Add new knowledge base page on causal DAGS for quasi-experiments #321
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
drbenvincent
changed the title
| @@ -0,0 +1,457 @@ | |||
| { | |||
There was a problem hiding this comment.
Optional: But you might want to mention that IV and RCT allow unconfounded inference for slightly different estimands. You can refer to the paper for details, but you might also want to distinguish ATE, ATT, ATC, and LATE kinds of estimands and how they relate to different DAGs?
Reply via ReviewNB
There was a problem hiding this comment.
+1
And maybe comment about the need of a 'strong' instrument. Becausee a weak relationship IV -> Z can lead to high variance (right?)
Also, add an example. My favorite on is https://matheusfacure.github.io/python-causality-handbook/09-Non-Compliance-and-LATE.html Maybee we can borrow it (with the proper citation)
There was a problem hiding this comment.
I think this is all fair enough. For the moment I've bundled this points into #327 so that they are definitely acted on. But feel free to let me know if you think these points need to be addressed in the first iteration.
There was a problem hiding this comment.
I think we can merge this one and then create an issue to add examples.
| @@ -0,0 +1,457 @@ | |||
| { | |||
There was a problem hiding this comment.
Good this is clear and I think useful. I think it's concise and we may want to avoid cluttering it with the discussion of ATE and LATE and ATC etc...
Reply via ReviewNB
| @@ -0,0 +1,457 @@ | |||
| { | |||
There was a problem hiding this comment.
For a next iteration we can show the problem with Bambi or PyMC with the do operator (maybe create a follow up issue?)
Also, it would be nice to give a concrete example for readers that see this for the same time.
Reply via ReviewNB
There was a problem hiding this comment.
I was thinking exactly this. I've created issue #327 to keep track of this.
| @@ -0,0 +1,457 @@ | |||
| { | |||
There was a problem hiding this comment.
If we add an example above, it would be nice to see how a RCT actually help us. Again, just to gain some intuition.
Reply via ReviewNB
There was a problem hiding this comment.
Exactly, and this would also allow some discussion of the do operator as being an intervention on the causal graph.
| @@ -0,0 +1,457 @@ | |||
| { | |||
There was a problem hiding this comment.
There was a problem hiding this comment.
Good idea. I've added an admonition box about parallel trends and also added it as a glossary term.
|
@drbenvincent, this looks very nice! I agree we should work on iterations. I suggest adding some examples for users who are new to the topic. This can be overwhelming at first (it was/is for me) |
|
Thanks for the comments @NathanielF + @juanitorduz. I've created #327 to keep track of some of your ideas for the second iteration. Hopefully you think the current changes are ok to merge for this first iteration, but let me know. |
|
@drbenvincent can you put the |
FYI: Recommended way to review is to build the docs locally and read the new page that way.