CAUSALITY
CAUSALITY by Judea Pearl
PREFACE (updated 9/99)

1.1 Introduction to Probability Theory
 1.1.1 Why probabilities
 1.1.2 Basic concepts in probability theory
 1.1.3 Combining predictive and diagnostic supports
 1.1.4 Random Variables and Expectations
 1.1.5 Conditional independence and graphoids

1.2 Graphs and Probabilities
 1.2.1 Graphical Notation and Terminology
 1.2.2 Bayesian Networks
 1.2.3 The dseparation criterion
 1.2.4 Inference with Bayesian networks

1.3 Causal Bayesian Networks
 1.3.1 Causal networks as oracles for interventions
 1.3.2 Causal relationships and their stability

1.4 Functional Causal Models
 1.4.1 Structural Equations
 1.4.2 Probabilistic predictions in causal models
 1.4.3 Interventions and causal effects in functional models
 1.4.4 Counterfactuals in functional models

1.5 Causal vs. Statistical Terminology

2.1 Introduction


2.2 The Causal Modeling Framework


2.3 Model Preference (Occam's razor)


2.4 Stable Distributions


2.5 Recovering DAG Structures


2.6 Recovering Latent Structures


2.7 Local Criteria for Causal Relations


2.8 NonTemporal Causation and Statistical Time


2.9 Conclusions


 2.9.1 On minimality, Markov, and stability


3.1 Introduction

3.2 Intervention in Markovian Models
 3.2.1 Graphs as models of interventions
 3.2.2 Interventions as variables
 3.2.3 Computing the effect of interventions
 3.2.4 Identification in causal models

3.3 Controlling Confounding Bias
 3.3.1 The backdoor criterion
 3.3.2 The frontdoor criterion
 3.3.3 Example: Smoking and the genotype theory

3.4 A Calculus of Intervention
 3.4.1 Preliminary notation
 3.4.2 Inference rules
 3.4.3 Symbolic derivation of causal effects: An example
 3.4.4 Causal inference by surrogate experiments

3.5 Graphical Tests of Identifiability
 3.5.1 Identifying models
 3.5.2 Nonidentifying models

3.6 Discussion
 3.6.1 Qualifications and extensions
 3.6.2 Diagrams as a mathematical language
 3.6.3 Translation from Graphs to Potential Outcomes
 3.6.4 Relations to Robin's Gestimation

4.1 Introduction
 4.1.1 Actions, acts, and probabilities
 4.1.2 Actions in decision analysis
 4.1.3 Actions and counterfactuals

4.2 Conditional Actions and Stochastic Policies

4.3 When is the Effect of an Action Identifiable?
 4.3.1 Graphical conditions for identification
 4.3.2 Remarks on efficiency
 4.3.3 Deriving a closedform expression for control queries
 4.3.4 Summary

4.4 The Identification of Plans
 4.4.1 Motivation
 4.4.2 Plan identification: Notation and assumptions
 4.4.3 Plan identification: A general criterion
 4.4.4 Plan identification: A procedure

4.5 Direct Effects and their Identification
 4.5.1 Direct vs. Total Effects:
 4.5.2 Direct effects, definition and identification
 4.5.3 Example: Sex Discrimination in College Admission
 4.5.4 Average direct effects

5.1 Introduction
 5.1.1 Causality in search of a language
 5.1.2 SEM: How its meaning got obscured
 5.1.3 Graphs as a mathematical language: An example

5.2 Graphs and Model Testing
 5.2.1 The testable implications of structural models
 5.2.2 Testing the testable
 5.2.3 Model equivalence

5.3 Graphs and Identifiability
 5.3.1 Parameter identification in linear models
 5.3.2 Comparison to nonparametric identification
 5.3.3 Causal effects: The interventional interpretation of structural equation models

5.4 Some Conceptual Underpinnings
 5.4.1 What do structural parameters really mean?
 5.4.2 Interpretation of effect decomposition
 5.4.3 Exogeneity, superexogeneity and other frills

5.5 Conclusion

6.1 Simpson's Paradox: An Anatomy
 6.1.1 A tale of a nonparadox
 6.1.2 A tale of statistical agony
 6.1.3 Causality vs. enchangeability
 6.1.4 A paradox resolved (or what kind of machine is man)

