An ICAR spatial model to reconstruct regional geographic variations in arrival times

Abstract

Reconstructing the speed and pattern of dispersals of cultural traits from the archaeological record is typically based on spatio-temporal distributions of dated artefacts. However, archaeological sampling biases often result in uneven datasets where some regions or time periods are well represented, while others are not. This hinders our ability to offer precise and accurate reconstructions of past dispersal processes. Here we are proposing an approach to address these issues, by introducing a new method to take into account local geographic variations in past movements and refine temporal estimates of regional arrival times by implementing a hierarchical Bayesian Intrinsic Conditional Auto-Regressive (ICAR) model that partially informs the estimated arrival time in a focal region by the estimated arrival times in neighbouring regions. The hierarchical Bayesian ICAR model makes two major advances over previous approaches. Firstly, it provides estimates of the arrival time for areas with little or no archaeological data. Secondly, the model is more flexible than existing phase models or regression analyses, as subjective hypotheses and assumptions about the origin and route of dispersal are not built in. Through a series of experimental applications on simulated archaeological datasets with different dispersal dynamics, we examine the behaviour, accuracy, and precision of the ICAR model against commonly applied phase models and the model's robustness to variation in sample size and unevenness in sampling intensity of archaeological datasets.