Bridges, brokers and boundary spanners in collaborative networks: a systematic review

The 24 articles that made up the final literature set were from a wide range of settings and contexts: from Italian Television production teams to Australian hospitals, Scandinavian telecommunication companies to US-Indian engineering projects. This enabled us to describe a rich picture of the brokerage phenomenon. Table 4 summarises the articles’ context, methods and findings on bridging, brokering or boundary spanning positions. More articles were generated in the last six years: 18 articles published between 2005 and 2011, while only six were from the decade 1994 to 2004. Nineteen of the 24 articles cite Burt’s seminal work on brokerage roles, Structural holes[5], and articles written in the last decade also cite Brokerage and closure[6]. Three studies not referring to Burt’s work are focused on boundary spanning and draw on different theoretical antecedents such as work by Tushman on boundary roles in the innovation process [17], Friedman and Polodny’s work on role conflict among boundary spanners during labour negotiations [14] and Allen’s work on ‘technological gatekeepers’ in R&D laboratories [33].

Levels of analysis differed. While the majority used individuals as the nodes describing relationships, two studies were at the interorganisational level looking at collaborations across an inter-firm alliance (the firms were the nodes). Eleven studies looked specifically at the interpersonal relationships within well-defined teams or project groups of less than 12 members. Some studies aggregated individual data from the teams to a single team score.

Methods of data collection varied with ten using surveys and seven using archival documentary data such as patents, emails or personnel records to record relational ties and attributes of the participants. Ethnographic methods such as observing interactions and interviewing participants were used in four studies. The majority of studies used regression analyses to test the impact of their chosen variables on brokerage. For example, Susskind and colleagues [51], testing the variable “team performance,” were able to ascertain that overall team performance in their study was negatively associated with more brokerage activity. Fifteen studies used SNA to compute network characteristics.

There was a marked difference in the way bridging, brokering or boundary-spanning positions were identified. Seven studies used Burt’s network constraint measure (or a measure derived from it), [5] which is used to reflect brokerage potential around a particular actor. It varies with the size, density and hierarchical nature of the network. High constraint means the actor’s egonet (the ties of that actor alone) has few nonredundant contacts, while low constraint has many. Eight studies used the algorithmic parameter betweenness centrality [53] that measures the extent to which an actor lies between other actors that would not otherwise be connected (that is are nonredundant). Seven studies used straight frequencies of external or bridging ties and four used direct observation or self-assessment by actors and their peers or supervisors.

We found four main themes addressed in the set of 24 articles. There were four studies that took the benefits of brokerage as a given and sought only to confirm brokerage opportunities in a particular network, or identify individuals with high brokerage potential [39, 41, 43, 44]. All four used betweenness centrality to identify the brokerage positions and the intention was to maximise efficient knowledge transfer, co-ordinate effort or to ensure the inclusion of people on the periphery. All four were based in a hospital or health promotion setting.

There were 12 articles that sought to define the network conditions and contexts in which brokerage facilitated the generation and integration of innovative ideas [6, 29, 30, 34, 35, 37, 40, 42],[46, 48, 49, 52]. The interplay of brokerage opportunities and network cohesion in this process of innovative performance, referred to by Burt [6] as brokerage and closure, is the main focus of seven of these articles.

Schumpeter’s [54] concept of innovation as the recombination of diverse understandings and knowledge is often cited to explain how brokerage can generate innovation. Burt [12] called it the “vision advantage:” the ability to select and synthesise different viewpoints and approaches from different, unlinked groups on either side of a structural hole. Hargadon and Sutton’s [42] ethnographic study of design engineers at IDEO, a US design firm, explored this process. The researchers saw the company itself as the broker embedded in a broad network of industries where there were gaps (structural holes) in the flow of technological knowledge. IDEO’s product designers deliberately brokered those gaps by looking for technological solutions from one industry that could be applied successfully to another, often resulting in an innovative, recombinant solution. For example, they described how the project team looking for a better design for a surgical skin stapler considered powering it with a gas engine from a model aeroplane.

“Good ideas” may involve declarative or factual knowledge, as well as tacit knowledge that grows up around highly specific contexts. However, while networks rich in structural holes are ideal to generate innovative ideas, they are the least well suited to integrate those ideas. The diffusion of innovations literature suggest the optimal network structure for spreading factual and especially tacit knowledge is cohesion (high density) not sparseness (more structural holes) [55]. These two competing aspects are referred to as the “idea problem” and the “action problem” [6, 30]. Seven authors show by their results that a combination of direct and indirect ties, in the form of closely knit teams linked by sparse ties is the optimal structure for generating and producing innovation [29, 30, 34, 35, 37, 46, 49] with evidence that increasing the number of structural holes eventually becomes counterproductive [34].

