1.5 - Causation in Observational Studies

In this part of the Introduction to Causal Inference course, we walk through what observational studies are and how we can measure causal effects in observational data. This is important because it is not always possible to run a randomized control trial. Please post questions in the RUclips comments section.
Introduction to Causal Inference Course Website: causalcourse.com
Course Lectures Playlist: • 1.5 - Causation in Obs...

[@adriangaldran4012]

I found the COVID-27 driving example in this section very illustrating for understanding do operators! Thank you so much for making all this available, I really think there was a need of this kind of course for people coming from applied ML with not much stats background; looking forward to see how this course unfolds

[@BradyNealCausalInference]

Thanks, Adrián. I'm looking forward to it as well!

[@edward_the_paddler]

Thanks Brady, love these lecture series! Agree with what @Adrian said here, I have relatively less stats background and this course is really useful for me.

[@BradyNealCausalInference]

@@edward_the_paddler Have you taken an introductory course in probability?

[@paulstevenconyngham7880]

Can you recommend any ​ @Brady Neal - Causal Inference ?

[@BradyNealCausalInference]

@@paulstevenconyngham7880 I like the one by John Tsitsiklis on edX, but it was awhile ago when I took it, so maybe there are better ones now.

[@choutycoh5828]

ASSUME A TABLE CAN BLOCK the impact from THE C to X(7::45)

[@chiaoyichao9904]

The mailing system doesn't work. It said, "We will be back online shortly
Our system is undergoing maintenance. Please come back in a few minutes."

[@michelspeiser5789]

Does the course also contain the calculation for the COV27 scenario where T -> C -> Y ?

[@connor-shorten]

Thank you so much for this! Really excited to see the rest of the course!!

[@softpeachhy8967]

Hi Brady, great videos! What does it mean when you say 'randomized t' at 6:01, removing edges of parents of t (conditioning on pa(t))? And why conditioning on pa(t) can block the path thus isolate the casual association?

[@MrWater2]

Great videos! I come from a economic perspective and I haven't seen the do operator before. Silly question sorry, what is M?

[@BradyNealCausalInference]

We'll define the do operator more precisely in week 3 (chapter 4 of book). M is just another variable in the data here. I give it the label "M" because it is a mediator (mediates the effect of T on Y).

[@zijingzhang172]

Thank you for putting this together! Wonderful content! Just one question regarding page 27- how would the causal death rate change if T is a cause of C rather than the other way around?

[@BradyNealCausalInference]

Welcome! It would change to be what's currently in the "Total" column. More details on this in week's 3 and 4.

[@zijingzhang172]

@@BradyNealCausalInference Got it! Thanks! Looking forward to week4:)

[@julius1841]

Super good video!!! Can't love more!!!

[@andysilv]

Hi Brady! Nice lecture, really enjoy them!
I have a question: on ruclips.net/video/MrZDBsS7hG4/видео.html I don't quiet grasp why do we have the second equality?
I mean, the order of the variables under expectation is confusing: shouldn't it be E[Y,W|t], not E[T|t,W] because this is how marginalization works?
UPD silly question, of course, we are talking about expectation here so E_wE[Y|t,w] = sum_w(E[Y|t,w]p(w)) so it's correct

[@Cindy-md1dm]

Thank you so much for sharing these awesome lectures! May I ask how to choose which nodes to block in sufficient adjustment set?

[@BradyNealCausalInference]

You'll see that in week 3 (with it being fully clear in week 4)!

[@Cindy-md1dm]

@@BradyNealCausalInference Awesome! Thanks a lot!

[@jake5549-r4k]

Is the mediating variable M necessary here? Can we control W and find the causal association without it?

[@megamarisa3930]

Thank you for this awesome content. I have a question.. in the simpson paradox example, we would chose treatment B if treatment is chosen due to conditions.. however, in this case, with the same examples, it seems like the causal association points that we should choose treatment a?

[@nstiarma8049]

This is such a gem. Thank you so much for providing us this invaluable course!

[@chiaoyichao9904]

Do you have pdf of your slides in the video?

[@zeeeeeeeeeavs]

ThanksBrady - love the course preview!

[@travissun6753]

Hello Brady,
I have a querstion: Does `causal effect` and `causal association` refer to differenct thing? Why did you use `causal association` ,instead of `causal effect`, in your causal diagram ?

[@BradyNealCausalInference]

Same thing. I use "association" to emphasize that the total association can be decomposed into both causal association and non-causal association (one part of which is confounding association).

[@CathrynSymons]

Thanks for this. The lecture is very clear, well done

[@neillpatterson]

I’m having a hard time appreciating the difference between Y(t) and Y|t. To my mind these are the exact same thing, but all the math seems to proceed based on these being fundamentally different quantities. Does someone have a good way of explaining this nuance?

[@michelspeiser5789]

They are fundamentally different. Y|t is the observational quantity. For example, does the sun rise (Y) when the clock shows 7AM (t)? The quantity Y(t) = Y|do(t) is different; if you set the clock to 7AM (do(t)) does the sun rise?

[@olhasobetska8066]

man, you are amazing

[@yannickprudent1232]

Great content, thank you so much !

[@suyogchandramouli7853]

This was a great preview! Looking forward to the rest of it :)

[@dylanyue5]

Q: What is the M here?

[@elementsofarah7918]

what is meant by the summation of c ?

[@BradyNealCausalInference]

Sum over all values c that the random variable C can take on