment

2. whether this emerging structure is influenced by a pre-existing social structure, and

3. if so, what is the nature of its effect on the development of macro-level social structure and individual-level social capital?

2 Review of literature

2.1 Communities of practice, social networks, and social capital

From the social network perspective, knowledge is a social and collective outcome and always embedded in a social context—both created and sustained through ongoing social relationships. Many researchers argue that learning is fundamentally a social process and the purpose of a knowledge community is to create and sustain knowledge, culture and social infrastructures (i.e. social networks) that foster seamless conversations and networks of connections and relations among members (Lave and Wenger [35], Haythornwaite [27]).

Assuming that learning activity is fundamentally situated in networks of multiple interactions and shared practices, many scholars have called for more research to examine knowledge creation and learning processes within the broader context of how communities are structured, and how individuals are situated in the larger social structure of the communities (Brown and Duguid [11], Lave and Wenger [35], Wenger [59]). For instance, Nahaphiet and Ghoshal ([39]) argue that, in order to understand how individuals attain and build knowledge, it is necessary to analyse how they are situated in networks of social relations, resource exchange and social support. In a similar vein, Nardi et al. ([40]) also suggest that the most fundamental unit of analysis for computer-supported cooperative/collaborative work and learning (CSCW/CSCL) should not be the group level, but at the collective social network level. While many have emphasised the theoretical importance of social and structural elements of communities of practice (see Brown and Duguid [11] for review), there is surprisingly little empirical work to directly examine how such social and communicative structures evolve over time in knowledge communities, and what social and technological elements influence such process (Woodruff [61]).

Studies have shown that social networks in a community serve as a conduit for information and resource exchanges within and across the larger social system (Wellman et al. [58], Haythornwaite [27]). For individuals, this provides a basis of social capital, the networks of crosscutting personal relationship that provide their members with cooperation, trust, opportunity and access to a set of resources 'collectively owned' by network members (Nahaphiet and Ghoshal [39]). For instance, Baldwin et al. ([3]) found that centrality in a network was positively correlated with satisfaction with a team-based learning program. With regard to information exchange among peers, proximity and the strength of ties between peers led to the exchange of more kinds of information and the use of more media (Haythornwaite and Wellman [28]). Haythornthwaite ([27]) found that a learner's network centrality was positively associated with a sense of belongingness in a learning community.

Previous studies show that social networks are an important antecedent for successful collaboration and socialisation in knowledge communities. However, studies examining the building process of community social infrastructure are surprisingly rare. Process-oriented research is imperative in that the community perspective holds strong assumptions about the ongoing nature of learning and knowing. The concept of legitimate peripheral participation in situated learning theory (Lave and Wenger [35]), for instance, emphasises that the learner gradually moves from peripheral participation towards full participation in the community of practice as they engage in over-time interactions and shared practices with multiple actors in a community.

Using a longitudinal analysis, we examine change and formation of a large collaborative learning network and individual social capital. As noted above, we pay special attention to the effect of a pre-existing social structure on this emergent pattern of knowledge community. In other words, we look at how a pre-existing social network develops into a CSCSN. In addition to this descriptive analysis, we also examine whether, to what extent, and how a pre-existing social network influences the formation of macro-level social structure and individual level social capital. In the subsequent sections, we specify the rationale for this research focus, review previous studies and summarise implications for this study.

2.2 A pre-existing network in a DLC

Social systems do not originate in a social vacuum (Cohen and Prusak [14]). As structuralists often argue, development of any social system is dependent upon and constrained by previous social structures, histories and ties (see Emirbayer and Goodwin [18] for review). That is, social networks 'exhibit aspects of both emergence, being called into existence to accomplish some particular work, and history, drawing on known relationships and shared experience' (Nardi et al. [40], p. 207). Therefore it is critical to examine whether and how pre-existing social networks and social ties influence the emerging pattern of collaborative social networks in DLCs.

In theory, a pre-existing network such as a friendship network is a relational foundation, providing a base for quick formation of multiple communication and social support networks in a DLC. Electronic communities often can be characterised by a lack of social bonds (especially in their initial phase), because the communication channels between members from distant locations are restricted to CMC, which may not be suitable for building strong relationships (for counter arguments see Walther [54], Walther et al. [55]). In this sense, pre-existing friendship networks are the building blocks for emerging social infrastructures, as people develop new social ties from 'friends of a friend' and 'friends of a friend of a friend', etc.

