PAPER ANTHROPOLOGY Louise Corron, 1 Ph.D.; Franc ßois Marchal, 1 Ph.D.; Silvana Condemi, 1 Ph.D.; Kathia Chaumo ^ ıtre, 1,2 Ph.D., M.D.; and Pascal Adalian, 1 Ph.D. A New Approach of Juvenile Age Estimation using Measurements of the Ilium and Multivariate Adaptive Regression Splines (MARS) Models for Better Age Prediction ABSTRACT: Juvenile age estimation methods used in forensic anthropology generally lack methodological consistency and/or statistical validity. Considering this, a standard approach using nonparametric Multivariate Adaptive Regression Splines (MARS) models were tested to predict age from iliac biometric variables of male and female juveniles from Marseilles, France, aged 0–12 years. Models using unidimensional (length and width) and bidimensional iliac data (module and surface) were constructed on a training sample of 176 individuals and validated on an independent test sample of 68 individuals. Results show that MARS prediction models using iliac width, module and area give overall better and statistically valid age estimates. These models integrate punctual nonlinearities of the relationship between age and osteometric vari- ables. By constructing valid prediction intervals whose size increases with age, MARS models take into account the normal increase of individ- ual variability. MARS models can qualify as a practical and standardized approach for juvenile age estimation. KEYWORDS: forensic science, anthropology, juvenile age estimation, MARS models, ilium, biometrics Anthropologists can choose from an array of methods to esti- mate age from juvenile remains (e.g. 1–10). These methods rely on various dental and skeletal development indicators expressing growth and/or maturation changes that occur between early fetal life and early adulthood. It is recognized that dental development is the most accurate age indicator (11–13), followed by bone growth (accessed via bone measurements, especially long bones) and bone maturation (ossification sequences) (14,15), with the reservation of possible sample/population specificities (13,16). Since the earliest studies on dental and skeletal growth and matu- ration, methodological precepts have come to change the standards required to construct valid age estimation methods. Moreover, a large number of articles have tested and compared methods, high- lighting the possible limitations of method construction or applica- tion (17–19). Ferembach in 1979 (20), the Arbeitsgruppe Forensische AltersDiagnostik (AGFAD) study group in the early 2000s (21), and several other authors more recently (22,23) have emitted recommendations for age estimation method construction in physical anthropology. They insist on respecting several statisti- cal prerequisites: sufficient reliability, precision, and accuracy of the estimates and recommend these parameters be mentioned explicitly in publications, along with other details about sample composition: adequate sample size; proven age of the subjects; even age distribution; clear definition of the examined features; data on the reference population regarding genetic/geographic ori- gin, socioeconomic status, and state of health (depending on authors); mean value and range of values for each examined fea- ture; and a detailed description of the methods used. Verifying residual (estimated age minus real age) homoscedasticity is rarely mentioned, but it could be added to this long list of prerequisites because it is the condition for constructing valid prediction inter- vals (PIs) of the estimates. Even though age estimation methods almost systematically provide PIs, homoscedasticity of the residual values is rarely explicitly mentioned. It is a fact that most juvenile age estimation methods used by anthropologists present several biases, in both method construc- tion and application. This is summarized by Stull and collabora- tors (24) who separate these biases into two categories: misapplication and the use of inappropriate samples. The “use of inappropriate samples” leads to the inability to obtain correct age estimates when using a method developed on a different population with intrinsic variability and secular trends, affecting the rates of growth and development. Added to the risk of “mis- application,” it translates into the current trend of constructing and using either “population-specific” methods (23,25,26) or “non-population-specific” methods (27), the choice of which is left to the authors. Another important and problematic limitation to practical method application is the lack of prediction intervals that provide the reliability of an age estimate (24). The fact that extant research is still actively going proves that sufficient relia- bility, accuracy, and precision have not yet been reached where 1 UMR 7268 ADES – Aix-Marseille Universit  e – EFS – CNRS, Facult e de M edecine de Marseille Secteur Nord, 51 Boulevard Pierre Dramard, Mar- seille Cedex 15 13344, France. 2 Service de Radiologie et Imagerie M edicale, CHU Nord, Assistance Pub- lique des H^ opitaux de Marseille, Chemin des Bourrely, Marseille Cedex 20 13915, France. Received 11 May 2015; and in revised form 9 Feb. 2016; accepted 6 Mar. 2016. 1 © 2016 American Academy of Forensic Sciences J Forensic Sci, 2016 doi: 10.1111/1556-4029.13224 Available online at: onlinelibrary.wiley.com