Carving model-free inference

DOI10.1214/23-AOS2318MaRDI QIDQ6183866zbMATH OpenOpenAlexFDO

Authors Snigdha Panigrahi

Publication date 4 January 2024

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

Full work available at URL https://arxiv.org/abs/1811.03142, https://projecteuclid.org/journals/annals-of-statistics/volume-51/issue-6/Carving-model-free-inference/10.1214/23-AOS2318.full

zbMATH Keywords

carvingconditional inferencemodel-freepost-selection inferencerandomizationselective inference

Mathematics Subject Classification ID

Asymptotic distribution theory in statistics (62E20) Nonparametric tolerance and confidence regions (62G15) Asymptotic properties of nonparametric inference (62G20)

Abstract: In many large-scale experiments, the investigator begins with pilot data to look for promising findings. As fresh data becomes available at a later point of time, or from a different source, she is left with the question of how to use the full data to infer for the selected findings. Compensating for the overoptimism from selection, carving permits a reuse of pilot data for valid inference. The principles of carving are quite appealing in practice: instead of throwing away the pilot samples, carving simply discards the information consumed at the time of selection. However, the theoretical justification for carving is strongly tied to parametric models, an example being the ubiquitous gaussian model. In this paper we develop asymptotic guarantees to substantiate the use of carving beyond gaussian generating models. In simulations and in an application on gene expression data, we find that carving delivers valid and tight confidence intervals in model-free settings.

Recommendations

• Multicarving for high-dimensional post-selection inference
• Statistical learning and selective inference
• Inferactive data analysis
• Valid post-selection inference
• Bayesian selective inference

Cites work

• A significance test for the lasso
• A unified framework for high-dimensional analysis of M-estimators with decomposable regularizers
• Exact post-selection inference, with application to the Lasso
• Inference on Selected Subgroups in Clinical Trials
• Integrative Bayesian Models Using Post-Selective Inference: A Case Study in Radiogenomics
• Integrative methods for post-selection inference under convex constraints
• Moderate Deviations and Associated Laplace Approximations for Sums of Independent Random Vectors
• Multicarving for high-dimensional post-selection inference
• Multivariate normal approximation using exchangeable pairs
• Normal Approximation by Stein’s Method
• On the length of post-model-selection confidence intervals conditional on polyhedral constraints
• Scalable methods for Bayesian selective inference
• Selective inference with a randomized response
• Splitting strategies for post-selection inference
• Uniform asymptotic inference and the bootstrap after model selection

Cited in

(5)

• Asymptotically-exact selective inference for quantile regression
• Treatment effect estimation with efficient data aggregation
• Selective inference using randomized group Lasso estimators for general models
• Selective inference with distributed data
• Exact selective inference with randomization

This page was built for publication: Carving model-free inference

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6183866)