Lecture 6. Causal Inference - Learning Causal Effects

Learning Goals: You should understand how causal effects can be modeled and learned from data through structured causal models (SCMs) and the corresponding directed acyclic graphs (DAGs). You should know about how independence relations can be determined directly from the graphical representation. You should know about open and blocked paths and be able to use this information to determine if two variable X and Y are independent given a set Z of other variables. You should know about valid adjustment sets and how to estimate $LaTeX: P(Y\mid \textbf{do}(X:=x))$ $P(Y\mid \textbf{do}(X:=x))$ from data if a valid adjustment set is available. You should know how estimation can be done using the backdoor criterion and frontdoor criterion. You should be able to understand and change python code presented at the lecture and exercise.

You find the Lecture 6 slides here Download Lecture 6 slides here (now updated with corrected numbers in the COPD example)

Video (Note, the recording is from 2020 when this was Lecture 8. This year, it will be Lecture 6).

(Correction: On slide "Bayesian Network..." the index i in the product should be changed to j)

Also available on Youtube here : https://youtu.be/xGnAz-k3RVc Links to an external site.

After watching this lecture you can work with Exercise 6.

Code: For slides 36-37 Links to an external site.

Interested persons might find the Dagitty - a tool for DAG analysis useful to experiment with, to further understand the concepts of causal inference in DAGs. However, we will not have time to get used to working with this tool, so this is outside the course.