Causal Inference in a Decision-Theoretic Framework Seminar

Aiming to make causal inference for the variables of interest, the DT framework differentiates between observational and interventional regimes using a non-stochastic variable to index the regimes. Typically, we consider the regimes under which data can be/is collected (observational regimes) and a number of interventional regimes that we want to compare. Appreciating the fact that we mostly have access to observational data, we focus on deducing information from the observational regime for the interventional regimes. The conditions that enable us to transfer information across regimes are expressed in the language of conditional independence. Using the language and calculus of conditional independence we discuss the regression discontinuity design and the more complicated case of dynamic treatment strategies (where we are concerned with the control of a variable of interest through a sequence of consecutive actions).