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.
