Thursday, 11 October, 6:00 pm -7.30pm
The University of Melbourne
G04/Theatre 1, Old Geology
- Drinks from 5.30pm in the Staff Tea Room, Peter Hall Building
The inaugural lecture in this series will be given by Professor Raymond J Carroll, Distinguished Professor of Statistics, Nutrition and Toxicology at Texas A&M University, Distinguished Professor at the University of Technology Sydney, Australia, and friend and collaborator of Peter. He was the first statistician to receive a U.S. National Cancer Institute MERIT Award. He served as editor of Biometrics and the Journal of the American Statistical Association. He has won many honours in the profession, including the COPSS Presidents’ Award and the Fisher Award and Lecture.
Title of the talk: Deconvolution for Episodically Consumed Dietary Intakes
Intakes: A Brief Personal Tour of Statistical Methods (with reference to Peter Hall’s fundamental contributions)
Abstract: Nutritional Surveillance is the estimation of the distribution of usual (longer-term average) dietary intakes, while Nutritional Epidemiology is about relating these usual intakes to diseases (cancers, cardiovascular) and to mortality. The crucial problem from both nutritional and statistical perspectives is that measuring longer-term dietary intakes is impossible, and only short term measurements or questionnaires are available. Remember, a bad nutrition day is not the same as a bad longer-term diet! This measurement error/uncertainty of measurement causes biases in both surveillance and epidemiology, and correcting for the biases is typically described as a deconvolution problem. My first paper with Peter Hall was on this general topic, and he went on to make numerous fundamental contributions to this area.
After personal remembrances of Peter Hall and his impact on me and this area, I will focus on the important surveillance question: What Percentage of U.S. Kids Have Alarmingly Bad Diets? The answer is surprising. I worked on this problem because my nutrition colleagues at the U.S. National Cancer Institute did not believe the published estimates arising from a naïve analysis that ignored the problem of what turns out to be severe measurement error. My biggest focus will be on the issue of how to handle measurement error involving multiple episodically consumed foods, i.e., those not consumed daily, such as whole grains, whole fruits, vegetables, etc. Answering the question above turns out to be a stunningly difficult deconvolution problem. I will describe, without any of the many technicalities, the gist of my original solution to the question, along with recent advances.