關於成功率信賴區間建構之探討 = A Study of Confiden...
國立高雄大學統計學研究所

 

  • 關於成功率信賴區間建構之探討 = A Study of Confidence Interval for Binomial Proportion
  • 紀錄類型: 書目-語言資料,印刷品 : 單行本
    並列題名: A Study of Confidence Interval for Binomial Proportion
    作者: 王珊珊,
    其他團體作者: 國立高雄大學
    出版地: [高雄市]
    出版者: 撰者;
    出版年: 民99[2010]
    面頁冊數: 55面圖,表 : 30公分;
    標題: 二項分佈
    標題: Binomial Distribution
    電子資源: http://handle.ncl.edu.tw/11296/ndltd/67381738380322992112
    摘要註: 二元資料是一種常見的資料型態,如何建構其成功率p的信賴區間,為一重要之統計問題。當樣本數n夠大時,藉由中央極限定理之應用,Wald’s interval 是最常被使用的區間建構方法。一般教科書會建議np及n(1-p)至少要達到5的情況下,使用Wald’s interval較為恰當;但是,即便在教科書建議的情況下,仍明顯看出在許多可能成功率下,區間覆蓋率遠低於設定的目標。針對此一問題,Wilson (1927)和Agresti et. al. (1998) 分別提出修正的區間方法。雖然這兩個方法都提升了區間覆蓋率,但是,當p靠近0或1時,前者的區間覆蓋率還是遠低於目標值,而後者的覆蓋率雖然有達到目標,但是又提升的太多,讓區間過度保守。事實上,這三種方法都是藉由常態分佈的近似而建構出來的,無法反應出二項分佈的變異數為成功率函數此一事實。顯而易見,最直接的方法應是透過二項分佈來建構成功率的信賴區間。Clopper and Pearson (1934)、Sterne (1954)、Clow(1956)、Clunies-Ross (1958) 和Blyth et. al. (1983) 都分別提出不同的區間建構方式。隨著計算機運算能力的進步,這些方法都變得可行。在本文中,我們一一介紹這些區間建構方式,並藉由數值結果討論他們的表現。而根據數值結果以及其性質之合理性,我們推薦使用Clunies-Ross區間。由於Clunies-Ross區間沒有公式,在早期計算速度不夠快時,區間不易取得。但是,現在的計算速度可以更快速的求得Clunies-Ross區間,而我們也提供網站讓使用者能更方便取得此區間。在本文最後,我們亦給出Clunies-Ross區間推廣到3+3實驗之應用。 Estimating a 95% confidence interval (CI) of the success probability of a treatment is an important statistical problem. The easiest way to create an approximated 95% CI of the success probability p can be done by the Wald’s interval which is based on central limit theorem. Although many textbooks suggest that the Wald’s interval is suitable when the sample size n is large enough so that both np and n(1 − p) are greater than 5; however, the coverage probability is still far away below the nominal confidence level in many cases. Later, Wilson (1927) and Agresti et. al. (1998) give an alternative interval, respectively, to fix the problem. However, the coverage probability is still not fixing as p near 0 or 1. If the computing time can be ignored, the 95% CI can be constructed through the underlying binomial distribution. The CI which is obtained by binomial distribution is usually called exact CI, which can ensure that the coverage probability is larger than the specified nominal confidence level. There are many articles give the method to build exact CI, for example, Clopper and Pearson (1934), Sterne (1954), Clow (1956), Clunies-Ross (1958) and Blyth et. al. (1983). In this paper, we will review these methods and discuss the performance of them based on limit numerical results. Due to the numerical results and also some of the monotone properties which we will discuss in this paper, we recommend the use of the Clunies-Ross’s interval. In the end of the study, we also apply the Clunies-Ross interval to a phase I cancer clinical study.
館藏
  • 2 筆 • 頁數 1 •
 
310002030156 博碩士論文區(二樓) 不外借資料 學位論文 TH 008M/0019 343201 1017 2010 一般使用(Normal) 在架 0
310002030164 博碩士論文區(二樓) 不外借資料 學位論文 TH 008M/0019 343201 1017 2010 c.2 一般使用(Normal) 在架 0
  • 2 筆 • 頁數 1 •
評論
Export
取書館別
 
 
變更密碼
登入