ingest cdrApp 2018-06-13T15:43:46.809Z 51cd2fe2-3fd7-401f-a923-a97bc3db68a2 modifyDatastreamByValue RELS-EXT fedoraAdmin 2018-06-13T16:11:07.332Z Setting exclusive relation addDatastream MD_TECHNICAL fedoraAdmin 2018-06-13T16:11:18.526Z Adding technical metadata derived by FITS addDatastream MD_FULL_TEXT fedoraAdmin 2018-06-13T16:11:41.332Z Adding full text metadata extracted by Apache Tika modifyDatastreamByValue RELS-EXT fedoraAdmin 2018-06-13T16:12:03.462Z Setting exclusive relation modifyDatastreamByValue MD_DESCRIPTIVE cdrApp 2018-07-16T21:03:01.489Z modifyDatastreamByValue MD_DESCRIPTIVE cdrApp 2018-07-18T16:37:36.315Z modifyDatastreamByValue MD_DESCRIPTIVE cdrApp 2018-08-22T15:17:50.289Z modifyDatastreamByValue MD_DESCRIPTIVE cdrApp 2018-09-28T18:04:56.798Z modifyDatastreamByValue MD_DESCRIPTIVE cdrApp 2018-10-12T16:57:20.870Z modifyDatastreamByValue MD_DESCRIPTIVE cdrApp 2019-03-22T20:18:34.535Z Brian Barkley Author Department of Biostatistics Gillings School of Global Public Health Causal Inference From Observational Studies in the Presence of Partial Interference Analyzing data to estimate the effect of treatment on health outcomes can play a major role in the fields of personal and public health. Interference occurs when the treatment of one individual affects the outcome of another individual. This work aims to develop statistical methodology for inference about causal effects from observational studies in the presence of interference. We assume partial interference throughout: interference may exist within clusters of individuals, but not between distinct clusters. In each paper we propose estimators that are consistent and asymptotically Normal; estimators for the asymptotic variance are also proposed. Finite-sample performance of each estimator is investigated, and each method is illustrated by analyzing a cholera vaccine study in Matlab, Bangladesh. In the first paper, we propose a method for inverse probability-weighted estimation of target estimands in the presence of partial interference that is more robust to model mis-specification than existing methods. This technique relies on an algorithm which combines machine learning and mixed effects methods to determine the relationship between treatment and covariates assuming a certain correlation structure. We employ the algorithm on a training sample as a data-adaptive model selection procedure. We recover the set of rules that the algorithm uses for prediction to transform the covariates in a testing sample. We proceed by fitting a model to the transformed covariates to estimate propensity scores for IPW estimation of target estimands. We propose a matching technique for estimating causal effects in the presence of partial interference in the second paper. These estimators extend methods that are commonly employed when no interference is assumed. The proposed methods can be carried out without modeling treatment, and may outperform existing IPW estimators in certain scenarios. Extensions of these estimators are discussed. In the third paper we propose new causal estimands for observational studies in the presence of partial interference. The proposed estimands describe counterfactual scenarios in which there may be within-cluster dependence in the individual treatment selections. These estimands may be more relevant for public health officials. These estimands are identifiable from observational data with parametric assumptions. Inverse probability-weighted estimators for these estimands are proposed. Spring 2018 2018 Biostatistics Statistics Epidemiology causal inference, interference, inverse probability weight, matching, observational study, spillover effects eng Doctor of Philosophy Dissertation University of North Carolina at Chapel Hill Graduate School Degree granting institution Biostatistics Michael Hudgens Thesis advisor Maurice Brookhart Thesis advisor Michael Emch Thesis advisor Michael Kosorok Thesis advisor Donglin Zeng Thesis advisor text Brian Barkley Author Department of Biostatistics Gillings School of Global Public Health Causal Inference From Observational Studies in the Presence of Partial Interference Analyzing data to estimate the effect of treatment on health outcomes can play a major role in the fields of personal and public health. Interference occurs when the treatment of one individual affects the outcome of another individual. This work aims to develop statistical methodology for inference about causal effects from observational studies in the presence of interference. We assume partial interference throughout: interference may exist within clusters of individuals, but not between distinct clusters. In each paper we propose estimators that are consistent and asymptotically Normal; estimators for the asymptotic variance are also proposed. Finite-sample performance of each estimator is investigated, and each method is illustrated by analyzing a cholera vaccine study in Matlab, Bangladesh. In the first paper, we propose a method for inverse probability-weighted estimation of target estimands in the presence of partial interference that is more robust to model mis-specification than existing methods. This technique relies on an algorithm which combines machine learning and mixed effects methods to determine the relationship between treatment and covariates assuming a certain correlation structure. We employ the algorithm on a training sample as a data-adaptive model selection procedure. We recover the set of rules that the algorithm uses for prediction to transform the covariates in a testing sample. We proceed by