6.2 Why there is no statistical test for confounding, why many think there is, and why they are almost right
 6.2.1 Introduction
 6.2.2 Causal and associational definitions

6.3 How the Statistical Criterion Fails
 6.3.1 Failing sufficiency due to marginality
 6.3.2 Failing sufficiency due to closedworld assumptions
 6.3.3 Failing necessity due to barren proxies
 6.3.4 Failing necessity due to incidental cancelations

6.4 Stable vs. Incidental Unbiasedness
 6.4.1 Motivation
 6.4.2 Formal definitions
 6.4.3 Operational test for stable noconfounding

6.5 Confounding, Collapsibility, and Exchangeability
 6.5.1 Confounding and collapsibility
 6.5.2 Confounding vs. confounders
 6.5.3 Exchangeability vs. structural analysis of confounding

6.6 Conclusions

6.7 Acknowledgment

7.1 Structural Model Semantics
 7.1.1 Definitions: Causal models, actions and counterfactuals
 7.1.2 Evaluating counterfactuals: Deterministic analysis
 7.1.3 Evaluating counterfactuals: Probabilistic analysis
 7.1.4 The twinnetworks method

7.2 Applications and Interpretation of Structural Models
 7.2.1 Policy analysis in linear econometric models: An example
 7.2.2 The empirical content of counterfactuals
 7.2.3 Causal explanations, utterances, and their interpretation
 7.2.4 From mechanisms to actions to causation
 7.2.5 Simon's Causal ordering

7.3 Axiomatic Characterization
 7.3.1 The axioms of structural counterfactuals
 7.3.2 Causal effects from counterfactual logic: An example
 7.3.3 Axioms of causal relevance

7.4 Structural and Similaritybased Counterfactuals
 7.4.1 Relations to Lewis' counterfactuals
 7.4.2 Imaging vs.conditioning
 7.4.3 Relation to NeymanRubin framework
 7.4.4 Exogeneity revisited with counterfactuals or On Errors, graphs and counterfactuals

7.5 Structural vs. Probabilistic Causality
 7.5.1 The reliance on temporal ordering
 7.5.2 The perils of circularity
 7.5.3 The closedworld assumption
 7.5.4 Singular vs.general causes
 7.5.5 Summary

8.1 Introduction
 8.1.1 Imperfect and indirect experiments
 8.1.2 Noncompliance and intenttotreat

8.2 Bounding Causal Effects
 8.2.1 Problem formulation
 8.2.2 The evolution of potentialpesponse variables
 8.2.3 Linear programming formulation
 8.2.4 The natural bounds
 8.2.5 Effect of treatment on the treated
 8.2.6 Example: The effect of cholestyramine

8.3 Counterfactuals and Legal Responsibility}

8.4 A Test for Instruments

8.5 Causal Inference From Finite Sample
 8.5.1 Gibbs sampling
 8.5.2 The effects of sample size and prior distribution
 8.5.3 Causal effects from clinical data with imperfect compliance
 8.5.4 Bayesian estimate of singleevent causation

8.6 Conclusion

9.1 Introduction

9.2 Necessary and Sufficient Causes: Conditions of Identification
 9.2.1 Definitions, notation, and basic relationships
 9.2.2 Bounds and basic relationships under exogeneity
 9.2.3 Identifiability under monotonicity and exogeneity
 9.2.4 Identifiability under monotonicity and nonexogeneity

9.3 Examples and Applications
 9.3.1 Example1: Betting against a fair coin
 9.3.2 Example2: The firing squad
 9.3.3 Example3: The effect of radiation on leukemia
 9.3.4 Example4: Legal responsibility from experimental and nonexperimental data
 9.3.5 Summary of results

9.4 Identification in Nonmonotonic Models

9.5 Conclusions

10.1 Introduction: The Insufficiency of Necessary Causation
 10.1.1 Singular causes revisited
 10.1.2 Preemption and the role of structural information
 10.1.3 Overdetermination and quasidependence
 10.1.4 Mackie's INUS condition

10.2 Production, Dependence, and Sustenance

10.3 Causal Beams and SustenanceBased Causation
 10.3.1 Causal beams: Definitions and implications
 10.3.2 Examples: From disjunction to general formulas
 10.3.3 Beams, preemption, and the probability of singleevent causation
 10.3.4 Pathswitching causation
 10.3.5 Temporal preemption

10.4 Conclusions