Combining Variables to Improve Subadult Age Estimation

Abstract

Anthropologists have reported that the combination of multiple variables and indicators generally increases precision and reduces bias in age estimates. However, endeavors specific to subadult age estimation have primarily focused on estimating age of the living and therefore on variables and indicators that are active later in ontogeny and easy to image. The current study aimed to determine if multivariable, single-indicator age-estimation models outperform single-variable age-estimation models throughout ontogeny using the three most common subadult age indicators: diaphyseal dimensions, epiphyseal fusion, and dental development. Data were collected from individuals from South Africa between birth and 12 years (N = 601) using Lodox Statscan radiographic images and from the United States between the ages of birth and 20 years (N = 1,277) using computed tomography images. Multivariate adaptive regression splines were used to build the multivariable, single-indicator, and single-variable models. Each subset used for model development had a unique training sample to build the model and testing sample to ensure that the results were generalizable. The multivariable models presented with increased precision and accuracy, reduced bias, and greater consistency across ontogeny compared to the single-variable models for both samples. Eighty percent of the independent test models (20/24) had ≥ 93% coverage, and 75% (18/24) of the independent tests models had ≥ 95% coverage. Besides providing more information to the resulting age estimate, multivariable models remove any a priori beliefs regarding variable importance and eliminate the requirement to contrive a final age estimate from multiple single-variable age-estimation models.