Inferring Age at Death for Japanese and Thai Skeletal Samples under a Bayesian Framework of Analysis: A Test of Priors and Their Effects on Estimation

Abstract

A critical step in age-at-death estimation is to identify the age-at-death distribution to which the unknown skeletal remains most likely belong. Age estimation based on a frequentist approach assumes that the age distribution of the target population is same as that of the reference sample. In a Bayesian framework, researchers have greater flexibility, with the freedom to specify mortality information via a prior distribution and integrate it with osteological data to produce more accurate age estimates. The selection of an optimal prior remains challenging, as forensic anthropologists analyze unidentified individuals often originating from unknown populations. Understanding the effects of the prior choice on the final age produced for various populations is essential to the interpretation of these estimates.In this article we investigate the effects of three different priors on age estimates using scores of three traditional methods collected on Asian skeletal samples representing peoples from Japan and Thailand. We test a uniform prior, which assumes the equal chance of death regardless of age, as well as various informative priors, derived from a Japanese mortality database and skeletal collections. We combine each of the priors with the parameters of the cumulative probit regression model to obtain age estimates. The results of our analyses show that the informative prior outperforms when it is carefully chosen to reflect the geographic and temporal origins of the target population. While the uniform prior produces the least-biasedage estimates, one should be cautious, as it can generate unrealistically old and inaccurate ages.

The accuracy of age estimation using transition analysis in the Hamann‐Todd collection

“…Thus, the informative prior distribution must be similar to the target sample for the method to show better accuracy, or alternative approaches to the prior, as proposed by Hens and Godde (2020), should be considered for age estimation. This is consistent with previous findings (Getz, 2020;Godde & Hens, 2012;Kim et al, 2019;Milner & Boldsen, 2012).…”

Current and emerging histomorphometric and imaging techniques for assessingage‐at‐deathand cortical bone quality

“…Estimating the skeletal age of adults, however, generally demonstrates a progressive decrease in accuracy as chronological age increases. The majority of methods for adult age‐estimation are based on degenerative morphological changes of the skeleton (Brooks & Suchey, 1990; Buckberry & Chamberlain, 2002; Hartnett, 2010; Iscan, Loth, & Wright, 1984, 1985; Lovejoy, Meindl, Pryzbeck, & Mensforth, 1985), which can vary between geographic populations (Atkinson & Tallman, 2019; Jayaraman et al, 2019; Schmitt, Murail, Cunha, & Rouge, 2002), ancestral groups (Go, Tallman, & Kim, 2019; Kim, Algee‐Hewitt, & Konigsberg, 2019; Uys, Bernitz, Pretorius, & Steyn, 2019), and within and between individuals (Britz, Thomas, Clement, & Cooper, 2009; Chan, Crowder, & Rogers, 2007; Crowder, 2005; Crowder & Pfeiffer, 2010; Gocha, 2014; Heinrich, 2015; Pfeiffer, Lazenby, & Chiang, 1995; Stout & Gehlert, 1980; Stout & Stanley, 1991; Tersigni, 2005; Thompson & Galvin, 1983).…”

New Approaches to Age Estimation Using Palatal Suture Fusion

“…The Gompertz parameters calculated from the age‐at‐death structure of the informative prior are combined with the ages‐of‐transition from the transition analysis sample. The informative prior should provide a good fit to the age‐at‐death structure of the transition analysis sample, usually by selection of similar mortuary profiles (see discussion in Konigsberg et al [20]), although Godde and Hens (17) found (and later validated by Kim et al [36] and Nikita et al [37]), some deviation from fit can be tolerated. In this paper, the cemetery source was used as our informative prior as it was an acceptable fit to the Donated and Luís Lopes collections age‐at‐death distribution (Fig.…”

“…The multivariate models use composite likelihood estimation (40,41), which differentiates them from the univariate models in this paper that employ maximum likelihood estimation. Consistent with Konigsberg et al (36,42,43), a Lagrange multiplier test for more than two categories (44–46) was applied to test the assumption of ordinal probit models that the unobserved error term is log‐normally distributed.…”