Department of Biostatistics

People

Prospective Students

Academics

Research

News & Events

Calendar

Consulting

Employment Opportunities

Resource Quicklinks

Computing Environment

Contact

THESIS DEFENSE ABSTRACT

Statistical Designs and Analyses for Partially Controlled Studies

 Fan Li,  PhD Candidate, Johns Hopkins Department of Biostatistics

Recently interest has been increasing for a broad class of studies where investigators do not directly control the mechanism of assignment of the treatment, but do control certain factors that affect the assignment of the treatment. The theme of this dissertation is to develop statistical designs and analyses for causal inference for such "partially controlled studies" in two parts. In the first part of the dissertation, we start with developing methodology to evaluate a partially controlled longitudinal treatment with a repeated outcome using principal stratification, and we apply our method to evaluate the Baltimore Needle Exchange Program (NEP). Motivated by a conflict often arising from parameter estimation in partially controlled studies, we then propose a general new design/analysis method, "polydesign", that combines inferences from different designs. Such "polydesign" methods can both identify the parameter of interest and achieve better robustness. As emphasized in the work of "polydesigns", a good design is essential of reliable data analysis. For this reason, we also provide guidelines for developing designs that can obtain larger treatment benefit and more accurate evaluation for location-controlled follow-up studies. In the second part of the dissertation,  we investigate the methods for linkage analysis using affected-sib-pair studies with covariates. We show that, when the covariates have genetic determinants, existing methods can declare linkage between the trait and marker loci even when no such linkage exists. By formulating the problem from the perspective of a partially controlled study, we show the need for  and propose an explicit causal framework for identifying the trait locus and the direct causal effect of a gene on the trait, when the covariates also have genetic determinants. We then indicate strategies to address the problem.