Bodong Chen

Crisscross Landscapes

Notes: Contractor et al. (2011). Multidimensional networks and the dynamics of sociomateriality



Citekey: @Contractor2011-vc

Contractor, N., Monge, P., & Leonardi, P. M. (2011). Multidimensional networks and the dynamics of sociomateriality: bringing technology inside the network. International Journal of Communication Systems, 5, 39. Retrieved from


Summarize: Fascinating work on multidimensional networks—multi-mode, multi-plex, dynamic networks (akin to Carley’s dynamic network analysis). This paper starts by talking about sociomateriality—arguing that tech and human are inseparable, so they should be kept in a same network. It then introduces a typology capturing a spectrum of networks starting from unidimensional (traditional) social networks and moving towards fully multidimensional networks. It presents an interesting empirical case (first analyzed using ethnographic approaches), and move onto discuss structuring signatures and use of ERGM/p* to test theories of multidimensional networks.

Reflect: This line of work shares similar motivations with DNA. A bit surprising they do not reference each other. This work is quite mind-opening for my ongoing work on KB discourse. Many ideas emerged from reading this article. Key ones include: discourse as a multidimensional networks; looking into sociology theories and their structural signatures that are applicable to CSCL; applying ERGM to test theories about CSCL (at both macro and micro levels, about both static and dynamic networks).


This article explores the theoretical implications of developing multidimensional social networks that include nonhuman technological elements. Using ideas from actor- network theory and sociomateriality, we develop a typology for multidimensional networks that includes multiple kinds of nodes and multiple kinds of relations. (p. 1)

exploration of how to develop multidimensional, multitheoretical, and multilevel models that include technological artifacts and relations. (p. 2)

By the late 20th century, researchers had accrued a good deal of empirical evidence to support hypotheses made in the 1950s and 1960s about the various ways that new technologies would alter social dynamics. Within formal organizational settings, for example, research by Barley (1990), Burkhardt and Brass (1990), Contractor and Seibold (1993), and others demonstrated the subtle, nuanced ways through which newly implemented computer-based technologies alter the flow of communication within networks and, hence, allow people to reconfigure formal organizational structures, decision making, and power relationships. (p. 2)

At the same time, scholars became interested in reversing the causal arrow, asking whether established networks could influence the effects of newly implemented technologies. A number of researchers showed, convincingly, that network dynamics could, indeed, shape what people thought about a new technology, as well as whether and how they would use it (Fulk, 1993; Karahanna, Straub, & Chervany, 1999; Kraut, Rice, Cool, & Fish, 1998; Rice & Aydin, 1991). (p. 2)

All three sets of studies share the ontological position that technologies exist separately from people’s social networks. In other words, they treat either technologies or networks as exogenous forces that impinge upon the functioning of the other. (p. 2)

this is getting really interesting. So signs embodies culture. And humans are essentially inseparable from tech. (p. 2)

In such situations , identifying what technology “causes” a particular network change becomes less intellectually interesting, because a given technology is often used interdependently with a wide swath of other pervasive and embedded technologies (Bailey, Leonardi, & Chong, 2010). Similarly, as technologies begin to store greater amounts of information that were once only held in the heads of people, individuals begin to “use” technologies in much the same ways that they “use” coworkers and friends (Su, Huang, & Contractor, 2010). These technologies are emerging as “social prosthetic systems” (Kosslyn, 2011, p. 182). (p. 2)

Thus, ubiquitous computing, both inside and outside formal organizations, is making it increasingly difficult to separate people’s interactions with other people from people’s interactions with technologies. Consequently, it may make more sense to begin treating technologies as endogenous to social networks, rather than as exogenous to them. (p. 3)

instead of asking how technologies might change social networks (or vice versa, or both), the more appropriate question is, “What happens when a new technology becomes a part of a social network?” (p. 3)

This move implies that researchers can no longer make an analytic distinction between technologies (or artifacts more generally) and people. They must begin to recognize that networks can be comprised of people and technologies, and that both types of nodes may, on occasion, play equivalent roles. Recently, proponents of a sociomaterial approach to studies of technology and communication have begun to provide us with the ontological foundations and theoretical language with which to make this conceptual shift (e.g., Leonardi & Barley, 2008; Orlikowski, 2007; Orlikowski & Scott, 2008; Pentland & Feldman, 2007). (p. 3)

same argument could be made for the context of CSCL. (p. 3)

