1Anthropology, University of Illinois at Chicago, 2Anthropology, University of Illinois at Urbana-Champaign
April 18, 2020 3:30PM, Diamond 8-9
The resolution of commingling requires the re-association of all skeletal elements. Typically, osteometric techniques of skeletal re-association involve linear regression analyses. These analyses are limited to two-bone comparisons at a time, lengthening and complicating the assessment of assemblages. This study describes and tests an algorithmic approach to osteometric sorting based on shape variables and multivariate distance in long bones that sorts multiple long bones simultaneously.
The algorithm produces Mosimann shape variables for every possible combination of bones and compares these variables to the reference sample centroid using Mahalanobis distances (D2)—a multivariate measure of distance. This method is evaluated in trials using a reference data set (N=2,271) of postcranial remains to test a separate sample data set (N=25). This new approach is compared to published sorting methods using the same reference data set. Receiver operating characteristic curves and Youden’s J statistic demonstrate the method’s diagnostic ability. For the paired models, the humerus-femur sorting models outperformed other combinations of the two-bone models (e.g., humerus-radius). For the three-bone matching model, the humerus-radius-femur model outperformed other combinations of the three-bone models. The four-bone model outperformed all two-bone and three-bone models.
This research is novel in that there are currently no automated tools for re-associating more than two bones at a time. The automation of simultaneous matching of multiple bones reduces the need for multiple paired comparisons and the amount of processing power and manpower needed for these analyses. These results have implications for greatly expediting the osteometric sorting process, especially with large assemblages.