A Novel Method for Osteometric Reassociation Using Hamiltonian Markov Chain Monte Carlo (MCMC) Simulation

Abstract

Traditional osteometric reassociation uses an error-mitigation approach, which seeks to eliminate possible matches, rather than a predictive approach, where possible matches are directly compared. This study examines the utility of a Bayesian approach for resolving commingling by using a probabilistic framework to predict correct matches.

Comparisons were grouped into three types: paired elements, articulating elements, and other elements. Ten individuals were randomly removed from the total sample (N = 833), acting as a small-scale, closed-population commingled assemblage. One element was chosen as the independent variable, with the ten possible matching elements representing the dependent variable. A Bayesian regression model was constructed using the remaining total sample, resulting in a distribution of possible values that were smoothed into a probability density, and probabilities were calculated. The element with the highest posterior probability was considered the best match. This process was repeated 500 times for each comparison. The correct match was identified 51.60% of the time. Paired elements performed the best, at 80.76%, followed by 42.10% for articulating and 33.63% for other comparisons.

These results suggest that metric analysis of commingled assemblages is complex and that both elimination-based and prediction-based approaches have a role in resolving commingling. In this regard, the strength of a Bayesian approach is versatility, allowing for prediction of the correct match and elimination of possible matches, as well as integration of independent lines of evidence within one cohesive model.