Detecting confounding in multivariate linear models via spectral analysis
Authors:
- Dominik JanzingDJDominik Janzing
- Bernhard SchoelkopfBSBernhard Schoelkopf
Preprint
Abstract
We study a model where one target variable Y is correlated with a vector X:=(X_1,...,X_d) of predictor variables being potential causes of Y. We describe a method that infers to what extent the statistical dependences between X and Y are due to the influence of X on Y and to what extent due to a hidden common cause (confounder) of X and Y. The method relies on concentration of measure results for large dimensions d and an independence assumption stating that, in the absence of confounding, the vector of regression coefficients describing the influence of each X on Y typically has `generic orientation' relative to the eigenspaces of the covariance matrix of X. For the special case of a scalar confounder we show that confounding typically spoils this generic orientation in a characteristic way that can be used to quantitatively estimate the amount of confounding.Keywords
All Related Versions
This publication has 12 references indexed in Scilit:
- Total positivity in Markov structuresThe Annals of Statistics, 2017
- Free Probability and Random Matrices2017
- Replacing Causal Faithfulness with Algorithmic Independence of ConditionalsMinds and Machines, 2012
- Causal Inference Using the Algorithmic Markov ConditionIEEE Transactions on Information Theory, 2010
- Justifying Additive Noise Model-Based Causal Discovery via Algorithmic Information TheoryOpen Systems & Information Dynamics, 2010
- Numerical Linear Algebra and Applications2010
- Estimation of causal effects using linear non-Gaussian causal models with hidden variablesInternational Journal of Approximate Reasoning, 2008
- Rank One Perturbations with Infinitesimal CouplingJournal of Functional Analysis, 1995
- C*–Algebras and Operator Theory1990
- Classes of orderings of measures and related correlation inequalities. I. Multivariate totally positive distributionsJournal of Multivariate Analysis, 1980