Sequential Truncation in Forensic Anthropology

Classification methods based on discriminant function analysis are widely used in forensic anthropology to assess population affinity and related questions. The probabilities generated during these analyses are commonly interpreted as indicators of classificatory strength. Despite regular use in forensic anthropological casework, the implications of how these probabilities are used remain underexplored. In this study, sequential truncation is defined as an inferential decision-making process where the posterior and/or typicality probabilities are used to eliminate reference groups and then recomputed in an iterative process to reduce the complexity of the classification model. Using a simulation-based analysis modeled on commonly taught discriminant function workflows, sequential truncation is revealed to violate rules of Bayesian coherence by repeatedly conditioning models on the same data without the introduction of new evidence. Results indicate this practice produces monotonic inflation of posterior probabilities, systematic collapse of posterior uncertainty, and a divergence between confidence and classification accuracy. Importantly, these effects persist even when the true group of origin is not among the reference groups. These findings indicate increasing posterior certainty under sequential truncation reflects procedure rather than improved evidentiary support. The study concludes by offering best-practice recommendations for interpreting and reporting probabilities to preserve inferential validity while remaining compatible with existing analytical tools and reporting conventions.