Craniometric sex estimation using a finite mixture missing data model

Accurate estimation of sex from human skeletal material is crucial for generating inferences regarding prehistoric populations. Metric approaches to sex assessment can provide an objective evaluation of skeletal material. However, commonly used procedures such as discriminant analysis require a reference sample or training set which may not be an appropriate analogy for prehistoric samples with differing levels of sexual dimorphism and/or overall size. Additionally, since most statistical analyses are developed for cases with a complete data set, such methods are hindered by unobservable measurements. This thesis tests the utility of applying finite mixture analysis, a form of cluster analysis that does not require a reference sample and also permits missing data, to craniometrics for the purpose of sex estimation. Finite mixture analysis is applied to two of Howells' known sex samples after measurements were randomly removed from the data set simulating missing data patterns found in 154 crania from the Late Mississippian Averbuch site. The analysis provided levels of accuracy comparable to discriminant analysis for the relatively dimorphic sample from Northern Japan both with and without missing data. However, the analysis performed poorly for the less dimorphic sample from South Africa both with and without missing data. When applied to an actual archaeological sample consisting of 94 Averbuch crania, the analysis performed relatively well considering the prevalence of missing measurements.