A little while ago I wrote a post titled “Visualisations Are Not Everything” . Here’s a good example, which uses data from Google Scholar to show collaboration between Iranian Nuclear Physicists. The data is visualised in UCINET/NetDraw and NetMiner III, and shows the dangers of incomplete data sets, poor interpretation, and poor understanding of how to use the tools.

Consider the network diagram below, which has been visualised in UCINET/NetDraw. The nodes are Iranian Nuclear Physicists, and the red nodes are the network cut points – that is the individuals who are holding the network together.

The map below is the same network, using the same colours and holding the same position. This time the nodes have been sized using betweenness centrality, which shows the ability of one scientist to "get to others".

It’s fairly obvious who the key players are, or is it? Now consider the network diagram below which has been produced in NetMiner 3. It’s the same network with a much a better layout. What becomes obvious is it isn’t one linked network at all! A proper analysis requires each network to be examined in its own right.

Now consider the diagram below – it’s the circled network.

Sattari_A looks interesting, and this appears to be confirmed by the size of the node which is sized by degree centrality. The diagram below reveals a different story – it’s sized by betweenness.

Sattari_A is now on the periphery. Closer examination of the previous map reveals why – all the links come to him, but there are no out links. The map below reveals how spurious the data really is, and how careful we have to be with interpretation. It shows Sattar_A’s ego network and the year of collaboration. The light blue links are 2004 and the red links are 2009. We all know human systems are dynamic, so it's reasonable to assume data will degrade and the network will change. This diagram illustrates the point and shows that once again there are actually two discrete networks.

I think this set of visualisations illustrates the primary pitfall of network analysis – it’s too easy to become enamoured by the visualisations. Simple visualisations do not equal analysis. Deep analysis is required, and one must understand the limitations of the data and tools – it matters!

Regards Graham