At an ontological level, a sociomaterial approach to technology and communication suggests that communicative behaviors and technologies are indistinguishable phenomena (Baptista, 2009; Orlikowski & Scott, 2008). As Leonardi suggests: “technologies are as much social as they are material (in the sense that material features were chosen and retained through social interaction) and [communication patterns] are as much material as they are social (in the sense that social interactions are enabled and constrained by material properties)” (2009a, p. 299). (p. 3)

a sociomaterial approach: … asserts that materiality is integral to organizing, positing that the social and the material are constitutively entangled in everyday life. A position of constitutive entanglement does not privilege either humans or technology (in one-way interactions), nor does it link them through a form of mutual reciprocation (in two-way interactions). Instead, the social and the material are considered to be inextricably related—there is no social that is not also material, and no material that is not also social. (2007, p. 1,437) (p. 3)

This sociomaterial approach draws heavily on the work of actor-network theory (Callon, 1986; Latour, 1987; Law, 1987) to stake this ontological claim of symmetry between human action and the actions of technology. Actor-network theory assumes the communicative actions that most social scientists would call “social” involve both people and technologies, and that the material features of a technology are developed and used in a system of social relationships. By abandoning the attempt to distinguish that which is social from that which is material, actor-network theorists consider the actions of humans and non-humans as part of a single network that is, itself, an actor, an “actor-network.” (p. 3)

Although the sociomaterial approach is appealing at the ontological level, it is somewhat problematic at the empirical level, because technologies and communication patterns are relatively easy to distinguish (Edmondson, Bohmer, & Pisano, 2001; Pentland & Feldman, 2008). (p. 4)

Thus, although the so ciomaterial approach provides an important way of thinking about technologies as parts of networks (as opposed to entities that exist independent of networks), this approach does not provide much guidance in specifying how researchers might depict sociomaterial relations empirically in ways that recognize these important differences. (p. 4)

In this article, we argue that making technologies endogenous to networks will offer researchers the ability to begin thinking about networks composed of different types of nodes (e.g., persons, databases, books, etc.), and about where the relationships among these varying nodes also differ (e.g., one might have a friendship relationship with another person, but an information-retrieval relationship with a database). We call these “multidimensional networks.” This approach stands in contrast to the more traditional approaches, outlined above, which treat networks and technologies as objects that exist separately. (p. 4)

That is, any non-human actor, no matter whether it is a policy, a routine, a chemical, or a drug, can be brought, conceptually, inside a social network, just like a technology. In so doing, the resultant network ceases to be a simple social network, and it should, instead, be considered a multidimensional network. (p. 4)

Conceptualizing Multidimensional Networks (p. 4)

unidimensionality is a significant oversimplification of the rich complexity that exists in most social networks. (p. 4)

However, in most instances, any single study typically looks at networks comprised of only one type of node and, at most, a handful of relations among these similar objects. (p. 5)

Unidimensional Networks (p. 5)

Unidimensional networks, sometimes called unimodal, uniplex networks, consist of a single type of object or node and a single type of relation. (p. 5)

There is nothing inherently wrong with unidimensional networks. Decades of work in the social sciences has revealed considerable information about how they operate (Easley & Kleinberg, 2010; Jackson, 2008; Monge & Contractor, 2003; Newman, 2010). But they do have a number of limitations. Perhaps most serious is the fact that they oversimplify reality. (p. 5)

Unimodal Multiplex Networks (p. 6)

Unimodal multiplex networks contain two or more kinds of relations on a single type of node. (p. 6)

Lee and Monge were able to examine how structural patterns in one network influenced structural patterns in the other network. A representative finding here was that “organizations with repeated collaboration in implementation networks are likely to have ties in knowledge-sharing projects as well.” (p. 7)

Multimodal Uniplex Networks (p. 8)

Multimodal networks contain two or more different kinds of nodes. (p. 8)

