The use of ordinal categorical variables in skeletal assessment of sex

Ordinal categorical variables are frequently used as a basis for assessing the sex of adult archaeological skeletal remains. In this thesis, seven commonly used pelvic and cranial characteristics are scored on a five point ordinal scale by two independent observers from a sample of 148 individuals taken from the Averbuch skeletal collection (40DV60). The scores are compared using ordinal log-linear models appropriate for measuring interobserver agreement. The results show that most of the trait observations are highly correlated between the two observers, but that a few traits do not exhibit marginal homogeneity (ie. the observers disagree in the proportions of individuals that should be placed in each ordinal class.)

The next step involved considering the seven characteristics across "known" sex as determined using Phenice's pubic bone characteristics, and through the addition of a modern collection of known sex. Logistic regression of sex onto the skeletal characters is used in order to find equations that provide posterior probabilities that a skeleton with a given set of characters was male as versus female.