學術產出-Theses

Article View/Open

Publication Export

Google ScholarTM

政大圖書館

Citation Infomation

  • No doi shows Citation Infomation
題名 失去部份訊息的類別資料之貝氏分析
Bayesian analysis for censored categorical data
作者 洪淑玲
貢獻者 姜志銘
洪淑玲
日期 1998
上傳時間 10-May-2016 16:00:44 (UTC+8)
摘要   失去部份訊息的類別資料之貝氏分析,早期的學者大都是針對1.失去的部份訊息是無價值性,2.誠實報告,與3.類別機率與報告的條件機率,二者的先驗分佈彼此互相獨立的假設條件來研究,直至近期雖有一些研究去除部份條件,但仍含有某些限制條件。文中將從不同的角度切入,並且不需上述的三個限制條件,同時先驗分佈族也由過去大部份學者所討論的Dirichlet分佈族擴展到Generalized Dirichlet分佈族,使能更廣泛包含各種先驗訊息。對於後驗平均數的估計,我們提出三種不同的計算方式:直接計算法、分解法與Quasi-Bayes法,並以幾個例子來比較說明。
  Bayesian methods for censored categorical data have been researched for decades. However, most of those are based on three restrictions that were discussed in Dickey, Jiang, and Kadane (1987), that is, noninformatively censored, truthful reporting, and prior independence between the categorical probability and conditional reported probability. Although some restrictions have been relaxed by some recent works, none has considered the cases without any restrictions. In this research, we shall remove all restrictions and extend prior distribution family to Generalized Dirichlet distribution family. In addition, three computational approaches for posterior means of parameters of the sampling population are presented: directed method, decomposition method, and quasi-Bayes method. Examples are given to illustrate and compare methods.
描述 博士
國立政治大學
統計學系
82354502
資料來源 http://thesis.lib.nccu.edu.tw/record/#A2010000642
資料類型 thesis
dc.contributor.advisor 姜志銘zh_TW
dc.contributor.author (Authors) 洪淑玲zh_TW
dc.creator (作者) 洪淑玲zh_TW
dc.date (日期) 1998en_US
dc.date.accessioned 10-May-2016 16:00:44 (UTC+8)-
dc.date.available 10-May-2016 16:00:44 (UTC+8)-
dc.date.issued (上傳時間) 10-May-2016 16:00:44 (UTC+8)-
dc.identifier (Other Identifiers) A2010000642en_US
dc.identifier.uri (URI) http://nccur.lib.nccu.edu.tw/handle/140.119/95946-
dc.description (描述) 博士zh_TW
dc.description (描述) 國立政治大學zh_TW
dc.description (描述) 統計學系zh_TW
dc.description (描述) 82354502zh_TW
dc.description.abstract (摘要)   失去部份訊息的類別資料之貝氏分析,早期的學者大都是針對1.失去的部份訊息是無價值性,2.誠實報告,與3.類別機率與報告的條件機率,二者的先驗分佈彼此互相獨立的假設條件來研究,直至近期雖有一些研究去除部份條件,但仍含有某些限制條件。文中將從不同的角度切入,並且不需上述的三個限制條件,同時先驗分佈族也由過去大部份學者所討論的Dirichlet分佈族擴展到Generalized Dirichlet分佈族,使能更廣泛包含各種先驗訊息。對於後驗平均數的估計,我們提出三種不同的計算方式:直接計算法、分解法與Quasi-Bayes法,並以幾個例子來比較說明。zh_TW
dc.description.abstract (摘要)   Bayesian methods for censored categorical data have been researched for decades. However, most of those are based on three restrictions that were discussed in Dickey, Jiang, and Kadane (1987), that is, noninformatively censored, truthful reporting, and prior independence between the categorical probability and conditional reported probability. Although some restrictions have been relaxed by some recent works, none has considered the cases without any restrictions. In this research, we shall remove all restrictions and extend prior distribution family to Generalized Dirichlet distribution family. In addition, three computational approaches for posterior means of parameters of the sampling population are presented: directed method, decomposition method, and quasi-Bayes method. Examples are given to illustrate and compare methods.en_US
dc.description.tableofcontents 謝辭
摘要
Abstract
目錄
圖目錄
表目錄
1. 緒論-----1
2. 預備知識與符號定義-----4
  2.1 預備知識-----4
  2.2 失去部份訊息的類別資料之概似函數-----5
  2.3 Dirichlet與Generalized Dirichlet分佈-----6
3. Bayes法-----8
  3.1 後驗分佈與參數估計-----8
  3.2 分解-----12
    3.2.1 定義與符號-----12
    3.2.2 Generalized Dirichlet分佈的分解性質-----15
    3.2.3 後驗分佈的分解性質-----18
4. Quasi-Bayes法-----31
  4.1 動機-----31
  4.2 參數估計-----32
  4.3 收斂性-----35
  4.4 模擬討論----Quasi-Bayes法與Bayes法的比較-----35
    4.4.1 限制條件-----36
    4.4.2 收斂性-Generalized Dirichlet分佈D(b,G,d),G=I-----37
    4.4.3 收斂性-Generalized Dirichlet分佈D(b,G,d),G≠I-----44
    4.4.4 Quasi-Bayes解使用時機-----60
5. 實例分析-----63
  5.1 實例一-----63
  5.2 實例二-----67
  5.3 實例三-----71
6. 結論-----74
附錄一-----76
附錄二-----78
參考文獻-----82