They can also be represented as a bimodal (or bipartite) matrix, where the two types of nodes are included into a single matrix with rows being the women and columns being the social events or vice versa. (p. 8)

Multidimensional Networks (Multimodal Multiplex) (p. 9)

Multidimensional networks have both multiple nodes and multiple relations, and thus, they are sometimes called multimodal multiplex networks. (p. 9)

Scholarly research on multidimensional networks is rare. One notable exception is the research on the evolution of the biotechnology industry by Powell, White, Koput, and Owen-Smith (2005). Their research examines six different sets of nodes (heximodal) and four different relations (quadriplex). The six types of nodes are dedicated biotechnology firms (DBFs), universities and other research and development firms, government regulators, pharmaceutical companies, venture capitalists, and others. The relations were research and development, finance, commercialization, and licensing. (p. 9)

Between Two (or more) Types of Nodes (Objects). (p. 10)

Instead, their omission reflects a limitation of traditional network analytic methods that were used to study bimodal networks. These methods typically required that relations only exist between different types of nodes (or modes) and not include relations among similar types of nodes. But most empirical contexts involving multidimensional networks would greatly benefit from the ability to depict ties among the same types of nodes. (p. 10)

Recent methodological developments (Robins & Wang, 2011) address the limitations indicated above that had restricted analysis only to relations between different types of nodes and highlight the potential for simultaneously analyzing networks where relations can also exist among nodes of the same type. (p. 10)

Multidimensional networks: Micro and macro variations. Historically, communication and other social networks have been examined from a static, single-point-in-time, cross-sectional perspective. In fact, each type of network described in this typology can be studied from a dynamic perspective (Breiger, Carley, & Pattison, 2003). (p. 11)

Kivran-Swaine, Govindan, and Naaman (2011) show how network structure—strength of ties, embeddedness and status—influenced when people dissolved ties with others on Twitter by “unfollowing” them. Powell et al.’s work, mentioned above, looked at two-mode networks to see how different types of firms (e.g., DBFs and pharmaceutical firms) changed in link structure over time. (p. 11)

all interesting studies (p. 11)

When network scholars have thought about network dynamics, they have tended to focus on the micro level. Research shows that networks grow by adding individual nodes, or links, or both, and that they decline the same way, by losing nodes, or links, or both. (p. 11)

However, rather than focus on individual nodes and links, we can focus on entire modes of objects and types of relations. In this macro approach, we can make networks grow or shrink by adding or omitting an entire mode of objects or an entire type of relation. (p. 11)

And, we could allow networks to grow by adding new types of relations, such as “provides directions to.” (p. 11)

Microlevel multidimensional network dynamics. (p. 11)

Th e types of nodes and the types of relations have not changed from Time 1 to Time 2. (p. 11)

Macrolevel multidimensional network dynamics. Figure 7 provides an example of a dynamic macrolevel multidimensional network. Here, a new mode (or type) of object, avatars, has been added to the network at Time 2. (p. 12)

this is interesting. consider the enter of a drawing in KF discourse? (p. 12)

Summary (p. 13)

In summary, unimodal networks comprise nodes that are all of the same type, while multimodal networks have nodes of different types. Likewise, uniplex networks are comprised of only one type of relationship among the nodes, while a multiplex network includes multiple types of relationships. A multidimensional network is one that is both multimodal and multiplex. Further, dynamic microlevel multidimensional networks are those where nodes or relations of existing types are added or eliminated over time. Dynamic macrolevel multidimensional networks are those where entire classes (or types) of nodes and relations might appear or disappear over time. (p. 13)

ll seven cases in this typology can be studied as dynamic, rather than static, networks. (p. 13)

Table 1. A Framework for Multidimensional Networks. (p. 14)

Empirical Example: How Multidimensional Networks Aid in the Explanation of Sociomaterial Dynamics (p. 15)

Autoworks (a pseudonym) is a large automobile manufacturer. The company designs vehicle systems, like body structures, fuel systems, and powertrains, and it analyzes the interactions among them on a number of parameters. One of the most important of these is how well these systems work together to protect the vehicle’s occupants during a collision. The idea behind “crashworthiness engineering” is that the best chance occupants have of surviving a crash with little or no injury is for the vehicle to absorb the energy of a collision (DuBois, 2004). (p. 15)

