Download my dissertation at: https://etda.libraries.psu.edu/catalog/13612smg5036
Skeletons are the most direct way to study long-term trends in longevity, mortality patterns, and disease experience for much of human existence. Adequate age estimates can be produced for children and young adults, but such estimates for most of adulthood remain elusive. This four-phase dissertation is a large-scale proof-of-concept that accurate and precise age estimates can be produced for all of adulthood without the two most widely used skeletal indicators—the pubic symphysis and auricular surface. This work builds on over two decades of research by an international research team and an existing age-estimation method called Transition Analysis (TA).
In Phase 1, more than 200 trait variants were investigated to identify and refine a set of age-informative features throughout the skeleton, and primary reference data were collected from four collections of modern, known-age North American skeletons (N=1010). In Phase 2, a simplified procedure based on existing TA was developed to produce age estimates from Phase 1 reference data. In Phase 3, standard age-estimation methods and new TA were applied to additional known-age samples—one modern and one historical. In both samples, the estimates produced by new TA have similar accuracy to traditional methods, but the ranges are, on average, half as wide and show essentially no systematic point estimate bias. In Phase 4, traditional methods and new TA were applied to two Danish archaeological samples.
Comparing the Phase 3 and 4 samples reveals that each traditional method produces a characteristic pattern of adult mortality that is practically independent of the age distribution of the sample. Thus, all mortality profiles generated for past populations using traditional techniques should be viewed with critical skepticism. In contrast, new TA produces mortality distributions that more closely approximate reality, including details that could not be detected using traditional techniques. This dissertation, in conjunction with a larger NIJ-funded research project using the same approach for geographically diverse modern populations, provides every indication that the new TA procedure may become the new gold standard for adult age estimation.
Investigation and Critique of the DiGangi et al. (2009)
Age-At-Death Estimation Method
Most commonly used methods of age estimation have several shortfalls. They tend to over-estimate the age of young individuals, under-estimate the age of older individuals, utilize terminal age categories, such as 50+, provide age ranges which are too precise or too wide to be of practical use in a forensic setting, and fail to provide prediction intervals based on an explicit probability. To address these issues, the DiGangi et al. (2009)1 first rib aging method utilizes transition analysis on features of the first rib previously investigated by Kunos et al. (1999)2 in the Hamann-Todd collection. The newly developed method was first applied to positively and presumptively identified males of Balkan ancestry collected in the former Yugoslavia (n=470). The application of the method, as described in the original publication, requires only that observers familiarize themselves with descriptions of the traits to be scored and the example photos found in the appendix, score the features of the ribs as described, and refer to the table of posterior densities provided in the article to find the appropriate age prediction range and the point estimate of age. The purpose of this study is to evaluate the performance of this method.
To assess inter- and intra-rater agreement, four graduate students with advanced osteological training scored 113 ribs of white males from the Hamann-Todd collection ranging from 21 to 88 years. Sub-samples of individuals were re-coded from the total sample by each observer to allow for the calculation of intra-observer agreement. The ‘irr’ package in R.2.10.10 (2009)3 was used to assess levels of agreement for the costal face, tubercle facet, and combined scores. The data were analyzed using tests for both nominal and ordinal data. Despite the fact that the published 95% probability intervals for each combination of scores range from 35 to 50 years in width, individuals were only placed into an age range that contained their true age on average 87% of the time. With the exception of four younger adults between 20 and 35 years of age who were problematic for all observers, all individuals incorrectly aged were above 55 years of age.