This episode examines methods for causal inference when unobserved variables, known as confounders, complicate identifying true causal relationships. It begins by discussing sensitivity analysis to assess how robust causal inferences are to such unobserved confounders. The text then introduces instrumental variables (IVs) as a technique to identify causal effects in the presence of these hidden factors, offering both partially linear and non-linear models. Furthermore, the chapter explores the use of proxy controls, which are observed variables that act as stand-ins for unobserved confounders, to enable causal identification, extending these methods to non-linear settings. Throughout, the document highlights practical applications and the role of Double Machine Learning (DML) in these advanced causal inference strategies.
Disclosure
