Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Estimation of Causal Treatment Effect for Clustered Observational Data with Unmeasured Confounding

Ou, Yang (2023) Estimation of Causal Treatment Effect for Clustered Observational Data with Unmeasured Confounding. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img] PDF
Restricted to University of Pittsburgh users only until 15 May 2025.

Download (735kB) | Request a Copy


The identification of causal average treatment effects (ATE) in observational studies requires data to satisfy the unconfoundness assumption which is not testable using the observed data. Violation of this assumption can undermine the accuracy of statistical conclusions derived under this assumption. Sensitivity analysis and instrumental variables (IV) methods are often used to address the issue of unmeasured confounding. In this dissertation, I proposed a novel sensitivity analysis and IV techniques for clustered observational data with binary or normally distributed outcomes. I further extended the newly proposed sensitivity analysis techniques to meta-analysis.

The first part of my dissertation describes the methods I developed to assess the robustness of confounded treatment effects in the presence of unmeasured confounders in clustered observational data. The proposed methods relax the restrictions imposed by sensitivity analyses in other settings. Assuming that the true treatment effect can be represented in a mixed model, our methods provide estimates of the minimal effect of unmeasured confounders that can explain away the confounded treatment effects. I further extended this approach to address the robustness of confounded treatment effects in multiple studies. Simulation results demonstrated the accuracy of the proposed methods. I applied the proposed methods to two real-world data.

In the second part, I proposed methods to estimate ATE in the presence of unmeasured confounders using IVs. The maximum likelihood estimator and a doubly robust estimator for ATE that I proposed can handle clustered data under the standard IV assumptions and additional identification assumptions. I also incorporated a newly proposed nuisance model to ensure that the estimator of ATE for binary outcomes falls between -1 and 1. Simulations were conducted to examine the performance of these methods. I applied the proposed methods to data from the National Longitudinal Study of Young Men.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Ou, Yangyao11@pitt.eduyao11
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairChang, Chung-Chou H.changj@pitt.educhangj
Committee MemberTang, Lulutang@pitt.edulutang
Committee MemberWang, Jiebiaojbwang@pitt.edujbwang
Committee MemberTalisa, Victorvit13@pitt.eduvit13
Date: 15 May 2023
Date Type: Publication
Defense Date: 6 April 2023
Approval Date: 15 May 2023
Submission Date: 25 April 2023
Access Restriction: 2 year -- Restrict access to University of Pittsburgh for a period of 2 years.
Number of Pages: 90
Institution: University of Pittsburgh
Schools and Programs: School of Public Health > Biostatistics
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Average treatment effect; Clustered observational data; Instrumental variables; Meta-analysis; Mixed model; Nuisance model; Sensitivity analysis; Unmeasured confounding
Date Deposited: 15 May 2023 14:17
Last Modified: 15 May 2023 14:17


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item