Dr. Lyle W. Konisgberg | It's not a Scandal. It's Statistics.

Date
Mon August 6th 2018, 5:30pm
Event Sponsor
The Wenner-Gren Foundation for Anthropological Research; co-sponsors, the Research Institute of the Center for Comparative Studies, CESTA, the Office of the Dean of Research, CLAS, and the Stanford Humanities Center.
Location
Terrace Room, Margaret Jacks Hall (Building 460, 4th floor)

"It's Not a Scandal. It's Statistics"

Public keynote address by Dr. Lyle W. Konisgberg (Department of Anthropology, University of Illinois at Urbana-Champaign)

The title of this talk is a quote from Christine Romans, host of CNN Money. The title harkens back to a discussion the evening of June 9, 1999 after the first day of the first Rostock Work-shop on “Mathematical Modelling for Paleodemography: Coming to Consensus.” At dinner one observer of the workshop referred to the day’s proceedings as “statistical smoke and mirrors.” They then stated that one simply needed to count the number of cementum annuli, add the average age of eruption, and one would know the age to within a year or two. This statement was made prior to the numerous advances that have been made in “cementochronology.” However, even with these recent developments, we are often faced with making age estimations in both the living and the dead that rely on general bone and/or dental development or senescence. This is necessarily a statistical endeavor. The goal is to model development or senescence as a function of age and then “invert” the problem using Bayes’ Theorem. For single traits, this is not a difficult endeavor, but the analysis becomes more complicated when there are multiple traits and particularly when the traits are a mix of binary, ordinal categorical, and continuous variables. It is expected that traits will not be conditionally independent. In other words, there are generally non-zero residual correlations between traits after conditioning on age. Although there have been ad hoc solutions to this problem, the use of Markov Chain Monte Carlo (MCMC) methods allows researchers to esti-mate the residual correlations without recourse to multivariate numerical integration. Ko-nigsberg and Herrmann suggested the use of MCMC methods at the first Rostock Work-shop. These methods have begun to receive greater acceptance and have been applied in paleodemography and in age estimation. MCMC comes with the additional benefit that one can produce posterior predictive distributions of the traits as a check on whether the refer-ence and target samples age in the same fashion. 

This talk is part of the workshop, "New Approaches to Skeletal Age Estimation for Diverse Populations," organized by Bridget Algee-Hewitt (Stanford CCSRE).

Public reception to follow.

*/