Skeletal age-at-death estimation: Bayesian versus regression methods

Area of Science:

• Bioarchaeology
• Forensic Anthropology
• Statistical Modeling

Background:

• Age-at-death estimation from skeletal remains can be inaccurate due to differing demographic profiles between training and target samples.
• Existing methods may not adequately correct for these demographic biases, impacting reliability.

Purpose of the Study:

• To develop and evaluate new regression-based methods for skeletal age-at-death estimation that minimize demographic bias.
• To compare the performance of these new methods against established Bayesian approaches.

Main Methods:

• Proposed two regression techniques: weighting by target sample demographics and creating a hypothetical training sample.
• Tested methods using 532 artificial skeletal systems with an eight-grade age marker expression.
• Evaluated performance based on various criteria.

Main Results:

• The proposed regression techniques, especially the hypothetical training sample method, showed superior performance compared to the Bayesian method in over 90% of tested systems.
• Careful selection of the training sample with a uniform or balanced demographic profile is crucial for optimal results.
• Direct regression with simple linear models is effective when training and target sample profiles are similar.

Conclusions:

• New regression-based approaches offer significant improvements for skeletal age-at-death estimation when demographic profiles differ.
• The hypothetical training sample method is particularly promising for enhancing accuracy.
• Method selection should consider the similarity of demographic profiles between training and target samples.