圖目錄
圖4.1 Bayes法之u的後驗平均數估計值(當u(0)=(0.5,0.3,0.2),E(u)=(0.45,0.4,0.15),b=(13.5,12,4.5))-----39
圖4.2 Bayes法之u的後驗平均數估計值(當u(0)=(0.5,0.3,0.2),E(u)=(0.3,0.4,0.2),b=(9,12,9))-----40
圖4.3 Bayes法之u的後驗平均數估計值(當u(0)=(0.7,0.23,0.07),E(u)=(0.675,0.27,0.05),b=(13.5,5.5,1))-----41
圖4.4 Bayes法之u的後驗平均數估計值(當u(0)=(0.7,0.23,0.07),E(u)=(0.75,0.15,0.1),b=(15,3,2))-----42
圖4.5 Bayes(B)法和Quasi-Bayes(Q.B.)法之後驗平均數估計值(當u(0)=(0.5,0.3,0.2),E(u)=(0.3,0.4,0.3),det(G)=0.0864,b=(9,12,9))-----46
圖4.6 Bayes法和Quasi-Bayes法之後驗報告機率估計值(當u(0)=(0.5,0.3,0.2),r=(0.37,0.31,0.32),det(G)=0.0864,b=(9,12,9))-----47
圖4.7 Bayes法和Quasi-Bayes法之u後驗平均數估計值與u(0)之相對誤差(當u(0)=(0.5,0.3,0.2),E(u)=(0.3,0.4,0.3),det(G)=0.0864,b=(9,12,9))-----47
圖4.8 Bayes法和Quasi-Bayes法之後驗報告機率估計值與r之相對誤差(當u(0)=(0.5,0.3,0.2),r=(0.37,0.31,0.32),det(G)=0.0864,b=(9,12,9))-----48
圖4.9 Bayes法和Quasi-Bayes法之u的後驗平均數估計值(當u(0)=(0.5,0.3,0.2),E(u)= (0.3,0.4,0.3),det(G)=0.4158,b=(9,12,9))-----49
圖4.10 Bayes法和Quasi-Bayes法之u的後驗平均數估計值(當u(0)=(0.5,0.3,0.2),E(u)=(0.4,0.35,0.25),det(G)=0.4158,b=(12,10.5,7.5))-----50
圖4.11 Bayes法和Quasi-Bayes法之u的後驗平均數估計值(當u(0)=(0.73,0.197,0.073),E(u)=(0.733,0.2,0.067),det(G)=0.0861,b=(22,6,2))-----52
圖4.12 Bayes法和Quasi-Bayes法之報告機率估計值(當u(0)=(0.73,0.197,0.073),r=(0.45,0.29,0.26),det(G)=0.0861,b=(22,6,2))-----52
圖4.13 Bayes法和Quasi-Bayes法之u的後驗平均數估計值(當u(0)=(0.73,0.197,0.073),E(u)=(0.85,0.1,0.05),det(G)=0.0861,b=(25.5,3,1,5))-----54
圖4.14 Bayes法和Quasi-Bayes法之後驗平均數估計值與u(0)之相對誤差(當u(0)=(0.73,0.197,0.073),E(u)=(0.85,0.1,0.05),det(G)=0.0861,b=(25.5,3,1.5))-----54
圖4.15 Bayes法和Quasi-Bayes法之報告機率估計值(當u(0)=(0.73,0.197,0.073),r=(0.45,0.29,0.26),det(G)=0.0861,b=(25.5,3,1.5))-----55
圖4.16 Bayes法和Quasi-Bayes法之報告機率估計值與r之相對誤差(當u(0)=(0.73,0.197,0.073),r=(0.45,0.29,0.26),det(G)=0.0861,b=(25.5,3,1.5))-----55
圖4.17 Bayes(B)法和Quasi-Bayes(Q.B.)法之u的後驗平均數估計值(當u(0)=(0.73,0.197,0.073),E(u)=(0.85,0.1,0.05),det(G)=0.33983,b=(25.5,3,1.5))-----57
圖4.18 Bayes法和Quasi-Bayes法之後驗平均數估計值與u(0)之相對誤差(當u(0)=(0.73,0.197,0.073),E(u)=(0.85,0.1,0.05),det(G)=0.3983,b=(25.5,3,1.5))-----57
圖4.19 Bayes法和Quasi-Bayes法之報告機率估計值(當u(0)=(0.73,0.197.0.073),r=(0.57,0.23,0.20),det(G)=0.3983,b=(25.5,3,1.5))-----58
圖4.20 Bayes法和Quasi-Bayes法之報告機率估計值與r之相對誤差(當u(0)=(0.73,0.197,0.073),r=(0.57,0.23,0.20),det(G)=0.3983,b=(25.5,3,1.5))-----58

