Contents lists available at ScienceDirect Journal of Archaeological Science: Reports journal homepage: www.elsevier.com/locate/jasrep Statistical artifacts: Critical approaches to the analysis of obsidian artifacts by portable X-ray fluorescence Luke Stroth a , Raquel Otto b , James T. Daniels Jr a , Geoffrey E. Braswell a, ⁎ a Department of Anthropology, University of California, San Diego, United States of America b Department of Anthropology, National Autonomous University of Honduras, Tegucigalpa, Honduras ABSTRACT In the past decade, chemical assay by portable X-ray fluorescence (pXRF) has become a common method of determining the geological origin of obsidian artifacts. We present source assignments made for 355 artifacts from the Late Postclassic site of Guadalupe, Honduras. Statistical studies of concentration data for six individual elements imply that obsidian from at least three sources (Güinope, La Esperanza, and Ixtepeque), and possibly two more (El Chayal and Otumba), was exploited in ancient times, but yield conflicting and ambiguous results for 5% of the collection. We argue that pXRF studies need to be conducted more critically and demonstrate how different statistical tools can yield contradictory source assignments for a significant portion of an obsidian collection. Visual sourcing supports discriminant function analysis over Mahalanobis distance assignments. Furthermore, element ratios imply that Otumba is not actually present. We are concerned that in obsidian studies, pXRF data are not consistently interpreted in ways that yield reproducible results. We propose an integrated methodology that employs pXRF, statistical analysis, visual sourcing, and—when needed and appropriate—destructive chemical assay to determine the source of “high-stake” artifacts. 1. Introduction Determining the geological origin of obsidian artifacts is an im- portant aspect of studies of lithic economy and exchange in ancient societies (e.g., Braswell, 2003; Cann and Renfrew, 1964; Forster and Grave, 2012; Healan, 1993; Knight et al., 2017; Matsumoto et al., 2018; Reepmeyer et al., 2016; Stross et al., 1983). During the past decade, chemical assay by portable X-ray fluorescence (pXRF) has become a widely applied technique used to make rapid and inexpensive source attributions for a wide variety of artifacts (e.g., Coronel et al., 2014; Craig et al., 2007; Johnson, 2014; Meanwell et al., 2013; Moholy-Nagy et al., 2013; Nazaroff et al., 2010; Schulze, 2013). Nonetheless, pXRF is subject to persistent critique regarding its unsuitability for hetero- geneous material, difficulties in assaying thin or angular artifacts, quantitative inaccuracies, and problems comparing datasets generated by different instruments (e.g., Coronel et al., 2014; Davis et al., 1998; Ferguson, 2012:401; Johnson, 2014; Nazaroff et al., 2010; Shackley, 2010). In addition to instrumental and measurement errors generated during the collection of compositional data by pXRF, it is important to consider interpretative errors introduced by the statistical analysis of such datasets. We demonstrate how these errors arise in the inter- pretation of new and previously unpublished geochemical data from a Honduran obsidian collection. Two distinct statistical methods—cano- nical discriminant and Mahalanobis distance analyses—were used to assign artifacts to known chemical groups and sources, and they yield conflicting and ambiguous results for about 5% of the collection. As a next step, we used visual analysis and comparison to make source as- signments for these ambiguous artifacts. Our results support canonical discriminant analysis over Mahalanobis distance calculations for this collection. Finally, we conducted secondary statistical analyses that consider ratios of concentrations rather than data for individual ele- ments. We propose an integrated source methodology for “high-stake” and “low-stake” artifacts based on their interpretative importance. This includes nondestructive pXRF assay, statistical inference, checking by visual sourcing, and destructive but accurate neutron activation ana- lysis (NAA), if appropriate. The Proyecto Arqueológico Guadalupe, directed by Markus Reindel and Franziska Fecher, is a collaboration between the German Institute of Archaeology and the Universidad Nacional Autónoma de Honduras. Guadalupe, in the municipality of Santa Fe, Department of Colón, consists of a low rectangular mound situated by an ancient river knoll (Figs. 1 and 2; Reindel and Fecher, 2016). Guadalupe dates to the Cocal period (1000–1530 CE), characterized by the intensification of maize agriculture and the emergence of incipient complexity. Exploitation of riverine and coastal resources was common, and many Cocal-period sites are located in defensive locations (Healy, 1984, 1992). Obsidian tools, although uncommon in Atlantic Honduras, were more available during this period than in earlier times (Healy et al., 1996). We ana- lyzed a sample of the obsidian collected from Guadalupe in order to better understand procurement patterns, technology, and the https://doi.org/10.1016/j.jasrep.2019.02.023 Received 30 December 2018; Accepted 14 February 2019 ⁎ Corresponding author. E-mail address: gbraswel@ucsd.edu (G.E. Braswell). Journal of Archaeological Science: Reports 24 (2019) 738–747 2352-409X/ © 2019 Elsevier Ltd. All rights reserved. T