Methods for aging adult skeletal remains typically use measurements from two bones in the pelvis: the ilium and the sacrum.[1] Since it is common for forensic anthropologists to receive incomplete skeletons that may not include the pelvis, they often rely on past experience to estimate age at death. While research has shown that well-trained forensic anthropologists can accurately age bones using past experience, this approach is not quantifiable or reproducible, making the results difficult to use in the courtroom.
Researchers from the Pennsylvania State University, supported by the National Institute of Justice, designed a computer program that “thinks” like an expert forensic anthropologist. Unlike its human counterparts, the algorithm is standardized and provides reproducible estimates of adult skeletal age at death that could be used as evidence in court. When tested with known-age skeletal collections, the program more accurately predicted age than traditional methods based on ilium and sacrum observations.
The researchers say their open-access computer program and accompanying manual can be used by nearly any forensic practitioner to accurately age skeletal remains based on “yes” or “no” questions about 20 bony traits.
Automating Skeletal Age Estimates
The Penn State team aimed to create a new procedure for producing unbiased estimates of skeletal age for adults, including remains from individuals over the age of 50 that cannot be aged accurately with traditional methods.
They gathered data for age-related bony traits from 1,774 adult human skeletons from known-age skeletal collections in Portugal, South Africa, Thailand, and the United States. (The geographic diversity increases the method’s forensic applicability.)
The researchers evaluated a number of potential skeletal age indicators. The indicators had to be binary — either present or absent — to simplify data collection and reduce observer error, and they had to be clearly informative about age. To assess a bony trait’s ability to indicate bone age, the researchers calculated a transition curve where they estimated the probability, at each age, of undergoing a change from present-to-absent or absent-to-present.
Next, they explored two approaches to automated age estimation: One involved estimating skeletal age from bony traits found on two pelvic joints and cranial sutures (seams in the skull). The other approach used machine learning to predict age from 20 bony traits. The machine learning approach estimated skeletal age better than the other automated method and traditional quantitative methods.
Program Dissemination and Impact
The researchers developed a free computer program, called Transition Analysis 3, and accompanying manual based on their machine learning protocol for aging adult skeletal remains from 20 bony traits. The manual features straightforward bony trait definitions, diagrams, and illustrations. The manual is paired with a data collection form to ensure accurate forensic observation recording.
“Perhaps the most important outcome of the project is an anticipated shift from observing a few parts of the skeleton for age estimation to a much wider array of bony features,” concluded the team. “In short, better methods coupled with better skeletal traits and better initial samples inevitably yield better results.”
Transition Analysis 3 is available as a beta version and will be updated as the team furthers their research and receives user feedback.
About this Article
The research described in this article was funded by NIJ grant 2014-DN-BX-K007, awarded to the Pennsylvania State University, University Park, Pennsylvania. This article is based on the grantee final report “Adult Age Estimated from New Skeletal Traits and Enhanced Computer-Based Transition Analysis” (July 2019) by George R. Milner, Stephen D. Ousley and Jesper L. Boldsen.