However, a pre-existing social network may also simultaneously act as a constraint on communication by restricting the evolution of network ties and structures to predefined, existing social circles (Granovetter [24]). Because long-term relations tend to be strong, symmetrical (i.e. reciprocated) ties, they tend to cluster in dense, interconnected groups (Krackhardt [34]). That is, strong ties tend to bond similar people together, such that they are all mutually connected and share similar or redundant resources (Granovetter [24]). Such tightly bound groups become insulated from outside information and resources. These types of groups are better suited for social support than for instrumental ends such as access to unique information and resources (Granovetter [24]). Considering that one of the main objectives of building a DLC is to produce an efficient flow of ideas, approaches, information and knowledge across different cultures, locations and social boundaries, an emergent network strongly rooted in a pre-existing friendship network is somewhat problematic as it functions to promote closed, rather than open, communication and resource exchanges (Robertson et al. [46]).

The above discussion indicates that pre-existing friendship networks could be expected to have both positive and negative influences on the formation of emerging CSCSN in a DLC. Few previous studies have empirically examined whether or how a pre-existing network influences the development of collaborative social networks in a distributed learning environment. Therefore, we ask the following research questions:

1. RQ 1: How does an existing social network develop into a CSCSN over time?

2. RQ 2: To what extent does an existing social network influence the formation of collaborative learning networks in a DLC?

3. RQ 3: Do existing social networks function to constrain or facilitate emergence of new collaborative networks and social capital?

2.3 The effect of a pre-existing network on CSCSN

With regard to the second research question, previous CMC literature suggests somewhat contradictory predictions on this question as follows. On one hand, early CMC studies generally held that CMC would liberate people from traditional social and physical constraints, fostering the formation of new, heterogeneous social relationships in CMEs. Researchers predicted that the use of CMC would radically reconfigure social relationships and structures, because anonymity, reduced social and contextual cues, and increased connectivity through CMC would make it easier for people to find social links and ties across hierarchical, social and organisational boundaries (Sproull and Kiesler [50] and [51], Jones [32]). CMC researchers also predicted that the advent of computer networks and applications would dramatically expand the geographical and temporal boundaries of interpersonal or small group communication, which would lead more frequent and heterogeneous information and resource exchange between weak social ties (Constant et al. [16]) and increase diversity of strong ties across organisations (Lievrouw et al. [37]).

The above discussion emphasises the role of CMC technology as an active agent helping members of DLCs form increasingly diverse and heterogeneous social and work relationships. Accordingly, pre-existing networks will minimally influence the formation of collaborative learning networks, as participants quickly and easily develop new 'network capital' due to the liberating effect of CMC technology (Wellman et al. [57]).

On the other hand, recent CMC studies suggest that a pre-existing network should have moderate or perhaps strong effects on the formation of emerging CSCSNs. Evidence suggests 'electronic links primarily enhance existing interaction patterns rather than creating new ones' (Bikson et al. [7], p. 102). Child and Loveridge ([12]) find that CMC is designed precisely to support ongoing hierarchical relations. More recently, Wellman et al. ([57]) found that people's interaction online supported pre-existing social capital by supplementing face-to-face contacts.

To summarise, previous studies on CMC have shown that it is unclear whether and how the effects of pre-existing social network manifest in a DLC setting. Some studies predict that the emergence of CSCSNs in a DLC is free from history as CMC liberates members from previous social, hierarchical, and relational boundaries and constraints. Others suggest that a pre-existing friendship network should have very strong influence on the formation of CSCSN. Hence we ask:

1. RQ 2a: How do CSCSNs evolve in a CME? Do existing social networks significantly influence the emergent patterns in CSCSNs?

2.4 Pre-existing social network and development of social capital

In addressing the third research question, we examine whether a pre-existing social network and network ties constrain or facilitate individuals' ability to maintain, refresh and activate their social capital in a new learning and work environment. A review of literature suggests contradictory predictions.

Studies on social capital generally assume that individuals with diverse social networks are capable of building greater social capital over time by utilising their existing network relationships. Social network studies also stress that an actor occupying central positions in a given social structure obtains a number of advantages including efficient access to new relational partners (Gulati, [26]). For instance, friendship network ties in organisations help people form new partnerships by providing access to information on the availability, competencies and reliability of potential partners, thus lowering searching costs and alleviating the risk of opportunism (Jehn and Shah [31], Dirks et al. [17]). These studies suggest actors occupying positions with high centrality in a pre-existing network should realise potential benefits of a DLC earlier than others, because they have greater accessibility and the flexibility to access different social groups than actors structurally located on the periphery of a network. A pre-existing friendship network thus is an enabler that helps people quickly form new links and ties, moving deep into emerging structures of a DLC. In this case pre-existing social ties are valuable relational assets for effective information exchange and knowledge gain in a learning community.

Conversely, researchers have found that strong network ties can function as a social liability, especially in dynamic environmental conditions like DLCs (Leenders and Gabbay [36]). In such cases, pre-existing social ties constrain one's ability to rejuvenate their network composition, which increases the ability to adapt to changing conditions. Gargiulo and Benassi ([22]) suggest that these negative effects operate in two ways: limited resources and relational inertia.

