Lynch Jeffrey James, Byrd John, LeGarde Carrie B
Defense POW/MIA Accounting Agency, 106 Peacekeeper Drive, Bldg 301, Offutt AFB, NE, 68113.
Defense POW/MIA Accounting Agency, 590 Moffett Drive, Bldg 4077, Joint Base Pearl Harbor-Hickam, HI, 96853.
J Forensic Sci. 2018 Mar;63(2):371-380. doi: 10.1111/1556-4029.13560. Epub 2017 May 26.
This study compares the original pair-matching osteometric sorting model (J Forensic Sci 2003;48:717) against two new models providing validation and performance testing across three samples. The samples include the Forensic Data Bank, USS Oklahoma, and the osteometric sorting reference used within the Defense POW/MIA Accounting Agency. A computer science solution to generating dynamic statistical models across a commingled assemblage is presented. The issue of normality is investigated showing the relative robustness against non-normality and a data transformation to control for normality. A case study is provided showing the relative exclusion power of all three models from an active commingled case within the Defense POW/MIA Accounting Agency. In total, 14,357,220 osteometric t-tests were conducted. The results indicate that osteometric sorting performs as expected despite reference samples deviating from normality. The two new models outperform the original, and one of those is recommended to supersede the original for future osteometric sorting work.
本研究将原始的配对骨测量分类模型（《法医科学杂志》2003年；48:717）与另外两个新模型进行比较，这两个新模型在三个样本上进行了验证和性能测试。样本包括法医数据库、美国海军俄克拉荷马号以及国防战俘/失踪人员身份查验局使用的骨测量分类参考资料。本文提出了一种计算机科学解决方案，用于在混合样本中生成动态统计模型。研究了正态性问题，结果表明该模型对非正态性具有相对较强的稳健性，并采用了数据变换来控制正态性。提供了一个案例研究，展示了国防战俘/失踪人员身份查验局一个实际混合案例中所有三个模型的相对排除能力。总共进行了14357220次骨测量t检验。结果表明，尽管参考样本偏离正态性，但骨测量分类的表现符合预期。两个新模型优于原始模型，其中一个被推荐在未来的骨测量分类工作中取代原始模型。
