Review and improve the Introduction to DAGS chapter

[mbcann01]

Overview

In the Fall of 2023, I moved over a bunch of stuff from PowerPoint slides (nearly) verbatim. I was in a rush, so I told myself to move it just move it over and improve it later.

• In this chapter, we just want to introduce what DAGs are and how to build them in R. In the chapters on bias and confounding, we will get into more complex DAG structures.

• In the lab session today (2023-10-10), we walked through the DAG chapter together. It didn't go over well at all. I think they are able to understand the structure and construction of DAGS; however, I did not do a good enough job explaining the big picture. How and why do we use DAGs in the first place?

• This chapter needs a lot of beefing up. Maybe just start with translating chapter 3 of Modern Epidemiology. Causal Inference in Statistics (on Kindle) has some good stuff too.

• You may want to copy over some of the material in the ggdag vignettes.

• The Socrative for the causal inference module actually asks some questions that could be used to generate good information for the chapter.

Left off at

2023-10-10: Finished the first draft. There is still a lot of cleaning up to do.

Tasks.

• [ ] Do a better job explaining the big picture. How and why do we use DAGs in the first place? Maybe start with Simpson's Paradox? How does Miguel Hernan's video series start out?
• [ ] Incorporate all of the summary bullets (in a comment below) about DAGs somewhere in the chapter.
• [ ] Add a section on common effects to the end of the Basic DAG structure section.
• [ ] Add code for finding parents of nodes from the ggdag documentation.
• [ ] The section on d-separation rules needs a lot more explanation. Why do these rules exist? What does it mean to block or be open? What does conditioning mean?
• [ ] Figure out how to manually declare adjustment variables in ggdag so that I can make the d-separation rule figures directly in R.
• [ ] Add a comparison of DAGs and SCC (see below)
• [ ] Add some examples of useful real-world DAGS.'
• [ ] Review for typos

[mbcann01]

Comparing DAGs and SCC

Answers two different questions.

“The sufficient component cause model considers sets of actions, events, or states of nature which together inevitably bring about the outcome under consideration. The model gives an account of the causes of a particular effect. It addresses the question, ‘Given a particular effect, what are the various events which might have been its cause?’” @Hernan2020-uu

“The potential outcomes or counterfactual model focuses on one particular cause or intervention and gives an account of the various effects of that cause. In contrast to the sufficient component cause framework, the potential outcomes framework addresses the question, ‘What would have occurred if a particular factor were intervened upon and thus set to a different level than it in fact was?’ Unlike the sufficient component cause framework, the counterfactual framework does not require a detailed knowledge of the mechanisms by which the factor affects the outcome.’ @Hernan2020-uu

[mbcann01]

Incorportate all of these bullets about DAGs

• Not magic, just a tool.
• Works because it encodes mathematical information, given assumptions.
• Association flows in both directions
• No distinction between Causal / Preventive (listens to).
• Made of nodes and edges (arrows).
• Arrows must go in one direction (i.e., not double-headed).
• Acyclic – Never get back to where you started if you follow arrows.
• Arrows indicate causal effects.
• Deleting arrows indicates a lack of a causal effect.
• Simultaneously qualitative causal models and statistical models.