First, actors holding many social relationships are less apt to develop new relations because maintaining such pre-existing ties takes up a significant portion of their limited time, energy and emotion. Investment in social capital is substantially more complex than investments in human capital (Coleman [15]). While a person can typically acquire new skills without having to discard previous ones, the same is not true for social capital. Since people have limited resources, pressures to maintain pre-existing relationships may hinder the ability to cultivate other relationships necessary to refresh their social capital. That is, unbalanced investment or overinvestment in pre-existing social capital can transform potentially productive assets into constraints and liabilities (Garguilo and Bernassi [22], Leenders and Gabbay [36]). Obligations and expectations for strong, long-lasting relationships may prevent a person from realising greater economic opportunities by constraining the search for, and development of, new trading partners (Granovetter [25]).

The second mechanism is relational inertia. People get used to dealing with their long-term partners. Individuals tend to keep strong ties with the same group of people (Quinn et al. [44]), and take similar positions and roles even when they belong to multiple social networks (Rice [45]). Similarly, research has shown that information exchange is heavily influenced by pre-existing friendships and personal contacts. People tend to be motivated to share information, and provide each other with early, frequent access to resources available within their initial social circle (Granovetter [24], Krackhardt [34]). In other words, people seek information that is the most easily accessed (such as asking co-workers), rather than searching for the best information (O'Reilly [41]).

Taken together, the above discussion suggests that people holding strong ties in a friendship network might be less motivated or able to explore new links and ties, as they might be satisfied with existing social networks or constrained by social liabilities. This is especially detrimental for knowledge building and learning in DLCs because positive outcomes are expected to emanate when heterogeneous actors share their diverse resources by engaging in shared practices.

In summary, previous literature suggests that pre-existing networks should operate as either a relational asset or social liability, facilitating or constraining the formation of new collaborative learning networks. Questions remain regarding how pre-existing social networks influence individuals' embeddedness in DLCs. Hence:

1. RQ 2b: Will a pre-existing social network constrain or enable the way individuals create, maintain and activate their new social capital?

To answer this question, we measured individuals' initial position in a pre-existing social network and tested whether these initial structural proprieties enable or constrain their ability to renew or expand social capital as they join in a collaborative learning environment. Because of the conflicted findings from previous literature, we do not specify predictions about the direction of association between a pre-existing social network and emerging CSCSN, but we do propose and test the following hypothesis regarding the magnitude of association in quality.

1. H1: Learners' initial structural positions will be significantly associated with the degree to which they develop new social ties and move into different social groups in an emerging CSCSN.

3 Methods

3.1 Study site and data collection

The data for this study were collected from a multi-year CSCL/CSCW research project. The goal of the project was to develop the capability for individuals at distributed geographic locations to interact effectively on development of future aerospace systems. A distributed engineering design class was co-hosted by engineering schools located at two universities. The two universities are separated by about 55 miles distance.[1] Thirty-one senior and graduate level students from two universities enrolled in a year-long design course (14 from University A and 17 from University B). Among the students, 23 were male (Univ. A = 9, Univ. B = 14) and eight were female (Univ. A = 4, Univ. B = 4).

The course emphasised distributed teamwork and collaboration at the group level, as well as collective learning and knowledge construction at the learning community level. As a means of achieving this goal, a web-based collaboration and communication system called the Advanced Interactive Discovery Environment (AIDE) was developed. The AIDE is a web-based portal providing a suite of integrated tools including simulation, application sharing, communication, networking, information retrieval, custom information storage, as well as instructor-provided material. The communication tools include realtime audio/video (AV) conferencing, chat and instant messaging (IM), email and discussion boards.

One key feature of the project was that distributed teams, consisting of students from two distributed locations, had to work closely together to design a future aerospace system. The group task focused on the design of the structural subsystem for the next-generation space shuttle, called a reusable launch vehicle (RLV). As members of the collaborative distance design team, students focused on materials and structure issues, as well as on thermal control and thermal protection. To create a full multidisciplinary experience, NASA engineers interacted with the class in teams addressing such disciplines as propulsion systems, hydraulics, aerodynamics, human factors and cost analysis. The task was highly interdependent, cooperative and multidisciplinary in nature. To create effective designs, students had to be aware of the overall system engineering but at the same time needed to cooperative and collaborate with other as each group member had to specialise in one area. In the first semester, the students considered alternative designs for elements and systems of the RLV. In the second semester, a detailed design was made, with virtual manufacturing, construction and testing. The course ended with a presentation to NASA.

The AIDE provided public social and knowledge space through which distributed students freely exchanged ideas and suggestions via email, IM and online discussion boards. Team level achievements were frequently posted on the shared web space so experiences, ideas and knowledge could be exchanged across the boundaries of design teams, classes and universities.

