The Book of Why

The Book of Why: The New Science of Cause and Effect (2019), by Judea Pearl and Dana Mackenzie_ (Read during 2022).

I hated this book so much.

Causal analysis is extremely important. While the correlation between smoking and lung cancer was studied for decades, it took years of statistical work to prove that one caused the other. As the authors explain, our usual (correlation) models can tell us that the rooster sings when the sun comes up. But how can we be sure that the rooster is not causing sunrise?

In an era when we are extremely proud of designing algorithms to draw conclusions directly from data, the causal approach argues for a different method. We should first gain a better understanding of the relationships between the relevant variables, and supplement it with a strong theoretical background to posit causal relationships which we should test with the data available. In that way we ascend the “ladder of causation” and are able to establish, precisely, that the sun is able to come up even if the rooster somehow refuses to sing anymore.

Admittedly, this book was my first approximation to causal inference and perhaps thinking that I should start from here was my first mistake. In my defence, I knew that the author was one of the leading figures in the subject, whose work was fundamental for the development of causal inference.

But this is not an academic book – we do not get a formal introduction to the subject here. This is also not strictly a science diffusion book because it has many equations and the reader is expected to be able to grasp mathematical concepts that might not be too obvious for the noninitiate. So it is a strange middle ground, which could be fine if the technical details and their implications were the focus of the analysis. Instead, the authors spend an awful amount of time in telling their own personal stories and those of their students, their “eureka” moments and discoveries. This anecdote could be attractive to some, but for me it derails the book from its main purpose. Which, on second thought, I am unsure what it was.

In the end, not a science book, not a diffusion book, not an book of anecdotes, but a weird mixture of the three. Also, the prose is dense and not fluid, making it very hard to follow and a struggle to finish. If a trained reader wants a proper introduction to causal inference, they should look elsewhere.