Brokerage strategies are considered by Obstfeld [30] in the context of innovation generation and integration. Burt [5] drawing on the earlier work of Simmel [27] described the brokerage strategy of tertius gaudens, literally “the third who enjoys.” In tertius gaudens the broker co-ordinates two distant parties who are not intending to meet, or else actively maintains or exploits the two parties’ separation. Tertius iungens, literally “the third who joins,” was presented by Obstfeld [30] to describe the alternative strategy where the broker deliberately introduces or facilitates ties already present between two parties, either staying to further facilitate the role or stepping away. Burt [12] showed that while a tertius gaudens strategy across structural holes led to good ideas, it did not guarantee the wider involvement of managers which would lead to the integration of those good ideas into organisational practice. Obstfeld [30] found that a tertius iungens orientation, social knowledge (who knows what in the team, where to find resources) and network density were all independent predictors of innovation involvement. Lingo and O’Mahoney [29] examined the process of innovation generation and integration by observing and interviewing music producers in Nashville and showed that producers use both strategies at different times to benefit the network’s outcomes: using a tertius gaudens strategy to look for song ideas across a wide range of individuals and tertius iungens to resolve team differences and to provide the team with the necessary expertise to pull it all together.

Heterogeneous knowledge held by members of a network is seen as an alternative source of novel information; diverse network content being as effective in some situations as a diverse network structure. Rodan and Galunic [48] found that the range and depth of knowledge and expertise held by members of their network and their network position (measured in terms of network sparseness) were independently significant for overall performance (including innovation performance) among managers. When innovation alone was considered, network sparseness failed to be significant when knowledge heterogeneity was added to the model. As networks became denser (and the number of structural holes and brokerage potential fell), knowledge heterogeneity still remained a significant predictor of innovation.

Information environments are obviously important, a point explored by Aral and van Alstyne [35]. Their model, the “diversity-bandwidth trade-off” states that the benefit of access to novel information from a structurally diverse network (rich in structural holes) is tempered by the resultant lower flow of information possible through these weak, bridging ties (the ties’ bandwidth). Stronger ties have a broader bandwidth and so carry more information, but overall the effects of the trade-off depend on the knowledge environment the actor occupies. Strong ties with actors holding heterogeneous knowledge would be more beneficial than weak, bridging ties in a context where the state of knowledge is fairly constant. However, if the knowledge environment is rapidly advancing, with new information appearing all the time, or one in which network members have overlapping knowledge sets, actors are better served by more bridging ties. The authors find support for their model in a study of email contacts between recruiting executives.

When the structural hole lies between two groups who could benefit from a transfer of knowledge, the actor in the bridging role may be thought of as a knowledge broker. Two main points were raised in the literature here. Firstly, performance of work teams is enhanced when knowledge is brokered across boundaries [28, 56]. Di Marco and colleagues [28] consider a specialised knowledge broker role which they call the cultural boundary spanner in their ethnographic study of two US-Indian engineering teams. Given the rise of collaborative, multinational teams they see the potential for conflict between team members arising from differences in language, education and training. Culture here is seen as a boundary that needs to be spanned in order to increase understanding and team performance. Di Marco and colleagues found that the team with a nominated boundary spanner (an Indian expatriate that had trained in the US) was seen to clear up mismatches in the members’ knowledge systems, such as different measurement conventions and unfamiliar terms, resulting in a more collaborative team.

Secondly, there are various costs and negative sides to the brokering of knowledge [25, 36‐38, 45, 51]. When actors in a work group had to go through their leader to get information or advice from another member, that is, if the leader needed to broker the transmission of knowledge between work group members, productivity and efficiency of the work group suffered [25], team conflict increased and team viability decreased [36]. Individuals perceived a personal cost to this brokerage role which was decreased when the role was spread across the whole team [45].

