Among those topics that is of perennial interest to me is archaeological epistemology–or finding out how we know what we know. Using citations listed on the blog archaeological morphometrics, I harvested data for each publication–and all of their cited references–from the Scopus and CrossRef databases. Since geometric morphometrics is a relative newcomer to archaeology, most of the publications were readily accessible.
After all citations and cited works were harvested, they were used to generate a directed bipartite network. This means that each citation and each cited work (nodes) are connected by edges (common citations). Since we know who cited what, the network is directed–so an arrow runs from each publication to the cited work. The ForceAtlas2 algorithm was used to generate the initial layout (below) before the network was filtered by the giant component, and included only those nodes with two degrees or more.
Once filtered, statistics for the network diameter (including betweenness centrality, closeness centrality and eccentricity) were calculated in advance of modularity and eigenvector centrality. The resulting network illustrates the variable schools of thought within geometric morphometrics in archaeology (noted by the different colors), inferred from differences in citations. The different colors (groups) were generated by the modularity algorithm, which looks for nodes that are more densely connected together than with the rest of the network.
In this layout, the nodes are sized by eigenvector centrality (or authority). This means that nodes with more citations (regardless of whether they are a publication or a cited reference) appear larger. Further, since this is a directed network, and because the interface is interactive, one need only hover over a publication (or reference) to highlight those publications that cite (or are cited by) it (below). When using the interactive options, the red arrows represent outgoing citations, the blue arrows incoming citations, and yellow arrows go both ways.
We can use modularity and size the nodes by out-degree (number of refs that each publication cited), then color the edges based upon the target node to illustrate patterns of between-group citation. Further, by using the circular layout, we can see that those groups defined by modularity share common authors; most notable here are Stephen J. Lycett and colleagues in blue, and Briggs Buchanan and colleagues in green.
These divisions make sense, since there are some notable differences in the analytical approaches of Lycett and Buchanan. For instance, Lycett works more regularly with 3D data, and Buchanan with 2D, and both use different software packages for their analyses. While both camps are focused upon lithic analysis, one better articulates with Acheulean handaxes, Victoria West cores and Levallois technology (blue), while the other is focused on Paleoindian projectile points from North America (green).
We can then shift a single attribute (from out-degree to in-degree) to view the same network in a different way, focusing on the publications that were cited most. At this point, it becomes evident that the brown group most readily articulates with citations related to the method and theory of geometric morphometrics. Names like Bookstein, Rohlf, Adams, O’Higgins, McLeod, Slice, Claude, and others dominate–an outcome which was expected.
There are many additional observations that can be made here, which help to illustrate the foundations of geometric morphometrics in archaeology. I am intentionally leaving much unsaid, primarily because this is the topic of a forthcoming paper; however, I think it evident that this approach can aid our efforts to explore the epistemological roots of our ideas, highlighting how each of the works cited stood on the shoulders of those works that were written before.
The method is not restricted to geometric morphometrics, and is also being used to explore the roots of our knowledge regarding scanning electron microscopy in archaeology (above), and I have just begun work on a similar citation network for predictive modeling.
In working with students, I have found that when citation networks are used as a part of their literature review, identifying resources (articles, books, papers, proceedings, etc.) that may have been missed during the original literature survey becomes much more straightforward. It also helps them to see where their work (and the work of others) fits within the larger field (forcing–in most cases–a more objective view). The citation networks also encourage students to think about whether they subscribe to a particular theoretical or methodological approach, and allows them to quickly identify practical examples that they can then use in their own efforts.
One final aspect of the citation networks that I have not yet touched upon is animation–and yes, we can animate these graphs since we know when each article was published. Our animations for the citation networks use the publication year, and illustrate the adoption and abandonment of specific citations through time. This last piece is of particular interest, since it may be possible to use the citation networks to predict potential shifts in practice.