2.2.3 Partial Network Designs

Between the extremes of ego and complete network designs exists a strategy that blends many of the design principles from each strategy, and incorporates practices that were initially developed for tracing infectious disease outbreaks (Klovdahl 1985). Partial network designs are particularly useful for settings where researchers do not have access to a reliable list of the members of a population, but would like to approximate the aims of complete network studies (Morris 2004).

In practice, partial network studies resemble elements of a snow ball sampling design. The researcher begins with a set of initial seeds from whom they gather information about their own personal network, incorporating most of the elements of an ego network design as described above. However, in addition to asking about the characteristics of an ego’s alters,46 partial network designs then develop a strategy to recruit some portion of those named alters as respondents themselves into the study. Where partial network studies especially differ from snow ball sampling, is in the more systematic nature of determining which alters to target for recruitment, and determination of how many “waves” of link tracing will be completed in a study (Woodhouse et al. 1994). If the partial network study is merely a sampling strategy intended to capture the entire (unknown) population of interest, these waves will be repeated until new nominations no longer identify anyone new. In this case, the “partial” nomenclature should be interpreted as indicating the researcher’s prior knowledge of the population of interest, and sampling was therefore adapted in a way that allowed for saturation.47

However, it is difficult to design a partial network study with certainty of reaching that saturation point; see for example the number of seeds that would be needed to contact all of the nodes in Figure 2.4. If we suppose that image represented our complete population of interest, we would require at least 4 seeds, more if any of them come from the same component. For this reason, many partial network designs operate with a rolling seed enrollment procedure. This can be especially useful when researchers gain information as the study progresses that that can potentially identify elements of the population that their existing link-tracing sample would not allow them to tap (adams et al. 2012). As a result, many partial network designs, even upon completion, should be thought of as capturing only a partial representation of the population (of nodes and ties) of interest.