In a complete network design, much of the information that researchers want to gather can potentially come from multiple sources. At the simplest level, any reported network tie involves—at minimum—two people. Therefore, if a census of the population is attained, and each respondent reports on each of their relationships, this would necessitate each member of a tie as having the opportunity to potentially report on it.41 Moreover, if complete network research designs followed all of the steps described above for ego network studies—particularly all three types of name interpreter questions—respondents could be asked to provide information that is more reliably collected from other sources. Most notably, gathering information about one’s alters’ characteristics and reporting on relationships among one’s alters (i.e., those not including the respondent), is much less commonly done in complete network designs, because it’s better to get such information from primary sources. In an ego network design, the primary reason to ask respondents about their alters’ characteristics is that information may not be available from any other sources. However, given the option to ask someone about their own characteristics or gather that information from people they’re connected to, in almost all cases the self-reports would be preferable.42 To reduce the complexity of the survey itself, and to rely on the more reliable sources of available information, complete network designs often therefore eliminate questions that would provide information redundantly available from primary sources.
I am not suggesting that complete network designs remove the efforts to gather any information that is potentially redundant. There are quite a few reasons that such information could be useful to the aims of a research study. Perceived features of one’s network (e.g., how many people hold a particular opinion) can be as influential in developing one’s own opinion as may be that person’s friends’ actual opinions (Young and Rees 2013; McCarty et al. 2019). As a result, some studies have aimed to directly compare these explanatory pathways (Acock and Bengtson 1980).
In the case of a complete network design, relational data not only could be reported from multiple sources; the nature of the format frequently necessitates multiple opportunities for a tie to be reported on—at minimum by both members of that relationship. For example, suppose the study gathers information about friendship ties, and we conduct a survey to elicit that information. If we are interested in evaluating the relationship between Frederick and Tim, numerous possibilities exist for what their respective reports might entail, and how that information should best be used in the study. At the simplest level, their respective reports could agree, either in describing the presence or absence of the relationship in the same way. It’s more complex to resolve when their reports disagree. Suppose Frederick names Tim as a friend, but Tim does not reciprocate that nomination. This could arise because of genuinely discrepant views of the relationship. It also could arise because Tim has more friends than Frederick, and simply doesn’t go far enough down his list to include Frederick among those he named (Aureo, Richards-Shubik, and Tamer 2018). Alternatively, they could both name each other as friends, but then disagree about the closeness or frequency of interaction within their relationship (Yamazaki, Strobino, and Ellen 2010).
These examples raise the question of how to treat such discrepant reports. Chapter 5 will dive more deeply into strategies forevaluating discrepant reports, and using those analyses for developing recommended practices. For now, it is sufficient to describe the general approaches available and most common practices. The intersection approach counts as “valid” only those reports that are corroborated by both members of the relationship. This strategy says that if Tim and Frederick do not both name each other as friends, we treat the tie as not present. The union approach counts a relationship as present if either of the members reports it as present.43 Studies employ different alternatives from these strategies for a variety of reasons, often dependent upon how the type of tie under study (see Table 1.2) is related to expected reporting patterns and discrepancies. Those differences noted, the union strategy is much more broadly used in network studies relying on self reports (Brewer 2000, but see [@LEE201855]).