What Should Be Typical about the Way We Calculate Typicality?

Abstract

The probability of correct classification, and ultimately identification, lies at the heart of forensic anthropological analyses. To this end, practitioners rely on a variety of ways to assess the error or uncertainty of their estimates, including the use of statistically based analytical packages such as FORDISC. This article addresses typicality probabilities and specifically examines issues and assumptions with calculating F-statistic typicalities both statistically and within FORDISC. It uses multiple methods to calculate F-test typicality from publicly accessible craniometric data drawn from the Howells data set, a data set also included as reference groups within FORDISC. While the results of these calculations agree across various F-tests proposed by different authors, the results do not match the “TypF” values generated by FORDISC when using the “resubstitution” option. Through additional calculations and various reproducibility exercises, the authors demonstrate how and why “TypF” in FORDISC produces erroneous typicality values with the “resubstitution” option. They also identify the correct equation to incorporate into the software to rectify this problem. This work represents the logical conclusion of a long-running debate the authors had with Stephen D. Ousley and a desire to improve the accuracy and interpretability of analyses generated in FORDISC.