References & Citations
Statistics > Methodology
Title: On Bootstrap Averaging Empirical Bayes Estimators
(Submitted on 27 Apr 2017)
Abstract: Parametric empirical Bayes (EB) estimators have been widely used in variety of fields including small area estimation, disease mapping. Since EB estimator is constructed by plugging in the estimator of parameters in prior distributions, it might perform poorly if the estimator of parameters is unstable. This can happen when the number of samples are small or moderate. This paper suggests bootstrapping averaging approach, known as "bagging" in machine learning literatures, to improve the performances of EB estimators. We consider two typical hierarchical models, two-stage normal hierarchical model and Poisson-gamma model, and compare the proposed method with the classical parametric EB method through simulation and empirical studies.
Submission history
From: Shonosuke Sugasawa [view email][v1] Thu, 27 Apr 2017 05:49:48 GMT (24kb,D)
Link back to: arXiv, form interface, contact.