Presenting multivariate statistical protocols in R using Roman wine amphorae productions in Catalonia, Spain

Several analytic techniques can provide data for characterizing archaeological ceramics. These data sources are not alternative but rather complementary to each other. They report on different aspects of ceramics concerning the origin of raw materials and the technological processes involved. However, when studies integrate more than one data source, they often do it through textual description and argument, not through a combined statistical analysis. We aim to help to overcome this situation by presenting four protocols for exploring data on archaeological ceramics. These protocols cover four different paths when interrogating ceramic samples. Protocol 1 aims to assist the definition of chemical reference groups using geochemical compositions, for instance, given by X-ray fluorescence analysis (WD-XRF). Protocol 2 focuses on fabric groups using petrographic examinations, such as in thin-section optical microscopy. Protocol 3 offers a hybrid assessment of provenance, using the integral sum of the two data sources. Last, Protocol 4 consists of the same approach as Protocol 3 but using geochemical data and a selection of petrographic variables that are considered indicative of the origin of raw materials and independent of human factors. We demonstrate their performance by applying them to a well-studied Roman wine amphorae dataset from Catalonia, NE Spain, and contextualising the results. Through a comparison of the results produced by these protocols, we restate the conclusion of Baxter et al. (2008) that a mixed mode approach is preferable to analysing data from different sources separately. Moreover, we argue that treating geochemical data as compositional and petrographic semi-quantitative observations as ordinal variables, when calculating dissimilarity, offers a more complete image of ceramic materials. The protocols are the synthetic product of several multivariate statistical methods developed for similar purposes in other disciplines, such as geology and ecology. To allow future users to replicate our analysis and apply the protocols, we published online two R packages containing all necessary procedures, from data cleaning to plotting. We also offer in the appendices a tutorial and the example scripts.