Crashworthiness analysts are dependent upon other analysts for information and advice on vehicle design. Analysts are also dependent upon numerous software programs (e.g., pre-processors, solvers, post-processors) to do their work, and those software programs also have computational interdependencies. Given the great number of social and technological interdependencies in crashworthiness engineering work, one would suspect that a small change in one set of relationships would quickly reverberate throughout the work system. (p. 15)

Details of the data collectio n and analysis procedures for this two year ethnographic study can be found in Leonardi’s work (2009b, 2010, 2011, in press). (p. 15)

In late 2006, Jerry finished development work on a small piece of software that he hoped would make this work much easier. The program, which he named “Intruder,” (see Figure 8) was a simple script. After the solver completed its calculations of the simulation model, analysts would now send information from the solver directly to Intruder to render it for analysis without using the pre-processor to make their intrusion calculations. (p. 16)

To determine which points to include as recommendations to the user in Intruder, Jerry consulted Balaji, another analyst who worked in the same department at Autoworks. Balaji was a senior analyst to whom Jerry often went for advice. (p. 16)

Balaji liked Intruder immensely. Analysts at Autoworks shared large cubicles with one another, and Balaji’s cube-mate, Damen, saw him working with Intruder on multiple occasions. After some conversation about its features, Damen asked Balaji if he might be able to secure a copy. Balaji told Damen to ask Jerry if he could try a copy, and Jerry agreed. Damen quickly became a fervent Intruder user. (p. 17)

Up to this point, this case reads like a fairly common diffusion of technology story: Intruder’s use spreads across a social network, and this diffusion is lubricated by various social forces. (p. 18)

These advice-seeking behaviors were highly correlated with proximity and friendship. (p. 18)

Perceptions of expertise were based predominantly on advice-seeking behaviors. (p. 19)

After analysts had used Intruder for two years, many changes were evident in the structure of their advice networks. Advice-seeking about what points to select became decoupled from advice about why one should select those points. Analysts who were both junior and senior stopped asking their colleagues for advice about what and turned their queries toward Intruder. Although Intruder was helpful at aiding analysts in deciding what points to select, it was of no help in instructing analysts as to why those were the correct points in the first place. (p. 19)

The logic of this shift in advice-seeking behaviors was simple: If someone relied on the technology to tell them what points to select, the software’s developer must know why those were the appropriate points. (p. 19)

The dynamics of this case, simple though they may be, are actually fairly difficult to capture with traditional network models. (p. 19)

An observer of th is longitudinal, unidimensional (unimodal, uniplex) network analysis would likely conclude that Intruder disturbed the existing fabric of social relationships in crashworthiness engineering at Autoworks, perhaps supplanting analysts as a source of information. (p. 19)

If we analyze this case using a unimodal multiplex network, we would uncover a very different story than what we saw in a unimodal uniplex network. (p. 20)

Did Intruder simply teach analysts how to select points, and now, they no longer need to seek advice on this behavior? The answer is unclear because technologies are not represented in the model. (p. 20)

An analysis of this case using a multimodal uniplex network (Figure 11) would help to more clearly explain the specific role that Intruder played at Autoworks. F (p. 21)

The story an observer of these networks might tell is one of deterministic technological change—a newer technology replacing an older technology. Due to the uniplex nature of the network, the observer would not know who of the analysts, if any, sought advice from each other about what points to select. (p. 21)

Figure 12 depicts the dynamics of the case using a multimodal, multiplex (or a partial multidimensional) network, in which multiple relations are represented between two types of nodes. (p. 22)

If observers used this partial multimodal, multiplex network to interpret the dynamics of the (p. 22)

case, they would likely conclude that, before Intruder, junior people did not seek advice at all, from anyone or any technology. (p. 23)

Figure 13 provides an illustration of a full multidimensional network for this case. (p. 23)

By examinin g multiplex ties among and between nodes of different types, we can see that advice-seeking practices and attributions of expertise that were directed toward senior analysts before the implementation of Intruder are now split between Intruder and Jerry. Intruder has become the central actor in what advice consultations, and it is seen by analysts as an expert in what knowledge. Jerry has become the central actor in why advice consultations, and he is seen as an expert in why knowledge. (p. 24)

