摘要: | We derive an explicit formula of S( n),where1-n is the Kaplan-Meier estimate of the incom-plete data X1,…,Xn,where X1,…,Xn are identical and independent random variables with common unknown distribution F and S is a given integral function.Since the X's are incomplete,S(n) is typically biased.Quenouille (1965) invented a method to estimate the bias nonparametrically and then to replace S(n) by a bias-corrected estimate: the jackknife.From this, the relationship between the bias-corrected jackknife estimate of S(F) and the observation data of X1,…,Xn is pointed out. 設X1,X2,…,Xn,為一獨立且具有相同分配函數F ( F為未知)的隨機變數樣本,設S為一已知的統計函數,我們想估計S(F)。當X1,X2,…,Xn可完全觀察到時,最典型的方法是用S(F,)估計S(F),此處Fn為X1,X2,…,Xn的經驗分配函數。當S非線性時, S(Fn)為一偏的估計量,因此 ; Quenouille於1956年提出用Jackknife的方法估t此偏差量,稱為Quenouille偏差估計量,並導出一個偏差校正估計量,此法有效的降低估S (F)的偏差量。當X1,X2,…,Xn的資料不能完全被觀察到時,一般用S(Fn)估S(F),此處l-F n為X1,X2,…,Xn,的Kaplan-Me ier估計量。因為X1,X2,…,Xn的資料不完全,即使S為一線性函數,S(Fn)仍為一偏的估計量。此篇論文,我們利用Kaplan-Meier估計量及Jackknife估計量的定義,準確地算出S(Fn)的Quenouille偏差估計量,並導出出S (F)的Jackknife偏差校正估計量與觀察值X1,X2,,…,Xn,的關係。 |