Multiple and dynamic treatments*

Abstract

Abstract This chapter generalizes binary treatments to multiple treatments, which can occur at a given time. Multiple treatments can occur over time, as well at each given time point having a binary treatment. The latter aspect leads to dynamic treatment effect, which is an important and evolving area of research. When a binary treatment is given consecutively over time, its duration can be handled as a single ‘cross-section’ cardinal treatment if the covariates are time-constant. Otherwise, if some covariates are time-variant, a ‘hazard-based’ causal framework may be used. The most difficult case in dynamic treatment effect arises when interim responses that are controlled for are affected by the earlier treatments, and affect future treatments; the ‘G algorithm’ is introduced for this case.