3.2 Measures

3.2.1 Social networks

We examined how community social infrastructure (CSCSN) emerged in this collaborative learning and work environment (RQ1) and the degree to which this emergent CSCSN was influenced by a pre-existing face-to-face (FtF) social network (RQ2). Social network data were gathered three times. The first survey was administered in the second week of the first semester measuring a pre-existing network. Students were asked to look carefully at the class roster and indicate up to five persons they most frequently communicated with, and how often during a typical month. Considering this network was measured before they participated in group aspects of the design project, and students belonged to the same departments or schools for years, it is assumed that the reported relationships were pre-existing friendships rather than any other type of instrumental relation.

In a subsequent data collection (at the end of the first and second semesters), students were asked to report names of people they talked to for two specific functions—information exchange and social support. Information exchange refers to communication about class, coursework, or design projects; social support refers to non-instrumental communication, or interactions that primarily provide social support and socialisation (Ibarra and Steven [30]). Two different types of networks are distinguished, considering multiple types of interaction networks coexist within the same organisation or community (Ibarra and Steven [30], Nardi et al. [40]).

3.2.2 Initial network positions

For RQ3 and H1, we examined whether pre-existing ties facilitated or constrained the way individual learners developed new social capital when they engaged in a distributed learning community. To test this hypothesis, an individual's initial network positions in a pre-existing network were measured in order to predict subsequent individual level social capital development (see below for the measures). Individual network positions were measured by two network centrality measures. Betweeness measures the frequency with which an actor falls between other pairs of actors on the shortest or geodesic paths connecting them (Freeman [21]). The higher the betweeness score of an actor, the greater the likelihood that actor serves as a structural conduit, connecting others in the network. Degree centrality refers to the number of social network ties that an actor holds in a given social network (Freeman [21]). High degree actors are the most active and strategically advantageous in the sense that they have the most ties to other actors in the network (Wasserman and Faust [56]). These network variables represent an actor's structural advantage (connections to other people) and disadvantage (liabilities) in terms of developing new social capital in a DLC.

3.2.3 Social/network capital

Two variables were developed to measure the extent individuals develop their social (network) capital in the emergent CSCSN. First, change propensity measures the degree to which an individual added new contacts in his/her ego network. An ego network is an individual level social network consisting of network partners directly connected to a given individual. For instance, if person 'x' initially reported a, <bold>b, c</bold>, d as her/his interaction partners in phase I (pre-existing network) and then reported <bold>b, c</bold>, f, g, t in phase II (information exchange network), then the change propensity for this particular individual in phase II is 0.6 (3/5). This variable measures how actively an individual acquired new relational resources and assets in his/her social circle.

Second, we counted how many cliques an individual belonged to in the final phase of network development. Clique membership represents how actively an actor accessed diverse social circles. A clique is defined as any group of at least three actors for which all pairs are adjacent to one another. Using the clique algorithm in UCINET V (Version 1.0; Borgatti et al. [9]), we identified 19 cliques in this collaborative learning community. On average, each student belonged to 2.34 cliques (minimum = 1, maximum = 5). It is assumed that those belonging to a number of different cliques (or in other words, those who are holding relationships spanning across multiple social circles) tend to have structural advantages in that they maintain diverse information resources otherwise unavailable to out-group members. As such, clique membership measures the extent to which an individual explored and moved around different subgroups emerging in a DLC. To distinguish clique membership from friendship networks, cliques were identified via the information exchange network in Phase III.

Given that collaborative learning and knowledge construction take place through shared practice and social interactions among multiple actors, the two variables represent the extent to which an individual explored and utilised new social resources as s/he participated in a new learning or work environment, such as a DLC.

3.3 Analysis

To test the influence of pre-existing friendship networks on the emergent collaborative networks (RQ2), the association between the pre-existing and emergent networks was measured. It was assumed that if the two social networks at two different timeframes (for example, the pre-existing network and the information exchange network in phase III) were highly correlated to each other, the internal characteristics of the first one significantly remained in the following network, indicating that the former significantly influenced the formation of the latter. The significance of the association between two social networks was calculated using the quadratic assignment procedure (QAP). QAP calculates Pearson's correlation coefficient as well as simple matching coefficient between corresponding cells of the two data matrices. By repeating such calculations thousands of times using random permutations, QAP can test if the observed association between the two networks is statistically significant (see Hubert and Schultz [29], Krackhardt [33] for reviews).

RQ3 and H1 were tested using multiple regression analyses. Each dependent variable was regressed onto two network variables to test whether initial network positions significantly influenced the way individuals developed their social/network capital.

4 Results

4.1 Descriptive results

At first, we examined how students from two distant locations built collaborative work and learning networks over time (RQ1). Figures 1 through 3 visually represent how social networks evolved into a CSCSN as students from two distant locations participated in a DLC. Note that the social networks were measured at three different periods. Figure 1 represents the social network of the class at the beginning of the semester (Phase I: Pre-existing friendship network). At this point students at different locations did not have any interaction, clearly indicated in the diagram. Of 31 students, five were identified as social isolates.

