Whereas sampling of individual observations within a single study is characterized by more than a century of research and development, sampling of studies into an IDA application has received virtually no attention. Nonetheless, we can draw on two well-developed methods commonly used in single-study designs to allow us to explicitly incorporate a model-based approach to sampling in IDA. First, from the field of multilevel modeling (MLM; Raudenbush & Bryk, 2002) we can use a two-tiered sampling method to define a random-effects IDA. More specifically, within the MLM it is possible to first randomly sample a finite number of groups from a population of groups (e.g., schools within a district, or hospitals within a county) and then randomly sample a finite number of individuals from a population of individuals who are nested within each group (e.g., students within a school, or patients within a hospital). This two-tiered sampling framework allows for the estimation of random components at both the level of the group (Level 2: contributing study) and of the individual (Level 1: individual observation within study). However, a key practical
