dw2005-305

January 17th 2005 1 The use of Social Network Analysis in Innovation Research: A literature review Fabrice Coulon, PhD Candidate1 Division of Innovation - LTH Lund University, Sweden Table of Contents 1 Introduction................................................................................................................. 1 2 Network analysis......................................................................................................... 2 2.1 Terminology........................................................................................................ 2 2.2 Structure.............................................................................................................. 3 2.3 Network Dynamics or Evolution ........................................................................ 5 2.4 Descriptive Measures.......................................................................................... 6 3 Methodology............................................................................................................. 10 4 Network analysis in innovation research .................................................................. 10 5 Conclusion ................................................................................................................ 15 Abstract The purpose of this paper is to review the innovation research literature which has made an explicit use of social network analysis methodology in order to provide empirical support to innovation theories or conceptual frameworks. The review introduces social network analysis then discusses why and how it has been used in innovation research so far. This paper argues that studies using social network analysis tend to focus too much on change in the relationships between interacting units or nodes of the network to the detriment of change within units/nodes. Therefore, a combination of case study and social network analysis can offer a solution to that problem by providing the best of both methodologies. 1 Introduction Social network analysis (SNA) is an interdisciplinary methodology developed mainly by sociologists and researchers in social psychology in the 1960s and 1970s, further developed in collaboration with mathematics, statistics, and computing that led to a rapid development of formal analyzing techniques which made it an attractive tool for other disciplines like economics, marketing or industrial engineering (Scott, 2000). SNA is based on an assumption of the importance of relationships among interacting units or nodes. These relations defined by linkages among units/nodes are a fundamental component of SNA (Scott, 2000). 1 Fabrice.coulon@innovation.lth.se, Phone: +46 462220248 January 17th 2005 2 Borgatti and Foster (2003) have shown that the exponential growth of the literature in social network research is part of a general shift, beginning in the second half of the 20th century, away from individualist, essentialist and atomistic explanations toward more relational, contextual and systemic understandings. This rapid increase of network research in several disciplines, and in innovation research in particular, has created the need for a review and a classification of studies done in this area. The purpose of this paper is to review the innovation research literature which has made an explicit use of social network analysis methodology in order to provide empirical support to innovation theories or conceptual frameworks. The review introduces social network analysis then discusses why and how it has been used in innovation research so far. This paper argues that studies using network analysis tend to focus too much on change in the relationships between interacting units or nodes of the network to the detriment of change within units/nodes. Therefore, a combination of case study and social network analysis can offer a solution to that problem by providing the best of both methodologies. The document is structured as followed: section 2 provides a very short introduction to network analysis which describes what it is, where it came from, the terminology used, and defines the concepts of structure and dynamics or evolution, and finally this section ends with the definition of the various measures offered by network analysis and their corresponding advantages and disadvantages. Section 3 presents the methodology used for searching, collecting and selecting the documents read for this literature review. Section 4 is the review itself, followed by Section 5, the conclusion and suggestions for further research. 2 Network analysis 2.1 Terminology For those not familiar with network analysis, I start by introducing a bit of terminology. A network is a set of nodes connected by a set of ties. The nodes can be anything persons/individuals, teams, organisations, concepts, patents, etc. In the case of social networks the nodes are individuals.2 Networks which are only made of one type of nodes are homogeneous, they are heterogeneous otherwise. Whereas ties connect pairs of nodes and can be directed (i.e., potentially one-directional, as in giving advice to someone) or undirected (as in being physically proximate) and can be dichotomous (present or absent, as in whether two people are friends or not) or weighted (measured on a scale, as in strength of friendship). It is important to note that as a matter of fact, all ties are weighted or have values, even dichotomous relations have binary values (either the tie exist and is assigned a value of 1 or it doesn’t and is assigned a value of 0). However, in this 2 Traditionally, Social Network Analysis (SNA) has focused on networks of individuals, the literature reviewed here includes studies which make use of measures developed in SNA but applied to networks of firms, other organisations, patents, and even whole sectors in some cases. Basically, the methodology is the same and the measures are the same but they should be called “network analysis” studies instead of SNA. January 17th 2005 3 document we will treat dichotomous ties as unweighted ties. When we focus our attention on a single node, we call that node the ego and call the set of nodes that ego has ties with alters. When network analysts collect data on ties from a set of nodes, they call it relational data. Relational data can be visualised in matrix form or in graphic form. Table 1, below, summarises this terminology. Table 1. Important terms and