Exact adaptive confidence intervals for linear regression coefficients

DOI10.1214/18-EJS1517MaRDI QIDQ131754zbMATH OpenOpenAlexFDO

Authors Peter Hoff, Chaoyu Yu, Peter D. Hoff

Publication date 1 January 2019

Published in Electronic Journal of Statistics (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1705.08331, https://projecteuclid.org/euclid.ejs/1546570943

zbMATH Keywords

adaptive confidence interval procedureBayes-optimalempirical Bayesfrequentist coverageridge regressionshrinkagesparsityspending functionuniformly most accurate unbiased confidence interval procedure

Mathematics Subject Classification ID

Nonparametric tolerance and confidence regions (62G15) Linear regression; mixed models (62J05) Ridge regression; shrinkage estimators (Lasso) (62J07)

Abstract: We propose an adaptive confidence interval procedure (CIP) for the coefficients in the normal linear regression model. This procedure has a frequentist coverage rate that is constant as a function of the model parameters, yet provides smaller intervals than the usual interval procedure, on average across regression coefficients. The proposed procedure is obtained by defining a class of CIPs that all have exact $1-alpha$ frequentist coverage, and then selecting from this class the procedure that minimizes a prior expected interval width. Such a procedure may be described as "frequentist, assisted by Bayes" or FAB. We describe an adaptive approach for estimating the prior distribution from the data so that exact non-asymptotic $1-alpha$ coverage is maintained. Additionally, in a "$p$ growing with $n$" asymptotic scenario, this adaptive FAB procedure is asymptotically Bayes-optimal among $1-alpha$ frequentist CIPs.

Review text

In this paper, the authors consider an alternative to the uniformly most accurate unbiased confidence interval procedure for the coefficients in the normal linear regression model \(\mathbf{y} \sim N_n(X\mathbf{\beta}, \sigma^2)\). They present an adaptive confidence interval procedure for adaptively estimating a normal prior distribution for \(\mathbf{\beta}\) from the data \(\mathbf{y}\) to reduce the potential risk of a poorly specified prior, and then using this estimated prior distribution to construct an approximately Bayes-optimal confidence interval procedure for each regression coefficient \(\beta_j,\) \(j = 1, \ldots, p\). The resulting adaptive confidence intervals maintain exact non-asymptotic \(1-\alpha\) coverage if the design matrix is full rank and the errors are normally distributed. Note that this procedure has a frequentist coverage rate that is constant as a function of the model parameters, yet provides smaller intervals than the usual interval procedure, on average across regression coefficients. No assumptions on the unknown parameters are necessary to maintain exact coverage. Additionally, in a ``\(p\) growing with \(n\) asymptotic scenario, this adaptive ``frequentist, assisted by Bayes procedure is asymptotically Bayes-optimal among \(1-\alpha\) frequentist confidence interval procedures.

Recommendations

• Confidence intervals for high-dimensional linear regression: minimax rates and adaptivity
• Constructing narrower confidence intervals by inverting adaptive tests
• Using adaptive weighted least squares to reduce the lengths of confidence intervals
• A fixed-width interval for \(1/\beta\) in simple linear regression
• Adaptive confidence intervals for regression functions under shape constraints

Cites work

• \(p\)-values for high-dimensional regression
• A second-order efficient empirical Bayes confidence interval
• Adaptive multigroup confidence intervals with constant coverage
• Approaches for Empirical Bayes Confidence Intervals
• Bayesian Variable Selection in Linear Regression
• Confidence intervals for low dimensional parameters in high dimensional linear models
• Confidence intervals for the normal mean utilizing prior information
• Confidence intervals in regression utilizing prior information
• Exact post-selection inference, with application to the Lasso
• scientific article; zbMATH DE number 3205757 (Why is no real title available?)
• scientific article; zbMATH DE number 3083069 (Why is no real title available?)
• Interval estimation via tail functions
• Least angle regression. (With discussion)
• On asymptotically optimal confidence regions and tests for high-dimensional models
• Relations between Weak and Uniform Convergence of Measures with Applications
• Shorter Confidence Intervals for the Mean of a Normal Distribution with Known Variance
• Sparse estimators and the oracle property, or the return of Hodges' estimator
• Statistical significance in high-dimensional linear models
• Using adaptive weighted least squares to reduce the lengths of confidence intervals

Cited in

(15)

• fabCI
• Smaller p-values via indirect information
• FABInference
• Admissibility of the usual confidence interval in linear regression
• Confidence intervals for high-dimensional linear regression: minimax rates and adaptivity
• Bayes-optimal prediction with frequentist coverage control
• Testing Sparsity-Inducing Penalties
• Using adaptive weighted least squares to reduce the lengths of confidence intervals
• Constructing narrower confidence intervals by inverting adaptive tests
• Admissible and optimal confidence bands in simple linear regression
• Tests of linear hypotheses using indirect information
• LIC criterion for optimal subset selection in distributed interval estimation
• Confidence intervals in general regression models that utilize uncertain prior information
• On Yu and Hoff's confidence intervals for treatment means
• Cultural consensus theory for the evaluation of patients' mental health scores in forensic psychiatric hospitals

This page was built for publication: Exact adaptive confidence intervals for linear regression coefficients

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q131754)