with National Bureau of Standards oxalic acid, 95 percent of the specific activity of which is taken to be that of a sample aged AD 1950, Nb = background count rate (cpm).As the sample, reference standard, and background are prepared and counted in the same way, it is generally assumed that any inaccuracies in experimental technique will be the same for each. For this reason radiocarbon dates are usually quoted with a precision (standard deviation, SD) based on the statistics of counting alone and makes no allowance for errors associated with technique apart from sample dilution and isotopic enrichment. However, in practice, sample, reference standard, and background are not counted in identical fashion, as the precision of a date is improved by accumulating counts from reference standard and background over a prolonged period. As a result, sample, reference standard, and background may be counted on different occasions, and there is a possibility that different errors may arise in each with a resulting loss of accuracy. It should be possible to eliminate this difficulty if constant preparatory and counting conditions can be provided by investigating (and correcting for) all foreseeable sources of error. This paper describes such an investigation, the object being to achieve an overall precision on a radiocarbon date of about ±20 yrs in order to make a detailed calibration of the radiocarbon time scale (Pearsorl and others, 1977).As the age of sample carbon is given by 8033 loge Z 095 (Callow, . Baker, and Hassall, 1965) The procedure adopted was to convert sample carbon to benzene according to the method of Barker, Burleigh, and Meeks (1969), but scaled up to produce at least 16 ml benzene per sample, and then assess the 14C content of the benzene in a liquid scintillation spectrometer. A preliminary outline account of the procedure has appeared elsewhere (Pearson and others, 1977).
EVALUATION OF ERRORS AND REQUIREMENTS OF THE COUNTING SYSTEMThe errors encountered in a liquid scintillation dating system fall into three distinct groups. There are, first, the errors associated with the statistics of counting; second, those associated with the uncertainty remaining after corrections have been made to improve the accuracy of measurement; and third, those consequent on the uncertainty surrounding the reproducibility of external source channels ratio (p 10) and the assumption that vial efficiency and background are constant (p 17). The last two groups of errors will be considered together and termed correction errors.If these correction errors are assumed to be independent of each other and specific to sample, reference standard, and background respectively, each count rate will have to be corrected in turn to give the count rate appropriate to the standard conditions. Errors associated with the corrections made to the components of N, N1, and N,, will each contribute to the total error in the ratio Z. To determine how much error can be allowed in each individual component and so assess what will be required ...