On the causal interpretation of acyclic mixed graphs under multivariate normality

In multivariate statistics, acyclic mixed graphs with directed and bidirected edges are widely used for compact representation of dependence structures that can arise in the presence of hidden (i.e., latent or unobserved) variables. Indeed, under multivariate normality, every mixed graph corresponds to a set of covariance matrices that contains as a full-dimensional subset the covariance matrices associated with a causally interpretable acyclic digraph. This digraph generally has some of its nodes corresponding to hidden variables. We seek to clarify for which mixed graphs there exists an acyclic digraph whose hidden variable model coincides with the mixed graph model. Restricting to the tractable setting of chain graphs and multivariate normality, we show that decomposability of the bidirected part of the chain graph is necessary and sufficient for equality between the mixed graph model and some hidden variable model given by an acyclic digraph.

[1]  M. Drton,et al.  Global identifiability of linear structural equation models , 2010, 1003.1146.

[2]  Jin Tian,et al.  Inequality Constraints in Causal Models with Hidden Variables , 2006, UAI.

[3]  Robin J. Evans,et al.  Graphical methods for inequality constraints in marginalized DAGs , 2012, 2012 IEEE International Workshop on Machine Learning for Signal Processing.

[4]  S. Orbom,et al.  When Can Association Graphs Admit A Causal Interpretation? , 1993 .

[5]  N. Wermuth.  PROBABILITY DISTRIBUTIONS WITH SUMMARY GRAPH STRUCTURE , 2010, 1003.3259.

[6]  S. Sullivant,et al.  Trek separation for Gaussian graphical models , 2008, 0812.1938.

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

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

[9]  Josephine Yu,et al.  On a Parametrization of Positive Semidefinite Matrices with Zeros , 2010, SIAM J. Matrix Anal. Appl..

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

[11]  P. Spirtes,et al.  MARKOV EQUIVALENCE FOR ANCESTRAL GRAPHS , 2009, 0908.3605.

[12]  D. Pollard.  A User's Guide to Measure Theoretic Probability by David Pollard , 2001 .

[13]  P. Games.  Correlation and Causation: A Logical Snafu , 1990 .

[14]  On the Markov equivalence of maximal ancestral graphs , 2005 .

[15]  Zoubin Ghahramani,et al.  The Hidden Life of Latent Variables: Bayesian Learning with Mixed Graph Models , 2009, J. Mach. Learn. Res..

[16]  Thomas S. Richardson,et al.  Learning high-dimensional directed acyclic graphs with latent and selection variables , 2011, 1104.5617.

[17]  Richard Scheines,et al.  Causation, Prediction, and Search, Second Edition , 2000, Adaptive computation and machine learning.

[18]  Michael Eichler,et al.  Computing Maximum Likelihood Estimates in Recursive Linear Models with Correlated Errors , 2009, J. Mach. Learn. Res..

[19]  Judea Pearl,et al.  When can association graphs admit a causal interpretation , 1994 .

[20]  N. Wermuth,et al.  Joint response graphs and separation induced by triangular systems , 2004 .

[21]  N. Wermuth,et al.  Sequences of regressions and their independences , 2011, 1110.1986.

[22]  M. Drton.  Discrete chain graph models , 2009, 0909.0843.

[23]  Seth Sullivant,et al.  Lectures on Algebraic Statistics , 2008 .

[24]  J. Koster.  Marginalizing and conditioning in graphical models , 2002 .

[25]  Thomas S. Richardson,et al.  Graphical Methods for Efficient Likelihood Inference in Gaussian Covariance Models , 2007, J. Mach. Learn. Res..

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

[27]  R. P. McDonald,et al.  Structural Equations with Latent Variables , 1989 .

[28]  S. Wright.  The Method of Path Coefficients , 1934 .

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

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