Introduction

Most approaches to age estimation currently used in paleodemography and forensic science are not based on formal (or even informal) statistical methods. Instead, various ad hoc procedures have been developed, based frequently on simple tabulations of skeletal markers by age. The classic methods of Todd (1920) and McKern and Stewart (1957), for example, involve a nonstatistical assignment of a skeleton's age-at-death according to documented changes in the pubic symphysis. These methods produce either a nonstatistical age range or a point estimate of age, without any assessment of the error structure of the estimate based on formal probability arguments. The individual ages produced in this way are then aggregated to estimate the age-at-death distribution for an entire sample. As discussed elsewhere in this volume, the age-at-death distribution produced by this procedure will usually be biased in the direction of the age distribution of whatever reference sample was used to generate the individual estimates in the first place. In addition, we are left with little understanding of the degree of estimation error involved, either in the individual age estimates or the estimate of the aggregate-level age-at-death distribution as a whole.

In this chapter we explore some statistical methods for estimating age-at-death distributions from skeletal samples, with special emphasis on recovering the parameters of parametric models of the age-at-death distribution (see Wood et al., Chapter 7, this volume). Only methods compatible with the Rostock protocol, described elsewhere in this book, are discussed.