Examining the Inherent Variability in ΔR: New Methods of Presenting ΔR Values and Implications for MRE Studies
MetadataShow full item record
CitationRussell, N., Cook, G. T., Ascough, P. L., Scott, E. M., & Dugmore, A. J. (2011). Examining the inherent variability in ΔR: New methods of presenting ΔR values and implications for MRE studies. Radiocarbon, 53(2), 277-288.
AbstractCurrently, there is significant ongoing research into the temporal and spatial variability of marine radiocarbon reservoir effects (MREs) through quantification of R values. In turn, MRE studies often use large changes in R values as proxies for changes in ocean circulation. R values are published in a variety of formats with variations in how the errors on these values are calculated, making it difficult to identify trends or to compare values, unless the method of calculating the R is explicitly described or all of the data are made available in the publication. This paper demonstrates the large range in R values (+34 to -122) that can be obtained from a single, secure archaeological context when using the multiple paired sample approach, despite the fact that the terrestrial entities were of statistically indistinguishable 14C ages, as were the marine samples. This demonstrates the inherent variability in the R calculations themselves and we propose that, together with calculation of mean R, the distribution of R values should be displayed, e.g. as histograms in order to illustrate the full data range. This spread is only apparent when employing a multiple paired sample approach as the uncertainty derived on a single pair of samples, taking account only of the errors on the individual 14C ages, will never truly represent the overall variability in R that results from the intrinsic variability in the population of 14C ages in samples that might have been used. Consequently, R values and the associated uncertainty calculated from single pairs should be treated with some caution. We propose that, where possible, when using paired archaeological samples, that a multiple paired approach should be employed as it will test the context security of the material used in the R calculations. When summarizing the values by the weighted average, we also propose that the standard error for predicted values should be employed as this will fully encompass the uncertainty of a future R calculation, using different samples for a similar time and location. Finally, we encourage future publishing of R values using the histogram format, making all of the data available. This will help ensure that R values are comparable across the literature and should provide a framework for standardization of publication methods.