서브메뉴
검색
Unified Analysis of Two-Stage Experiments From the Design-Based Perspective
Unified Analysis of Two-Stage Experiments From the Design-Based Perspective
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202103554
- ISBN
- 9798288863745
- DDC
- 574
- 서명/저자
- Unified Analysis of Two-Stage Experiments From the Design-Based Perspective
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 87 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-01, Section: B.
- 주기사항
- Advisor: Ding, Peng.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약In this thesis, we introduce a comprehensive class of two-stage experiments that not only encapsulates most standard designs but also offers flexibility to accommodate a myriad of others, particularly those with clustered structures. Despite the proliferation of diverse experiments, their effective analysis often remains a challenge.A prevailing approach is to apply the ordinary least squares (OLS) regression of the outcome on the treatment indicators, supplemented by robust standard errors clustered at the appropriate level. However, this approach fails to deliver consistent estimators for treatment effects, especially when the in cluster sizes vary. Another common strategy is to fit a linear mixed-effects model, embedding normal random effects and errors. However, this model-based method makes strong parametric assumptions that may often be violated in practice.Central to our approach is a design-based methodology. By sidestepping modeling assumptions, we accept a trade-off: while unbiased estimators for the variance of our contrasts may be impossible without extra assumptions, we can consistently achieve conservative estimators, even if the sizes of clusters vary. Our inference is directly grounded in the experiment's controlled randomization mechanism.We discuss design-based inference for both the Horvitz-Thompson estimator and the Hajek estimator and show how we can not only recover both of them from linear models, but also get asymptotically conservative standard errors using the associated cluster-robust covariances estimates. We then extend the results to accommodate covariate adjustments. This approach serves as a unifying framework, drawing together various works that have previously proposed methods to derive conservative estimators.Finally, we show how to recover traditional conservative estimators for known designs from our method but also how to derive conservative estimators for other designs for which the existing literature has not provided guidance yet. We also include some examples of non-standard designs that are inspired by real-world data and show that they fit our framework.
- 일반주제명
- Biostatistics
- 일반주제명
- Statistics
- 일반주제명
- Applied mathematics
- 키워드
- Causal inference
- 키워드
- Linear models
- 기타저자
- University of California, Berkeley Biostatistics
- 기본자료저록
- Dissertations Abstracts International. 87-01B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017357746
■00520260202103554
■006m o d
■007cr#unu||||||||
■020 ▼a9798288863745
■035 ▼a(MiAaPQ)AAI32041999
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a574
■1001 ▼aBenac, Kevin Kiane Jacques.
■24510▼aUnified Analysis of Two-Stage Experiments From the Design-Based Perspective
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a87 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-01, Section: B.
■500 ▼aAdvisor: Ding, Peng.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aIn this thesis, we introduce a comprehensive class of two-stage experiments that not only encapsulates most standard designs but also offers flexibility to accommodate a myriad of others, particularly those with clustered structures. Despite the proliferation of diverse experiments, their effective analysis often remains a challenge.A prevailing approach is to apply the ordinary least squares (OLS) regression of the outcome on the treatment indicators, supplemented by robust standard errors clustered at the appropriate level. However, this approach fails to deliver consistent estimators for treatment effects, especially when the in cluster sizes vary. Another common strategy is to fit a linear mixed-effects model, embedding normal random effects and errors. However, this model-based method makes strong parametric assumptions that may often be violated in practice.Central to our approach is a design-based methodology. By sidestepping modeling assumptions, we accept a trade-off: while unbiased estimators for the variance of our contrasts may be impossible without extra assumptions, we can consistently achieve conservative estimators, even if the sizes of clusters vary. Our inference is directly grounded in the experiment's controlled randomization mechanism.We discuss design-based inference for both the Horvitz-Thompson estimator and the Hajek estimator and show how we can not only recover both of them from linear models, but also get asymptotically conservative standard errors using the associated cluster-robust covariances estimates. We then extend the results to accommodate covariate adjustments. This approach serves as a unifying framework, drawing together various works that have previously proposed methods to derive conservative estimators.Finally, we show how to recover traditional conservative estimators for known designs from our method but also how to derive conservative estimators for other designs for which the existing literature has not provided guidance yet. We also include some examples of non-standard designs that are inspired by real-world data and show that they fit our framework.
■590 ▼aSchool code: 0028.
■650 4▼aBiostatistics
■650 4▼aStatistics
■650 4▼aApplied mathematics
■653 ▼aCausal inference
■653 ▼aDesign-based inference
■653 ▼aExperimental design
■653 ▼aLinear models
■690 ▼a0308
■690 ▼a0463
■690 ▼a0364
■71020▼aUniversity of California, Berkeley▼bBiostatistics.
■7730 ▼tDissertations Abstracts International▼g87-01B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17357746▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