definitions Network Analysis Terms Definitions Node The basic element of a network Tie / Edge A set of two nodes. Ties can be dichotomous (unweighted) or weighted/valued, directed or not (undirected) Directed Tie An ordered set of two nodes, i.e., with an initial/source and a terminal/destination node Network A set of nodes connected by a set of ties Valued Network A network whose ties/edges are associated with a measure of magnitude or strength Ego A node which receives particular focus Alters The set of nodes that has ties with the ego but not including the ego itself Network Size The total number of nodes of a network Relational data The set of ties of a network Following this terminology, Table 2 below summarizes the four types of networks that will be considered in this review, depending whether ties are weighted or not, and directed or not. Table 2. Four network types Types of networks Weighted Unweighted Directed (a) Directed & Weighted ties (b) Directed & Unweighted ties Undirected (c) Undirected & Weighted ties (d) Undirected & Unweighted ties Network analysis is very different from other methodologies, in that, several levels/units of analysis are embedded in the network analysis itself. Measures are available at the node-level, the group or local-level and at the network-level. The choice of the appropriate measure depends on what the network analyst wants to show. 2.2 Structure For social network analysts, there is a sharp distinction between information about the social actors and information concerning the social structures within which these actors are located. Wellman (1988) clearly emphasize this paradigm: “behavior is interpreted in terms of structural constraints on activity rather than in terms of inner forces within January 17th 2005 4 [actors].” (Wellman, 1988: 20). For some social network analysts (Doreian, 2001: 83), the “rather than” can be replaced by “in addition to.” Therefore social network analysts have developed two strands of thought in one, they focus only on structure to interpret behaviour, but in the other, they focus on both structure and actor-diversity to interpret behaviour. The nodes of the networks in their studies are often individuals or members of a social group. The first strand deals with the relationship between network structure, i.e., the observed set of ties linking the members of a population like a firm, a school, or a political organization, and the corresponding social structure, according to which individuals can be differentiated by their membership in socially distinct groups or roles. The combination of network structure and social structure is the social network. A substantial array of definitions and techniques have been introduced over the years, like blockmodels (e.g., DiMaggio, 1986), hierarchical clustering (e.g., Lorrain and White, 1977) and multidimensional scaling (Bailey, 1976). But in short, they are essentially designed to extract information about socially distinct groups from purely relational data, either in terms of some direct measure of social “distance” between nodes or by grouping nodes in the network.3 According to this view, networks are the signature of social identity/role – the pattern of relations between individuals reflects the underlying preferences and characteristics of the individuals themselves (Watts, 2003: 48). The second strand of techniques bears a more mechanistic flavour. In this strand, the network is viewed as a conduit for the propagation of information or the exertion of influence, and an individual’s place or position in the overall pattern of relations determines what information that person has access to or, correspondingly, whom he or she is in a position to influence. A person’s social identity/role therefore depends not only on the groups to which the individual belongs but also on the individual’s position within these groups. Similarly to the first strand, a number of metrics, i.e., measures of social “distance”, have been developed to quantify individuals’ network positions relatively to others and to explain observable differences in individual performance in terms of difference in these metrics (Watts, 2003: 48-49). An exception to these strands is Granovetter (1973), which introduced the distinction between strong and weak ties, e.g., contractual/formal and informal ties, or friend and acquaintance. Grannoveter shows that effective social coordination does not arise from densely and strongly interconnected networks but from the presence of occasional weak ties between individuals who frequently didn’t know each other that well or have much in common. According to Granovetter’s “strength of weak ties” theory, in order for an individual to get a job, it is not its close friends who are important and who will inform about that job but casual acquaintances who can give access to information that would never have been received otherwise (Scott, 2000: 34-35). 3 Social “distance” or “proximity” is a metric (a mathematical formula) that allows social network analysts to measure a distance between individuals. This distance can be dependent on the number of nodes or ties that has to be traversed in order to go from one individual (or ego) to another (or alter). The average of all these distances calculated for the whole network gives an estimate of the efficiency of a network. January 17th 2005 5 After this important finding the question that Granovetter put on the research agenda was how to distinguish strong and weak ties. He claimed that by observing the structure (i.e., the social network) in which the individuals are embedded it would be possible to make this distinction. Granovetter’s study distinguishes itself from previous works, because Granovetter suggested that in order to define relations at the individual level as strong or weak, it is necessary to observe the group or the whole network (Watts, 2003: 49). So far, not so much work has been done on weak, indirect, ties probably because of their empirical intractability. Most of the studies reviewed here are focused on strong and direct ties. The critique that has emerged in parallel with these two strands of literature is that they are static descriptions of structure – they do not consider change but apply their techniques to “frozen” networks, in other s, there is no dynamics. Instead of thinking of social networks as entities that evolve under the influence of social forces, network