RESUMEN
The use of advanced statistical methods, such as transition analysis, has transformed adult age estimation into a systematically and statistically appropriate practice. The method developed by Boldsen and colleagues (2002) uses 36 features from the cranial sutures, pubic symphysis and auricular surface to calculate maximum likelihood point estimates and 95% confidence intervals, using the ADBOU computer software. However, when using the method in a geographically and contextually distinct sample, such as South Africa, accuracy and precision is of concern. This study aimed to test the repeatability, accuracy and precision of the transition analysis method, using the ADBOU computer software, on a South African sample. Age estimations were generated, for 149 black individuals from the Pretoria Bone Collection, using three individual components as well as different combinations of components and prior distributions (uniform and informative). The informative prior distributions represented both an archaeological and a forensic context. Cohen's kappa statistic uncovered some failings in the scoring procedure. While the accuracy compared favourably with existing methods, the method lacked satisfactory precision. Although combining the components improved accuracy and precision, removing the cranium from the combination was beneficial in some instances. The influence of population variation was observed in the scoring procedure, reference sample and the prior distributions. Validity may be improved for a South African sample by adding age-related components that have been developed on a relevant population. A prior distribution based on South African mortality rates might also be beneficial.