Structural Causal Models: Cycles, Marginalizations, Exogenous Reparametrizations and Reductions

Structural causal models (SCMs), also known as (non-parametric) structural equation models (SEMs), are widely used for causal modeling purposes. In particular, acyclic SCMs, also known as recursive SEMs, form a well-studied subclass of SCMs that generalize causal Bayesian networks to allow for latent confounders. In this paper, we investigate SCMs in a more general setting, allowing for the presence of both latent confounders and cycles. We show that in the presence of cycles, many of the convenient properties of acyclic SCMs do not hold in general: they do not always have a solution; they do not always induce unique observational, interventional and counterfactual distributions; a marginalization does not always exist, and if it exists the marginal model does not always respect the latent projection; they do not always satisfy a Markov property; and their graphs are not always consistent with their causal semantics. We prove that for SCMs in general each of these properties does hold under certain solvability conditions. Our work generalizes results for SCMs with cycles that were only known for certain special cases so far. We introduce the class of simple SCMs that extends the class of acyclic SCMs to the cyclic setting, while preserving many of the convenient properties of acyclic SCMs. With this paper we aim to provide the foundations for a general theory of statistical causal modeling with SCMs.

[1]  J. Koster.  On the Validity of the Markov Interpretation of Path Diagrams of Gaussian Structural Equations Systems with Correlated Errors , 1999 .

[2]  J. Koster,et al.  Markov properties of nonrecursive causal models , 1996 .

[3]  Thomas S. Richardson,et al.  A Discovery Algorithm for Directed Cyclic Graphs , 1996, UAI.

[4]  Frederick Eberhardt,et al.  Discovering Cyclic Causal Models with Latent Variables: A General SAT-Based Procedure , 2013, UAI.

[5]  Robin J. Evans,et al.  Graphs for Margins of Bayesian Networks , 2014, 1408.1809.

[6]  Frederick Eberhardt,et al.  Combining Experiments to Discover Linear Cyclic Models with Latent Variables , 2010, AISTATS.

[7]  D. A. Kenny,et al.  Correlation and Causation , 1937, Wilmott.

[8]  Gregory F. Cooper,et al.  A Simple Constraint-Based Algorithm for Efficiently Mining Observational Databases for Causal Relationships , 1997, Data Mining and Knowledge Discovery.

[9]  P. Spirtes,et al.  Using Path Diagrams as a Structural Equation Modeling Tool , 1998 .

[10]  S. J. Mason.  Feedback Theory-Further Properties of Signal Flow Graphs , 1956, Proceedings of the IRE.

[11]  Judea Pearl,et al.  Complete Identification Methods for the Causal Hierarchy , 2008, J. Mach. Learn. Res..

[12]  J. Pearl,et al.  Studies in causal reasoning and learning , 2002 .

[13]  Christopher Meek,et al.  Strong completeness and faithfulness in Bayesian networks , 1995, UAI.

[14]  Bernhard Schölkopf,et al.  Elements of Causal Inference: Foundations and Learning Algorithms , 2017 .

[15]  R. Penrose.  A Generalized inverse for matrices , 1955 .

[16]  M. Maathuis,et al.  Estimating high-dimensional intervention effects from observational data , 2008, 0810.4214.

[17]  Bernhard Schölkopf,et al.  Causal Consistency of Structural Equation Models , 2017, UAI.

[18]  Gregory F. Cooper,et al.  A bayesian local causal discovery framework , 2005 .

[19]  O. D. Duncan,et al.  Introduction to Structural Equation Models. , 1977 .

[20]  T. Richardson.  Markov Properties for Acyclic Directed Mixed Graphs , 2003 .

[21]  Judea Pearl,et al.  A Constraint-Propagation Approach to Probabilistic Reasoning , 1985, UAI.

[22]  Bernhard Schölkopf,et al.  From Ordinary Differential Equations to Structural Causal Models: the deterministic case , 2013, UAI.

[23]  T. Richardson.  Discovering cyclic causal structure , 1996 .

[24]  T. Haavelmo.  The Statistical Implications of a System of Simultaneous Equations , 1943 .

[25]  Jiji Zhang,et al.  On the completeness of orientation rules for causal discovery in the presence of latent confounders and selection bias , 2008, Artif. Intell..

[26]  Kenneth A. Bollen,et al.  Structural Equations with Latent Variables , 1989 .

