Biology Reference
In-Depth Information
TABLE 3.2 Pubic Symphysis Reference Table
Median Age within Stage
(exponentiated mean age)
Stage
Mean Log Age
Precision (1/var)
1
2.708094
15.00
10.932196
2
2.844875
17.20
8.625380
3
3.154550
23.44
10.341600
4
3.491685
32.84
8.948107
5
3.836729
46.37
9.964530
6
4.110183
60.96
13.583607
CASE STUDY: BAYESIAN THEORY APPLIED TO THE
MULTIFACTORIAL AGE INDICATOR PROBLEM
Uhl et al. (2011)
made use of a more diverse, and possibly more appropriate, reference
sample and familiar skeletal scoring techniques to estimate age-at-death from multiple indi-
cators when combined with an appropriate prior age-at-death distribution. Their data set
consisted of age indicator scores for pubic symphysis (six phases;
Brooks and Suchey,
1990
), auricular surface (eight phases;
Lovejoy et al., 1985a
), and sternal rib end (eight phases;
I¸can et al., 1984a, b, 1985
) for 623 individuals from four modern collections: the Hamann-
Todd Collection, the Bass Collection, the Terry Collection (all mentioned earlier), and the Pre-
toria Bone Collection.
14
One initial issue addressed in studies employing transition analysis (
DiGangi et al., 2009;
Langley-Shirley and Jantz, 2010; Uhl et al., 2011
) is whether the original scoring follows
a particular transition model. A Lagrange multiplier test assesses these transition models
by testing whether the addition of more variables improves the fit of the model.
Uhl et al.
(2011)
found that the original six-phase pubic symphysis scoring and the eight-phase rib
end scoring fit well in a cumulative log-probit model. The auricular surface scoring did
not fit well, so the first four phases in the
Lovejoy et al. (1985a)
system were collapsed into
a single phase. After making this collapse, the scoring did fit well in a cumulative log-probit
model.
Following initial testing, Uhl and colleagues randomly sampled 100 individuals struc-
tured on age-at-death using a Gompertz hazard model of mortality estimated from the
ages at death for Judy Suchey's Los Angeles County male forensic data. This Gompertz
model was also used as the informative prior in estimating ages for the 100 individuals. After
forming this "hold out" sample of 100, log-normal transition models (to normalize the data)
were fit using the remaining 523 individuals, and the 95% highest posterior density region
was found for each of the 240 morphological patterns (6 pubic symphyseal phases times 5
auricular surface phases times 8 rib phases) combined with the informative prior. The left
and right boundaries of the distribution were placed in a reference table and then compared
14
Curated by the Department of Anatomy at the University of Pretoria, Pretoria, South Africa.
