Validating the accuracy and repeatability of transition analysis for age estimation in South Africa

Abstract

Transition analysis transforms skeletal traits with an invariant, unidirectional series of stages into a likelihood function with a maximum likelihood value and a 95% confidence interval. Boldsen et al. used transition analysis to develop an adult age estimation method employing components of the cranial sutures, pubic symphysis and ilial portion of the sacroiliac joint, used either in combination or individually. This validation study aimed to use the 36 transition analysis numerical, categorical scores for the anatomical features in conjunction with the ADBOU computer program to assess the accuracy and precision of the age estimates for 149 black individuals from the Pretoria Bone Collection. In addition, the effect of observer variability in scoring of these traits was assessed. Six age estimations were generated by the ADBOU computer program using 1) the cranial sutures only, 2) the pubic symphysis only, 3) the auricular surface of the ilium only, 4) all three features combined, 5) all three features combined and modified by a forensic prior distribution and 6) all three features combined and modified by an archaeological prior distribution. The six point estimate categories, calculated from the maximum likelihood values, were evaluated for accuracy using mean absolute values. The 95% confidence intervals were evaluated for range width and accuracy. Cohen’s Kappa statistics were used to analyse repeatability of the scoring procedure through inter- and intra-observer agreement and Kruskal-Wallis ANOVA statistics to determine the effect of observer differences on the final age estimates. The usefulness of the age ranges were diminished by large widths encompassing up to 95 years. The accuracy for the point estimates fared better for the combined skeletal indicators and overall accuracy was improved by using the archaeological prior distribution. The archaeological prior distribution was also responsible for narrowing the age ranges, especially in the older ages (over 70 years). Age estimates did not differ significantly when using inter- and intra-observer scores, but experience with the method did seem to improve results. Overall, age ranges were too wide, but accuracy could potentially be improved by adding more skeletal components to the method and using a population-specific prior distribution. The method would need considerable adjustments to make it usable in a South African setting.