Bayesian Canine Sex Estimation in the Coimbra Identified Skeletal Collection

Objectives: The aim of this research is to analyze the usefulness of human canines for sex estimation following the basic principles of Bayesian theory. This approach surpasses the frequentist statistical approach since it does not consider fixed thresholds (such as p=0.05 or 0.01) to assign an individual to one of the groups (male or female), but calculates posterior probabilities of correct estimation. The sample is composed of 115 subadult and adult individuals that belong to the Collection of Identified Skeletons of the University of Coimbra (Portugal).

Methods: The mesiodistal and buccolingual diameters of permanent canine crowns and necks were measured with a digital odontometric caliper. As skulls with intact canines were chosen, 16 measurements were obtained for each individual, and no calculation of missing data was needed. Firstly, intraobserver error was analyzed using the Intraclass Correlation Coefficient. 40 teeth were measured twice, with a difference of at least two weeks, and the results were compared. Secondly, the likelihoods of correct allocation (or the probability that an individual is correctly sexed) were calculated from a Bayesian approach, considering the different dimensions measured for each variable.

Results: Intraobserver error is not statistically significant for all the variables. Posterior probabilities using the section points range between 0.713 and 0.832 for females and between 0.755 and 0.847 for males. Upper canines show slightly lower values than lower canines, the same as those of the crowns with respect to those of the neck. The results obtained for the different possible measurements show that it is easy to define which is the most probable sex regarding the associated probability.

Conclusions: The Bayesian approach offers an adequate alternative to classical statistics for canine sex estimation in the sample, without considering arbitrary statistical thresholds and can be successfully used in forensic cases.