서브메뉴
검색
Experimental Design Methods for Treatment Effect Estimation Under Constraints
Experimental Design Methods for Treatment Effect Estimation Under Constraints
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104746
- ISBN
- 9798290652559
- DDC
- 790
- 서명/저자
- Experimental Design Methods for Treatment Effect Estimation Under Constraints
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 125 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-01, Section: A.
- 주기사항
- Advisor: Owen, Art.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약Across tech platforms, medical trials, and a host of other settings, causal experiments are a pop-ular procedure to obtain "gold standard" estimates of treatment effects. Despite their ostensible simplicity, such experiments can present a host of challenges. For instance, there may be ethical or economic constraints on who can receive treatment, there may be non-compliance issues that render direct treatment assignment impossible, and there may be a desire to maintain constant marginal treatment probabilities for fairness and interpretability.Moreover, experiments are quite resource intensive. The process of collecting and tracking study participants can be time-consuming and expensive, and many companies employ whole teams of statisticians who are responsible for design implementation. Hence, it is beneficial to ensure that each experiment is as powerful as possible and results in a final treatment effect estimate with low variance. To that end, this thesis studies three problems in experimental design that each involve deriving a low-variance estimate of a treatment effect under constraints.1.1 Multivariate tie-breaker designsIn Chapter 2 we introduce multivariate tie-breaker designs, which are inspired by the univariate work of Owen and Varian (2020). In a tie-breaker design (TBD), subjects with high values of a running variable are given some (usually desirable) treatment, subjects with low values are not, and subjects in the middle are randomized. TBDs are intermediate between regression discontinuity designs (RDDs) and randomized controlled trials (RCTs). TBDs allow a tradeoff between the resource allocation efficiency of an RDD and the statistical efficiency of an RCT.We study a model where the expected response is one multivariate regression for treated subjects and another for control subjects. We propose a prospective D-optimality, analogous to Bayesian op-timal design, to understand design tradeoffs without reference to a specific data set. For given covariates, we show how to use convex optimization to choose treatment probabilities that optimize this criterion. We can incorporate a variety of constraints motivated by economic and ethical con-siderations. In our model, D-optimality for the treatment effect coincides with D-optimality for the whole regression, and, without constraints, an RCT is globally optimal.We show that a monotonicity constraint favoring more deserving subjects induces sparsity in the number of distinct treatment probabilities. We apply the convex optimization solution to a semi-synthetic example involving triage data from the MIMIC-IV-ED database.1.2 Constrained design of a binary instrument in a partially linear modelIn Chapter 3 we study the question of how best to assign an encouragement in a randomized encouragement study. In our setting, units arrive with covariates, receive a nudge toward treatment or control, acquire one of those statuses in a way that need not align with the nudge, and finally have a response observed. The nudge can be modeled as a binary instrument if one assumes that it affects the response only via the treatment status. Our goal is to assign the nudge as a function of covariates in a way that best estimates the local average treatment effect (LATE).We assume a partially linear model, wherein the baseline model is non-parametric and the treat-ment term is linear in the covariates.
- 일반주제명
- Design optimization
- 일반주제명
- Convex analysis
- 일반주제명
- Ethics
- 일반주제명
- Sample variance
- 일반주제명
- Blood pressure
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-01A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017358748
■00520260202104746
■006m o d
■007cr#unu||||||||
■020 ▼a9798290652559
■035 ▼a(MiAaPQ)AAI32149754
■035 ▼a(MiAaPQ)Stanfordym492tn8726
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a790
■1001 ▼aMorrison, Timothy.
■24510▼aExperimental Design Methods for Treatment Effect Estimation Under Constraints
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a125 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-01, Section: A.
■500 ▼aAdvisor: Owen, Art.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aAcross tech platforms, medical trials, and a host of other settings, causal experiments are a pop-ular procedure to obtain "gold standard" estimates of treatment effects. Despite their ostensible simplicity, such experiments can present a host of challenges. For instance, there may be ethical or economic constraints on who can receive treatment, there may be non-compliance issues that render direct treatment assignment impossible, and there may be a desire to maintain constant marginal treatment probabilities for fairness and interpretability.Moreover, experiments are quite resource intensive. The process of collecting and tracking study participants can be time-consuming and expensive, and many companies employ whole teams of statisticians who are responsible for design implementation. Hence, it is beneficial to ensure that each experiment is as powerful as possible and results in a final treatment effect estimate with low variance. To that end, this thesis studies three problems in experimental design that each involve deriving a low-variance estimate of a treatment effect under constraints.1.1 Multivariate tie-breaker designsIn Chapter 2 we introduce multivariate tie-breaker designs, which are inspired by the univariate work of Owen and Varian (2020). In a tie-breaker design (TBD), subjects with high values of a running variable are given some (usually desirable) treatment, subjects with low values are not, and subjects in the middle are randomized. TBDs are intermediate between regression discontinuity designs (RDDs) and randomized controlled trials (RCTs). TBDs allow a tradeoff between the resource allocation efficiency of an RDD and the statistical efficiency of an RCT.We study a model where the expected response is one multivariate regression for treated subjects and another for control subjects. We propose a prospective D-optimality, analogous to Bayesian op-timal design, to understand design tradeoffs without reference to a specific data set. For given covariates, we show how to use convex optimization to choose treatment probabilities that optimize this criterion. We can incorporate a variety of constraints motivated by economic and ethical con-siderations. In our model, D-optimality for the treatment effect coincides with D-optimality for the whole regression, and, without constraints, an RCT is globally optimal.We show that a monotonicity constraint favoring more deserving subjects induces sparsity in the number of distinct treatment probabilities. We apply the convex optimization solution to a semi-synthetic example involving triage data from the MIMIC-IV-ED database.1.2 Constrained design of a binary instrument in a partially linear modelIn Chapter 3 we study the question of how best to assign an encouragement in a randomized encouragement study. In our setting, units arrive with covariates, receive a nudge toward treatment or control, acquire one of those statuses in a way that need not align with the nudge, and finally have a response observed. The nudge can be modeled as a binary instrument if one assumes that it affects the response only via the treatment status. Our goal is to assign the nudge as a function of covariates in a way that best estimates the local average treatment effect (LATE).We assume a partially linear model, wherein the baseline model is non-parametric and the treat-ment term is linear in the covariates.
■590 ▼aSchool code: 0212.
■650 4▼aDesign optimization
■650 4▼aConvex analysis
■650 4▼aEthics
■650 4▼aSample variance
■650 4▼aBlood pressure
■690 ▼a0394
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-01A.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358748▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


