1 / 11

CONFOUNDING EQUIVALENCE

CONFOUNDING EQUIVALENCE. Judea Pearl – UCLA, USA Azaria Paz – Technion, Israel (www.cs.ucla.edu/~judea/). L. T. Z. W 1. W 4. ?. Z  T. W 2. W 3. V 1. V 2. X. Y. Bias =. CONFOUNDING EQUIVALENCE. PROBLEM: When would two measurements be

trista
Download Presentation

CONFOUNDING EQUIVALENCE

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. CONFOUNDING EQUIVALENCE Judea Pearl – UCLA, USA Azaria Paz – Technion, Israel (www.cs.ucla.edu/~judea/)

  2. L T Z W1 W4 ? Z T W2 W3 V1 V2 X Y Bias = CONFOUNDING EQUIVALENCE PROBLEM:When would two measurements be equally effective for reducing confounding bias?

  3. CONFOUNDING EQUIVALENCE DEFINITION AND SOLUTION • DEFINITION: • T and Z are c-equivalent if • DEFINITION (Markov boundary): • Markov boundary Sm of S (relative to X) is the minimal subset of S that d-separates X from all other members of S. • THEOREM: • Z and T are c-equivalent iff • Zm=Tm, or • Z and T are admissible (i.e., satisfy the back-door • condition)

  4. T Z Z T ARE Z AND T CONFOUNDING EQUIVALET? ANSWER:Yes! Because both are admissible W1 W4 W2 W3 V1 V2 X Y

  5. L T Z Z T ARE Z AND T CONFOUNDING EQUIVALET? ANSWER:No! None is admissible and W1 W4 W2 W3 V1 V2 X Y

  6. A PUZZLE: The definition of c-equivalence is purely statistical. The graphical criterion we found invokes “admissibility,” which is causal – why? Corollary 1: c-equivalence is an invariant property of M-equivalent graphs Corollary 2: There ought to be a graphical criterion that does not invoke “admissibility,” based ond-separation only. Indeed, Theorem:Z~ T iff and

  7. EXAMPLE: TESTING IF Z~T WITHOUT INVOKING ADMISSIBILITY L Z T Z T Z T W1 W4 W2 W3 V1 V2 X Y Z T W1 W4 W2 W3 V1 V2 X Y

  8. V W1 W2 V W1 W2 Z1 Z2 Z1 Z2 X Y X Y APPLICATIONS TO MODEL TESTING Are these two models Markov-equivalent? (a) (b) Answer:No,  a CI that holds in (b) but not in (a) Answer:No, because {Z1,W1,W2}~ {W1,W2,Z2} holds in (b) but not in (a) Which is easier to detect? Which is easier to test?

  9. WHICH IS EASIER TO TEST WHEN Z IS HIGH DIMENSIONAL? V W1 W2 V W1 W2 Z1 Z2 Z1 Z2 X Y X Y    (a) (b) Conditional independence? or c-equivalence? c-equivalence can invoke propensity scores (Scalars): And we can ignore the outcome process

  10. CONCLUSIONS • c-equivalence provides: • A simple, polynomial time test for deciding • whether one set of measurements has the • same bias-reducing potential as another. • A powerful tool for: • Testing from sampled data, the validity of a • given graphical model and, eventually, and • systematic search for graph structures.

  11. CONCLUSIONS • c-equivalence provides: • A simple, polynomial time test for deciding • whether one set of measurements has the • same bias-reducing potential as another. • A powerful tool for: • Testing from sampled data, the validity of a • given graphical model and, eventually, and • systematic search for graph structures.

More Related