analysts have tended to treat them effectively as the static embodiment of those forces (Watts, 2003: 50). Purely structural and static measures of network structure cannot account for whatever action is taking place in the network – social network analysis offer no systematic way to translate the output of various metrics into meaningful statements about outcomes (Watts, 2003: 51). This is why it is only an analytical tool and not theory (Scott, 2000: 37). Without a corresponding theory of agency or behaviour, i.e., including the dynamics, the metrics remain essentially un-interpretable and of little practical use. In the rest of this text, when not dealing with individuals but, for example, with organizations, we will talk about “network analysis” and not SNA. However, all measures developed by social network analysts can be adapted to networks of firms or other organizations, since the network nodes can represent anything from humans and organizations to technologies. 2.3 Network Dynamics or Evolution According to the definition of network structure introduced previously, there are two types of dynamics that can be defined, i.e., first, dynamics of the network and second, dynamics on the network (Watts, 2003: 54-55). In the first type, dynamics refer to the evolving or changing structure of the network itself, i.e., the making and breaking of ties. The network structure of network analysts explained previously are snapshots taken during this ongoing process of evolution. However, a dynamic view of networks claims that existing network structure can only be properly understood in terms of the nature of the process that led to it. In the second type, the individuals (or firms, etc.) represented by the nodes of the network are doing something. They can search for information, learn, spread a rumour, make decisions, etc., the outcome of their actions is influenced by what their neighbours are also doing and therefore, to some extent by the network structure either locally from the nearest neighbours or globally from distant neighbours. January 17th 2005 6 In reality, both dynamics are taking place concurrently. For example, an individual can meet new friends and lose contact with old ones and simultaneously learn or make decisions. If you do not like the behaviour of a friend you can either decide to alter his or her behaviour or choose to spend time with another friend instead, both examples illustrate the two types of dynamics taking place in a personal network of friends. In the rest of this text, examples and explanations will be given for networks of individuals but similar examples can be found for a network of firms or other organisations since the measures developed by SNA apply to these other networks. However, one must be careful on terminological issues. Two problems emerge when applying SNA to other type of nodes than individuals which are important for describing the change or evolution of any network. The first problem is that, since firms, or more generally, organisations, are already made of individuals involved in social networks (in the social network analyst sense), is there a need to talk about “network organizations”? One can just call them “networks” and claim that in the 21st century, firms must transform themselves from organizations into networks (Palmer and Richards, 1999), confusing those who think, like social network analysts, already in terms of social networks of individuals. To be terminologically correct, they should be called “network analysis” of organisations. Another type of confusion appears in innovation research with for example “networks of innovators” (Powell, 2004) and “networks of innovation” (Tuomi, 2002). The former sees innovators as firms or other organisations, therefore talks about homogeneous networks in which nodes are organisations and the ties between them are contractual or informal relations. Whereas the latter is about heterogeneous networks in which nodes can be programmers or technologies and ties are relationship of use, e.g, a programmer using a text editor. This distinction is important since social network analysts have been mainly interested in homogenous networks, whereas actor-network theorists, e.g., Callon (2001), have particularly been interested in heterogeneous networks. As we will see later in this review, the metrics defined in SNA are not directly applicable to studies of heterogeneous networks (one need to transform them into multiple-mode networks4), and often these studies are limited to visualization of the network only. 2.4 Descriptive Measures This section starts by briefly describing the different measures that have been encountered during the review of the literature. Some or all of these measures are often present in any network analysis and their understanding is fundamental for the comprehension of the empirical work reviewed here. I also include a short discussion of the methodological problems associated with each of these measures. I do not present the mathematical formulas behind them the reader should consult Scott (2000) for further details. 4 See Scott (2000) for more details about multiple-mode networks January 17th 2005 7 The four most important concepts used in network analysis are network density, centrality, betweenness and centralization. Under these concepts are grouped several measures (or mathematical formulas) with various corresponding advantages and disadvantages regarding their use. Additionally, there are four measures of network performance: robustness, efficiency, effectiveness and diversity. Whereas the first set of measures concerns structure, the second set concerns the dynamics and thus depends on a theory explaining why certain agents do certain things (e.g., access to information). Most of the definitions are adapted (so that they use the terminology previously defined) from Scott (2000) and Burt (1992). Network Density Intuitively density is a measure of the connectedness in a network. Density is defined as the actual number of ties in a network, expressed as a proportion of the maximum possible number of ties. It is a number that varies between 0 and 1.0. When density is close to 1.0, the network is said to be dense, otherwise it is sparse. When dealing with directed ties, the maximum possible number of pairs is used instead. The problem with the measure of density is that it is sensible to the number of network nodes, therefore, i