Sex Allocation from Long Bone Measurements using Logistic Regression

Logistic regression equations are more suitable for the prediction of sex than other multivariate statistical techniques such as least squares linear regression or discriminant function analysis. Logistic regression equations for sex determination were derived from measurements of the major long bones from a nineteenth century archaeological sample. Accuracy of prediction was validated using two holdout samples of individuals of documented sex. Accuracy of sex allocation ranges from 71% to 95% with measures of articular surfaces, followed by shaft measures. Bone lengths provide the lowest levels of correct allocation. Logistic models developed from this nineteenth century sample show levels of accuracy that are not significantly different from those attained in the Forensic Database holdout sample. Some articular surface measurements appear to best differentiate sex while remaining relatively insensitive to secular or population differences.