Lectures and seminars PhD student seminar: Arvid Sjölander

Title 

Confounder adjustment with Directed Acyclic Graphs

Abstract

The aim of epidemiologic research is often to assess the causal effect of a particular exposure on a particular outcome. It has been long recognized that in ideal randomized experiments, observed associations can be given causal interpretations. However, due to practical, economical, or ethical constraints, most epidemiologic studies are purely observational (i.e. non-randomized). Informally, people have argued that causal effects can be estimated from observational data, if all “confounders” are adjusted for in the statistical analysis. However, for most of the 20th century, there was no formal definition of the term “confounder”, or a satisfactory strategy to select confounders to adjust for in the analysis. During the last decades, a formal theory of causal inference was developed, based on counterfactuals and directed acyclic graphs (DAGs). In this lecture we will discuss how DAGs can be used to formally define the term “confounder”, and to select what confounders to adjust for in the statistical analysis. 

Contact persons

Daniela Skalt&Antoine Créon