Graph: Figure 1. Pre-existing social network.

Graph: Figure 2. CSCSN in Phase II.

Graph: Figure 3. CSCSN in Phase III.

Figure 2 shows students from both universities formed large communication networks in Phase II. Note that this study measured two different types of social networks—social support and information exchange as described in the network diagrams. With few exceptions, students indicated that they exchanged social support and information with the same partners. Finally, figure 3 shows the network structures of Phase III. Students formed several subgroups, mostly with co-located partners. It is interesting to note that most students did not attempt to interact with remote partners at this point. Inter-group communication only occurred through liaisons connecting those subgroups.

4.2 The effects of a pre-existing friendship network on the formation of CSCSN

This study questioned to what extent a pre-existing friendship network influences formation and change of CSCSNs in a DLC (RQ3). Table 1 reports the results of QAP correlation analyses testing the association between two network matrices at different timeframes. Results show that all five networks measured at different periods are significantly correlated with each other. This is not surprising given that the composition of network members remained constant throughout the study period, except for one participant who dropped the course after the first semester (a2 in the figures). However, it is interesting to note that the pre-existing friendship network significantly influenced the emerging structures of collaborative learning networks (r = 0.17, r = 0.33, p < 0.01 for information exchange networks in Phase II and III, and r = 0.32, r = 0.42, p < 0.01 for social support networks). The results indicate that, despite abrupt changes in communication structures (i.e. convergence of two distributed sub-networks as the students participated in a collaborative learning project) internal characteristics of the pre-existing friendship network remained.

Table 1. Matrix correlation results using quadratic assignment procedure (QAP).

<table><thead valign="bottom"><tr><td /><td>Phase I</td><td>Phase II</td><td>Phase III</td></tr><tr><td /><td>Friendship</td><td>Social</td><td>Information</td><td>Social</td><td>Information</td></tr></thead><tbody><tr><td>Friendship</td><td>&#8211;</td><td>0.322**</td><td>0.171**</td><td>0.425**</td><td>0.332**</td></tr><tr><td>Social II</td><td>0.26&#8201;&#8211;&#8201;0.38</td><td>&#8211;</td><td>0.850**</td><td>0.366**</td><td>0.260**</td></tr><tr><td>Information II</td><td>0.11&#8201;&#8211;&#8201;0.23</td><td>0.83&#8201;&#8211;&#8201;0.87</td><td>&#8211;</td><td>0.214**</td><td>0.164**</td></tr><tr><td>Social III</td><td>0.37&#8201;&#8211;&#8201;0.48</td><td>0.31&#8201;&#8211;&#8201;0.42</td><td>0.15&#8201;&#8211;&#8201;0.27</td><td>&#8211;</td><td>0.861**</td></tr><tr><td>Information III</td><td>0.27&#8201;&#8211;&#8201;0.39</td><td>0.20&#8201;&#8211;&#8201;0.32</td><td>0.10&#8201;&#8211;&#8201;0.23</td><td>0.84&#8201;&#8211;&#8201;0.88</td><td>&#8211;</td></tr></tbody></table>

4225826762 Upper diagonal: Pearson correlation coefficients based on QAP analysis. Lower diagonal: Confidence interval (CI: 95%) using Fisher's z-Transformation method. ** p⩽0.01.

Another interesting finding in this analysis is that the associations between Phase I and III networks (r = 0.34, 0.42) are greater than those between Phases I and II (r = 0.17, 0.32) and Phases II and III (r = 0.16, 0.37). This indicates that the network structures became more similar to the pre-existing friendship network in the final Phase III, than in Phase II. In other words, networks regressed into the initial pre-existing friendship network as time passed. To examine whether the associations between Phases I and III are statistically more significant than any other combinations, Fisher's z transformation method was used to test the statistical significance of the difference between two correlation values. Since the sampling distribution of Pearson r is not normally distributed, r is converted to Fisher's z according to the r to z transformation formula [z = 0.5log[(1 + r)/(1 − r)] for computing the confidence intervals of the given correlation values. The values of Fisher's z in the confidence interval were then converted back to Pearson's r using the equation r = [(e<sups>2</sups><sups>z</sups>−1)/(e<sups>2</sups><sups>z</sups> + 1)]. If the confidence intervals of the different correlation values overlap each other, there is no significant difference between them. The lower diagonal of Table 1 reports the confidence intervals for each correlation coefficient.

