Member-only story
Unlock the Secrets of Causal Inference with a Master Class in Directed Acyclic Graphs
A step-by-step explanation of Directed Acyclic Graphs from the basics through to more advanced aspects
Objective
Having spent a lot of time researching causal inference I began to realise that I did not have a full grasp of Directed Acyclic Graphs (DAGs) and that this was hampering my efforts to develop my understanding to a point where I could apply it in order to solve real-world problems.
This objective of this article is to document my learning journey and to share everything you need to know about DAGs in order to take your understanding of Causal Inference to the next level.
Background
I would like to start by proposing a definition for causal inference -
Causal inference is the process of reasoning and the application of conclusions drawn from cause-and-effect relationships between variables while taking into account potential confounding factors and biases.
That is quite a mouthful, but it does encapsulate the key points -
- It is the study of cause-and-effect.
- The point is to draw conclusions that can be applied to solve real-world problems.
- Any bias or “confounding” must be taken account of and compensated for.
Moving beyond the definition, there is an age old saying that “correlation does not imply causation” which leads to the question “so what does then?”
It turns out that causation cannot be inferred or calculated from a set of data in isolation. That data needs to be extended and supplemented with additional information that can propose, visualise and represent the causal relationships and one common approach to the is to use a “Directed Acyclic Graph”.
A Simple DAG
At the most basic level DAGs are very simple indeed. The example below is representing the proposed relationship between taking a drug “D” and recovery “R” and the arrow is stating that taking the drug has a causal effect on recovery …