fitting a model to the transformed covariates to estimate propensity scores for IPW estimation of target estimands. We propose a matching technique for estimating causal effects in the presence of partial interference in the second paper. These estimators extend methods that are commonly employed when no interference is assumed. The proposed methods can be carried out without modeling treatment, and may outperform existing IPW estimators in certain scenarios. Extensions of these estimators are discussed. In the third paper we propose new causal estimands for observational studies in the presence of partial interference. The proposed estimands describe counterfactual scenarios in which there may be within-cluster dependence in the individual treatment selections. These estimands may be more relevant for public health officials. These estimands are identifiable from observational data with parametric assumptions. Inverse probability-weighted estimators for these estimands are proposed. Spring 2018 2018 Biostatistics Statistics Epidemiology causal inference, interference, inverse probability weight, matching, observational study, spillover effects eng Doctor of Philosophy Dissertation University of North Carolina at Chapel Hill Graduate School Degree granting institution Biostatistics Michael Hudgens Thesis advisor Maurice Brookhart Thesis advisor Michael Emch Thesis advisor Michael Kosorok Thesis advisor Donglin Zeng Thesis advisor text Brian Barkley Author Department of Biostatistics Gillings School of Global Public Health Causal Inference From Observational Studies in the Presence of Partial Interference Analyzing data to estimate the effect of treatment on health outcomes can play a major role in the fields of personal and public health. Interference occurs when the treatment of one individual affects the outcome of another individual. This work aims to develop statistical methodology for inference about causal effects from observational studies in the presence of interference. We assume partial interference throughout: interference may exist within clusters of individuals, but not between distinct clusters. In each paper we propose estimators that are consistent and asymptotically Normal; estimators for the asymptotic variance are also proposed. Finite-sample performance of each estimator is investigated, and each method is illustrated by analyzing a cholera vaccine study in Matlab, Bangladesh. In the first paper, we propose a method for inverse probability-weighted estimation of target estimands in the presence of partial interference that is more robust to model mis-specification than existing methods. This technique relies on an algorithm which combines machine learning and mixed effects methods to determine the relationship between treatment and covariates assuming a certain correlation structure. We employ the algorithm on a training sample as a data-adaptive model selection procedure. We recover the set of rules that the algorithm uses for prediction to transform the covariates in a testing sample. We proceed by fitting a model to the transformed covariates to estimate propensity scores for IPW estimation of target estimands. We propose a matching technique for estimating causal effects in the presence of partial interference in the second paper. These estimators extend methods that are commonly employed when no interference is assumed. The proposed methods can be carried out without modeling treatment, and may outperform existing IPW estimators in certain scenarios. Extensions of these estimators are discussed. In the third paper we propose new causal estimands for observational studies in the presence of partial interference. The proposed estimands describe counterfactual scenarios in which there may be within-cluster dependence in the individual treatment selections. These estimands may be more relevant for public health officials. These estimands are identifiable from observational data with parametric assumptions. Inverse probability-weighted estimators for these estimands are proposed. Spring 2018 2018 Biostatistics Statistics Epidemiology causal inference, interference, inverse probability weight, matching, observational study, spillover effects eng Doctor of Philosophy Dissertation University of North Carolina at Chapel Hill Graduate School Degree granting institution Biostatistics Michael Hudgens Thesis advisor Maurice Brookhart Thesis advisor Michael Emch Thesis advisor Michael Kosorok Thesis advisor Donglin Zeng Thesis advisor text Brian Barkley Author Department of Biostatistics Gillings School of Global Public Health Causal Inference From Observational Studies in the Presence of Partial Interference Analyzing data to estimate the effect of treatment on health outcomes can play a major role in the fields of personal and public health. Interference occurs when the treatment of one individual affects the outcome of another individual. This work aims to develop statistical methodology for inference about causal effects from observational studies in the presence of interference. We assume partial interference throughout: interference may exist within clusters of individuals, but not between distinct clusters. In each paper we propose estimators that are consistent and asymptotically Normal; estimators for the asymptotic variance are also proposed. Finite-sample performance of each estimator is investigated, and each method is illustrated by analyzing a cholera vaccine study in Matlab, Bangladesh. In the first paper, we propose a method for inverse probability-weighted estimation of target estimands