As reported, for information exchange networks, the association between Phases I and III is significantly stronger than those between Phases II and III or I and II. For social support networks, the difference between correlation values was not significant. The results indicate that for social support networks, internal characteristics of the pre-existing friendship network remained essentially constant, regardless of development phase of the DLC. On the contrary, for information exchange networks, the influence of the pre-existing friendship on the emergent collaborative networks is less significant in Phase II, and stronger in Phase III. This indicates that members in this DLC, at first, explored and added new information exchange partners in Phase II, but reverted back to their old friends and social circles in the final phase. Note that all correlations reported in Table 1 are significant at the 0.01 level. Therefore, the discussions here refer to relative strengths of associations.

Since this result was unexpected, additional analyses were conducted to confirm the finding. A variable measuring the degree to which an individual changed the composition of his/her ego network partners in phases II and III was created. We counted the number of new network partners a given individual added in his/her ego network in Phases II and III compared to Phase I network (the pre-existing friendship network). Similarly, we also counted the number of friends (identified in Phase I) who remained in an ego's network in Phases II and III. Using those numbers, we tested at what point people kept more friends or added more new contacts in their social circles. Figure 4 shows the results of these comparisons. On average, students kept their old friends quite consistently (M = 2.72 (phase II), M = 2.76 (phase III); t = 0.157, p > 0.05) throughout the study period. However, the way they managed their new social capital differed significantly across time. That is, students substantially reduced the number of new contacts in later phases (M = 3.80 (phase II), M = 2.08 (phase III), t = 6.42, p < 0.01). These results are consistent with the QAP analysis results. People explored and added new social ties in the second phase, but in the final phase they removed those new ties/links from their ego networks. As a result, the final network structure became similar to the initial pre-existing network.

Graph: Figure 4. Comparison of change propensity between Phases II and III.

In summary, the pre-existing friendship network significantly influenced the emerging patterns of collaborative learning and working networks during all time periods, but the magnitude of influence varied over time (less significant in Phase II and more significant in Phase III). This indicates that the effects of pre-existing friendship networks are significant, but also dependent on the developmental phase of the DLC.

4.3 The effect of pre-existing network on individual social capital

For RQ3 and H1, we investigated whether a pre-existing network would facilitate or constrain an individual's ability to develop network capital as they became embedded in the emergent CSCSN. Regression analyses were performed to predict each of the dependent variables (change propensity and clique membership) from two network centrality variables. Note that this study centered the centrality variables in order to correct for the multicollinearity problem given that network variables tend to have high intercorrelations. To check the severity of multicollinearity among the independent variables, we examined the conditioning index and variance proportions associated with each independent variable (see Belsley et al. [4], for a discussion). According to Tabachnik and Fidell ([52], pp. 86 – 87), a conditioning index greater than 30 and at least two variance proportions greater than 0.50 indicates serious multicollinearity. The multicollinearity diagnostics showed that none of regression analyses had serious multicollinearity problems after the measures were centered.

Table 2 reports the results of the regression analyses used to predict three dependent variables (change propensity in Phases I and II, and clique membership). Note that the analyses excluded six people who were identified as social isolates or close to an isolate (those who had only one connection) in the pre-existing network, because computing the change propensity (i.e. adding new members in their ego networks) for these people would be less meaningful and would bias the results of the analyses.[2] As shown in the table, degree centrality displayed significant associations with the dependent variable, change propensity at two timeframes. Note that the direction of the associations is negative (b =− 0.569, p < 0.01 at T1 and b =− 0.410, p = 0.052 at T2). That is, those who had many relational partners in the pre-existing friendship network were less likely to form new ties and links in later periods. Although these centrality variables did not display any significant relationships with the other dependent variable, clique membership, the directions were the same, indicating that central actors in a pre-existing network did not move into different social circles in the emerging social structures. Thus, H1 was partially supported.

Table 2. Regression analyses predicting change propensity and clique membership.

<table><thead valign="bottom"><tr><td /><td><italic>B</italic></td><td><italic>p</italic></td><td><italic>R</italic>-Square</td></tr></thead><tbody><tr><td>Dependent variable: change propensity (T1)</td><td /><td /><td /></tr><tr><td>Degree centrality</td><td>&#8722;0.410</td><td>0.052</td><td /></tr><tr><td>Betweeness</td><td>&#8722;0.135</td><td>N.S.</td><td>0.224</td></tr><tr><td>Dependent variable: change propensity (T2)</td><td /><td /><td /></tr><tr><td>Degree centrality</td><td>&#8722;0.569</td><td>0.008</td><td /></tr><tr><td>Betweeness</td><td>&#8722;0.034</td><td>N.S.</td><td>0.340</td></tr><tr><td>Dependent variable: clique membership</td><td /><td /><td /></tr><tr><td>Degree centrality</td><td>&#8722;0.059</td><td>N.S.</td><td /></tr><tr><td>Betweeness</td><td>&#8722;0.006</td><td>N.S.</td><td>0.004</td></tr></tbody></table>

5 Discussion

This study first examined how social networks emerged over time in a distributed learning environment. Special attention was placed on how pre-existing friendship networks influenced the formation of emergent collaborative learning and work relationships in a DLC. We discovered that pre-existing friendship networks significantly influenced the formation of community level CSCSN, and individual level social capital in an emerging DLC. Results demonstrate that network structures were somewhat fixed, from the start, even though the distributed learning environment strongly encouraged students to form new relational ties using collaborative learning projects and technologies.

