<sarcasm>I'm not normally one to wade into a controversial topic</sarcasm> but having just read the recent #TallElHammam / biblical Sodom paper in Nature: Sci. Rep., I see one serious flaw in how they use their 14C dates. A thread (1/10):
2/10 The authors state "The age of the destruction layer at TeH was modeled using the OxCal radiocarbon calibration program...with the ‘Combine’ computer routine". If you don't understand, cool. Apparently the reviewers didn't either. Let explain it:
3/10 They used an OxCal command to assume from the outset that all the "combined" dates are from the same event. This is NOT a model, rather, it is a specific command to OxCal to take many age estimates of a single event, and calculate a single age of that event.
4/10 In the words of OxCal manual "The simplest form of analysis which one might wish to perform on a series of samples is the combination of several dates to give one measurement with smaller errors associated with it...
5/10 ... Combination of dates should clearly only be carried out if there is good reason to assume that the events being dated all occurred within a short period." tinyurl.com/bfwhu8
6/10 This last bit is critical to the interpretation of the output: It means that whatever the input, OxCal will give you an estimated single range, but the onus is on the researcher to have good reason to believe that a set of dates represents a single event.
7/10 Ex: You have 20 14C dates obtained from a mammoth skeleton. You KNOW that all of the dates MUST reflect the age at which the mammoth died. But, b/c 14C dating produces date ranges, you want to know what the consensus age is for ALL the dates. You would use "combine."
8/10 The output is not a VALIDATION that all of the dates represent a single event; rather, the output reflects the assumption-made by the user-that it MUST BE a single event, and provides the estimated age of that assumed event.
9/10 This is also NOT a "Bayesian-modeled age" as the authors claim. The authors seem to have no clear understanding of what Bayesian modeling of 14C dates is. There is no "model" in the paper, just a presentation of computer output that supports their preconceived notions.
10/10 TL;DR The date(s) reported reflect the a priori assumption that all of the 14C dates MUST represent a single event, and not (as presented) a test of whether the dates DO represent a single event. This is the definition of affirming the consequent, and it's #badscience
• • •
Missing some Tweet in this thread? You can try to
force a refresh
