A0279
Title: An efficient variance estimator for cross-validation under partition-sampling
Authors: Qing Wang - Wellesley College (United States) [presenting]
Xizhen Cai - Williams College (United States)
Abstract: The problem of variance estimation of cross-validation is concerned. It considers an unbiased cross-validation risk estimator in the form of a general U-statistic. The proposal is an efficient variance estimator under a half-sampling design, where the estimator's bias can be expressed explicitly. Furthermore, one can approximate the bias by a two-layer Monte Carlo method so that a bias-corrected variance estimator can be obtained. In the simulation study and real data examples, the performance of the proposed variance estimator, in comparison to the commonly used bootstrap and jackknife methods, is evaluated in the context of model selection. The numerical results suggest that the proposal yields identical or similar conclusions for model selection compared to its counterparts, and it is much more efficient to calculate than its competitors. The generalization of the methodology to other partition-sampling scenarios is also discussed.
