Introduction

In this chapter, we present some thoughts on how to think about age estimation problems in paleodemography. After this has been addressed, we offer statistical solutions to the problems that arise, expanding on the results of Müller et al. (2001). Finally, we point out several problems that may be of future interest to researchers, both in practice and in theory.

The problem

One problem that arises from paleodemographic data is that they do not lend themself to the typical methods of demography. Since the age-at-death is not known, classical demographic and statistical methods encounter problems when the aim is the construction of a mortality schedule.

The available data typically consist of a classification of age indicator stages from skeletal remains. Such characteristics are scored by the physical anthropologist, and are often assigned to categorical stages or classes. Ideally the mapping from skeleton to stage is such that each age class is assigned to a unique stage. In practice, however, stages overlap substantially and are fraught with error, and information is sparse. The only information available to the paleodemographer is the observed frequency counts resulting from the classification by the physical anthropologist.

For the proposed approach we require two distinct datasets. The first dataset, the reference or training set, contains both actual age-at-death and the age indicator stage to which the skeletal remain is assigned. Ideally there would exist a standard reference/training set, i.e., a scoring technique would be evaluated based on a single reference set.