Test for conditional independence with application to conditional screening (Q2293386)

From MaRDI portal
Revision as of 23:39, 28 July 2024 by Daniel (talk | contribs) (‎Created claim: Wikidata QID (P12): Q127101900, #quickstatements; #temporary_batch_1722205759638)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
scientific article
Language Label Description Also known as
English
Test for conditional independence with application to conditional screening
scientific article

    Statements

    Test for conditional independence with application to conditional screening (English)
    0 references
    0 references
    0 references
    0 references
    5 February 2020
    0 references
    In this paper the authors present a novel distribution-free test for conditional independence. Let \(X \in \mathbb{R}^{1}\), \(Y \in \mathbb{R}^{1}\) be random variables and \(\mathbf{u}=\left(U_{1},\ldots,U_{d} \right)^T \in \mathbb{R}^{d}\) be a random vector. Under mild conditions, they show that conditional independence between \(X\) and \(Y\) given \(\mathbf{u}\) is equivalent to the unconditional independence between the respective conditional distributions \(V=F_{1}\left(X|\mathbf{u} \right) \) and \(W=F_{2}\left(Y|\mathbf{u} \right) \). The authors use the Blum-Kiefer-Rosenblatt correlation \[ \rho = \rho_{V,W} \stackrel{\mathrm{def}}{=} \int_{\mathbb{R}^{1}} \int_{\mathbb{R}^{1}}\left[F_{V,W}\left(v,w \right) - F_{V}\left(v \right) F_{W}\left(w \right) \right]^{2} dF_{V}\left(v \right) d F_{W}\left(w \right) \] to test the independence of \(V\) and \(W\). The BKR correlation is powerful to capture nonlinear dependence and is robust to heavy-tailed errors. From the authors' abstract: ``We obtain explicit forms for the asymptotic null distribution which involves no unknown tunings, rendering fast and easy implementation of our test for conditional independence. With this conditional independence test, we further propose a conditional screening method for high dimensional data to identify truly important covariates whose effects may vary with exposure variables. We use the false discovery rate to determine the screening cutoff. This screening approach possesses both the sure screening and the ranking consistency properties. We illustrate the finite sample performances through simulation studies and an application to the gene expression microarray dataset.''
    0 references
    conditional independence
    0 references
    feature screening
    0 references
    high dimensional data
    0 references
    independence
    0 references
    sure screening property
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references

    Identifiers