ABSTRACT
This chapter introduces the general themes that motivate the methods used for carrying out meta-analysis and provides a map with which the reader can navigate the rest of the book. Meta-analyses use one of two types of data structures. Data collected at the study or arm-level may be reported as summaries (often means or counts) of outcomes and other study characteristics, often by treatment group, and can address questions relating to variation between studies. Individual participant data include information on each study participant and so can address within-study variation as well. Statistical models can be formulated to analyze study-level summaries (two-stage models) or to analyze data at an individual level (one-stage models). Models can also be characterized by whether they allow treatment effects to vary randomly between studies (random-effects models) or assume that all studies estimate the same effect (common-effect models). A third distinction relates to whether the effects in the reference (or control) group in each study derive from a common distribution or are fixed parameters on which the analysis is conditioned. Models may be fit in both frequentist and Bayesian frameworks and should always be checked to verify their assumptions.
