Bioarchaeological Age Estimation Bias Correction

What this research found

Chronic bone and dental pathologies distort the very skeletal features used to estimate age at death. A Bayesian transition-analysis model was extended with a disease covariate and applied to a 387-adult working sample calibrated to the Christ Church Spitalfields documented collection, where surviving coffin plates record the true ages. The correction barely moves the population-level age distribution, but it shifts individual estimates by as much as 1.75 years, and in the worked example it cut absolute error from 1.42 years to 0.17 years.

• Periodontal disease is the largest single distorter of the traditional age estimate at +0.52 years, ahead of vertebral wear (+0.48 years) and osteoarthritis (+0.47 years). All three carrier-versus-non-carrier contrasts give Cohen's d near 1.
• At the population level the correction is negligible. Across the 387 individuals mean signed error moves from +0.119 to +0.109 years and mean absolute error from 5.78 to 5.82 years.
• The effect concentrates in individuals rather than the aggregate. In the high-disease subgroup — two or more pathologies present, 240 skeletons — the corrected estimate is pulled 0.24 years younger on average and up to 1.75 years for a single skeleton, while low-disease skeletons shift the opposite way.
• The pubic symphysis held up better as an age clock than the auricular surface. Its fitted disease shift was +0.195 years per standard deviation of disease load (likelihood-ratio p = 0.86), against −1.126 years for the auricular surface (p = 0.074).
• Severity, not mere presence, drives the correction. Regressing the per-individual correction on the three standardised severity grades explains 68.3% of its variance, against 44.3% for the binary presence indicators.

How it was done

The Spitalfields crypt collection is access-controlled, and no open tabular file linking documented ages to indicator stages and pathology scores was retrievable, so the analysis used a 387-individual working dataset statistically calibrated to the collection's published descriptive statistics. Standard transition analysis treats each ordinal age indicator — Suchey-Brooks pubic-symphysis stages and Lovejoy auricular-surface stages — as a cumulative-normal function of age; here it was augmented with a disease covariate that linearly shifts every transition mean, following the framework Fuchs and colleagues published in February 2026. Disease-blind and disease-conditioned parameter sets were both fitted by maximum likelihood, combined with a Gompertz-Makeham mortality prior fitted to the documented ages, and integrated on a 343-point age grid to give per-individual posterior means. The output is an 18-page tutorial that walks a single skeleton through all six Bayesian steps alongside the distortion regressions and figures.

Data sources

• Christ Church Spitalfields documented skeletal collection, London, 1729–1859 — 968 named individuals; working sample of 387 adults reconstructed from published summary statistics
• Fuchs et al., International Journal of Paleopathology (2026) — the perspective paper that names disease-related age
• Molleson & Cox, The Spitalfields Project Volume 2: The Anthropology (1993)
• Brooks & Suchey pubic-symphysis method (1990); Lovejoy et al. auricular-surface method (1985)
• Boldsen et al. transition analysis (2002) and the Milner & Boldsen known-age validation study (2012)