Reading Group, Week 18: Fun with Bayesian modeling!

This week we dared to venture into the world of Bayesian statistics. Luckily, we had the marvelous Dr. Julie Hoggarth as our guide. See below for summaries of the articles and the questions that directed our 2.5-hour discussion.

Summaries


Hamilton, W. Derek, and Anthony M. Krus. (2018) The myths and realities of Bayesian chronological modeling revealed.” American Antiquity 83.2: 187-203.
Hamilton and Krus review the history of using Bayesian chronological modeling within archaeology, offering several myths and misconceptions about the method based on their experience with colleagues. They show that there has been a boom in the use of Bayesian modeling over the past 5 years in particular, particularly applied within British archaeology as well as American archaeology. They list the following misconceptions: 1. Bayesian statistics are not scientifically objective; 2. Old radiocarbon measurements with large errors should be ignored; 3. Stratigraphic relationships between samples are needed to make a Bayesian chronological model; 4. The dates for diagnostic artifacts or time ranges should be included in models as constraints; 5. Agreement indices are useful tools for determining which models are probable; 6. Bayesian modeling is not necessary if you have an accepted site/regional chronology. Throughout their discussion of the general practice of Bayesian modeling, the authors ask chronological questions based on when activities began, ended, and the length of time the activity took place. These are all feasible questions when working in Bayesian models, given that prior information are used to model dates. The authors take the readers through the whole process from selecting samples to setting up models.


Hoggarth, Julie A., Brendan J. Culleton, Jaime J. Awe, Christophe Helmke, Sydney Lonaker, J. Britt Davis, and Douglas J. Kennett. (Accepted) Buliding High-Precision AMS 14C Bayesian Models for the Formation of Peri-Abandonment Deposits at Baking Pot, Belize. Accepted in Radiocarbon.
Hoggarth et al. present a case study on the timing of the depositional processes of peri-abandonment deposits at Baking Pot, Belize, to assess the timing for the end of political activity at the site during the Late to Terminal Classic period. Multiple hypotheses have been presented to explain these deposits, which often feature large amounts of broken pottery, along with faunal remains, figurines, and in some cases musical instruments, in corners of plazas and courtyards in sites across the Maya lowlands. These have been interpreted as rapid events, such as warfare, with the assumption that the deposits were created in a single short-lived event (day or two) in which the entire deposit was formed (e.g., the palace was sacked and the material in it were deposited in these features). Other interpretations stress more protracted processes for the formation of deposits, such as the ritual termination of ceremonial spaces or pilgrimage by post-abandonment populations. To test these ideas, the authors dated faunal remains from distinct layers of 3 deposits, using hieroglyphic texts with calendar dates as priors to constrain dates when available. They used their stratigraphic positions to place the dates into the Bayesian model, using three models to assess rapid, medium, and protracted depositional processes. The results show low statistical agreement for rapid and medium models, with very high agreement with protracted processes, and show that for Baking Pot at least, these processes included at least 3 depositional events spanning around 150 years in total.

Questions:

  1. Do we agree with Hamilton and Krus’ list of misconceptions of Bayesian modeling? Are there any points here that you think might still be problematic, particularly within their application in Maya archaeology?
  2. Given the expansion of Bayesian chronological modeling within British and North American archaeology,why don’t we see more use of these methods in the Maya area? Is this a remnant of how Maya archaeologists are trained, are large-scale dating projects too expensive, or is it not that necessary given that Maya ceramic sequences are tied to the Gregorian calendar by association with Long Count dates on monuments (or a combination of these factors)?
  3. Do we all agree that ceramic types that have been well-dated for one site/region date to the same time when they are found at other sites? Using our example of Belize Red from the last meeting, does the Belize Valley ceramic sequence nail down the the timing of Belize Red in southern Belize (or another region)?
  4. In my [Julie’s] article on Baking Pot, we are able to use Long Count and Calendar Round dates on ceramics in the deposits as a terminus post quem, to constrain the ‘time after which’ we know the deposit could have formed. Is this an under-explored area for chronology building in the Maya region or are these types of artifacts so rare that this type of method will not be likely to catch on much in Maya archaeology?
  5. Inomata and colleagues have had quite a bit of push-back from regional specialists on their Bayesian chronological revision for Kaminaljuyu and other sites. What are the main issues with applying these methods when one is not a specialist working in that region? Should archaeologists stick to those sites where they conduct excavation or can specialists working in other areas lend information or perspectives that might otherwise be missed?
  6. What do you feel would be the most important information that you’d like to learn from these types of studies and how they can be applied to the Maya area? Do you think that those studies that have been attemped (e.g. Inomata et al.) have been successful?