Ugh, I have spent so much energy trying to understand the definition of exchangeability --- and I think your video finally got me over the finish line. Thanks!
Exchangeability is independence of each potential outcome (Y^a=1 and Y^a=0) and treatment but that doesn't mean that the causal effect is zero. This is because the causal effect is a contrast of the two potential outcomes whereas exchangeability makes a statement about the relationship between each potential outcome and treatment in turn. (The video's example is actually one in which exchangeability holds but there is a nonzero causal effect.)
at 9:45 , when you divided data into two strata by education, are the graphs correct? when you select low education strata, plot Y(a=1) and Y(a=0) (treatment and no treatment), still you cannot observe contra factual, how come you have double line graph again in both cases Ya=1 and Ya=0? How are you observing both effects (doubly line graph) within both treatment and control group? Was that a hypothetical graph?
Hello miss, great video, but just one thing. How do we assume that the potential outcomes are independent of the treatment? Isn't this a bit counterintuitive? If we want to measure the causal effects between the treatment and the outcome then why are we assuming them to be independent? Thanks again for the video.
Ugh, I have spent so much energy trying to understand the definition of exchangeability --- and I think your video finally got me over the finish line. Thanks!
Totally agree!! Thankyou!!
i want to thank you so much!!! your videos are very understandable and helpful for me!!! please make more of them. you're a very good teacher!!
If the potential outcomes are independent of treatment, then why given treatment to intervene? Thanks.
Exchangeability is independence of each potential outcome (Y^a=1 and Y^a=0) and treatment but that doesn't mean that the causal effect is zero. This is because the causal effect is a contrast of the two potential outcomes whereas exchangeability makes a statement about the relationship between each potential outcome and treatment in turn. (The video's example is actually one in which exchangeability holds but there is a nonzero causal effect.)
This is the best expalnation for ignorability I have ever heard!
are there any code examples with this tutorial?
at 9:45 , when you divided data into two strata by education, are the graphs correct? when you select low education strata, plot Y(a=1) and Y(a=0) (treatment and no treatment), still you cannot observe contra factual, how come you have double line graph again in both cases Ya=1 and Ya=0?
How are you observing both effects (doubly line graph) within both treatment and control group?
Was that a hypothetical graph?
Hello miss, great video, but just one thing.
How do we assume that the potential outcomes are independent of the treatment? Isn't this a bit counterintuitive? If we want to measure the causal effects between the treatment and the outcome then why are we assuming them to be independent?
Thanks again for the video.
Thanks for the explanation!
🤩 understandable!
👍👍👍
You're so cool.