Causal Analysis Using DAGs: Theory

This chapter presents parts of the theory of using DAG models to study causality. This theory is a major achievement over other approaches to study causal structures in a data analytic environment and is the subject of lots of praise but also of lots of criticism. The chapter does not side with either opinion instead, it formulates three fundamental assumptions with respect to the causal structure, its response to interventions, and possible observations from it and how these three components are related. Many of the important results of the theory may be derived from these assumptions. Checking the plausibility of the three assumptions provides the applied researcher with a tool to decide about the relevance of this theory to describe the actual research problem. The results of this chapter justify the use and the study of the Markov models associated with directed acyclic graphs in the previous chapter. Also, the results provide tools to learning the causal system not only when interventions to the causal system are possible, but to a certain degree also in cases when only an observation is possible without intervention.