The patterns of associations between networks at different timeframes are noteworthy. Network structures became more similar to the pre-existing friendship network in Phase III, in comparison to Phase II. In other words, networks regressed into the initial pre-existing friendship network over time. Intuitively, it was predicted that the effects of pre-existing networks on the formation of emerging collaborative networks would be greater in the initial phase and then would diminish later as people incrementally develop new ties and links over time.

Developing social or trustful information exchange links is time-consuming, and may be especially true in a DLC where members are physically located at different places and faced with high levels of uncertainty. As Uncertainty Reduction Theory (URT) suggests (Berger and Calabrese [6], Berger and Bradac [5]), people develop interpersonal relationships after they reduce uncertainties about each other via ongoing interactions and observations. Members of this DLC were expected to keep their old social circles (friends) for a while, and then expand their networks as they reduced uncertainty. Hence, it was expected that the internal characteristics of pre-existing networks would be less likely to remain in later phases of DLC development.

However, we found that the direction of change was opposite. That is, structural characteristics of the emergent social networks became more similar to (in other words, reverted back to) pre-existing networks as time passed. Although this finding was unexpected, URT assumes that people interact more frequently when they have higher uncertainties about each other and when they expect that they will develop future relationships. In this distributed learning community, for instance, students joined an electronic community with remote participants. Because they had higher uncertainties with the remote participants and they expected future relationships with these people (for example, collaborating on design projects), they might have been quickly involved in frequent interactions to reduce uncertainty about each other. In this case, high-uncertainty may have pushed people to explore new social ties in the early phase of the project. But as uncertainty levels declined via interaction, so did information seeking behaviours. People may have become less motivated to explore or maintain new social and information links. In some cases, they may have either gone back to their old friends if the interactions with new partners had not been satisfactory, or tried to keep a smaller number of network partners for future interactions (combining old and new partners).

Previous research typically tested whether or not people could develop strong interpersonal relationships using CMC channels (Walther [54], Walther et al. [55]). However, they rarely investigated how people maintain CMC relationships for an extended period of time. Findings in this study show that CMC relationships are fragile and task-oriented, rather than resilient. Since the finding is descriptive, future studies would benefit from analysing this research topic with more rigorous research design and measurement.

RQ3 and H1 investigated the effects of initial structural position on development of social capital in emerging networks. Central actors in the pre-existing network were less likely to form new relationships and explore new social worlds. On the other hand, those who were least connected with other members early on quickly formed new links with other members. Central actors in a pre-existing friendship network, by contrast, maintained their old social circles throughout the study period. This indicates that, contrary to our intuition, an actors' centrality in pre-existing networks acted as a social liability that significantly constrained an actor's ability to explore new social contacts and resources. Strong personal relationships had negative impacts on network capital development by 'locking' individuals in pre-existing social relationships.

Many researchers and practitioners highlight the importance of strong interpersonal relationships among community members, since they are a building block of a community (Preece [43]). However, our finding suggests that the qualities that make a community an ideal structure for learning and work—shared perspective, trust, communal identity based on long-standing, strong personal relationships—are the same qualities that can suppress its potential for success. That is, the community can become an 'ideal structure for avoiding learning' (Wenger et al. [60], p. 254). As Wenger et al. ([60]) note, too much intimacy can create a barrier to newcomers, a blinder to new ideas, or a reluctance to critique each other. Granovetter ([24]) also argues that tight bonds tend to create closed social circles and these can become exclusive and present an insurmountable barrier to new social ties.

6 Conclusion and directions for future studies

It is often said that traditional approaches to collaborative learning or cooperative work suffer from too narrow theoretical and methodological orientations which, in turn, have led researchers to examine individuals or task groups, rather than considering the larger social structures and how they constrain or enable individual as well as collective behaviour (Brown and Duguid [11], Nardi et al. [40], Woodruff [61]). By adopting a relatively new analytical tool in this field—social network analysis—we examined the building processes of emergent social infrastructures in a distributed learning community. Also, the study tested whether pre-existing social networks had functional or dysfunctional effects for individual learners in terms of developing social/network capital. Although the above findings are only a first step, they do provide a model for understanding the processes of emergent collaborative social networks in relatively new social systems like DLCs.

