/blog/2021/09/07/do-calculus-backdoors/

[andrewheiss]

Do-calculus adventures! Exploring the three rules of do-calculus in plain language and deriving the backdoor adjustment formula by hand | Andrew Heiss
https://www.andrewheiss.com/blog/2021/09/07/do-calculus-backdoors/

[andrewheiss]

Originally posted by Huy Nguyen on 2021-10-06 08:38:18

Thanks Andrew,
I had been looking for an easy way to understand the do-calculus (even
in the sources you also listed) . But your blog post with illustration
and code was the first to help me understand it. You did an awesome job
making the content accessible.
I have some questions:

1) In the application of rule 3 to derive the back door adjustment. I
think it should be P(z| do(x) ) = P (z | nothing) . Instead of p(y |
nothing) as in currently, but perhaps Im wrong.
2) I wonder if you would consider a next blog post on the intuition
behind the validity of do-calculus. To me at this stage, these are only
rules to memorize and I can't justify for them.
For example, in the application of rule 2 above, in the unmodified
graph, we see that intervene on X would affect y. What allows us to do
the graph surgery then?
Or I would appreciate if you have any pointer on this topic. Thank you

Huy

[andrewheiss]

Originally posted by Long Luu on 2022-01-01 04:14:39

I think he made a typo there. Yes it should be P(z|nothing).