Notes Dump

[cthoyt]

Here's the remaining notes from my work document that weren't categorized:

probability of Y given X -> regress Y on x
probability of Z

1. how to incporporate the adjustment set
   how do you figure out what parameters of linear structural causal model?
   do this for every edge:
2. what's source/target
3. regress target on source + adjustment set
4. coefficient associated with source is the one you use
5. any conditionals as union on to the adjustment set

---

front door adjustment that's more complicated
napkin is identifiable, but you can still get an estimand
estimand to linear regression -> you actually have to pile several of them together
the generic approach is the plugin method. For discrete data, you can just go for it. Continuous data is harder
chiro made a bayesian version of doubly robust estimation - that's the general way of most > efficiently making use of data