Bayes factor: A useful tool to quantitatively evaluate and compare performance of multiple stature estimation...

Area of Science:

• Forensic Anthropology
• Bioarchaeology
• Bayesian Statistics

Background:

• Accurate stature estimation from incomplete skeletal remains is crucial in forensic and archaeological contexts.
• Traditional methods often rely on frequentist approaches with limitations in model comparison.
• Selecting the optimal estimation equation for a specific population is a persistent challenge.

Purpose of the Study:

• To propose the Bayes factor as a quantitative tool for evaluating and comparing stature estimation equations.
• To demonstrate the utility of the Bayes factor in identifying superior equations for specific target samples.
• To identify the best-performing stature estimation equations for Korean War casualties.

Main Methods:

• Generated 33 sets of stature estimates using various equations applied to Korean War casualty osteometric data.
• Employed the Bayes factor and posterior probabilities, calculated using R codes in the LearnBayes package, for statistical comparison.
• Compared the distribution of estimated statures from each equation against the known population distribution (Korean servicemen during the Korean War).

Main Results:

• The Bayes factor effectively quantified the similarity between estimated stature distributions and the population distribution.
• Choi et al.'s (1997) humerus equation yielded the highest Bayes factor (bf=9.84), indicating the closest fit.
• The same authors' femur equation also performed well, with a Bayes factor of 5.3.

Conclusions:

• The Bayes factor offers a robust, quantitative method for comparing the performance of stature estimation equations.
• This Bayesian approach provides practical interpretations and allows for model prioritization, surpassing traditional frequentist p-value reliance.
• Utilizing equations with higher Bayes factors can lead to more accurate stature estimations for target skeletal samples.