Inferences about and Inferences from: A Comparison of Kernel Density Estimation and Latent Mixture Modeling in Demographic Temporal Frequency Analysis

Summary

This is an abstract from the "Novel Statistical Techniques in Archaeology I (QUANTARCH I)" session, at the 84th annual meeting of the Society for American Archaeology.

Temporal frequency analysis (TFA) comprises methods both for the characterization of temporal distributions of archaeological samples and for drawing inferences about their underlying data generating processes (DGPs). In motivation, these two activities resemble descriptive and inferential statistics, respectively. However, several sources of uncertainty confront archaeological chronometry, necessitating engagement in statistical inference at both levels. Perhaps because of this, the demarcation between ‘inferring about’ and ‘inferring from’ in TFA has frequently been blurred, accounting for some of the contention surrounding interpretations of temporal distributions rendered through histogram aggregation, probability summation, and most recently kernel density estimation (KDE). In TFA’s demographic applications, ‘inferring from’ is arguably the more relevant activity, incentivizing efforts to identify methods specifically well-suited to DGP recovery. While KDE has often been favored over parametric methods in this capacity for it purportedly superior ability to accommodate inconveniently complex DGPs, it also only yields single-shot estimates thereof. Furthermore, it requires single-point measures of, or guesses at, the timestamps anchoring archaeological samples to the timeline, at odds with probabilistic chronometry. Here I advocate an alternative approach – latent mixture modeling – which resembles KDE in form, flexibility, and output, yet which is better-suited to probabilistic chronometry, likelihood-based model-fitting, and uncertainty quantification.

This Resource is Part of the Following Collections

• Society for American Archaeology 84th Annual Meeting, Albuquerque, NM (2019)   •
• Novel Statistical Techniques in Archaeology I (QUANTARCH I)

Keywords

Spatial Coverage

Record Identifiers

Abstract Id(s): 23813