Distal Humerus Morphological Variation and Sex Estimation in Modern Thai Individuals,

Abstract

Due to differential preservation, it is necessary to develop sex estimation methods on varied anatomical regions, including the distal humerus. Sexually dimorphic differences of the medial epicondyle angle, olecranon fossa shape, trochlear extension, and trochlear constriction have been documented in several non‐Asian groups. This study examines distal humerus morphological variation in 616 modern Thai individuals (f = 198; m = 418) 18–96 years old and tests the methods on a hold‐out sample of 152 individuals (f = 91; m = 61). The results indicate that population‐specific binary probit regression performs the best (74.1–100%), followed by composite scoring (77.0–90.1%), binary logistic regression (63.3–92.2%), and univariate sectioning points (37.7–90.1%). Age minimally effects the expression of the medial epicondyle angle in males, and trait scoring is susceptible to high intra‐ and interobserver error. While demonstrating relatively high sex biases, distal humerus morphology can be used to estimate the sex of Thai individuals when more sexually dimorphic regions are absent.

Getting Rid of Dichotomous Sex Estimations: Why Logistic Regression Should be Preferred Over Discriminant Function Analysis

“…We recommend the use of LogR due to the flexibility of the method. Similar recommendations have also previously been made for LogR and probit (2,5,24). The performances may vary when additional assumptions are violated, and for different samples.…”

“…The performances may vary when additional assumptions are violated, and for different samples. Tallman and colleagues (2) found both probit and LogR to outperform DFA in modern Thai individuals, and Walker (5) found LogR to produce a lower sex bias than LDA. LogR also outperformed DFA in the Athens collection (6), while the predictions accuracies in this study were consistent between the two methods.…”

“…Other discriminant alternatives to LDA are quadratic discriminant analysis (QDA) and nearest neighbor discriminant analysis (NNDA), both of which have fewer assumptions than LDA and have previously seen success in sex estimation (41). While the use of LDA can be theoretically justified when the assumptions are not violated, LogR is more appropriate for predicting a small number of groups, and to develop models for sex estimation that include traits scored on a discrete scale due to the absence of distribution assumptions for the predictor variables (2,3,5). Additionally, if certain assumptions are not met, the reliability of a method may be called into question by a court, for example, the Daubert Standard (42).…”

“…It is important to have many different reference samples (i.e., anatomical collections and/or archaeological samples with known sex) given that the expression of sexually dimorphic traits varies between populations (7,8). While there have been many sex estimation equations provided from a variety of populations to deal with the inter‐population differences in sexual dimorphism (2,9–11), there are still issues when it comes to the biological overlap that occurs between the sexes within a population (12). One way to offset intrasexual variability is by providing estimates associated with a probability, which can readily be obtained from DFA and LogR; however, the ability to obtain probabilities for predictions does not always carry over from the original publication to future users of the model.…”

“…Discriminant function analysis (DFA) and logistic regression (LogR) are common statistical methods for estimating sex in both forensic (1–4) and osteoarcheological contexts (3,5,6). Statistical models are built from reference samples, which can then be applied to future cases for sex estimation.…”