Sampling variability and centrality score comparisons in archaeological network analysis: A case study of the San Pedro Valley, Arizona

Archaeological network analysis often focuses on networks in which ties between sites reflect some sort of similarity, such as in artifact assemblages. Site centrality is often of interest, but an apparent difference in two sites’ centrality may not be meaningful once sampling variability is considered. We investigate bootstrap assessments of sampling variability in centrality scores of a set of late pre-Hispanic archaeological sites in the San Pedro Valley, U.S. Southwest, for which ceramic assemblage data can be transformed into networks of ceramic similarity. We considered a variety of bootstrap confidence intervals for site eigenvector centrality scores and the implications of these intervals for interpretation of the site's structural importance. In analysis of the San Pedro Valley for CE 1300–1349, small differences among site centrality were not statistically distinguishable, but moderate to large differences were, with conclusions consistent across methods of constructing bootstrap confidence intervals. Similar patterns were evident when examining a broader region in which the Valley is located. It appears that substantive interpretation of site centrality differences often will be justified.