One paper in our final set of articles addresses trust and brokerage. This is a common theme in brokerage theory in the wider social network literature and is discussed at length in Brokerage and Closure[12]. Marsden [20] defines brokers as “intermediary actors who facilitate transactions between actors lacking access to or trust in one another.” In other words, a broker must be seen as a trustworthy intermediary by the two being brokered. Luo [26] looked at the effect of one’s network structure on one’s perceived trustworthiness within Chinese and Taiwanese technology firms. He found that actors who are go-betweens and bridges in collaborative advice networks are seen as being more trustworthy than people who are not. Social capital theory would further argue that the power and benefits of brokerage (having access to unique sources of information or resources) would be lost if the broker proved untrustworthy [12] so there is social pressure on the broker to maintain trust. Luo [26] also hypothesised that bridges may increase cooperation and general trust in the company through their perceived trustworthiness but this was not supported by the data.

There were a range of brokerage parameters (e.g. constraint, betweenness centrality, boundary spanning ties) and methodologies (e.g. archive mining, interviews, observation) used in the studies but all were seen as valid empirical means of identifying brokers. Most of the studies that used SNA took a structural approach to brokerage by evaluating actors’ relational positions within their network and the opportunities and constraints for brokerage behaviour those positions gave [57]. In contrast, qualitative studies identified brokerage behaviour, observing how actors were using their position within the network [58]. Both approaches are based on the relationships surrounding the actors of interest and the key feature of that actor lying on the pathway between two other unlinked actors.

Another notable difference between studies was how relational data was collected. Objective methods such as counting emails or patent data, or observing actual interactions contrasted with more subjective surveys and interviews reliant on respondents’ perceptions and recall of interactions. Our decision to include widely different methods of data collection arose from the recognition that brokerage as we defined it - one actor linking two unlinked others - operates basically the same within any network. However, to ensure subjective methods were appropriately managed, all the studies were assessed for quality (e.g. checking accuracy of self reports by considering reciprocity). Low response rates for whole network surveys and studies that did not discuss how they maximised data accuracy were not included.

A hole, says Burt [6], is an empty space and actors at the edge of a group viewing that hole may be unaware or uninterested in what is on the other side. Various group processes such as homophily (the tendency for actors with similar attributes or tasks to be linked together) and repeated interaction, strengthen the bonds between actors in a group and increase its introspection. Brokers, he argues, can see the value of bridging that hole. Burt [12] showed in his study of managers in a large electronics firm that there is a personal value to brokering. Managers whose ties bridged more structural holes were better paid, received more positive job evaluations and were more likely to be promoted, a tangible manifestation of the broker’s social capital.

The value to the network that brokers bring by crossing that hole or boundary is also considerable. They can generate innovative ideas and increase the quality of creative work. Brokerage can mediate and resolve conflict, make advice and knowledge more accessible, and can act synergistically with network cohesion and strong ties to produce environments in which collaboration can flourish.

The concept of a cultural boundary spanner [28] suggests other settings in which a boundary of language, knowledge systems or expectations may need to be deliberately mediated. Gray [15] refers to brokers as conflict resolvers in transdisciplinary teams such as those working collaboratively to translate scientific health discoveries into clinical practices. The boundary of mismatched knowledge, paradigms and experience between scientists and clinicians is best brokered by a clinician-researcher with experience of both worlds [59]. As well as pre-empting conflict by knowing what each side does not know about the other, their familiarity with both sides increases trust and collaboration.

The costs of brokerage are that bottlenecks in information flow may form at the broker who risks being overloaded and stressed by others’ reliance on them. There may also be a decrease in productivity as the “vision advantage” of a team high in brokerage is tempered by the cost of a dispersed focus. In addition, individuals must also bear the costs involved in maintaining bridging ties. Actors outside your cluster are likely to be different to you: involved in different work, located somewhere geographically distant or from a different profession. Since similar actors find it easier to communicate and predict one another’s behaviour, trusting ties are easier to form and maintain [60] and bridging ties require more work. Bridging ties are also harder to keep viable over time and were shown by Burt [61] to decay faster than ties to actors within one’s own cluster. Moreover, bridging ties have a short shelf life with time rendering many bridging ties and the information they broker as obsolete. Soda and colleagues [49] showed that old bridging ties in the Italian television production industry are not effective for generating innovative ideas since their usefulness is so dependent on the ever-changing context of the industry. Negative network outcomes arising from bridging or brokering ties were the potential to hoard or distort information, bottlenecks in the flow of information, and individual role overload, all resulting in a decline in overall network efficiency [62]

Rangachari’s study [47] on the positive effect of knowledge brokering on the quality of hospital coding led directly to a later paper where she suggests [56] an optimal model for such a knowledge sharing network. Her comparative models varied the three network measures of brokerage, density and hierarchy across a network made up of a management group and two or more subgroups of different professionals. She argued that the optimal network to produce performance outcomes is one high in brokerage and hierarchy with some density as seen in the high quality hospital coding teams. This model lies between the two extremes of a network in which the management group brokers across unconnected groups (high brokerage and hierarchy but low density) or one in which all teams and managers interact (low brokerage and hierarchy but high density).

