Statistical methods for evidence synthesis

In many empirical disciplines, scientific discovery is modularized into discrete papers each investigating one or more hypotheses. Synthesizing these modules of evidence is critical to inform a balanced and appropriately evolving view of the overall evidence on a topic as well as to identify where substantial uncertainty remains. This dissertation considers three realms in which such synthesis can occur: (1) when meta-analyzing multiple studies; (2) when subjecting a single study to independent replications; and (3) when testing related hypotheses within a study. We consider specific methodological challenges within each of these realms and propose statistical methods to address each. All proposed methods are implemented in R packages.