An Introduction to Estimating Equations

When: -

Where: 160 Cunz Hall

Paul Zivich, assistant professor at University of North Carolina at Chapel Hill, and Ohio State alumnus ‘15 BSPH, ‘16 MPH, visits for this talk hosted by the Division of Epidemiology.

Abstract

Epidemiologists are often tasked with using analytic methods to account for systematic errors (e.g., confounding, measurement error, selection bias). Many of those methods require estimating nuisance parameters, parameters that are not of direct interest but are needed to estimate the interest parameter (e.g., sensitivity and specificity to correct for measurement error). As standard approaches to estimate the variance are no longer valid generally, epidemiologists often resort to the bootstrap. Estimating equations provide an alternative to the bootstrap that removes the associated computational burden. In this presentation, I will highlight how I have found estimating equations to be useful both conceptually and practically. I draw from examples of their use to account for informative missing outcome data, adjust for confounding in preterm birth research, and compare treatment arms across different trials on HIV treatment.