Understanding Population-Specific Age Estimation Using Multivariate Cumulative Probit Regression For Asian Skeletal Samples

For many years, the field of anthropology has encouraged anthropologists to assume that population variation exists in skeletal aging although interpretations of population specificity in skeletal aging have been inconsistent. This project investigates age progressive changes in modern East and Southeast Asian populations, and attempts to quantify the magnitude of differences or similarities in skeletal aging between different Asian groups as a first step to develop a more inclusive age estimation method for Asian populations. Specifically, this study explores the utility of currently available age estimation methods for Asian populations, asks whether a population-specific aging method should be region-specific (Thai vs. Japan) or continental-specific (Asian), and investigates whether population specificity in age estimation is needed at all.

To achieve these goals, data from four skeletal collections representing 20th century Japanese and modern Thais were collected. The four age estimation methods were applied: Transition Analysis (TA) (Boldsen et al. 2002b), plus three ‘conventional’ methods of Suchey and Katz (1998), Lovejoy et al. (1985a), and Meindl and Lovejoy (1985). To develop age estimation models for the Asian samples, while minimizing methodological error a multivariate ordered probit regression model was fitted to the Asian skeletal data under a Bayesian framework.

The results of this study show that Japanese-specific age estimation models do not necessarily increase error in age estimates of Thai populations, and vice versa. Thus, multivariate ordered probit models fitted on a pooled sample of Japanese and Thais yielded significantly improved age estimates. Although TA and conventional methods yielded reasonable age estimates for Japanese and Thai populations, the probit models of pooled Asian groups outperformed those methods. The results are even more promising when the scores of TA and conventional methods were combined to fit a multivariate ordered probit model.

Broader implications of this research is that, at least for Japanese and Thais, a single age estimation method can be developed. In addition, any error and bias in ages at death estimation may not be due to differences in population-specificity in skeletal aging and age estimation error induced by biased reference samples may be greater than expected.