Cyclical Regression Modeling of δ18O Isotopic Profiles on Sparse Samples with Bayesian Multilevel Modeling

Summary

This is an abstract from the "The Expanding Bayesian Revolution in Archaeology" session, at the 88th annual meeting of the Society for American Archaeology.

Profiles of stable oxygen isotopic values (δ18O) from archaeofaunal tooth enamel provide in-depth information about the past environments in which animals lived while their teeth mineralized. Cyclical regression models can fit a specimen’s isotopic profile to a particular sinusoidal curve to estimate aspects of past environments and animal behavior. These include birth seasonality and seasonal environmental variation. Existing models, however, rely on densely sampled specimens, limiting the number of specimens to be analyzed within a particular budget. I present a Bayesian multilevel model to fit cyclical regression curves to enamel δ18O data. These models use prior distributions of the regression parameters to avoid overfitting to sparsely sampled specimens. I test the model’s efficacy by modeling first a set of densely sampled specimens, then removing observations to create artificially sparse datasets. The model effectively reconstructs the isotopic profile for sparse datasets. Furthermore, the multilevel structure of the model provides a natural framework for summarizing assemblages, providing a valuable baseline for inter-assemblage comparisons. This Bayesian model helps researchers shift their interpretive focus from specific animal histories to assemblage-level questions. Finally, I discuss how this approach may open new avenues for understanding past herd management.

This Resource is Part of the Following Collections

• The Expanding Bayesian Revolution in Archaeology   •
• Society for American Archaeology 88th Annual Meeting, Portland, OR (2023)

Keywords

Record Identifiers

Abstract Id(s): 36906.0