EVENT MODEL: A ROBUST BAYESIAN TOOL FOR CHRONOLOGICAL MODELING

We propose a new modeling approach for combining dates through the Event model by using hierarchical Bayesian statistics. The Event model aims to estimate the date of a context (unit of stratification) from individual dates assumed to be contemporaneous and which are affected by errors of different types: laboratory and calibration curve errors and also irreducible errors related to contaminations, taphonomic disturbances, etc, hence the possible presence of outliers. The Event model has a hierarchical structure which makes it possible to distinguish between date of an Event and dates of the artifacts involved. Prior information on the individual irreducible errors is introduced using a uniform shrinkage density with minimal assumptions about Bayesian parameters. The model is extended in the case of stratigraphic sequences which involve several Events with temporal order constraints (relative dating). Calculations are based on MCMC numerical techniques and can be performed using the ChronoModel software which is freeware, open source and cross-platform. This modeling provides a very simple way to automatically penalize outlying data without having to remove them from the dataset. This approach is compared to alternative approaches implemented in Oxcal or BCal software: we show that the Event model is more robust but generally yields less precise credibility intervals. Mathematical formulations are explained in detail and comparisons are done thanks to synthetic examples. Three application examples are shown: the radiocarbon dating of the shroud of Turin, the dating of a medieval potter's kiln in Lezoux (Auvergne, France) by using radiocarbon, archaeomagnetism and thermoluminescence, and the OSL dating of the Shi'bat Dihya 1 sequence in Wadi Surdud middle paleolithic complex (western Yemen)