An experiment in writing in public, one page at a time, by S. Graham, I. Milligan, & S. Weingart

When Not To Use Networks

1 Leave a comment on paragraph 1 0 Networks analysis can be a powerful method for engaging with a historical dataset, but it is often not the most appropriate. Any historical database may be represented as a network with enough contriving, but few should be. One could take the Atlantic trade network, including cities, ships, relevant governments and corporations, and connect them all together in a giant multipartite network. It would be extremely difficult to derive any meaning from this network, especially using traditional network analysis methodologies. The local clustering coefficient, for example, would be meaningless, as it would be impossible for the neighbors of any node to be neighbors with one another. Network scientists have developed algorithms for multimodal networks, but they are often created for fairly specific purposes, and one should be very careful before applying them to a different dataset, and using that analysis to explore a historiographic question.

2 Leave a comment on paragraph 2 1 Networks, as they are commonly encoded, also suffer from a profound lack of nuance. It is all well to say that, because Henry Oldenburg corresponded with both Gottfried Leibniz and John Milton, he was the short connection between the two men. However, the encoded network holds no information of whether Oldenburg actually transferred information between the two, or whether that connection was at all meaningful. In a flight network, Las Vegas and Atlanta might both have very high betweenness centrality because people from many other cities fly to or from those airports, and yet one is much more of a hub than the other. This is because, largely, people tend to fly directly back and forth from Las Vegas, rather than through it, whereas people tend to fly through Atlanta from one city to another. This is the type of information that network analysis, as it is currently used, is not well-equipped to handle. Whether this shortcoming affects a historical inquiry depends primarily on the inquiry itself.

3 Leave a comment on paragraph 3 0 When deciding whether to use networks to explore a historiographic question, the first question should be: to what extent is connectivity specifically important to this research project? The next should be: can any of the network analyses I or my collaborators know how to employ be specifically useful in the research project? If either answer is negative, another approach is probably warranted.

Page 94

Source: http://www.themacroscope.org/?page_id=449