New random generalized linear model for sex determination based on cranial measurements

Abstract

The estimation of the sex of the individual is a parameter of great value in forensic contexts and, above all, in archaeological contexts, where it is more difficult to apply genetic studies. In contrast with methods based on non-metric variables, we propose the use of a random generalized linear model for the determination of sex, starting from the Howells craniometric database and testing them on the dataset of known sex of the Forensic Data Bank, with 2524 and 1314 individuals respectively. After eliminating the individual’s considered outliers or with missing data, we proceeded to analyse which variables were more dimorphic between sexes (bizigomatic width, ZYB; bijugal width, JUB; mastoid height, MDH; glabela-occipital length, GOL; bifrontal width, FMB); these were used to build the statistical model. Subsequently, a comparison was made between the functions proposed by other authors and our model to determine their capacity in absolute terms, as well as by sex. The result is a random generalized linear model made up of 300 bags that, based on the five measures mentioned, reached 86.26% precision classifying the sex of individuals from the Forensic Data Bank (89.7% in the male sample and 82.82% in the female one). Although the method presented here should be taken with caution and not as the only way to estimate sex, it has proven to be statistically accurate in addition to having a non-regional vocation.