Since probit has not been applied to ancestry estimation in forensic anthropology, the current study (1) evaluated the performance of probit analysis as a classification tool for ancestry estimation using ordinal data and (2) expanded the current understanding of human cranial variation for an understudied population.
Multivariate probit models were used to classify the ancestral affiliation of Filipino crania using morphoscopic traits. Ancestral reference populations represented Africa, Asia, and Europe in a three-group model, with the addition of Hispanics in a four-group model. Posterior probabilities across these groups were interpreted as admixture proportions of an individual. Model performance was also evaluated for individuals with missing data. The overall correct classification rates for the three-group and four-group models were 72.1 percent and 68.6 percent, respectively. Filipinos classified as Asian 52.9 percent of the time using three ancestral reference groups and 48.6 percent using four groups. A large portion of Filipinos also classified as African. There were no significant differences in classification trends or accuracy rates between complete crania and crania with at least one missing variable. The study concluded that multivariate probit models using morphoscopic traits perform well when populations are represented in both training and test samples. Probit can also accommodate individuals with missing data. Classifying Filipinos showed only moderate success. Filipinos are more phenotypically similar to Africans than the other Asian samples used here, but still affiliate most closely as Asian. Ancestry methods would benefit from including Filipinos as a reference sample given the additional variation they provide to the continental category of Asian. (publisher abstract modified)
Downloads
Similar Publications
- Optimizing the Use of Video Technology to Improve Criminal Justice Outcomes
- Facilitators and Impediments to Designing, Implementing, and Evaluating Risk-Based Policing Strategies Using Risk Terrain Modeling: Insights From a Multi-City Evaluation in the United States
- Regularization-based learning of the Choquet integral