By examining this multidimensional network longitudinally, we can also begin to make inferences about how these shifts in network dynamics occurred. (p. 24)

Summary (p. 25)

This illustration of the sociomaterial dynamics of crashworthiness engineering at Autoworks shows that many of the alternatives in the network typology presented in the previous section fail to capture the richness of these data. As we moved progressively through each analytic alternative, we sought to demonstrate how this narrative can be best understood when the new technology and other non-human artifacts are considered to be part of the multidimensional network, rather than as separate entities influencing and being influenced by the social network. We used the fully multidimensional network framework to understand the structure and dynamics of networks involving different types of nodes (people and technology) and different types of relations both among and between people and technologies. (p. 25)

The rich interplay between people and technology was progressively revealed as we considered a unimodal uniplex representation (Figure 9), a unimodal multiplex representation (Figure 10) or a multimodal uniplex representation (Figure 11), a multimodal multiplex representation (Figure 12) with relations only between nodes of different types (people and technologies), and finally, a multimodal multiplex “fully multidimensional” representation (Figure 13) with multiple types of relations among and between nodes of different types (people and technologies). (p. 25)

Although many authors recommend longitudinal, ethnographic approaches to capture the sociomaterial dynamics of organizing (e.g., Leonardi & Barley, 2010; Orlikowski & Scott, 2008), this type of study is not always possible. (p. 25)

complex dynamic sociomaterial phenomena (p. 25)

KB discourse as such (p. 25)

However, when attempting to identify and explain sociomaterial dynamics at scale, it is helpful to have certain heuristics with which to be able to detect particular patterns within the data. In the next section, we explore these multidimensional representations. (p. 25)

Developing Multi-Theoretical Multilevel Models of Multidimensional Networks (p. 26)

The development of analytic techniques to explain the emergence of networks is often motivated by multi- theoretical multilevel (MTML) models (Monge & Contractor, 2003). These models are multi-theoretical because of a growing recognition among social networks researchers that the emergence of a network can rarely be adequately explained by a single theory. Therefore, these models combine disparate theoretical generative mechanisms, such as self-interest, collective action, social exchange, balance, homophily, proximity, contagion, and co-evolution. These models are multilevel because the emergence of a network can be influenced, for instance, by theories of self-interest that refer to characteristics of actors (at the individual level), theories of social exchange that describe ties between pairs of actors (at the dyadic level), theories of balance that explain configuration of ties among three actors (at the triadic level), and theories of collective action that explain configurations among larger aggregates of actors (at the group or network level). (p. 26)

Of particular interest from an analytic perspective is that each of these theoretical generative mechanisms has a “structural signature” that is unique to that theory. (p. 26)

Recent statistical advances, such as exponential random graph models (also known as p*, see Robins, Snijders, Wang, Handcock, & Pattison, [2007] for a recent review) are able to assess the degree to which these structural signatures are observed in a network, above and beyond what might be expected in comparable random networks. As such, these methods provide the means to test multi-theoretical multilevel hypotheses about mechanisms that explain the emergence of networks (Contractor, Wasserman, & Faust, 2006). (p. 26)

check the reference (p. 26)

Figure 14. Examples of Structural Signatures Associated with Theories of Network Emergence. (p. 27)

Although MTML models have been used to understand the emergence of unidimensional networks among unimodal nodes (such as people or organizations), there is considerable potential for developing and testing new theories to explain the emergence of multidimensional networks. Software tools now exist to test multi-theoretical multilevel hypotheses for cross-sectional and longitudinal partially multidimensional networks—those that are multiplex or multimodal but without relations among different types of nodes (Seary & Richards, 2000; Huisman & Van Duijn, 2005;; Handcock, Butts, Goodreau, & Morris, 2008; Steglich, Snijders, & West, 2006). (p. 27)

This opens up the possibility of inscribing a vast number of new structural signatures that capture the dynamics of how multiple network relations among and between individuals and technologies will enable or constrain their dynamics. (p. 28)

the concept of structural signature is very interesting. No work has been done for CSCL yet. Healthy discourse could have a (or several) structural signature. Food for thought. (p. 28)

