Second iteration on DAGS for quasi-experiments knowledge base page

[drbenvincent]

The first iteration of the knowledge base page is currently a PR, #321. But we already have plans to improve on it further.

This issue can be dealt with in stages by multiple PR's (and people) if needed.

⚠️WARNING: Because this docs page quasi_dags.ipynb is based on a notebook, care will need to be taken to avoid notebook conflicts if multiple people work on this at once. ⚠️

---

Fix DAG image rendering issues

The DAG images (produced by daft) come up as warped on my phone (EDIT: not that great on my laptop on readthedocs either). Maybe there is a setting that can fix this, possibly a kwarg of pgm.render()?. Otherwise we may need to change pgm.render() to export an image to file (~/docs/source/_static) instead then simply embed a saved png. We could also increase the resolution of the images a bit.
Fixed by #332

Confounding & RCT's

Provide a concrete and numerical example of confounding, and then follow that up with the RCT solution using the do operator.

IV's

@NathanielF suggests:

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?

@juanitorduz follows up with:

And maybe comment about the need of a 'strong' instrument. Because 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)

Regression discontinuity

We could potentially add discussion of ATE and LATE and ATC etc if we can do that while keeping the section clear and concise. Maybe with drop down admonition boxes or something.

Add DAG for Synthetic control

The first iteration of the knowledge base page will not have a DAG for synthetic control. It would be nice to add this if possible.

Add DAG for propensity scores

Once we merge #311 then we'll have propensity score methods available. So it would be good to add a new section for propensity methods. A very good reference is

Peter M Steiner, Yongnam Kim, Courtney E Hall, and Dan Su. Graphical models for quasi-experimental designs. Sociological methods & research, 46(2):155–188, 2017

but can of course use whatever is useful. Tagging @NathanielF - don't think I have strong feelings about it, but it could be good to follow up #311 with this update to the knowledge base so that a new causalpy release would be ultra impressive.

Add DAG for ANCOVA

We are also currently missing a DAG for the ANCOVA / non-equivalent group design with pre and post test. It would be good to add a concise section for this.

[juanitorduz]

@drbenvincent regarding the diagrams see executablebooks/MyST-NB#588 Hre is a quick solution pymc-labs/pymc-marketing#667