A performance test of archaeological similarity-based network inference using New Guinean ethnographic data

Network analysis has become increasingly common within archaeological practice, yet little consensus exists as to what networks based on material culture actually reveal about ancient social life. One common approach to archaeological network inference relies on constructing similarity networks based on shared material types or stylistic categories between archaeological sites or contexts. Many studies implicitly or explicitly assume that the topology of similarity networks is a useful proxy for underlying patterns in ancient social networks, yet this basic assumption has not been rigorously evaluated. Here, I present a preliminary test of how well network measures inferred from material culture—in this case, bone daggers made on the island of New Guinea between 1845 and 2002—predict network measures derived from ethnographic accounts of social engagement between 1720 New Guinea communities. In this case study network distance partially predicts material similarity, and neighborhood/cluster identification algorithms partially identify similar patterning in underlying patterns of inter-community engagement. However, most commonly applied network measures of centrality are not strongly predicted by material cultural similarity. Similarity based network analysis is a powerful means of visualizing and exploring data, and can help in formulating archaeological hypotheses, but may be problematic as a direct inference procedure.