[27]  Judea Pearl,et al.  Probabilistic Evaluation of Counterfactual Queries , 1994, AAAI.

[28]  Joseph Y. Halpern.  Axiomatizing Causal Reasoning , 1998, UAI.

[29]  T. Richardson,et al.  Markovian acyclic directed mixed graphs for discrete data , 2013, 1301.6624.

[30]  D. Rubin.  Estimating causal effects of treatments in randomized and nonrandomized studies. , 1974 .

[31]  A. Kechris.  Classical descriptive set theory , 1987 .

[32]  Thomas S. Richardson,et al.  Causal Inference in the Presence of Latent Variables and Selection Bias , 1995, UAI.

[33]  David Lewis.  Counterfactual Dependence and Time's Arrow , 1979 .

[34]  Kevin P. Murphy,et al.  Exact Bayesian structure learning from uncertain interventions , 2007, AISTATS.

[35]  Peter Bühlmann,et al.  CAM: Causal Additive Models, high-dimensional order search and penalized regression , 2013, ArXiv.

[36]  Peter Spirtes,et al.  Directed Cyclic Graphs, Conditional Independence, and Non-Recursive Linear Constructive Equation Models , 1993 .

[37]  Frederick Eberhardt,et al.  Learning linear cyclic causal models with latent variables , 2012, J. Mach. Learn. Res..

[38]  Bernhard Schölkopf,et al.  Distinguishing Cause from Effect Using Observational Data: Methods and Benchmarks , 2014, J. Mach. Learn. Res..

[39]  Thomas S. Richardson,et al.  Automated discovery of linear feedback models , 1996 .

[40]  Peter Spirtes,et al.  Conditional Independence in Directed Cyclic Graphical Models for Feedback , 1994 .

[41]  J. Mooij,et al.  Markov Properties for Graphical Models with Cycles and Latent Variables , 2017, 1710.08775.

[42]  M. Drton,et al.  Half-trek criterion for generic identifiability of linear structural equation models , 2011, 1107.5552.

[43]  Bernhard Schölkopf,et al.  From Deterministic ODEs to Dynamic Structural Causal Models , 2016, UAI.

[44]  P. Spirtes,et al.  Ancestral graph Markov models , 2002 .

[45]  A. Dawid.  Influence Diagrams for Causal Modelling and Inference , 2002 .

[46]  Samuel J. Mason,et al.  Feedback Theory-Some Properties of Signal Flow Graphs , 1953, Proceedings of the IRE.

[47]  Gene H. Golub,et al.  Calculating the singular values and pseudo-inverse of a matrix , 2007, Milestones in Matrix Computation.

[48]  A. Goldberger,et al.  Structural Equation Models in the Social Sciences. , 1974 .

[49]  Jin Tian,et al.  Causal Discovery from Changes , 2001, UAI.

[50]  Barry D. Hughes,et al.  A correspondence principle , 2016 .

[51]  Bernhard Schölkopf,et al.  Causal discovery with continuous additive noise models , 2013, J. Mach. Learn. Res..

[52]  Steffen L. Lauritzen,et al.  Independence properties of directed markov fields , 1990, Networks.

[53]  Peter Spirtes,et al.  Directed Cyclic Graphical Representations of Feedback Models , 1995, UAI.

[54]  Joris M. Mooij,et al.  Cyclic Causal Discovery from Continuous Equilibrium Data , 2013, UAI.

[55]  D. Geiger.  Graphoids: a qualitative framework for probabilistic inference , 1990 .

[56]  J. Pearl.  Causality: Models, Reasoning and Inference , 2000 .

[57]  R. Evans.  Margins of discrete Bayesian networks , 2015, The Annals of Statistics.

[58]  P. Spirtes,et al.  Causation, prediction, and search , 1993 .

[59]  Patrik O. Hoyer,et al.  Discovering Cyclic Causal Models by Independent Components Analysis , 2008, UAI.

[60]  Radford M. Neal.  On Deducing Conditional Independence from d-Separation in Causal Graphs with Feedback (Research Note) , 2000, J. Artif. Intell. Res..

[61]  Herbert A. Simon,et al.  Causality and Model Abstraction , 1994, Artif. Intell..

[62]  James M. Robins,et al.  ACE Bounds; SEMs with Equilibrium Conditions , 2014, 1410.0470.