Seminarium i matematisk statistik 
måndagen den 7:e december 13.30 i MH227

Goeff Nicholls, Auckland University

Statistical modeling and inference for radiocarbon dating

Abstract

A Bayesian method has been proposed (by Buck, Kenworthy,
Litton and Smith) for analysing radiocarbon dates. The
method takes into account stratigraphic constraints on
recovered calendar dates.  We find that the non-informative
priors in use in the literature apply a bias towards wider
date ranges which is not in general supported by substantial
prior knowledge.  We give a model of the process of artifact
depostion (by ancient people) and recovery (by
archaeologists). This defines a simple prior distribution
for the unknown true artifact ages.  Our prior can be
thought of as defining a shrinkage estimator for a point
estimate of the unknown true carbon dates, though we do not
attempt to compute this estimator. Subsequent analysis is
based on fairly efficient simulation of the posterior using
the algorithm of Metropolis, Hastings and Green.  We apply
the method to relatively large synthetic and real data sets,
examining the effect that various priors have on the
reconstructed dates.