C05_01

Statistical challenges and opportunities when modelling multiple radiocarbon dates

Heaton T¹

¹University Of Sheffield, Sheffield, United Kingdom

The recent explosion in the availability of radiocarbon dates has been accompanied by an ever-increasing interest in the application of data science techniques within the archaeological and environmental science communities. Detailed modelling and computational analyses of large sets of dates have the potential to provide unprecedented inference on our past, on rates of change, and on population dynamics. It is essential however that the methods underpinning these “big-data” analyses are rigorous and robust. This concern is particularly relevant for radiocarbon dating since the need for calibration of the determinations introduces considerable, and complex, uncertainties in our dates that must be incorporated into any inference.

This talk will discuss some of the opportunities, and challenges, for the modelling of multiple radiocarbon dates. We will introduce statistically-rigorous alternatives to summed probability distributions (SPDs) that provide robust predictive calendar age summaries, with accompanying uncertainty bands that are essential to aid inference, as well as improved calibration accuracy. We will also discuss how radiocarbon users might obtain more from the IntCal20 radiocarbon calibration curve. Current approaches to calibration consider the calibration curves as normally-distributed around their published pointwise mean. This simplification results in some potentially crucial information about the calibration curve, such as on its covariance, being entirely lost to calibration users. The new IntCal20 methodology generates multiple possible calibration curve realisations, each representing an entire plausible ¹⁴C history from 55,000 – 0 cal yr BP. Using the collection of these realisations, rather than pointwise means, may provide improved inference for complex modelling.