Social Networks and Resilience

Networks have structures, and their structural properties and variations can be tested for vulnerabilities. But unlike physical systems, social networks are made up of people whose own behaviours, perceptions, and choices introduce additional layers of complexity.

Social Networks

A network has two basic elements: a set of nodes and a set of ties connecting particular pairs of those nodes. Anything, potentially, might count as a node or tie if it is meaningful to define it as such for research purposes. In specifically social networks, nodes are often either human actors or organisations, and ties are some kind of relationship or transaction between them. Together, ties and nodes form a structure which has measurable properties, and which can be visualised as a graph (see Figure One).

Figure One: A Basic Network Graph

Network Resilience

Network resilience is usually defined by reference to the likelihood that removing a small proportion of nodes and/or ties would significantly alter a network’s structure, thereby preventing it from delivering the desired outcomes. If removal of a few nodes would fragment a network into disconnected ‘components’ and this would significantly disrupt the flow of desired goods through it, for example, then the network would be said to lack resilience. In the abstract this makes sense but when we add details of concrete examples, particularly involving human beings, things get more complicated.

Centralisation vs Decentralisation

Resilience depends on network flow and threat type. For example, if a virus is spreading, rapid fragmentation into disconnected ‘components’ may actually help protect many nodes and ties. While this would break the integrity of the network as a whole, it could allow parts to continue functioning. In this scenario, a network that fragments easily might be considered more resilient than one that remains fully connected, since full connectivity would expose every node to infection.

Studies of network resilience often focus upon ‘random attacks’. Researchers remove nodes and/or ties randomly and consider the impact upon a network’s structure and functioning. This is often inappropriate for social networks, particularly in a national security context where attacks are frequently targeted.

To give an illustration, in his popular book, Linked, Barabási argues that ‘scale-free networks’, that is, very large networks involving a tiny number of ‘hubs’ which account for a disproportionately large number of connections, are highly resilient. Hubs hold scale-free networks together, indirectly connecting the many other nodes which connect to them. Their loss would have a major negative impact on the structure and functioning of the network. However, because they form such a small minority of nodes within a scale-free network (perhaps one in several million), the probability of one of them being removed in a random process is vanishingly small. Random attacks are much more likely to affect other nodes, which have little significance for the structure and functioning of the network.

This is good news if chance is our main adversary but not if we are dealing with sophisticated cyber-terrorists who will attack hubs deliberately. (Barabási’s subsequent work acknowledges some such complexities).

Decentralisation comes with trade-offs: while it may reduce vulnerability to attack, it can also slow communication and hinder coordination, introducing different kinds of weaknesses.

Scale-free networks are highly centralised, built around a small number of nodes with disproportionately many connections (see Figure Two for an idealised model). By contrast, more decentralised networks (where nodes have a roughly similar number of ties) may be more resilient to targeted attacks, as they lack obvious single points of failure (Figure Two). This resilience is often cited in the case of malign COM networks such as 764, and it also underpinned the shift in some Islamic terror networks towards a model of “leaderless jihad”. However, decentralisation comes with trade-offs: while it may reduce vulnerability to attack, it can also slow communication and hinder coordination, introducing different kinds of weaknesses.

Figure Two: Centralisation and Decentralisation

The Role of Network Density

Increasing network density, that is, increasing the proportion of all pairs of nodes which are connected, would potentially increase speed of communication without necessarily making a network any more centralised. It would also increase the number of indirect paths of connection between nodes, reducing their dependence upon any one path and making connections more resilient to damage. However, higher density requires nodes to maintain more connections which can be costly and which increases node exposure to dangers within the network. In addition, it allows disinformation, viruses, and other ‘bads’ to spread more effectively through a network, again reducing resilience.

As a final point, in contrast to the physical systems which are often the focus in discussions of resilience, the human nodes of social networks typically perceive and attach meaning to ‘attacks’; repairing damage, addressing the source of their vulnerability, and rounding upon their attacker. Removing key players from a terrorist network is not the same as removing a fuse from a circuit board. Activity may be frustrated in the short-term but not stopped.

Download the article

Nick Crossley is a Professor of Sociology at Manchester University and is a co-founder and co-director of the Mitchell Centre for Social Network Analysis.