3.9 - The Flow of Association and Causation in Graphs
ฝัง
- เผยแพร่เมื่อ 6 ก.พ. 2025
- In this part of the Introduction to Causal Inference course, we bring everything together to fully understand the flow of assocition and causation in graphs. Please post questions in the TH-cam comments section.
Introduction to Causal Inference Course Website: causalcourse.com
Course Lectures Playlist: • 3.9 - The Flow of Asso...
Nice conclusion at the end: "... and that's why association is not causation"
Great content! I am studying in the university now Causality subject and your youtube videos just give me the right amount of information to understand the topics that were more abstract to me (given just by the definitions).
I just got confused though by the slide on the timestamp 0:25. You talk there about association (red dotted arrow), and it seems like it should two a two sided arrow, since association is purely statistical notion and doesn't have a direction. So I guess its either a typo or I misinterpreted something 😁
This lecture and the previous one (3.8) a really important in understanding the full picture!
The lectures are fabulous
it's amazing
thanks for another week of lectures~ just to clarify: only the one way path from cause T to response Y is a causal association? And following up: when there exists multiple one way paths between T and Y, we can say that multiple causal associations exist between T and Y?
"only the one way path from cause T to response Y is a causal association?"
Yes, causal association only flows along directed paths.
"And following up: when there exists multiple one way paths between T and Y, we can say that multiple causal associations exist between T and Y?"
Yeah, I would say "causal association flows from T to Y along mutliple paths."
Thank you!
Just wanted to say that you're doing a great job and that it's very much appreciated. Just a few months ago, when asked about CI MOOCs or resources on Twitter, Judea said there wasn't any. Now, I'm doing Hernán's "Draw your assumptions before you conclusions" on edX, following your awesome lectures on TH-cam, and having in-line next Paul Hünermund's Udemy course, "Causal Data Science". What motivated you to do these lectures? Is CI finally taking off?
Thanks! What motivated me? Well, I learned machine learning online from Andrew Ng, and I learned deep learning online from Andrej Karpathy. Both of them put quality graduate level lectures on TH-cam, and I thought it was insane that no analog existed for causal inference.
I'm unsure if CI is taking off. It seems like some of the machine learning hype might be rubbing off on causal inference.
@@BradyNealCausalInference Well, thank you very much for doing this! I'm sure it will be a landmark for CI and you'll be up there next to Karpathy and Andrew Ng. I found out about your course thanks to a tweet from Judea himself so what better endorsement could you ask for?
I just finished a master's in CS doing NLP and none of my peers knows about CI even though all of them are into ML. To be honest, I don't even remember how I came across "The Book of Why" a year ago, but I think CI it's a game-changer. We'll see what the future holds.
Sorry for being late but is a non-causal association always a confounding association?
No.
If you can guarantee that no colliders are conditioned on (in your adjustment or even in the selection of the data), then I would say all non-causal association under those conditions is confounding association, since it all must be flowing along backdoor paths (that are collider-free).
I think association flowing out of T, through a conditioned collider, and into Y should definitely not being called "confounding association." For example, "selection bias" is what Hernán & Robins (2020) use for that, and I think that's a better name (though, that term is used for other things, which can cause confusion across disciplines).
And then what about non-causal association that's flowing along backdoor paths that were unblocked by conditioning on colliders? What do we call that? Who knows. "Non-causal" certainly seems like the safest name in that case.
@@BradyNealCausalInference Got it! Thank you for your insight.