I encountered a causal inference problem in practice and want to find if there is a previously established statistical toolset that can be applied to my problem.

My problem is characterized as follows:

  • My goal is to characterize the causal effects of each $T$ treatment $X_1, \cdots X_T$ on outcome $Y$ where the values of treatments and outcomes are binary (0 or 1).
  • Unlike a typical multiple treatment setting, multiple treatments can be active simultaneously. For example, it is possible for a sample to have $X_1 =1$ and $X_2 =1$.
  • If necessary, the additivity assumption can be introduced. For example, the causal effect of having $(X_1, X_2)=(1, 1)$ is the sum of the causal effects of the two cases: $(X_1, X_2)=(1,0)$ and $(X_1, X_2)=(0, 1)$.
  • As in typical causal inference settings such as that of propensity score matching, there can be multiple common cause variables.

Q1. Is there a specific term describing the problem setting described above? If there is such a term, could you cite a few pedagogical materials?

Q2. If the problem is not studied before, how can it be tackled using existing causal inference methods?


1 Answer 1


A1: A term sometimes used is "joint interventions". However, joint interventions explicitly refers to multiple treatments, not multiple treatments with the additivity assumption. This chapter may be a useful resource to refer to. Ideally, we would avoid the additivity assumption, and I wouldn't recommend using a different term to shorten the phrase (it is clearer to describe the assumption being made rather than use some singular term).

A2: This problem has been tackled through a variety of approaches. As detailed in the above chapter. One of my favorite applied examples is Taubman et al. 2009. The authors apply the parametric g-formula in the time-varying treatments/exposures (which is more complicated but demonstrates the concept). McCaffrey et al. 2013 describe multiple treatments in the context of propensity scores


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.