As social networks and social capital come to be more important in learning and work places, social network analysis has been increasingly utilised by researchers to uncover the social dynamics of CSCW (Fisher and Dourish [20], McDonald [38]), CSCL (Haythorthwaite [28], Cho et al. [13]), knowledge management (Cohen and Prusak [14], Seufert et al. [48]), information visualisation (Terveen et al. [53], Gloor et al. [23]) and so forth. We believe that this study extended the theoretical and empirical basis of social network analysis in several ways. First, this study explored the building processes of a DLC in a longitudinal context. Social network analysis is fundamentally process-oriented and emphasises the emergent nature of social systems. However, few previous social network studies adopted a longitudinal approach to examine how networks emerge over time, and what influences such processes of change. As a result, social network studies have been criticised for not having a clear explanation about the 'origins of the networks' (Brass [10], Emirbayer and Goodwin [18]). The incorporation of a time dimension in this study enabled the researcher to examine the way two disconnected pre-existing friendship networks evolved into new collaborative learning and working networks, as students from two universities joined design project groups using online collaboration tools.

Further, the study examined whether pre-existing social ties exerted functional or dysfunction effects on the emerging pattern of individual level social capital. While a large body of social network research focuses on the benefits of social capital, the literature on its risks, or negative effects, is much sparser (Adler and Kwon [1]). In general, many researchers posit that social networks can be an important resource providing structural advantages to individuals, such as power and influence in decision-making (Brass [10]), and better performance (Baldwin et al. [3], Sparrowe et al. [49]). However, others increasingly see this position as one-sided (Leenders and Gabbay [36]). Social structures change over time and so do their effects on individuals. Hence, relationships beneficial to goal attainment in the past may thwart goal attainment in the future. By using a longitudinal approach, we demonstrated how the social network that once conferred social capital and relational assets could become a constraint, or social liability, restricting one's ability to adapt to changing conditions.

Overall, the findings presented herein provide practical implications for managers and designers of collaborative learning and work environments. As discussed above, members in this distributed community tended to treat interactions as simple transactions. When topics changed or the task was completed, distant relationships tended to quickly disappear. While the real power of 'intensional social networks' is their ability to quickly form and disintegrate (Nardi et al. [40]), too much refreshment is also problematic in terms of building and sustaining cohesive communities. That is, community members need a meaningful 'sense of shared identity', one that binds members beyond specific exchanges. To overcome the limitation of task-oriented interactions, community members or sponsors may employ team building exercises and socialisation activities. Assuming that members in new DLCs tend to feel disconnected and experience high uncertainty, studies often suggest providing community-building exercises in the beginning phases of community development (Preece [43]). However, the findings of this study suggests that the relationships in DLCs tend to be more ephemeral in the later phase, suggesting that at least an equal amount of attention should be directed toward the later period.

Finally, this study contains several methodological and theoretical limitations, which warrant attention in future studies. This research has positioned itself at a considerable distance from cognitive or motivational research. While the importance of building social infrastructures is clearly identified in this study, these merely create opportunity for collaboration; building a DLC requires not only establishing social ties but also nurturing motivation and providing resources (Haythornwaite [27]). Future research would benefit from a more interdisciplinary approach that links structural and psychological factors, and tests how these combine to affect various processes and outcomes in a DLC. For instance, future studies may test how social embeddedness affects a sense of community membership, and how this cognitive element, in turn, leads to actors' voluntary participation in, and contribution to, the development of collective knowledge.

This study also revealed that social networks could have both functional and dysfunctional effects on individual learners, and the learning community as a whole. While we understand much about market failures and technology failures, we still lack knowledge about how and when a social network or social capital fails (Leenders and Gabbay [36]). An important task for future research is to identify when and how social networks serve different functions using increasingly rigorous research plans. Similarly, this study did not test whether the constraint effect of pre-existing social ties ultimately led to poor outcomes such as poor learning performance. Future research would benefit from directly testing the relationships between network constraints and task coordination failures or performance outcomes.

Finally, the particular sites that this study examined consisted of relatively small, homogeneous learning groups. Having a small sample size is not unusual in social network or longitudinal studies, due to the extreme difficulties gathering rich and complex information. Yet, the small sample size restricts researchers from generalising the findings to broader social settings. Similarly, the two research sites were situated in an educational context; participants were all college students. Although this subject group accounts for a significant portion of learning and knowledge communities, care should be taken when generalising these findings to other settings where social dynamics/structures might be significantly different. For instance, knowledge sharing and collaboration in business firms would be very different from those in an educational setting because of high competition, hierarchical structures, and rules commonly found in these environments. The findings in the current study might be further validated or modified when researchers test the structural model of this study in different contexts.

Acknowledgement

The authors gratefully acknowledge the support of NASA Langley Research Center, through Cooperative Agreement No. NCC-1-01004. Additional support was provided by the State of New York and the AT&T Foundation.

By H. Cho; J.-S Lee; M. Stefanone and G. Gay

Reported by Author; Author; Author; Author