The Hardness of Conditional Independence Testing and the Generalised Covariance Measure

DOI10.1214/19-AOS1857zbMATH Open1451.62081arXiv1804.07203MaRDI QIDQ118262FDOQ118262

Authors: Rajen D. Shah, Jonas Peters, Rajen D. Shah, Jonas Peters 

Publication date: 19 April 2018

Published in: The Annals of Statistics (Search for Journal in Brave)

Abstract: It is a common saying that testing for conditional independence, i.e., testing whether whether two random vectors $X$ and $Y$ are independent, given $Z$, is a hard statistical problem if $Z$ is a continuous random variable (or vector). In this paper, we prove that conditional independence is indeed a particularly difficult hypothesis to test for. Valid statistical tests are required to have a size that is smaller than a predefined significance level, and different tests usually have power against a different class of alternatives. We prove that a valid test for conditional independence does not have power against any alternative. Given the non-existence of a uniformly valid conditional independence test, we argue that tests must be designed so their suitability for a particular problem may be judged easily. To address this need, we propose in the case where $X$ and $Y$ are univariate to nonlinearly regress $X$ on $Z$, and $Y$ on $Z$ and then compute a test statistic based on the sample covariance between the residuals, which we call the generalised covariance measure (GCM). We prove that validity of this form of test relies almost entirely on the weak requirement that the regression procedures are able to estimate the conditional means $X$ given $Z$, and $Y$ given $Z$, at a slow rate. We extend the methodology to handle settings where $X$ and $Y$ may be multivariate or even high-dimensional. While our general procedure can be tailored to the setting at hand by combining it with any regression technique, we develop the theoretical guarantees for kernel ridge regression. A simulation study shows that the test based on GCM is competitive with state of the art conditional independence tests. Code is available as the R package GeneralisedCovarianceMeasure on CRAN.

Full work available at URL: https://arxiv.org/abs/1804.07203

Recommendations

• A GENERALIZATION OF TESTING INDEPENDENCE OF SETS OF VARIATES
• General tests of conditional independence based on empirical processes indexed by functions
• Expected conditional characteristic function-based measures for testing independence
• Testing conditional independence via empirical likelihood
• A flexible nonparametric test for conditional independence
• A consistent characteristic function-based test for conditional independence
• Nonparametric tests for conditional independence using conditional distributions
• On the power of conditional independence testing under model-X
• On some tests of the covariance matrix under general conditions

zbMATH Keywords

hypothesis testingwild bootstrapconditional independencekernel ridge regressiontestability

Mathematics Subject Classification ID

Nonparametric regression and quantile regression (62G08) Nonparametric hypothesis testing (62G10) Hypothesis testing in multivariate analysis (62H15) Ridge regression; shrinkage estimators (Lasso) (62J07)

Cites Work

• Causation, prediction, and search
• A projection-based conditional dependence measure with applications to high-dimensional undirected graphical models
• Sparse graphical models for exploring gene expression data
• Causal inference by using invariant prediction: identification and confidence intervals. With discussion and authors' reply
• Asymptotic normality and optimalities in estimation of large Gaussian graphical models
• Title not available (Why is that?)
• Statistics for high-dimensional data. Methods, theory and applications.
• Double/debiased machine learning for treatment and structural parameters
• A Non-Parametric Test of Independence
• Gaussian approximations and multiplier bootstrap for maxima of sums of high-dimensional random vectors
• Higher order influence functions and minimax estimation of nonlinear functionals
• Title not available (Why is that?)
• Title not available (Why is that?)
• Nonparametric independence testing via mutual information
• Causality. Models, reasoning, and inference
• Probabilistic graphical models.
• Goodness-of-Fit Tests for High Dimensional Linear Models
• Elements of causal inference. Foundations and learning algorithms
• Uniformly powerful goodness of fit tests
• Minimax estimation of a functional on a structured high-dimensional model
• Semiparametric minimax rates
• Hypothesis testing for densities and high-dimensional multinomials: sharp local minimax rates
• Convergence of estimates under dimensionality restrictions
• Measure theory and functional analysis
• Title not available (Why is that?)
• The Nonexistence of Certain Statistical Procedures in Nonparametric Problems
• Partial association measures and an application to qualitative regression
• Title not available (Why is that?)
• On Non-parametric Testing, the Uniform Behaviour of the t-test, and Related Problems
• Two new properties of mathematical likelihood
• On the equivalence between kernel quadrature rules and random feature expansions
• On the testability of identification in some nonparametric models with endogeneity
• Higher order inference on a treatment effect under low regularity conditions

Cited In (40)

• comets
• Minimax optimal conditional independence testing
• Conditional independence testing via weighted partial copulas
• Anytime-Valid Tests of Conditional Independence Under Model-X
• Sufficient variable screening with high-dimensional controls
• Local permutation tests for conditional independence
• The conditional permutation test for independence while controlling for confounders
• On the power of conditional independence testing under model-X
• A double-robust test for high-dimensional gene coexpression networks conditioning on clinical information
• Conditional feature importance for mixed data
• Reconciling model-X and doubly robust approaches to conditional independence testing
• Nonparametric conditional local independence testing
• Title not available (Why is that?)
• Test of conditional independence in factor models via Hilbert-Schmidt independence criterion
• Extending greedy feature selection algorithms to multiple solutions
• The Holdout Randomization Test for Feature Selection in Black Box Models
• Title not available (Why is that?)
• Efficient and multiply robust risk estimation under general forms of dataset shift
• Assumption-lean falsification tests of rate double-robustness of double-machine-learning estimators
• From statistical to causal learning
• Testing Directed Acyclic Graph via Structural, Supervised and Generative Adversarial Learning
• Yet another look at the omitted variable bias
• On Azadkia-Chatterjee's conditional dependence coefficient
• Asymptotic distributions of high-dimensional distance correlation inference
• GeneralisedCovarianceMeasure
• A survey of some recent developments in measures of association
• Demystifying Statistical Learning Based on Efficient Influence Functions
• Title not available (Why is that?)
• Causal structure learning: a combinatorial perspective
• A new covariate selection strategy for high dimensional data in causal effect estimation with multivariate treatments
• Double-estimation-friendly inference for high-dimensional misspecified models
• Testing conditional independence in supervised learning algorithms
• weightedGCM
• Game-theoretic statistical inference: optional sampling, universal inference, and multiple testing based on e-values. Abstracts from the workshop held May 5--10, 2024
• A simple measure of conditional dependence
• Optimal rates for independence testing via \(U\)-statistic permutation tests
• Comment: Reflections on the Deconfounder
• On universally consistent and fully distribution-free rank tests of vector independence
• General tests of conditional independence based on empirical processes indexed by functions
• Learning to increase the power of conditional randomization tests

Uses Software

• R
• CRAN
• TETRAD
• xgboost
• XGBoost
• ppcor
• HdBCS

This page was built for publication: The Hardness of Conditional Independence Testing and the Generalised Covariance Measure

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q118262)