On the Identification of Individual Level Interactive, Direct, and Always-Survivor Causal Effects
Access StatusFull text of the requested work is not available in DASH at this time ("dark deposit"). For more information on dark deposits, see our FAQ.
MetadataShow full item record
CitationZaidi, Jaffer. 2019. On the Identification of Individual Level Interactive, Direct, and Always-Survivor Causal Effects. Doctoral dissertation, Harvard University, Graduate School of Arts & Sciences.
AbstractMost effects of treatments of interventions are evaluated at the population level. This dissertation instead examines whether it is possible to first detect the effect of treatments at the individual level, and also characterizes the type of individual level effect. As a consequence of this work, individual level interactive effects, direct effects and always survivor casual effects are defined, and empirical conditions to detect such individual level effects are derived under baseline randomization. We also demonstrate that should these empirical conditions hold, then proportion of individuals that display the relevant individual level interactive, direct, or always survivor causal effect must be greater than zero and greater than other combinations of individual level effects.
In the first paper, tests for sufficient cause interaction for ordinal outcomes are developed to give understanding of mechanism of a treatment or intervention. We also illustrate the usefulness of our proposed methodology from publicly available data. In our second paper, we develop results for individual level natural and principal stratum direct effects and show how the birth-weight paradox dissolves using our conditions, where it otherwise appears as though low birthweight babies will have a lower risk of infant mortality should their mothers smoke than low-birthweight babies whose mothers who do not smoke.
Finally, in the third paper, we establish a new methodology that enables to scientists to detect whether there exist individuals in the population would survive until a given time period regardless of treatment and whether, for these individuals, treatment alters some other outcome. We use this method on data from prostate cancer to illustrate that our method detects whether a new treatment prevents an individual’s cancer from progressing. I hope that the material contained herein proves useful for scientists, statisticians, epidemiologists, and researchers in different disciplines that use data to draw conclusions about our shared world.
Citable link to this pagehttp://nrs.harvard.edu/urn-3:HUL.InstRepos:42106935
- FAS Theses and Dissertations