Colloquium: Barking up the Wrong Tree with Causal Inference: The limits of confounding, the boundlessness of measurement error
Join us March 24 for our next invited speaker of the semester!
Dr. Donna Spiegelman from Yale University will be presenting at 11 AM in the Z. Smith Reynolds (ZSR) Auditorium, Room 404.
Donna Spiegelman is the Susan Dwight Bliss Professor of Biostatistics at Yale School of Public Health and founding director of the Center for Methods in Implementation and Prevention Science, with additional appointments in statistics, medicine, and global oncology. Her work focuses on developing statistical methods and implementation science approaches to improve public health, particularly in areas like cancer, HIV, and cardiovascular disease prevention.
Dr. Spiegelman’s talk is entitled “Barking up the Wrong Tree with Causal Inference: The limits of confounding, the boundlessness of measurement error.”
Abstract: It has been claimed that confounding is the major source of bias in epidemiology, and why epidemiology ‘gets it wrong’. Upon infinite replicates of the same trial – an unrealistic scenario — and in large single trials, randomization will balance measured, unmeasured, and still to be discovered confounders. In disputing claims by the tobacco industry that denied cigarette smoking as the primary cause of lung cancer, Cornfield (1954) showed that confounding bias in non-randomized studies is limited by the minimum of the confounder-outcome relative risk, the confounder-exposure association, and the confounder prevalence. These bounds are tight, explaining why confounding adjustment often has little impact on results, whether traditional or modern causal inference methods are used. On the other hand, measurement error is widespread in observational and experimental research, including in randomized nutrition and other trials where non-adherence may feature prominently. Bias due to measurement error bias has no bounds. These remarks will be elaborated upon and illustrated by prominent examples in nutritional and environmental epidemiology.