Bandwidth selection for kernel density estimate for randomly right-censored data

• Main Authors: Hui-ChunYu, 游惠群
• Other Authors: Tiee-Jian Wu
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/99235596557868886332

博士 === 國立成功大學 === 統計學系碩博士班 === 101 === Based on randomly right-censored sample of size n, the problem of selecting the global bandwidth in kernel density estimation of lifetime density f is investigated. A stabilized bandwidth selector, which is an extension to censored data of the complete-sample selector of Chiu (1992), is proposed. The key idea of our selector is to modify the weighted sample characteristic function beyond some cut-off frequency to reduce the sample variations without significantly inflating the bias. It is shown that under some smoothness conditions on f and the kernel, our selector is asymptotically normal distributed with the optimal root n relative convergence rate and attains the (conjectured) information bound. The excellent performances of the proposed selector at practical sample sizes are clearly demonstrated in simulation studies. In particular, the proposed selector performs conclusively better than the one selected by cross-validation.