Why read ‘The Book of Why’? Pearl, Pipes & missed chances in Time

‘If only Abe and Yak had been able to summon the Causal Revolution to clear the air!’ (p. 168)


Prof Judea Pearl loves people, stories, causality and diagrams. That much is clear from meeting him only through the 370 pages of ‘The Book of Why’, which is targeting the smart layperson interested in causality but not in math. People are center-stage in Pearl’s theatre, from his students and misguided colleagues to his family (‘My wife was right’).  Stories abound on Pearl’s heroes and anti-heroes and on how he and they came about their viewpoints. I found them interesting (though long-winded) but caution they are re-interpreted in Pearl’s self-proclaimed ‘Whig history’ – so advise the reader to read their original work. Finally, causality and diagrams go together likes peas and carrots in Pearl’s view, and I very much enjoyed the combination of words and graphs in thinking about causality. Let’s talk about the pearls, pipe(r)s and missed chances in this book.

Screen Shot 2018-08-27 at 12.41.02 PM

Pearls: THEORY is king in Pearl’s world, as reflected by the title of his first chapter ‘Mind over Data’ and his self-identification as a ‘neat’ instead of a ‘scruffy’ on his key expertise area of AI. Indeed, theory is often ignored in today’s ‘Big Data’ hype, even though I agree with Pearl it is more needed now than ever. Steeped in theory, my favorite chapters are ‘the ladder of causation’ (go straight to page 28), ‘the conquest of Mount Intervention’, ‘Mediation: The Search for a Mechanism’ and ‘Big Data, Artificial Intelligence and the Big Questions’. The latter’s critique of deep learning as a black box, and discussion of the transportability problem (with an application of online ad response in different populations) will especially interest business readers. In contrast, statisticians and scientists may, like me, get annoyed by the winding road Pearl takes to equations and calculations, and instead read his ‘Causality: Models, Reasoning and Inference’ book as recommended by Kevin Gray (https://www.kdnuggets.com/2018/06/gray-pearl-book-of-why.html)


Pipes: I found Pearl most useful in thinking through what domain experts know and agree on regarding causal structure, and in quantifying these agreed-upon causal effects (the chapter on Mediation Analysis is especially illuminating). I found Pearl least useful when domain experts disagree on the causal structure – in fact I saw nothing in these 370 pages that would help come to such agreement. Pearl uses the example of smoking causing cancer, implying the counterfactual that, if only his causal diagrams would have been used in the 1950s-1960s, agreement on the causal link would have been reached much earlier, saving thousands of lives (this blog’s opening quote). I find such rung 3 on the ladder of causation unlikely, and believe neither Yak nor Fisher would have piped down. In my own domain of expertise, the causality of brand perceptions and brand purchases is the subject of a lively debate (https://analyticdashboards.wordpress.com/2018/08/15/do-perceptions-drive-sales-or-sales-drive-perceptions-for-your-brand/), with little apparent relevance of Pearl’s perspective. Even as Pearl discusses the current debate on climate change, I found myself wondering whether he could convince even a fan of his causal approach. Indeed, a friend of mine makes his living selling causal diagram solutions to business clients and is positively cited in Pearl’s book. He is also a climate change denier. Which data could we collect for the causal diagram on p. 294 to settle our debate? For a most interesting take, see https://www.slideshare.net/SAMSI_Info/program-on-mathematical-and-statistical-methods-for-climate-and-the-earth-system-methods-for-causality-analysis-in-climate-science-imme-ebertuphoff-aug-23-2017

Screen Shot 2018-08-27 at 12.38.46 PM.png

TIME is the usual suspect missing from Pearl’s conception of causality: the only time his causal diagrams show 2-way arrows is when the domain experts insist on dual causality (e.g. Price and Supply on page 251). This suspicious absence is especially harmful in my areas of business and government, as decision makers typically collect time series data and care about when a proposed intervention should yield results (hopefully before their term/position ends J) . Pearl dismisses ‘Granger Causality’ without even letting the reader know what it means – in sharp contrast to the many pages devoted to other causality approaches Judea disagrees with. A variable X Granger Causing a variable simply means that knowing the past of X helps you predict Y better than only knowing the past of Y.  In other words, X changes before Y in time. How can one write a book of 370 pages capturing ‘natural’ human views of causality without including or at least discussing this natural tendency? Seven years ago, White, Chalak and Lu (2011) showed that Pearl’s and Granger’s causality concepts are closely related and provide practical methods to test for direct causality using tests for Granger causality. They also showed the important limitations of Pearl’s causal diagrams, which apply to e.g. the Prisoner’s Dilemma (which has a Nash Equilibrium) but not to even simple games such as Hide and Seek (eg penalty kicks or Hitter vs Pitcher) or Battle of the Sexes, nor to the econometric workhorse of the Vector Autoregressive (VAR) model. VanderWeele and Shpitser (2011) discuss how Pearl’s “backdoor path criterion” presupposes that the structure of a causal diagram is known and that one needs to have knowledge not simply on whether each covariate affects the treatment and the outcome but also on how each of the covariates are causally related to one another: “Often, this is implausible”. Pearl does not incorporate, refute or even acknowledge any of this in ‘The book of why’. Why? I find this a missed opportunity to offer the world a more comprehensive view of causality. Talking about missed chances, I would have loved to have met Pearl during my UCLA PhD. I bet I would have liked and learned from him. In this counterfactual, could causal diagrams and time series econometrics have found a happy integration?  One can only imagine…

Screen Shot 2018-08-27 at 12.33.36 PM

“Who really can face the future? All you can do is project from the past, even when the past shows that such projections are often wrong. And who really can forget the past? What else is there to know?” Robert Pirsig ‘Zen and the Art of Motorcycle Maintenance’


Leave your comment here

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s