東華大學圖書館 |

Advanced Methods in Bayesian Variable Selection and Causal Inference.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Advanced Methods in Bayesian Variable Selection and Causal Inference./
作者:	Cui, Can.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	138 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-06, Section: B.
Contained By:	Dissertations Abstracts International83-06B.
標題:	Confidence intervals. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28814865
ISBN:	9798494454799

Advanced Methods in Bayesian Variable Selection and Causal Inference.
Cui, Can.

Advanced Methods in Bayesian Variable Selection and Causal Inference. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 138 p.

Source: Dissertations Abstracts International, Volume: 83-06, Section: B.

Thesis (Ph.D.)--North Carolina State University, 2021.

This item must not be sold to any third party vendors.

Variable selection is a fundamental statistical tool to identify important covariates that are associated with the outcome of interest. Because of the increasing availability of high-dimensional data, it plays a pivotal role in detecting sparse signals in the presence of noise. Bayesian variable selection methods are favorable when researchers have prior belief or would like to fully quantify the uncertainty. However, complex data structure, especially in high-dimensional spaces, often poses challenges to variable selection. In the first part of this thesis, we contribute a Bayesian variable selection method to cope with high-dimensional microbiome data. In the second part, we develop an asymmetric shrinkage prior that takes the likely sign of covariates underlying the data into consideration. Finally, we turn our attention from the Bayesian variable selection to another statistical field, causal inference. Rather than studying association, causal inference deduces causality between the set of covariates and the outcome. Observational studies have been widely used to infer causal effects in various practical problems. In the last part of this thesis, we focus on the clustered observational data to study the causal effect using matching estimators with an application to environmental research.Chapter 1 is dedicated to the Bayesian variable selection and causal inference. We begin with an introduction to variable selection. Next, several Bayesian variable selection methods and their theoretical properties are reviewed. We then provide background on causal inference and briefly extend our overview to the matching techniques, which are commonly applied in causal inference and further discussed in subsequent chapters.In Chapter 2, we propose a Bayesian rank model to identify influential covariates that affect microbiome composition. Microbiome data are usually summarized as abundance counts. The counts are high-dimensional, overdispersed, often zero-inflated, and exhibit complex dependence structures. To detect important external factors that are associated with compositional counts under this data structure, we transform abundance counts into ranks and develop a Bayesian variable selection model that uses ranks to identify important covariates. We show by simulation that the proposed model outperforms competitors across various settings and particular improvement in recall for small magnitude and low prevalence covariates. The method is used to analyze a global topsoil dataset to investigate environmental factors that affect the topsoil microbiome composition.Chapter 3 describes a novel asymmetric shrinkage prior for Bayesian variable selection. The Bayesian variable selection methods usually assume a symmetric prior distribution for the regression coefficients with a high concentration near 0 to shrink unimportant ones, while heavy tails to identify important covariates. However, a symmetric distribution precludes formally incorporating prior information about the likely sign of the regression coefficients, which may be important for difficult high-dimensional problems. To address this issue, we propose an asymmetric shrinkage prior that accounts for such information on the sign of regression coefficients. We conduct simulation studies to demonstrate the advantages of the proposed asymmetric shrinkage in estimation accuracy and apply it to two real-world data examples in which the underlying distributions of coefficients are possibly skewed.In Chapter 4, we study the matching estimators in clustered observational studies. Rather than using propensity score methods, we focus on matching as a nonparametric approach and extend the matching estimator under the potential outcomes framework to clustered data. Large sample properties are explored for two aggregate estimands of interest, i.e., the average treatment effect and the average treatment effect on the treated. A cluster weighted bootstrap strategy is proposed to estimate the variance of matching estimators. We evaluate the performance of matching estimators in numerical studies. Finally, we apply the method to a motivating environmental dataset to study the causal effect of different marine protected area policies on fish biodiversity.

