The conditional independence assumption in dental developmental age estimation

Residual correlations between developing teeth, correlations that persist after accounting for the effects of chronological age, pose a challenge to estimating age intervals using multiple teeth. If the teeth are assumed to be conditionally independent, meaning that there are no residual correlations, deriving age intervals is mathematically straightforward. If the conditional independence assumption is false, however, estimated age intervals will be narrower than is supported by the data, resulting in errors that are larger than expected and potentially biasing estimates of age. Bioarchaeologists need methods that are unbiased in order to conduct accurate paleodemographic reconstruction in past populations. Forensic anthropologists need age estimation methods that have known error rates in order to meet the Daubert standard for expert testimony. Overly narrow age intervals will result in false exclusions, compromising the process of identifying deceased children. This dissertation produces a method of dental developmental age estimation that accounts for residual correlations between teeth. The extent of residual correlations and the axes along which these correlations vary are tested. Using these findings, a method that incorporates residual correlations is developed, validated, and tested. First, the conditional independence assumption is tested in a single large dataset of dental development scores from living individuals (N=2606). Residual correlations are non-zero and incorporating these correlations into estimates of age via a multivariate cumulative probit model reduces age interval error rates from ~22% to the expected value of 5%. Second, residual correlations are consistent with theories of dental development previously established using morphology and eruption data. Patterns are observed by sex but not by ancestry. Finally, validation of the age estimation model with nine different possible correlation matrices on a second dataset of deceased individuals shows that correlation matrices with lower measures of variability result in fewer false exclusions without increasing the width of the age interval. Using an appropriate informative prior produces narrower age intervals but more bias in residuals relative to an uninformative prior. The final test error of the best-performing combination of correlation matrix and prior is 14%. Test error of the multivariate model was therefore intermediate between the uncorrected model error rate of 22% when conditional independence was assumed and the expected value of 5%. This study demonstrates that accounting for residual correlations when estimating age from multiple teeth improves age estimation performance. In addition, the residual correlation matrix contains meaningful information about dental development that would otherwise be lost. The increase in error from training data to test data suggests that interobserver error in dental scoring is an avenue of future study.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Anthropology

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Sgheiza, Valerie

Contributors dc:contributor

• Konigsberg, Lyle W
• Shackelford, Laura
• Hughes, Cris
• Stull, Kyra

• Dental Development
• Age Estimaton

Statement dc:rights

• Copyright 2022 Valerie Sgheiza

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/117645

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F117645