With the progression of multivariate statistics, the creation of population specific equations is on the rise. Multivariate analysis generally revolves around metric methods or geometric morphometrics, not on morphoscopic features. A total of eight samples were analysed spanning from prehistoric American to modern day South African and ranged between pygmy populations from the Andaman Islands to medieval British populations. With a sample size of more than 1100 individuals, each os coxa was scored using eight morphoscopic features most commonly used by physical anthropologists and osteoarchaeologists. Trait frequencies were compiled and compared between each of the eight samples. Then, the samples were placed into two groups: a known age and sex group (Christ Church Spitalfields, South African White, South African Black, and South African Coloured), and an unknown archaeological group (Poulton, St. Owens, Chumash, and Andaman). When comparing trait frequencies, slight differences between the samples could be seen. Ordinal Logistic Regressions (OLR) were applied onto each of the four samples from the known age and sex group to create population specific sexing equations (cross-validated). Results from these four equations ranged from 90.24% (South African Black population specific equation) to 96.38% (Christ Church, Spitalfields population specific equation). Population specifity was tested by applying all of the equations onto each sample in this group. In an attempt to reduce this, two new equations were created by combining samples together resulting in a South African specific equation (92.54% accuracy) and a "Summary Sex" equation (92.98% accuracy). After applying each of the six new OLR equations onto the four archaeological samples, high percentage accuracies (ranging from 92.59% to 100.00%) were found when comparing them to the previous records. The only sample that did not produce as high of an accuracy was the Chumash sample with 82.35%. In the attempt to analyse fragmented remains, three avenues were taken. Firstly, all missing values were replaced by the median score. Secondly, the original six OLR equations were 'sectioned' to make three smaller sets of equations. Lastly, to mirror the sectioned equations, three new sets of OLR equations were generated. This study shows that when using morphoscopic traits for sex estimation, applying multivariate techniques can be used to obtain a high accuracy even when dealing with fragmented samples.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:762861 |
| Date | January 2018 |
| Creators | Rennie, S. R. |
| Contributors | Gonzalez, S. ; Clegg, M. |
| Publisher | Liverpool John Moores University |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://researchonline.ljmu.ac.uk/9471/ |
Page generated in 0.0019 seconds