in the presence of partial interference that is more robust to model mis-specification than existing methods. This technique relies on an algorithm which combines machine learning and mixed effects methods to determine the relationship between treatment and covariates assuming a certain correlation structure. We employ the algorithm on a training sample as a data-adaptive model selection procedure. We recover the set of rules that the algorithm uses for prediction to transform the covariates in a testing sample. We proceed by fitting a model to the transformed covariates to estimate propensity scores for IPW estimation of target estimands. We propose a matching technique for estimating causal effects in the presence of partial interference in the second paper. These estimators extend methods that are commonly employed when no interference is assumed. The proposed methods can be carried out without modeling treatment, and may outperform existing IPW estimators in certain scenarios. Extensions of these estimators are discussed. In the third paper we propose new causal estimands for observational studies in the presence of partial interference. The proposed estimands describe counterfactual scenarios in which there may be within-cluster dependence in the individual treatment selections. These estimands may be more relevant for public health officials. These estimands are identifiable from observational data with parametric assumptions. Inverse probability-weighted estimators for these estimands are proposed. Spring 2018 2018 Biostatistics Statistics Epidemiology causal inference, interference, inverse probability weight, matching, observational study, spillover effects eng Doctor of Philosophy Dissertation Biostatistics Michael Hudgens Thesis advisor Maurice Brookhart Thesis advisor Michael Emch Thesis advisor Michael Kosorok Thesis advisor Donglin Zeng Thesis advisor text University of North Carolina at Chapel Hill Degree granting institution Brian Barkley Creator Department of Biostatistics Gillings School of Global Public Health Causal Inference From Observational Studies in the Presence of Partial Interference Analyzing data to estimate the effect of treatment on health outcomes can play a major role in the fields of personal and public health. Interference occurs when the treatment of one individual affects the outcome of another individual. This work aims to develop statistical methodology for inference about causal effects from observational studies in the presence of interference. We assume partial interference throughout: interference may exist within clusters of individuals, but not between distinct clusters. In each paper we propose estimators that are consistent and asymptotically Normal; estimators for the asymptotic variance are also proposed. Finite-sample performance of each estimator is investigated, and each method is illustrated by analyzing a cholera vaccine study in Matlab, Bangladesh. In the first paper, we propose a method for inverse probability-weighted estimation of target estimands in the presence of partial interference that is more robust to model mis-specification than existing methods. This technique relies on an algorithm which combines machine learning and mixed effects methods to determine the relationship between treatment and covariates assuming a certain correlation structure. We employ the algorithm on a training sample as a data-adaptive model selection procedure. We recover the set of rules that the algorithm uses for prediction to transform the covariates in a testing sample. We proceed by fitting a model to the transformed covariates to estimate propensity scores for IPW estimation of target estimands. We propose a matching technique for estimating causal effects in the presence of partial interference in the second paper. These estimators extend methods that are commonly employed when no interference is assumed. The proposed methods can be carried out without modeling treatment, and may outperform existing IPW estimators in certain scenarios. Extensions of these estimators are discussed. In the third paper we propose new causal estimands for observational studies in the presence of partial interference. The proposed estimands describe counterfactual scenarios in which there may be within-cluster dependence in the individual treatment selections. These estimands may be more relevant for public health officials. These estimands are identifiable from observational data with parametric assumptions. Inverse probability-weighted estimators for these estimands are proposed. Biostatistics Statistics Epidemiology causal inference; interference; inverse probability weight; matching; observational study; spillover effects eng Doctor of Philosophy Dissertation Biostatistics Michael Hudgens Thesis advisor Maurice Brookhart Thesis advisor Michael Emch Thesis advisor Michael Kosorok Thesis advisor Donglin Zeng Thesis advisor text University of North Carolina at Chapel Hill Degree granting institution 2018 2018-05 Brian Barkley Author Department of Biostatistics Gillings School of Global Public Health Causal Inference From Observational Studies in the Presence of Partial Interference Analyzing data to estimate the effect of treatment on health outcomes can play a major role in the fields of personal and public health. Interference occurs when the treatment of one individual affects the outcome of another individual. This work aims to develop statistical methodology for inference about causal effects from observational studies in the presence of interference. We assume partial interference