High-Dimensional Single-Index Models: Link Estimation and Marginal Inference
Authors:
Kazuma Sawaya,
Yoshimasa Uematsu,
Masaaki Imaizumi
Abstract:
This study proposes a novel method for estimation and hypothesis testing in high-dimensional single-index models. We address a common scenario where the sample size and the dimension of regression coefficients are large and comparable. Unlike traditional approaches, which often overlook the estimation of the unknown link function, we introduce a new method for link function estimation. Leveraging…
▽ More
This study proposes a novel method for estimation and hypothesis testing in high-dimensional single-index models. We address a common scenario where the sample size and the dimension of regression coefficients are large and comparable. Unlike traditional approaches, which often overlook the estimation of the unknown link function, we introduce a new method for link function estimation. Leveraging the information from the estimated link function, we propose more efficient estimators that are better aligned with the underlying model. Furthermore, we rigorously establish the asymptotic normality of each coordinate of the estimator. This provides a valid construction of confidence intervals and $p$-values for any finite collection of coordinates. Numerical experiments validate our theoretical results.
△ Less
Submitted 27 April, 2024;
originally announced April 2024.
Moment-Based Adjustments of Statistical Inference in High-Dimensional Generalized Linear Models
Authors:
Kazuma Sawaya,
Yoshimasa Uematsu,
Masaaki Imaizumi
Abstract:
We developed a statistical inference method applicable to a broad range of generalized linear models (GLMs) in high-dimensional settings, where the number of unknown coefficients scales proportionally with the sample size. Although a pioneering inference method has been developed for logistic regression, which is a specific instance of GLMs, we cannot apply this method directly to other GLMs becau…
▽ More
We developed a statistical inference method applicable to a broad range of generalized linear models (GLMs) in high-dimensional settings, where the number of unknown coefficients scales proportionally with the sample size. Although a pioneering inference method has been developed for logistic regression, which is a specific instance of GLMs, we cannot apply this method directly to other GLMs because of unknown hyper-parameters. In this study, we addressed this limitation by developing a new inference method designed for a certain class of GLMs. Our method is based on the adjustment of asymptotic normality in high dimensions and is feasible in the sense that it is possible even with unknown hyper-parameters. Specifically, we introduce a novel convex loss-based estimator and its associated system, which are essential components of inference. Next, we devise a moment-based method for estimating the system parameters required by the method. Consequently, we construct confidence intervals for GLMs in a high-dimensional regime. We prove that our proposed method has desirable theoretical properties, such as strong consistency and exact coverage probability. Finally, we experimentally confirmed its validity.
△ Less
Submitted 23 May, 2024; v1 submitted 28 May, 2023;
originally announced May 2023.