表目錄
表2.1 聯合機率矩陣[μij]-----5
表3.1 首m行為對角矩陣之聯合機率矩陣 [μij]-----19
表4.1 u之後驗平均數收斂情形(當u(0)=(0.5,0.3,0.2),E(u)=(0.45,0.4,0.15)時)-----39
表4.2 u之後驗平均數收斂情形(當u(0)=(0.5,0.3,0.2),E(u)=(0.3,0.4,0.3)時)-----40
表4.3 u之後驗平均數收斂情形(當u(0)=(0.7,0.23,0.07),E(u)=(0.675,0.27,0.05)時)-----41
表4.4 u之後驗平均數收斂情形(當u(0)=(0.7,0.23,0.07),E(u)=(0.75,0.15,0.1)時)-----42
表4.5 u之後驗平均數收斂情形(當u(0)=(0.5,0.3,0.2),E(u)=(0.3,0.4,0.3))-----46
表4.6 u之後驗平均數收斂情形(當u(0)=(0.5,0.3,0.2),E(u)=(0.3,0.4,0.3))-----49
表4.7 u之後驗平均數收斂情形(當u(0)=(0.5,0.3,0.2),E(u)=(0.4,0.35,0.25))-----50
表4.8 u之後驗平均數收斂情形(當u(0)=(0.73,0.197,0.073),E(u)=(0.733,0.2,0.067))-----51
表4.9 u之後驗平均數收斂情形(當u(0)=(0.73,0.197,0.073),E(u)=(0.85,0.1,0.05))-----53
表4.10 u之後驗平均數收斂情形(當u(0)=(0.73,0.197,0.073),E(u)=(0.85,0.1,0.05))-----56
表4.11 gm,使得Pr(det(G)≧g)大約大於0.98之最大g值-----62
表4.12 使得(請參見全文資料),當N*-----62
表4.13 (請參見全文資料)-----62
表5.1 維他命C及感冒與否之列聯表(Pauling [1971])-----64
表5.2 維他命C及感冒與否的聯合機率矩陣[μij]-----64
表5.3 θ(維他命C及感冒與否)的後驗平均數-----65
表5.4 以分解法計算θ(維他命C及感冒與否)的先驗平均數-----66
表5.5 以分解法計算θ(維他命C及感冒與否)的後驗平均數-----66
表5.6 BIE的次數分配表-----68
表5.7 BIE的聯合機率矩陣[μij]-----68
表5.8 θ(BIE)的後驗平均數(標準差)-----69
表5.9 以分解法計算θ(BIE)的先驗平均數(標準差)-----70
表5.10 以分解法計算θ(BIE)的後驗平均數(標準差)-----70
表5.11 齲齒嚴重程度的聯合機率矩陣[μij]-----73
表5.12 Quasi-Bayes法與Bayes法之θ(齲齒嚴重程度)的後驗平均數(α=(1,1,1,1,1,1,1))-----73
表5.13 Quasi-Bayes法與Bayes法之θ(齲齒嚴重程度)的後驗平均數(α*=(10,8,10,3.5,4,4,5))-----73
zh_TW
dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#A2010000642en_US
dc.title (題名) 失去部份訊息的類別資料之貝氏分析zh_TW
dc.title (題名) Bayesian analysis for censored categorical dataen_US
dc.type (資料類型) thesisen_US