Brokers are key players in the sense that they can be vital to the integrity and viability of the network. The importance of the pharmacist as a knowledge broker was revealed by Creswick and Westbrook’s study [39] of a hospital ward’s medication advice-seeking network. Likewise, identifying the bridges and brokers enabled a health promotion initiative to increase its likelihood of success in a high school network, ensuring that the message could reach right to the periphery of the network [43].

While in some circumstances brokering across structural holes may not be as efficient as being embedded in an environment rich in heterogeneous knowledge this may be dependent on the age of the network. March [63] puts forward the idea of the difference between exploration and exploitation as a firm’s strategy: exploitation uses in-house knowledge (akin here to the network rich in heterogeneous knowledge) while exploration looks elsewhere (across structural holes) for ideas. March argues that in the early stages of a firm’s (or network’s) development it may be more important to exploit the knowledge within than seek it elsewhere. This also ensures that within team ties are also fostered, building cohesion among people still learning to work with one another. Hargadon and Sutton [42], drawing on organisational memory theory [64] talk of the importance of storing and being able to retrieve knowledge acquired earlier or brought in by network members. Without an adequate process for this retrieval, the knowledge remains locked away and cannot be integrated. The authors speak of brainstorming sessions and the scheduling of deliberate opportunities to explore this mental archive among the designers; a process facilitated by their strong, cohesive ties. Cross and colleagues [65] also stress the importance of knowing “who knows what” in a collaborative network in order to maximise the use of in-house intellectual capital.

Two work group studies had contrasting results from their assessment of brokerage activity on team performance that may be explained by the nature of their knowledge contexts. Marrone and colleagues [45] looked at the benefit of boundary spanning behaviours in 31 consulting teams of MBA students who were on work experience. Susskind and colleagues [51] assessed boundary spanning behaviour in 11 project teams that included senior academics in an alliance of six major hotels and 11 hospitality based university programs. Boundary spanning improved team performance in the former and decreased performance in the latter. For Marrone’s students, this brokerage activity included advice-seeking from faculty members, and in the context of their lack of experience and advanced knowledge was effective in improving team performance. On the other hand, Susskind’s academics already had all the knowledge, expertise and experience needed within the hand-picked project team. Where a team has adequate heterogeneous knowledge held between its members, boundary spanning to outside contacts becomes a distraction which negatively affects team performance.

This paper argues that a knowledge of brokerage roles and their ability to improve connectivity and function across disparate groups has wide implications for healthcare. Studies in healthcare settings have shown how SNA can reveal patterns of communication between groups – both connections and gaps - and highlight key actors and areas for intervention [66]. Heng and colleagues’ [44] examination of the structure of communication between health facility management departments shows the potential a well-placed broker has to increase efficiency, and Rangachari [47] showed how bridging subgroups of professionals increased the quality of coding. The identification of the pharmacist as a key knowledge broker in Creswick and Westbrook’s paper [39] could be used to inform an intervention supporting the role. It also has implications for how medication advice-seeking should be handled after the introduction of electronic medication management systems. Will medication advice be sought as readily from a computer as from a person? Hawe and Ghali’s [43] identification of brokers among high school staff was aimed at maximising the reach and effectiveness of a health promotion intervention. This strategy of identifying bridges could be used across healthcare settings to enhance uptake of new practice guidelines or other initiatives needing to be disseminated widely and has a different focus from similar work looking at the role of opinion leaders in this process [67‐69]. There is also the potential for research identifying bottlenecks in communication flows or instances of information hoarding or inappropriate gatekeeping. Future research on the enactment of brokerage roles in specific healthcare contexts [4, 15] and evaluating interventions to support or introduce brokers will further inform this promising area.

Useful as SNA is to describe communication patterns and identify key actors, data collection through surveys or interviews can involve a significant time investment for already stretched clinicians and achieving a high response rate from mobile staff on rotating shifts can be difficult. This review has shown that qualitative methods such as observation and the use of documentary data are also useful and may be a preferred alternative.