CRISP: Cremated remains inference of sex probabilities – A software for Bayesian sex estimation in human cremated remains

Abstract

Sex assessment in cremation contexts represents a key step for reconstructing funerary rituals and demographic profiles. However, the high degree of fragmentation and thermal alteration of skeletal elements significantly hampers this operation. Most often, the morphological traits of pelvis and skull are not preserved while the application of metric criteria follows those developed from modern reference series. We used a prehistoric Italian sample ( n  = 155) with gender specific grave goods as a proxy for sex and developed binary logistic regression and Bayesian models based on 21 postcranial metric variables. We compared the results with the cut-off point method and validated them using an independent Austrian prehistoric sample (n = 45) sex estimation based on morphological sex traits. The Bayesian model achieved the highest accuracy (89%), outperforming cut-off points and regression models (68–88%, depending on the variable). In the Austrian sample, cut-off points produced the highest proportion of classified individuals but misclassified four times more cases than the other methods. Bayesian and regression models yielded higher rates of ambiguous classifications but maintained low error rates. Inconsistencies between the Bayesian sex prediction and morphological sex estimation can be explained by less reliable morphological traits in the morphological assessments. We also present CRISP, a new open-source software using the Bayesian approach for sex prediction. Multivariate statistical methods proved both more reliable and more adaptable for sex estimation in human cremated remains than univariate cut-off approaches. Validation demonstrated that metric methods substantially increase the number of individuals for which sex can be estimated and retain robustness when morphological indicators are ambiguous. The tool can further be used to identify deposits of cremated human remains composed of more than one individual. Minimal interpopulation differences between the Italian training and Austrian validation samples suggested CRISP is suitable for both Mediterranean and Central European contexts.