Figure 1.
Directed acyclic graphs (DAG) indicating the relationships between an instrumental variable (G), exposure (X), confounder (U) and outcome (Y). Selection (S) leads to bias if it is a function of the exposure (left panel) or the outcome (right panel), as both exposure and outcome are causally downstream of the genetic variant and confounder, and hence conditioning on selection induces an association between the genetic variant and confounder in both cases.