Biology Reference
In-Depth Information
STAT
ISTICS AND ADULT AGE-AT-DEATH ESTIMAT
ION
Traditional Statistical Methods
Many of the most popular methods for estimating age-at-death use phases for categorizing
remains. Large skeletal samples (for example, the Suchey
e
Brooks method was initially devel-
oped on a sample of almost 800 individuals) are studied for trends in morphological change
and then those changes are described and divided into phases. In other words, a particular
bone will be chosen based on observations that a particular feature seems to demonstrate
age-related change, and all the bones in the collection will be seriated based on age cohorts
(15
e
20; 21
e
30; etc.) so that the changes can be easily described and grouped into phases.
Typicallyworkerswill publish themean age of individuals in each stage, which is oftenused
as thepoint estimate, alongwith the 95%confidence intervals for eachphase for constructing an
age range. Sometimes authorswill instead include the standarddeviation or standard error for
each phase for use in constructing age ranges. Understanding the statistical theory underlying
age-at-death estimates (mostly based on regression analysis) is very important for anthropol-
ogists employing these methods, especially in medicolegal cases that could require trial testi-
mony. The Methods section of any age-at-death method paper will have a discussion of the
statisticalmethodsutilized, andthereforeknowledgeof statisticswill facilitate comprehension.
Transition Analysis and Bayesian Theory
Transition analysis uses a “known” reference sample (skeletons for which sex, age-at-
death, stature, etc. is known), preferably from the same human population as the individual
being analyzed, and a hazard model (information that models survivorship) to estimate at
what age those known individuals transition from one phase to another. This type of analysis
provides a highest posterior density distribution (the probability after a priori (prior) infor-
mation has been taken into account in Bayesian analysis) of age-at-transition; this informa-
tion can be used for age-at-death estimation of an unknown individual. Because this
analysis uses “prior” information (the reference sample) it is considered Bayesian in nature.
Bayesian theory and the statistical methods developed from it use a priori information to alter
the final probabilities.
In the context of age-at-deathestimation theprior information includes the ages-at-transition
of the known reference sample and a model of mortality for that population (see Konigsberg
and Frankenberg [Chapter 11], this volume). To estimate an individual's age-at-death, one
can calculate a percentage of the distribution (similar to a confidence interval, but it is distinct
because the distribution is asymmetrical so this range is not centered on a mean), which gives
a range. An anthropologist using a 95% estimate would report a larger (more conservative) age
range, while an anthropologist using a 50% estimate would report a smaller range.
Accuracy and Precision
There has been much discussion as to the trade-off between accuracy and precision in
reporting age-at-death estimates. In this context, an accurate age-at-death estimate refers
to an estimate (range) that
includes the actual age-at-death of
the decedent. While
