Total positivity in exponential families with application to binary variables

DOI10.1214/20-AOS2007zbMATH Open1473.60046arXiv1905.00516OpenAlexW3191412046WikidataQ115517711 ScholiaQ115517711MaRDI QIDQ820806FDOQ820806

Authors: Caroline Uhler, Piotr Zwiernik, Steffen Lauritzen 

Publication date: 28 September 2021

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

Abstract: We study exponential families of distributions that are multivariate totally positive of order 2 (MTP2), show that these are convex exponential families, and derive conditions for existence of the MLE. Quadratic exponential familes of MTP2 distributions contain attractive Gaussian graphical models and ferromagnetic Ising models as special examples. We show that these are defined by intersecting the space of canonical parameters with a polyhedral cone whose faces correspond to conditional independence relations. Hence MTP2 serves as an implicit regularizer for quadratic exponential families and leads to sparsity in the estimated graphical model. We prove that the maximum likelihood estimator (MLE) in an MTP2 binary exponential family exists if and only if both of the sign patterns $(1,-1)$ and $(-1,1)$ are represented in the sample for every pair of variables; in particular, this implies that the MLE may exist with $n=d$ observations, in stark contrast to unrestricted binary exponential families where $2^d$ observations are required. Finally, we provide a novel and globally convergent algorithm for computing the MLE for MTP2 Ising models similar to iterative proportional scaling and apply it to the analysis of data from two psychological disorders.

Full work available at URL: https://arxiv.org/abs/1905.00516

Recommendations

• Total positivity properties of the bivariate diagonal natural exponential families
• Positive Dependence of a Class of Multivariate Exponential Distributions
• On a generalization of the positive exponential family of distributions and the estimation of reliability characteristics
• Exponential Families in Theory and Practice
• On empirical Bayes tests in a positive exponential family
• Estimation of P(X>Y) for the Positive Exponential Family of Distributions
• Modeling and Inference for Multivariate Binary Data with Positive Dependence

zbMATH Keywords

graphical modelsIsing modelpositive dependenceexponential familieslog-supermodular distributions

Mathematics Subject Classification ID

Multivariate analysis (62H99) Inequalities; stochastic orderings (60E15) Positive matrices and their generalizations; cones of matrices (15B48)

Cites Work

• Title not available (Why is that?)
• Title not available (Why is that?)
• Title not available (Why is that?)
• Title not available (Why is that?)
• On the toric algebra of graphical models
• Title not available (Why is that?)
• Probabilistic Networks and Expert Systems
• Classes of orderings of measures and related correlation inequalities. I. Multivariate totally positive distributions
• A likelihood ratio test for \(\text{MTP}_ 2\) within binary variables.
• Some notions of multivariate positive dependence
• Correlation inequalities on some partially ordered sets
• Exact sampling with coupled Markov chains and applications to statistical mechanics
• Tensors of nonnegative rank two
• Total positivity in Markov structures
• Palindromic Bernoulli distributions
• High-dimensional Gaussian graphical model selection: walk summability and local separation criterion
• A recursive algorithm for Markov random fields
• Covariance selection and multivariate dependence
• Title not available (Why is that?)
• On the effective implementation of the iterative proportional fitting procedure
• A general central limit theorem for FKG systems
• Maximum likelihood estimation in Gaussian models under total positivity
• Learning with submodular functions: a convex optimization perspective
• Geometry of log-concave density estimation
• Estimation of positive definite \(M\)-matrices and structure learning for attractive Gaussian Markov random fields
• Discrete Gaussian Distributions via Theta Functions
• Maximum likelihood estimation of the latent class model through model boundary decomposition

Cited In (8)

• Total positivity in multivariate extremes
• Dependence in elliptical partial correlation graphs
• Algebraic structures in statistical methodology. Abstracts from the workshop held December 4--10, 2022
• Total positivity properties of the bivariate diagonal natural exponential families
• A likelihood ratio test for \(\text{MTP}_ 2\) within binary variables.
• Total positivity in Markov structures
• Locally associated graphical models and mixed convex exponential families
• Some aspects of non-standard multivariate analysis

This page was built for publication: Total positivity in exponential families with application to binary variables

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q820806)