throughout: interference may exist within clusters of individuals, but not between distinct clusters. In each paper we propose estimators that are consistent and asymptotically Normal; estimators for the asymptotic variance are also proposed. Finite-sample performance of each estimator is investigated, and each method is illustrated by analyzing a cholera vaccine study in Matlab, Bangladesh. In the first paper, we propose a method for inverse probability-weighted estimation of target estimands in the presence of partial interference that is more robust to model mis-specification than existing methods. This technique relies on an algorithm which combines machine learning and mixed effects methods to determine the relationship between treatment and covariates assuming a certain correlation structure. We employ the algorithm on a training sample as a data-adaptive model selection procedure. We recover the set of rules that the algorithm uses for prediction to transform the covariates in a testing sample. We proceed by fitting a model to the transformed covariates to estimate propensity scores for IPW estimation of target estimands. We propose a matching technique for estimating causal effects in the presence of partial interference in the second paper. These estimators extend methods that are commonly employed when no interference is assumed. The proposed methods can be carried out without modeling treatment, and may outperform existing IPW estimators in certain scenarios. Extensions of these estimators are discussed. In the third paper we propose new causal estimands for observational studies in the presence of partial interference. The proposed estimands describe counterfactual scenarios in which there may be within-cluster dependence in the individual treatment selections. These estimands may be more relevant for public health officials. These estimands are identifiable from observational data with parametric assumptions. Inverse probability-weighted estimators for these estimands are proposed. Spring 2018 2018 Biostatistics Statistics Epidemiology causal inference, interference, inverse probability weight, matching, observational study, spillover effects eng Doctor of Philosophy Dissertation University of North Carolina at Chapel Hill Graduate School Degree granting institution Biostatistics Michael Hudgens Thesis advisor Maurice Brookhart Thesis advisor Michael Emch Thesis advisor Michael Kosorok Thesis advisor Donglin Zeng Thesis advisor text Brian Barkley Creator Department of Biostatistics Gillings School of Global Public Health Analyzing data to estimate the effect of treatment on health outcomes can play a major role in the fields of personal and public health. Interference occurs when the treatment of one individual affects the outcome of another individual. This work aims to develop statistical methodology for inference about causal effects from observational studies in the presence of interference. We assume partial interference throughout: interference may exist within clusters of individuals, but not between distinct clusters. In each paper we propose estimators that are consistent and asymptotically Normal; estimators for the asymptotic variance are also proposed. Finite-sample performance of each estimator is investigated, and each method is illustrated by analyzing a cholera vaccine study in Matlab, Bangladesh. In the first paper, we propose a method for inverse probability-weighted estimation of target estimands in the presence of partial interference that is more robust to model mis-specification than existing methods. This technique relies on an algorithm which combines machine learning and mixed effects methods to determine the relationship between treatment and covariates assuming a certain correlation structure. We employ the algorithm on a training sample as a data-adaptive model selection procedure. We recover the set of rules that the algorithm uses for prediction to transform the covariates in a testing sample. We proceed by fitting a model to the transformed covariates to estimate propensity scores for IPW estimation of target estimands. We propose a matching technique for estimating causal effects in the presence of partial interference in the second paper. These estimators extend methods that are commonly employed when no interference is assumed. The proposed methods can be carried out without modeling treatment, and may outperform existing IPW estimators in certain scenarios. Extensions of these estimators are discussed. In the third paper we propose new causal estimands for observational studies in the presence of partial interference. The proposed estimands describe counterfactual scenarios in which there may be within-cluster dependence in the individual treatment selections. These estimands may be more relevant for public health officials. These estimands are identifiable from observational data with parametric assumptions. Inverse probability-weighted estimators for these estimands are proposed. 2018-05 2018 Biostatistics Statistics Epidemiology causal inference; interference; inverse probability weight; matching; observational study; spillover effects eng Doctor of Philosophy Dissertation University of North Carolina at Chapel Hill Graduate School Degree granting institution Michael Hudgens Thesis advisor Maurice Brookhart Thesis advisor Michael Emch Thesis advisor Michael Kosorok Thesis advisor Donglin Zeng Thesis advisor text Barkley_unc_0153D_17700.pdf uuid:f735f3ca-ecb1-4aa6-a422-9a910671fbff 2020-06-13T00:00:00 2018-04-19T19:08:48Z proquest application/pdf 1267844