Causal Analysis in Theory and Practice

January 10, 2018

2018 Winter Update

Filed under: Announcement,General — Judea Pearl @ 10:07 pm

Dear friends in causality research,

Welcome to the 2018 Winter Greeting from the UCLA Causality Blog. This greeting discusses the following topics:

1.  A report is posted, on the “What If” workshop at the NIPS conference  (see December 19, 2017 post below). It discusses my presentation of: Theoretical Impediments to Machine Learning, a newly revised version of which can be viewed here. []

2. New posting: “Facts and Fiction from the Missing Data Framework”. We are inviting discussion of two familiar mantras:
Mantra-1. “The role of missing data analysis in causal inference is well understood (eg causal inference theory based on counterfactuals relies on the missing data framework).
Mantra-2. “while missing data methods can form tools for causal inference, the converse cannot be true.”

We explain why we believe both mantras to be false, but we would like to hear you opinion before firming up our minds.

3. A review paper is available here:
Titled: “Graphical Models for Processing Missing Data.” It explains and demonstrates why missing data is a causal inference problem.

4. A new page is now up, providing information on “The Book of Why”
It contains Table of Contents and excerpts from the book.

5. Nominations are now open for the ASA Causality in Education Award. The nomination deadline is March 1, 2018. For more information, please see

6. For those of us who were waiting patiently for the Korean translation of Primer — our long wait is finally over. The book is available now in colorful cover and in optimistic North Korean accent.

Don’t miss the gentlest introduction to causal inference.

Enjoy, and have a productive 2018.


  1. Symbols can be causes in the sense of Interventionist Causality. For example, in Causality, Pearl writes that the leader of a firing squad ording the squad to fire can be considered the _cause_ of the prisoner’s death. Similarly, a traffic light changing color _causes_ cars to stop/go. But, of course, symbols are causes only when there is an interpreter. I call this autonomous causality. It is explained briefly in Section 7 (bottom of p. 6) of this overview paper on complex systems. (Dropbox link:

    I would be interested in readers’ comments.


    Comment by Russ Abbott — January 10, 2018 @ 11:51 pm

  2. Russ,
    I totally agree with you that symbols could be causes of agents and mechanisms that respond to those symbols.
    This was indeed the reasoning behind the mediation formula, which attempted to emulate the legal definition of
    “discrimination” in counterfactual language. The result was the recognition that some interventions do not
    “fix” a variable to a constant, but prevent one variable (the mediator) from responding to another (the primary cause).
    But I do not understand why you call it “autonomous causality”, Isnt it true for non-autonomous respondents as well.
    Isn’t our metabolism responding to signals from enzymes and other cheminal constituents in the blood?

    On another topic, thanks for pointing us, through your paper, to an enterprise called “general complex systems”.
    I once contributed to this enterprise, but I left it (in the 1970’s) when I felt that it was too
    general, and was primarily made up of taxonomies, rather than tools.
    Can you educate us on what I lost by this move?
    In other words, what would I be able to do today that I can’t do already had I not ignored
    advances in “general complex systems”. Do you have any canonical example that would make me feel
    sorry for abandoning the field?

    Comment by Judea Pearl — January 14, 2018 @ 9:34 am

RSS feed for comments on this post. TrackBack URI

Leave a comment

Powered by WordPress