Improved Variance Estimation in Homoskedastic Nonparametric Random-Design Regression via a Two-Scale Approach
Abstract
We study estimation of a constant conditional variance σ2 in nonparametric regression with a d-dimensional random design. This is an important problem, and similar questions arise in causal inference. The regression function is βb-Hölder smooth, the design density is βg-Hölder smooth and bounded above and away from zero, and we consider the nonparametric regime βb>1 and d>4βb. Set βg⋆=βb(1−4βb/d)/{1+2βb/d+8(βb/d)2}. We give an estimator whose mean squared error is upper bounded by Cn−4(βb+1)/(d+4) in the low-regularity regime when 0<βg≤βg⋆. The low-regularity branch is based on a new two-scale construction: the covariate space is partitioned into cells, the local polynomial trend is projected out within each suitable cell, and the squared normalized contrast from one eligible close pair per cell is averaged across cells. In the high regularity regime when βg>βg⋆, a higher-order influence function estimator of Robins, Li, Tchetgen Tchetgen, and van der Vaart (2008) provides the rate Cn−8βb/(d+4βb). We also give an all-pairs ridge extension, which achieves the same two-scale rate, and evaluate the methods alongside a range of existing estimators in simulations.
Related
- Nonparametric shrinkage estimation of high-dimensional variance component matrices
- Scalable Nonparametric Change-Plane Regression with Random Features: Estimation and Inference
- Optimal Neural Estimation for High-Dimensional Nonparametric Regression under Multisource Covariate Shift
- Variance estimation for generalized linear regression with binary outcome when using case-cohort data
- Nonparametric Variance-Penalized Actor-Critic: Statistical Inference for Risk-Sensitive Reinforcement Learning
BibTeX
@misc{dobriban2026improved,
title = {{Improved Variance Estimation in Homoskedastic Nonparametric Random-Design Regression via a Two-Scale Approach}},
author = {Edgar Dobriban and Rajarshi Mukherjee and James M. Robins and Zixiao Wang},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.08783},
eprint = {2609.08783},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.08783},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Improved Variance Estimation in Homoskedastic Nonparametric Random-Design Regression via a Two-Scale Approach
AU - Edgar Dobriban
AU - Rajarshi Mukherjee
AU - James M. Robins
AU - Zixiao Wang
PY - 2026
DA - 2026/09/08
PB - arXiv
DO - 10.48550/arxiv.2609.08783
UR - https://arxiv.org/abs/2609.08783
AB - We study estimation of a constant conditional variance σ^2 in nonparametric regression with a d-dimensional random design. This is an important problem, and similar questions arise in causal inference. The regression function is β_b-Hölder smooth, the design density is β_g-Hölder smooth and bounded above and away from zero, and we consider the nonparametric regime β_b>1 and d>4β_b. Set β_g^⋆=β_b(1−4β_b/d)/{1+2β_b/d+8(β_b/d)^2}. We give an estimator whose mean squared error is upper bounded by Cn^−4(β_b+1)/(d+4) in the low-regularity regime when 0<β_g≤β_g^⋆. The low-regularity branch is based on a new two-scale construction: the covariate space is partitioned into cells, the local polynomial trend is projected out within each suitable cell, and the squared normalized contrast from one eligible close pair per cell is averaged across cells. In the high regularity regime when β_g>β_g^⋆, a higher-order influence function estimator of Robins, Li, Tchetgen Tchetgen, and van der Vaart (2008) provides the rate Cn^−8β_b/(d+4β_b). We also give an all-pairs ridge extension, which achieves the same two-scale rate, and evaluate the methods alongside a range of existing estimators in simulations.
ER -