A Fully Bayesian Approach to Adult Skeletal Age Estimation: Multivariate Latent Trait Modeling With Markov Chain...

This paper investigates the potential of Bayesian transition analysis for improving age-at-death estimation in osteological datasets, utilizing the software NIMBLE and JAGS (Just Another Gibbs Sampler). We aim to explore the applicability of this method to different skeletal datasets including different age markers.

Transition analysis models estimations of skeletal age markers based on age-at-transition and population priors. This study extends the methodology by integrating multiple latent traits into a Bayesian framework. Our probit regression model estimates age-at-death and Gompertz distribution parameters directly and without relying on predefined reference populations for age-at-transition or population priors. The analysis includes data from three distinct sources, which contain diverse sets of age markers with known age-at-death.

The model demonstrates good performance across the datasets and successfully estimates mortality patterns. Sample size does matter, but even smaller samples can still yield useful results. The same is true for the number of traits. A simple multiple-probit regression will underestimate age ranges, but calibration improves upon this. Even with missing data, our model produces robust estimates.

The results highlight the model's effectiveness in improving age-at-death estimations and mortality patterns across diverse populations. This Bayesian transition analysis offers a reliable and flexible method for age estimation in bioarchaeological applications on datasets with varying qualities of morphological age indicators. Even compatibility between populations is not mandatory. It is only necessary that there be an observer-consistent age marker evaluation per skeletal assemblage and that the trait(s) in question change monotonically with biological age.