The four (and a half) families
In this section, we’ll present an overview of the four families of causal discovery methods. By the end of this section, you should have a good grasp of the four families and their core properties.
We use the word families rather than a more formal term such as type, as the distinction between the four families we’ll use might be slightly vague. We follow the categorization proposed by Glymour et al. (2019) and extend it slightly to include more recent causal discovery methods not mentioned in Glymour and colleagues’ paper.
The four streams
The origins of modern causal discovery can be traced back to the works of Judea Pearl and the 1987 paper he co-authored with George Rebane. The paper described a method that recovers causal structure from statistical data, assuming that the data-generating process has a poly-tree structure (Rebane and Pearl, 1987).
The ideas that followed from this research led to two parallel research...
