We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Methodology

Title: Accurate directional inference in Gaussian graphical models

Abstract: Directional tests to compare nested parametric models are developed in the general context of covariance selection for Gaussian graphical models. The exactness of the underlying saddlepoint approximation leads to exceptional accuracy of the proposed approach. This is verified by simulation experiments with high-dimensional parameters of interest, where the accuracy of standard asymptotic approximations to the likelihood ratio test and some of its higher-order modifications fails. The directional p-value isused to illustrate the assessment of Markovian dependencies in a dataset from a veterinary trial on cattle. A second example with microarray data shows how to select the graph structure related to genetic anomalies due to acute lymphocytic leukemia.
Subjects: Methodology (stat.ME); Computation (stat.CO)
Cite as: arXiv:2103.15394 [stat.ME]
  (or arXiv:2103.15394v1 [stat.ME] for this version)

Submission history

From: Claudia Di Caterina [view email]
[v1] Mon, 29 Mar 2021 07:42:53 GMT (113kb,D)

Link back to: arXiv, form interface, contact.