Publications

Detailed Information

Empirical Bayes Estimation of Negative Binomial Models And Its Applications : 음이항분포의 경험적 베이즈 추정 및 응용

DC Field Value Language
dc.contributor.advisor장원철-
dc.contributor.author최재철-
dc.date.accessioned2019-05-07T04:33:40Z-
dc.date.available2019-05-07T04:33:40Z-
dc.date.issued2019-02-
dc.identifier.other000000155858-
dc.identifier.urihttps://hdl.handle.net/10371/151611-
dc.description학위논문 (석사)-- 서울대학교 대학원 : 자연과학대학 통계학과, 2019. 2. 장원철.-
dc.description.abstractData such as the number of accidents in a location and the number of children per women is characterized by the facts that one or a small number of observations are available in each subject and the observation values are nonnegative integers. When statistical inference is performed based on these types of data, empirical Bayes estimation methods based on Poisson distribution have been commonly used. However, the Poisson distribution has the limitation that the mean and the variance are the same. The negative binomial distribution and the zero-inflated Poisson distribution have the same support as that of the Poisson distribution, but with two parameters, a more flexible model fit is possible. In this thesis, we explore and compare various methods that fit the negative binomial distribution model and zero-inflated Poisson model in a situation where there is only one datum per a subject. We also applied them to some case studies including automobile insurance claims data and fatal traffic accidents data.-
dc.description.abstract특정 지역과 시간대에 일어난 사건의 수, 여성 1인당 자녀 수 등의 자료는 대상별로 관측치가 음 아닌 정수 형태로 하나밖에 없다는 특징이 있다. 이러한 자료를 바탕으로 통계적 추정을 할 때, 포아송 분포를 가정한 경험적 베이즈 추정 방식을 많이 사용해 왔다. 그러나 포아송분포는 평균과 분산이 항상 같다는 제약이 있다. 음이항분포와 0이 기대보다 많이 관측된 포아송분포는 모수가 두 개라는 점에서 보다 유연한 모형적합이 가능하다. 본 논문에서는 1인당 자료가 하나밖에 없는 상황에서 음이항분포 모형과 0이 기대보다 많이 관측된 포아송 분포 모형을 적합하는 다양한 방법을 탐구하고 비교해 보았다. 또한, 이를 보험 신고 자료와 교통사고 자료에 응용해 보았다.-
dc.description.tableofcontents1 Introduction 5
2 Models 7
2.1 PoissonModel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Properties of PoissonModel . . . . . . . . . . . . . . . . . . . 8
2.1.2 Estimation Strategies for PoissonModel . . . . . . . . . . . 8
2.1.3 Limitations of PoissonModel . . . . . . . . . . . . . . . . . . 9
2.2 Negative BinomialModel . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Constant Shape ParameterMethod . . . . . . . . . . . . . . 10
2.2.2 Estimation Strategies for The Constant Shape Negative BinomialModel
. . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.3 Conditional PriorMethod . . . . . . . . . . . . . . . . . . . . 13
2.2.4 Estimation Stratigy for NB PriorModels . . . . . . . . . . . . 15
2.3 Zero-Inflated PoissonModeling . . . . . . . . . . . . . . . . . . . . 15
2.3.1 Zero-Inflated Poisson density . . . . . . . . . . . . . . . . . . 15
2.3.2 Estimation Strategy for Zero-Inflated PoissonModel . . . . 16
3 Simulations 17
3.1 Estimation and Inference . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.1 Beta Negative Binomial . . . . . . . . . . . . . . . . . . . . . 19
3.1.2 Gamma Poisson . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1.3 Zero-Infalted Poisson . . . . . . . . . . . . . . . . . . . . . . 21
3.1.4 UniformPoisson . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.5 Known UniformPoisson . . . . . . . . . . . . . . . . . . . . . 22
4 Case Studies 24
4.1 Automobile Insurance Claims Data . . . . . . . . . . . . . . . . . . 24
4.2 Fatal Traffic Accidents Data . . . . . . . . . . . . . . . . . . . . . . . 26
5 Conclusion 28
2
-
dc.language.isoeng-
dc.publisher서울대학교 대학원-
dc.subject.ddc519.5-
dc.titleEmpirical Bayes Estimation of Negative Binomial Models And Its Applications-
dc.title.alternative음이항분포의 경험적 베이즈 추정 및 응용-
dc.typeThesis-
dc.typeDissertation-
dc.contributor.AlternativeAuthorCHOI JAECHEOL-
dc.description.degreeMaster-
dc.contributor.affiliation자연과학대학 통계학과-
dc.date.awarded2019-02-
dc.contributor.major고차원 대용량 자료-
dc.identifier.uciI804:11032-000000155858-
dc.identifier.holdings000000000026▲000000000039▲000000155858▲-
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