Byrd John E, LeGarde Carrie B
Laboratory, Defense POW/MIA Accounting Agency, Joint Base Pearl Harbor-Hickam, HI, USA.
Forensic Sci Res. 2019 Feb 7;3(4):343-349. doi: 10.1080/20961790.2018.1535762. eCollection 2018.
Evaluation of method performance involves the consideration of numerous factors that can contribute to error. A variety of measures of performance can be borrowed from the signal detection literature and others are drawn from statistical science. This article demonstrates the principles of performance evaluation by applying multiple measures to osteometric sorting models for paired elements run against data from known individuals. Results indicate that false positive rates are close, on average, to expected values. As assemblage size grows, the false positive rate becomes unimportant and the false negative rate becomes significant. Size disparity among the commingled individuals plays a significant role in method performance, showing that case-specific circumstances (e.g. assemblage size and size disparity) will determine method power.
方法性能评估涉及对众多可能导致误差的因素的考量。可以从信号检测文献中借鉴多种性能度量方法，其他方法则源自统计学。本文通过将多种度量方法应用于针对已知个体数据运行的成对元素的骨测量分类模型，展示了性能评估的原则。结果表明，平均而言，误报率接近预期值。随着集合规模的增大，误报率变得不重要，而漏报率变得显著。混合个体之间的大小差异在方法性能中起着重要作用，这表明特定案例的情况（如集合规模和大小差异）将决定方法的效能。
