Reevaluation of radiocarbon counting statistics on the MICADAS AMS system: Evidence and implications of non-Poisson distributions on robust uncertainty calculation

A reassessment of radiocarbon counting statistics in accelerator mass spectrometry (AMS) at the Andre E. Lalonde National Facility revealed that the traditionally assumed Poisson distribution may not always apply. An extensive analysis of 2.5 years of 14 C and 12 C data was conducted on a MICADAS™ AMS. This study found that only 63% of results adhered to Poisson statistics, while 34.2% showed slight deviations, and 2.8% exhibited strong non-Poisson behavior. This finding challenges the classic assumption that radiocarbon AMS is inherently a Poisson process. This study recommends considering non-Poisson models, specifically quasi-Poisson and negative binomial models, to better account for internal error and improve the accuracy of the reported error. Integrating 12 C current noise into error calculations is also suggested as it plays a significant role in measurement variability. We would like to ignite curiosity on other AMS laboratories to test the non-Poisson error framework with the broader aim of assessing its applicability in improving conventional statistical models, error expansion methods, and in ensuring more accurate and reliable 14 C results.