Friday, 18th December, 2015

15.30-16.30

Room BG10, Ground Floor, Oakfield/Barley House

An instrument is not enough: Monotonicity, homogeneity, and instrumental variable methods

Abstract

Epidemiologists increasingly use instrumental variable methods to estimate causal effects. These methods rely on investigators proposing a variable, known as an instrument, that (1) is associated with treatment, (2) causes the outcome at most through treatment, and (3) shares no causes with the outcome. However, even if an investigator can find a true instrument, an instrument alone is generally insufficient for obtaining a point estimate for a causal effect. Despite acknowledgment in the methodological literature, the additional conditions necessary for obtaining point estimates are infrequently mentioned in IV applications. In this presentation, Dr. Swanson will discuss potentially pervasive violations of these additional conditions in common IV applications, and propose tools for understanding and quantifying biases when the conditions do not hold. 

Biography

Dr. Swanson is an Assistant Professor in the Department of Epidemiology at Erasmus MC, Rotterdam, the Netherlands. She previously completed her doctoral and post-doctoral training in the Department of Epidemiology at the Harvard T. H. Chan School of Public Health. Her methodological research interests include developing tools for understanding potential biases when estimating causal effects, with a recent focus on instrumental variable methods.

ALL WELCOME