Publications

Detailed Information

Prior Elicitation Methods in the Normal Linear Model : 정규선형회귀모형에서의 사전확률 도출방법

DC Field Value Language
dc.contributor.advisor이재용-
dc.contributor.author권용찬-
dc.date.accessioned2017-07-14T00:30:49Z-
dc.date.available2017-07-14T00:30:49Z-
dc.date.issued2013-08-
dc.identifier.other000000012525-
dc.identifier.urihttps://hdl.handle.net/10371/121142-
dc.description학위논문 (박사)-- 서울대학교 대학원 : 통계학과, 2013. 8. 이재용.-
dc.description.abstract베이지안 정규선형회귀모형에서의 새로운 사전확률분포 도출절차를 제시하고, 이를 실제 자료분석에 적용한다. 연구의 목적은 기존 연구들에서 제시된 방법들에 비해 확률분포 도출에 사용되는 설문을 응답자들이 보다 직관적으로 파악하여 쉽게 응답할 수 있도록 구성하고 보다 간편한 계산과정을 통해 해당 모형의 모수추정이 가능하게 하는 절차를 제안하는 것이다.
모수추정방법들은 설문자료에 대하여 도출과정에서의 불명확성이 반영된 도출오차를 가정한 통계적인 모형에 기반하여 제시된다. 제안된 추정방법들은 Ad-hoc방법, 최대가능도추정법, 로버스트 추정법 그리고 마르코프연쇄 몬테카를로(MCMC)를 통한 완전 베이지안 추정법이다. 이들 방법들은 여러 가지 상황을 고려한 모의실험을 통해 그 정확성이 비교‧분석된다.
제안된 방법은 실제 국내 무기체계 – 어뢰, 전차 - 연구개발실험평가(RDTE) 비용모델 구축에 사용되며, 적용절차에 대해서 구체적으로 제시한다. 결론을 통해 비용추정모델 및 비용추정 결과에 대해 논의하고 토론한다.
-
dc.description.abstractWe propose a new procedure for the prior elicitation under the Bayesian normal linear model and apply it to the real data analyses. Object of the study is to elicit prior probabilities of the model with simple and intuitive questionnaire for interviewees and provide estimation methods of model hyperparameters under the proposed procedure.
The estimation methods are explained under the assumption of elicitation error
involved in the process of elicitation and that the elicited summaries are to be statistically modelled. The proposed methods are Ad-hoc method, maximum likelihood estimation under the assumption of elicitation errors, robust(L1) estimation and full Bayesian estimation via Bayesian computation such as Markov Chain Monte Carlo. The performance of proposed estimates including Bayesian estimation is compared and discussed under the simulated data set for various parameter values.
Details of data analysis procedure are provided in the perspective of the research, development, test and evaluation (RDTE) cost model building for Korean weapon development with the estimated predictive cost distributions for two weapon systems - the torpedo and the tank. Review and discussion of the overall result are provided in the discussion and conclusion chapter of the study.
-
dc.description.tableofcontentsAbstract i
1 Introduction 1
2 Literature Review 5
3 Model and Data 10
3.1 Model and Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Data to be Elicited . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4 Parameter Estimation 17
4.1 Review of Previous Studies . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 Ad-hoc Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.1 Estimation of b and δ . . . . . . . . . . . . . . . . . . . . . . 21
4.2.2 Estimation of w and R . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Maximum Likelihood Estimation . . . . . . . . . . . . . . . . . . . . 28
4.3.1 Likelihood Approach . . . . . . . . . . . . . . . . . . . . . . . 28
4.3.2 Estimation of Parameters . . . . . . . . . . . . . . . . . . . . 32
4.4 Robust Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.5 Bayesian Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.5.1 Bayesian Setup . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.5.2 MCMC Scheme and Parameter Estimation . . . . . . . . . . 38
4.6 Design Point Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5 Simulation Study 43
5.1 Ad-hoc method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2 Maximum Likelihood Estimation . . . . . . . . . . . . . . . . . . . . 50
5.3 Robust Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.4 Mean Squared Error Comparison . . . . . . . . . . . . . . . . 58
5.5 Bayesian Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6 Application 87
6.1 Procedure of Cost Model Building . . . . . . . . . . . . . . . . . . . 88
6.1.1 Construction of The GWBS of a Weapon System . . . . . . . 91
6.1.2 Cost Estimation Procedure of Total System . . . . . . . . . . 92
6.2 Real Data Analysis : Torpedo System . . . . . . . . . . . . . . . . 94
6.2.1 GWBS of a Torpedo System . . . . . . . . . . . . . . . . . . 94
6.2.2 Elicitation of Cost Drivers and Modeling . . . . . . . . . . . . 96
6.2.3 Cost Estimation and Evaluation . . . . . . . . . . . . . . . . 97
6.3 Real Data Analysis : Tank System . . . . . . . . . . . . . . . . . . . 111
6.3.1 GWBS of a Tank System . . . . . . . . . . . . . . . . . . . . 111
6.3.2 Elicitation of Cost Drivers and Modeling . . . . . . . . . . . . 112
6.3.3 Cost Estimation and Evaluation . . . . . . . . . . . . . . . . 114
7 Discussion and Conclusion 136
Bibliography 139
Abstract(in Korean) 152
-
dc.formatapplication/pdf-
dc.format.extent3868800 bytes-
dc.format.mediumapplication/pdf-
dc.language.isoen-
dc.publisher서울대학교 대학원-
dc.subjectPrior Elicitation-
dc.subjectMultivariate t-distribution-
dc.subjectCost Model-
dc.subjectData Absence-
dc.subjectMCMC(Markov Chain Monte Carlo)-
dc.subjectImprecision in Elicitation-
dc.subject.ddc519-
dc.titlePrior Elicitation Methods in the Normal Linear Model-
dc.title.alternative정규선형회귀모형에서의 사전확률 도출방법-
dc.typeThesis-
dc.contributor.AlternativeAuthorYongchan Kwon-
dc.description.degreeDoctor-
dc.citation.pages152-
dc.contributor.affiliation자연과학대학 통계학과-
dc.date.awarded2013-08-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share