Collaborative networks – such as the patterns of links between scientists who have worked together – are good candidates for study because they tend to be well defined. The links between researchers are well documented and the dates of the associations are clear.

In his study, Newman took several large databases containing information about scientific papers that had been published in physics, maths and biology over a five-year period. He then constructed networks between the papers, in which the nodes are scientists. Two nodes are connected together if the corresponding scientists have co-authored one or more papers. Newman then undertook a statistical analysis of the data using a large parallel computer at the Santa Fe Institute in New Mexico.

He found that publication patterns were very different in the three fields: biology had large groups of co-authors, mathematics had either single authors or pairs, while physics lay somewhere in-between. An exception is high-energy particle physics, in which authors had an average of 173 collaborators over five years.

Newman also found that most researchers produce few papers and have few collaborators. However, a small number of scientists collaborate with many others – up to thousands in some cases – and produce huge numbers of papers. Although the number of collaborators a scientist has does not necessarily reflect the quality of his or her work, researchers identified as well-connected in the analysis tended to be better known in their fields.

It turns out that the distance between people - the so-called degree of separation or small-world effect - in the networks is very short. In biology, there are only about 4 steps from one scientist to another, in physics there are about 6 and in maths about 7. Moreover, Newman noted what he calls the "funnelling effect": most people’s connections to the rest of the research world go through only one or two collaborators. "They may collaborate with many people, but in general their contacts with others come through just a couple of highly influential co-workers," he told PhysicsWeb.

Finally, clustering – where two people are more likely to be connected if they are both linked to a third individual – was very apparent in Newman’s study. Clustering coefficients were highest for physics (43%) and lowest for biology (7%) but the reasons for these differences are unclear.

Newman says that his results could provide new insights into the way science is conducted, published and even funded. "It’s fascinating having a window like this on a particular community – especially a community one is part of," he said.