Abstract

Establishing a chronology is part and parcel of archaeological research. Seriation refers to a family of methods that aim to order archaeological contexts along a relative chronological sequence. Among these, multivariate seriation approaches use abundance data to recover an underlying temporal dimension, under the assumption that temporal change is a main driver of variation between contexts. In practice, the first extracted dimension (or an equivalent one-dimensional ordering) is often interpreted as the relative chronological order. In this paper, we use simulations to evaluate the performance of several seriation techniques—Correspondence Analysis (CA), Principal Component Analysis (PCA), Multidimensional Scaling - SMACOF (MDS-SMACOF), Uniform Manifold Approximation and Projection (UMAP), and two dendrogram-based orderings (Optimal Leaf Ordering and GW)—across a range of experimental archaeological setups. Overall, the results identify which methods perform best on average, which aspects of assemblage-type distributions and contextual structure most strongly affect seriation accuracy, and whether some methods are advantageous only under specific data-generating conditions. By providing a systematic, simulation-based comparison of multivariate ordering techniques, this study offers practical guidance for selecting methods for seriation—and, more broadly, for any application where the goal is to recover a meaningful one-dimensional order from multivariate abundance data.

Authors: Kafetzaki, D., Garcia-Angulo, A., Groenen, P.J.F., Poblome, J.