This is the monthly thread for posting media of various types that you've found that you enjoy. Post what you're reading, listening to, watching, and your opinion of it. Post recommendations to blogs. Post whatever media you feel like discussing! To see previous recommendations, check out the older threads.
Rules:
- Please avoid downvoting recommendations just because you don't personally like the recommended material; remember that liking is a two-place word. If you can point out a specific flaw in a person's recommendation, consider posting a comment to that effect.
- If you want to post something that (you know) has been recommended before, but have another recommendation to add, please link to the original, so that the reader has both recommendations.
- Please post only under one of the already created subthreads, and never directly under the parent media thread.
- Use the "Other Media" thread if you believe the piece of media you want to discuss doesn't fit under any of the established categories.
- Use the "Meta" thread if you want to discuss about the monthly media thread itself (e.g. to propose adding/removing/splitting/merging subthreads, or to discuss the type of content properly belonging to each subthread) or for any other question or issue you may have about the thread or the rules.
Causal Inference in Statistics: A Primer by Judea Pearl et al.
I've always enjoyed reading about causality, ever since EY turned us on to Pearl back in The Fabric of Real Things. Perhaps I'm inclined to enjoy this small (136 page) primer on a purely romantic, intellectual basis alone. I've read Causality four or five times now and I must confess: outside of a few toy examples, I really have no concept of how to apply this field to an actual problem. It seems probable to me, after a cursory reading of the primer, that it is more likely to teach a person this applied skill.
Part of the reason that I've never picked up causal analysis from the far more encyclopedic Causality is that I've never had a legitimate use case. There surely must be some deep, irrevocable link between the extra-statistical concept of causality and the structure of certain Lorentzian manifolds, which are the setting of the mathematical study of general relativity. That link, however tenuous, seems remote and philosophical, far removed from the endless examples of lawns made wet by either rain or sprinklers. And yes, all the usual suspects, including the wet lawn example, return in abbreviated form in the primer, but thankfully there are a few additional examples proposed, I assume, by coauthors and attendant grad students.
The authors do a much better job injecting intuition into their exposition. In my opinion, the largest obstruction to causal education, aside from the field being an unlikely combination of directed graph theory and probability, is the large number of technical terms with varying degrees of clarity. The three building blocks of a directed acyclic graph (chains, forks, and colliders) are each given substantial room and multiple examples, which is more than I've ever seen Pearl put into a single text. It is unfortunate then that d-separation (a complicated inductive definition on paths between subsets of nodes) gets only one page of abstract discussion and no concrete example. On the other hand, I know full well that I could never motivate that particular concept myself, so perhaps my criticism is unfair.
The section on interventions is more visually compelling than I recall it being in earlier books and papers, practically showing a cartoon of someone severing the edges connecting an intervention with its node's ancestors. The "adjustment formula", which updates a graphical probabilistic model in the wake of an intervention, is presented first in a general way and then specialized to intervention. The discussion of the back-door and front-door criteria is again too short, in my opinion, but I appreciated the connection to Lord's paradox.
Sadly, the chapter on counterfactuals does not appear to be substantially new, save a prolonged example and some study questions toward the end.
Generally speaking I found the exposition to be a step above the typical undergrad math textbook; I assume "advanced statistics undergraduate or first-year graduate student" is their target demographic. (The extravagant price also points in this direction.) The historical notes and errata and the end of each section are interesting, including some links to computational resources (though I cringe at URLs printed in dead tree books). If, unlike me, you're able to read and apply the results from Causality directly, you have no need for this book. I found the frequently interspersed "study questions" nauseating but perhaps they're helpful to the less mathematically inclined. I am frankly confused at Wiley's pairing the book with a "companion website" that appears to be nothing more than a placeholder.