The case study described in the previous section offers several provocative examples of structural signatures that might be particularly relevant in understanding the emergence of the multidimensional network depicted in Figure 13. (p. 28)

Cons ider how extant theories of network emergence might be used in the present example. The theory of transactive memory systems developed by Moreland (1999) and Wegner (1995), for example, has been used extensively to explain the emergence of knowledge networks (p. 28)

This structural signature serves to valida te a hypothesis offered by transactive memory theory in a multidimensional network. (p. 28)

However, the above example based on the theory of transactive memory systems also illustrates the potential of extending existing theory from unidimensional to multidimensional networks. (p. 29)

Given that the co-presence of an expertise relation and a retrieval relation might occasionally occur in any random network, the ERGM/p* analysis would likely prove to be statistically insignificant. (p. 30)

In fact, the network at Time 1 in Figure 13 suggests a novel structural signature with potentially interesting theoretical insights. Notice that, though Balaji recognizes the expertise embodied in the government document, he is more inclined to retrieve that information from Jerry, who retrieved that information from the document. This pattern might suggest a structural signature that, when given a choice to retrieve information from a technology or an individual, people are more likely to retrieve the information from other people who have already retrieved it from the document. (p. 30)

Figure 13 also provides several illustrations of how structural signatures can be used to theorize the dynamics of networks—in this case, from Time 1 to Time 2. One focal point is the emergence of the new Intruder software as an important network node. (p. 31)

The introduction of this new technology by Jerry resulted in several individuals within the network adopting the new technology based on contagion, social exchange, proximity, homophily, or friendship ties with Jerry and his colleagues. The adoption of this technology by an individual is represented by the presence of a retrieval link from the individual to the Intruder technology. A structural signature for this adoption based on contagion would suggest that a person would create a retrieval link to Intruder if the person had a friendship link to another person who already had a retrieval link to Intruder. (p. 31)

A similar approach can be used to test the hypothesis that proximity leads to adoption of a new technology. In this case, the structural signature would suggest that a person is more likely to retrieve information from Intruder if they have a proximity tie to another individual who retrieves information from this same new technology. (p. 31)

Finally, structural signatures can also be used to develop new theories about how the introduction of technology, a new node in a multidimensional network, can transform the structure of relations in that multidimensional network. (p. 32)

This modest case study illustrates the distinctive structural signatures that emerge from an analysis of a multidimensional network. Our goal here has been to develop a multidimensional framework, and to demonstrate how it can be used to tease out novel structural signatures that capture the richness missed by unidimensional analyses. (p. 33)

Furthermore, the multidimensional network approach and methods will enable us to develop novel theories, and to more precisely test extensions to existing theories with a degree of inferential certitude that will benefit from, and contribute to, the careful case studies and in-depth ethnographies that have been the mainstay of scholarship based on actor-network theory and sociomateriality. (p. 33)

In summary, there are three reasons for the development of MTML models of multidimensional networks. First, articulation of new multidimensional structural signatures will enable us to examine unique network patterns. Second, it enables us to empirically investigate the extent to which multiple structural signatures (which, in turn, reflect multiple logics of attachment) might simultaneously be considered in understanding the emergence of the multidimensional networks. Third, it opens the possibility of attempting to understand the emergence of multidimensional networks where there is a large corpus of digital data tracing the network relations within and between human actors and technologies. In such cases, MTML models can be used to posit and detect structural signatures that have previously been proposed or tentatively identified using theory or ethnographies. (p. 33)

The promise of this approach is further accentuated by the potential of applying these frameworks to address interesting and intriguing questions about the emergence of multidimensional networks that are on the scale of the Web (Lazer et al., 2009). The increasingly easy access to large amounts of multidimensional network data from the Web within the past decade make this challenge eminently addressable. Specifically, the recent exponential growth in the development and utilization of the Semantic Web (especially the Linked Open Data initiative) offers considerable promise in capturing, collating, and reasoning about large-scale multidimensional networks (Berners-Lee et al., 2006; Hall, this special section; Shadbolt, Hall, & Berners-Lee, 2006). (p. 33)