As mentioned above, often times social science research begins by defining the population boundaries, then constructs a sampling strategy within that population, followed by measurement design for members of that sample. As may have become apparent in the previous sections, the delineation between these stages is a bit more blurred for social networks research. Sometimes the way a study goes about measuring relationships in turn becomes the means of sampling—e.g., for a partial network design, the link-tracing only happens once some initial number of seeds have reported information about their alters.
It is for these reasons that I have left what is labeled in social networks research as the “boundary specification problem” for last—because in practice it requires a combination of sampling and measurement considerations described above. The primary difference between boundary specification in social networks research—as compared to general questions of bounding the sample within the study population of interest in other research designs—comes from the fact that social networks research must simultaneously determine which nodes and relationships fall within the study’s boundaries. As was mentioned above with the example of adolescents’ friendships as measured from a school based sample, sometimes a boundary defined on the node level provides a similarly useful boundary for relationships (e.g., students in “Jefferson” HS), while others the relationships of interest extend beyond the node-based population boundary. As such, the extreme “ideal type” strategies for sampling within network studies—ego and complete—conceptually bound these possibilities, but reality often requires significant adaptations from these precise forms.
Specifically, boundary specification problem (BSP) in social network research is the notion that how we determine study inclusion must simultaneously address which people (or nodes) and which relationships (or ties) among those nodes will constitute the population of interest. (Laumann, Marsden, and Prensky 1994). Researchers have developed a variety of strategies for resolving the BSP. A “realist” approach allows members within the study population to inductively determine which nodes and/or ties from within a population should be included (or excluded48) when enumerating the network. Conversely, a “nominalist” approach begins with some pre-determined identification of the population boundary–e.g., via a membership roster.49
It is important to highlight two additional details about either of these potential solutions to the BSP. First, these alternatives only identify which ties could potentially be identified and retained within the study design. It says nothing about which relationships will actually be present and reported on. Second, as with the delineation between ego, partial, and complete network designs, both nominalist and realist strategies may require some adaptation in practice.50 There is no single set of “best practices” for BSP solutions that apply across all study designs. Instead, the selection between these, and how best to adapt them should draw on a combination of the design principles described above, the empirical details of the case to be studied, and the theoretical aims for the study to be conducted.