I thank Takeshi Inomata, Bárbara Arroyo, and Eugenia J. Robinson for their comments.
My use of “uninformative priors” had two senses (Love Reference Love2018). In the technical sense (applying Bayes theorem or “Little Bayes”), radiocarbon “dates” are probability distributions of the age of a sample, not raw data observations. In calibration, a distribution based on radiocarbon years serves as a prior that is combined with new information (the calibration curve) to calculate a posterior probability. If a sample is not properly assigned to the ceramic phase it purportedly dates, the prior is uninformative. A suite of such distributions can serve as a prior as well. Imprecise definition of phases, poor stratigraphy, or questionable pottery analysis can assign samples to the same phase that are not, in fact, estimates of the same thing. The suite can therefore yield a poor estimate of phase boundaries, and that model is an uninformative prior.
As a broader approach to integrating old data with new (“Big Bayes”), I adopted Bayesian terminology to characterize older data (pottery, stratigraphy) from Kaminaljuyu as uninformative. That usage was metaphorical, as explained. The metaphor is apt because stratigraphy and materials analysis are integral parts of constructing chronological models. In that sense I argued broadly for avoiding the use of dubious prior information and suggest patience while awaiting more data.
The links between Kaminaljuyu and Ujuxte are robust, making the Ujuxte chronology a good proxy for that of Kaminaljuyu. The major wares at Ujuxte are very closely related to the highest-frequency types at Kaminaljuyu. Many vessel forms, decorative attributes, and manufacturing techniques—all detailed in my article—link the sites. Arroyo (personal communication 2017) commented that the two sites appeared to be “married.”
In relating the Ujuxte sequence to Kaminaljuyu, Inomata, Arroyo, and Robinson present alternatives that would link a given phase at Ujuxte with one at Kaminaljuyu that is earlier than my proposals. Their chronology, however, would predict the opposite. Space does not allow complete discussion of this variance.
“Providencia” was used to describe the post-Charcas occupation at Naranjo before there were radiocarbon dates for those contexts (Arroyo et al. Reference Arroyo, Pereira, Cossich, Paiz, Arévalo, de León, Alvarado, Quiroa, Laporte, Arroyo and E. Mejía2007).
I do not oppose Bayesian methods or their use on recent assays obtained from gas-counting methods. Chronometric hygiene is always needed because some contexts are clearly better than others and even powerful tools cannot transform bad data into good. None of the data analyzed by Inomata and colleagues (Reference Inomata, Raúl Ortiz and Robinson2014)—at Kaminaljuyu or elsewhere—match the quality of the data from Ujuxte in terms of stratigraphic context, the features sampled for dating, or the detailed ceramic analysis.
You have
Access
I thank Takeshi Inomata, Bárbara Arroyo, and Eugenia J. Robinson for their comments.
My use of “uninformative priors” had two senses (Love Reference Love2018). In the technical sense (applying Bayes theorem or “Little Bayes”), radiocarbon “dates” are probability distributions of the age of a sample, not raw data observations. In calibration, a distribution based on radiocarbon years serves as a prior that is combined with new information (the calibration curve) to calculate a posterior probability. If a sample is not properly assigned to the ceramic phase it purportedly dates, the prior is uninformative. A suite of such distributions can serve as a prior as well. Imprecise definition of phases, poor stratigraphy, or questionable pottery analysis can assign samples to the same phase that are not, in fact, estimates of the same thing. The suite can therefore yield a poor estimate of phase boundaries, and that model is an uninformative prior.
As a broader approach to integrating old data with new (“Big Bayes”), I adopted Bayesian terminology to characterize older data (pottery, stratigraphy) from Kaminaljuyu as uninformative. That usage was metaphorical, as explained. The metaphor is apt because stratigraphy and materials analysis are integral parts of constructing chronological models. In that sense I argued broadly for avoiding the use of dubious prior information and suggest patience while awaiting more data.
The links between Kaminaljuyu and Ujuxte are robust, making the Ujuxte chronology a good proxy for that of Kaminaljuyu. The major wares at Ujuxte are very closely related to the highest-frequency types at Kaminaljuyu. Many vessel forms, decorative attributes, and manufacturing techniques—all detailed in my article—link the sites. Arroyo (personal communication 2017) commented that the two sites appeared to be “married.”
In relating the Ujuxte sequence to Kaminaljuyu, Inomata, Arroyo, and Robinson present alternatives that would link a given phase at Ujuxte with one at Kaminaljuyu that is earlier than my proposals. Their chronology, however, would predict the opposite. Space does not allow complete discussion of this variance.
“Providencia” was used to describe the post-Charcas occupation at Naranjo before there were radiocarbon dates for those contexts (Arroyo et al. Reference Arroyo, Pereira, Cossich, Paiz, Arévalo, de León, Alvarado, Quiroa, Laporte, Arroyo and E. Mejía2007).
I do not oppose Bayesian methods or their use on recent assays obtained from gas-counting methods. Chronometric hygiene is always needed because some contexts are clearly better than others and even powerful tools cannot transform bad data into good. None of the data analyzed by Inomata and colleagues (Reference Inomata, Raúl Ortiz and Robinson2014)—at Kaminaljuyu or elsewhere—match the quality of the data from Ujuxte in terms of stratigraphic context, the features sampled for dating, or the detailed ceramic analysis.