Semiparametric method for estimating paleodemographic profiles from age indicator data

Hans‐Georg Müller, Brad Love, Robert D. Hoppa

TL;DR

This paper introduces a novel semiparametric statistical method to accurately estimate age-at-death distributions (paleodemographic profiles) from skeletal remains. It overturns the traditional approach by demonstrating that estimating individual ages first leads to biased population profiles, proposing instead to estimate the population profile first using "weight functions" derived from reference skeletal data.

What Problem Does This Paper Address?

Paleodemography, the study of past populations using skeletal remains, relies heavily on accurately determining the age-at-death distribution of individuals within a sample. However, estimating an individual's age from skeletal indicators is inherently imprecise, leading to significant challenges and criticisms in paleodemographic reconstructions. A major issue is "age mimicry," where different age groups can appear similar based on skeletal indicators, and the underestimation of ages for older adults. The traditional approach of first estimating individual ages and then aggregating them to form a population profile has been shown to be flawed, particularly due to the influence of the reference sample's age distribution on the target population's estimated distribution. This paper aims to resolve these issues by proposing a statistically sound method that correctly accounts for the uncertainties in age estimation and the nature of skeletal age indicators.

Methods

The core of the proposed method lies in the concept of weight functions. These functions represent the conditional probability that a skeleton of a specific age-at-death will be assigned to a particular osteological age-indicator stage. The authors advocate for estimating these weight functions nonparametrically from a reference skeletal sample where both age-at-death and age-indicator stage are known. This nonparametric approach allows the weight functions to adapt to complex shapes and avoid biases introduced by misspecified parametric models.

Once the weight functions are estimated, the paper proposes a semiparametric approach for estimating the paleodemographic profile of the target sample. This involves assuming a parametric family for the age-at-death distribution of the target population and then using maximum likelihood estimation to find the parameters of this distribution, incorporating the estimated weight functions. This two-stage process—nonparametric estimation of weight functions followed by parametric estimation of the demographic profile—is what gives the method its semiparametric nature.

Crucially, the paper argues that individual age-at-death estimation should be performed after the population demographic profile has been established, using the derived weight functions and the estimated population distribution. This reverses the traditional order of operations.

Datasets

The paper utilizes a reference dataset of 737 individuals with known ages for the Suchey-Brooks pubic symphysis age-indicator scheme. This dataset was previously published and is described as originating from forensic data (likely from the Los Angeles Coroner's Office, as hinted in the text). The authors also use simulated data to demonstrate the effectiveness of their proposed method and compare it to existing approaches. No new benchmark datasets are introduced.

Key Insights and Ideas

  • The "Osteological Paradox" Reversed: The paper highlights a counterintuitive finding: to accurately estimate individual ages from skeletal remains, you first need a good estimate of the population's age distribution. This is the inverse of the traditional approach.
  • Weight Functions as the Key: The concept of "weight functions" is central. They encapsulate all the necessary information about how skeletal age indicators relate to actual age, and importantly, they can be estimated from reference data and applied to target samples.
  • Nonparametric Estimation for Flexibility: Estimating weight functions nonparametrically is crucial because the relationship between skeletal indicators and age can be complex and vary significantly, making rigid parametric models insufficient.
  • Avoiding Age Mimicry: By directly modeling the probability of an age-indicator stage given age, rather than the other way around, the method avoids the biases associated with age mimicry that plague traditional approaches.
  • Individual Age Estimation as a Secondary Step: The paper firmly establishes that estimating individual ages should be the last step, derived from the population profile, not the first.

Key Results

  • The proposed semiparametric method, when applied to simulated data, accurately recovers the true age-at-death distribution of the target population, demonstrating robustness against age mimicry.
  • In a simulated example, the method's estimate for the population mean age-at-death (62.610 years) was very close to the true value (63.399 years), significantly outperforming the "direct" method (52.080 years), which was heavily biased.
  • Analysis of the Suchey-Brooks pubic symphysis method using weight functions revealed substantial overlap between age-indicator categories, suggesting inherent ambiguity in this scheme and highlighting the need for better indicators or multivariate approaches.
  • The paper provides formulas for constructing confidence regions and a goodness-of-fit test, demonstrating the statistical rigor of the proposed framework.

Limitations and Future Work

  • Data Sparsity: The paper acknowledges that with a limited number of age-indicator stages (I), a fully nonparametric estimation of the demographic profile is not feasible, as the number of estimable parameters is constrained by I-1.
  • Weight Function Overlap: The significant overlap observed in the weight functions for the Suchey-Brooks system indicates limitations in its discriminatory power, suggesting that categories 2 and 3 might need to be combined or that alternative, more refined age-indicator schemes (like multivariate approaches) should be explored.
  • Future Work: The authors suggest further investigation into using weight functions to compare and improve age-indicator schemes, checking the invariance assumption with multiple reference samples, and adapting the methodology for multidimensional age-indicator schemes. They also mention that detailed statistical consistency and algorithmic implementation will be addressed in a separate paper.

Summary updated

Abstract

This paper addresses the problem of estimating an age-at-death distribution or paleodemographic profile from osteological data. It is demonstrated that the classical two-stage procedure whereby one first constructs estimates of age-at-death of individual skeletons and then uses these age estimates to obtain a paleodemographic profile is not a correct approach. This is a consequence of Bayes' theorem. Instead, we demonstrate a valid approach that proceeds from the opposite starting point: given skeletal age-at-death, one first estimates the probability of assigning the skeleton into a specific osteological age-indicator stage. We show that this leads to a statistically valid method for obtaining a paleodemographic profile, and moreover, that valid individual age estimation itself requires a demographic profile and therefore is done subsequent to its construction. Individual age estimation thus becomes the last rather than the first step in the estimation procedure. A central concept of our statistical approach is that of a weight function. A weight function is associated with each osteological age-indicator stage or category, and provides the probability that a specific age indicator stage is observed, given age-at-death of the individual. We recommend that weight functions be estimated nonparametrically from a reference data set. In their entirety, the weight functions characterize the relevant stochastic properties of a chosen age indicator. For actual estimation of the paleodemographic profile, a parametric age distribution in the target sample is assumed. The maximum likelihood method is used to identify the unknown parameters of this distribution. As some components are estimated nonparametrically, one then has a semiparametric model. We show how to obtain valid estimates of individual age-at-death, confidence regions, and goodness-of-fit tests. The methods are illustrated with both real and simulated data. Am J Phys Anthropol 117:1–14, 2002. © 2002 Wiley-Liss, Inc.
BibTeX
@inproceedings{muller2002semiparametric,
  author = {Hans‐Georg Müller and Brad Love and Robert D. Hoppa},
  title = {{Semiparametric method for estimating paleodemographic profiles from age indicator data}},
  booktitle = {American},
  year = {2002},
  doi = {10.1002/ajpa.10000.abs},
}
Figure