Linking behavior in the physics education research coauthorship network

There is considerable long-term interest in understanding the dynamics of collaboration networks, and how these networks form and evolve over time. Most of the work done on the dynamics of social networks focuses on well-established communities. Work examining emerging social networks is rarer, simply because data are difficult to obtain in real time. In this paper, we use thirty years of data from an emerging scientific community to look at that crucial early stage in the development of a social network. We show that when the field was very young, islands of individual researchers labored in relative isolation, and the coauthorship network was disconnected. Thirty years later, rather than a cluster of individuals, we find a true collaborative community, bound together by a robust collaboration network. However, this change did not take place gradually—the network remained a loose assortment of isolated individuals until the mid 2000s, when those smaller parts suddenly knit themselves together into a single whole. In the rest of this paper, we consider the role of three factors in these observed structural changes: growth, changes in social norms, and the introduction of institutions such as field-specific conferences and journals. We have data from the very earliest years of the field, a period which includes the introduction of two different institutions: the first field-specific conference, and the first field-specific journals. We also identify two relevant behavioral shifts: a discrete increase in coauthorship coincident with the first conference, and a shift among established authors away from collaborating with outsiders, towards collaborating with each other. The interaction of these factors gives us insight into the formation of collaboration networks more broadly.

Figure 1

The number of PER authors, and the number of PER papers. Papers are broken down by journal. Important events in the field are highlighted.

Figure 2

Distribution of paper publication on a log-log scale: 1981–2010. A small number of authors produce a disproportionate fraction of papers written in the field.

Figure 3

Degree distribution for the PER network on a log-log plot: 1981–2010. Degree on the coauthorship network is the number of coauthors an author has. A small number of authors publish with a large fraction of the field, while most authors publish with very few.

Figure 4

The largest connected component of the PER coauthorship network (1981–2010). The color represents degree and the size represents eigenvector centrality.

Figure 5

The shape of the collaboration network in the 1980s, 1990s, and 2000s.

Figure 6

The fraction of nodes in the largest connected component of the network. The emergence of a cohesive core occurs in the period from 2004 to 2005.

Figure 7

The fraction of papers with multiple authors. The fraction of papers coauthored exhibits a discrete jump in 1997. The dotted lines indicate the average rate of coauthorship before and after this break.

Figure 8

The change in the log odds of a paper in AJP being coauthored in the four years before and after a given point. Gray lines indicate the 95% confidence interval on the point estimate of the log odds ratio. A log odds ratio above 0 indicates a persistent increase in the odds of a paper being coauthored. This occurs in only one time period for this journal: 1997.

Figure 9

Fraction of PER coauthor relationships that are between existing members of the field (bottom, blue), between new members of the field (middle, dark green), and between existing and new members (top, pale green).

Figure 10

The fraction of coauthorships that occur between members of the community (blue), compared with the number that would be expected if linking behavior were random (red).