Technical note: prediction of sex based on five skull traits using decision analysis (CHAID)

Area of Science:

• Forensic Anthropology
• Biological Anthropology
• Bioinformatics

Background:

• Osteologists traditionally assess skeletal sex using subjective ordinal scores of dimorphic traits.
• Logistic regression (LR) models have been developed for probabilistic sex determination but often neglect trait interactions.

Purpose of the Study:

• To evaluate Chi-square automatic interaction detection (CHAID) as an alternative classification strategy for skeletal sex determination.
• To analyze the predictive accuracy of trait combinations using CHAID, focusing on interactions.
• To compare CHAID performance against logistic regression using known-sex skeletal samples.

Main Methods:

• CHAID decision trees were employed to analyze 31 combinations of five sexually dimorphic skull traits.
• Known-sex samples (n=304) of English, African American, and European American ancestry were utilized.
• Selection criteria for optimal sex predictive trait combinations (SPTCs) included >=75% accuracy for both sexes and <5% sex bias.

Main Results:

• CHAID produced suitable SPTCs for English and European American samples, but not for the African American sample.
• Out-of-sample testing indicated that trees derived from the English and combined (all groups) samples exhibited the best predictive performance.
• CHAID's focus on variable interactions offers a potentially more robust approach to sex determination.

Conclusions:

• CHAID is a valuable tool for skeletal sex determination, especially when trait interactions are significant.
• The effectiveness of CHAID models may vary across different ancestral populations.
• Further research is warranted to refine CHAID applications for diverse forensic and archaeological contexts.