ISBN: 9798494454799Subjects--Topical Terms:

566017
Confidence intervals.

Advanced Methods in Bayesian Variable Selection and Causal Inference.
LDR:05287nmm a2200313 4500 001 2346912
005 20220706051322.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798494454799
035 $a (MiAaPQ)AAI28814865
035 $a (MiAaPQ)NCState_Univ18402039054
035 $a AAI28814865
040 $a MiAaPQ $c MiAaPQ
100 1 $a Cui, Can. $3 1267688
245 1 0 $a Advanced Methods in Bayesian Variable Selection and Causal Inference.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 138 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-06, Section: B.
500 $a Advisor: Reich, Brian.
502 $a Thesis (Ph.D.)--North Carolina State University, 2021.
506 $a This item must not be sold to any third party vendors.
520 $a Variable selection is a fundamental statistical tool to identify important covariates that are associated with the outcome of interest. Because of the increasing availability of high-dimensional data, it plays a pivotal role in detecting sparse signals in the presence of noise. Bayesian variable selection methods are favorable when researchers have prior belief or would like to fully quantify the uncertainty. However, complex data structure, especially in high-dimensional spaces, often poses challenges to variable selection. In the first part of this thesis, we contribute a Bayesian variable selection method to cope with high-dimensional microbiome data. In the second part, we develop an asymmetric shrinkage prior that takes the likely sign of covariates underlying the data into consideration. Finally, we turn our attention from the Bayesian variable selection to another statistical field, causal inference. Rather than studying association, causal inference deduces causality between the set of covariates and the outcome. Observational studies have been widely used to infer causal effects in various practical problems. In the last part of this thesis, we focus on the clustered observational data to study the causal effect using matching estimators with an application to environmental research.Chapter 1 is dedicated to the Bayesian variable selection and causal inference. We begin with an introduction to variable selection. Next, several Bayesian variable selection methods and their theoretical properties are reviewed. We then provide background on causal inference and briefly extend our overview to the matching techniques, which are commonly applied in causal inference and further discussed in subsequent chapters.In Chapter 2, we propose a Bayesian rank model to identify influential covariates that affect microbiome composition. Microbiome data are usually summarized as abundance counts. The counts are high-dimensional, overdispersed, often zero-inflated, and exhibit complex dependence structures. To detect important external factors that are associated with compositional counts under this data structure, we transform abundance counts into ranks and develop a Bayesian variable selection model that uses ranks to identify important covariates. We show by simulation that the proposed model outperforms competitors across various settings and particular improvement in recall for small magnitude and low prevalence covariates. The method is used to analyze a global topsoil dataset to investigate environmental factors that affect the topsoil microbiome composition.Chapter 3 describes a novel asymmetric shrinkage prior for Bayesian variable selection. The Bayesian variable selection methods usually assume a symmetric prior distribution for the regression coefficients with a high concentration near 0 to shrink unimportant ones, while heavy tails to identify important covariates. However, a symmetric distribution precludes formally incorporating prior information about the likely sign of the regression coefficients, which may be important for difficult high-dimensional problems. To address this issue, we propose an asymmetric shrinkage prior that accounts for such information on the sign of regression coefficients. We conduct simulation studies to demonstrate the advantages of the proposed asymmetric shrinkage in estimation accuracy and apply it to two real-world data examples in which the underlying distributions of coefficients are possibly skewed.In Chapter 4, we study the matching estimators in clustered observational studies. Rather than using propensity score methods, we focus on matching as a nonparametric approach and extend the matching estimator under the potential outcomes framework to clustered data. Large sample properties are explored for two aggregate estimands of interest, i.e., the average treatment effect and the average treatment effect on the treated. A cluster weighted bootstrap strategy is proposed to estimate the variance of matching estimators. We evaluate the performance of matching estimators in numerical studies. Finally, we apply the method to a motivating environmental dataset to study the causal effect of different marine protected area policies on fish biodiversity.
590 $a School code: 0155.
650 4 $a Confidence intervals. $3 566017
650 4 $a Signal to noise ratio. $3 3681178
650 4 $a Firearm laws & regulations. $3 3682013
650 4 $a Normal distribution. $3 3561025
650 4 $a Firearms. $2 fast $3 3509397
650 4 $a Statistics. $3 517247
690 $a 0338
690 $a 0463
710 2 $a North Carolina State University. $3 1018772
773 0 $t Dissertations Abstracts International $g 83-06B.
790 $a 0155
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28814865