One advantage when evaluating social network data (as compared to data focused on the individual level), is that by its very nature, we can obtain first-hand reporting from multiple sources.100 Whereas only one person can self identify their own demographic characteristics or attitudes,101 relationships necessarily involve at least two people. Therefore, each relationship has the potential to be reported on by both members of that relationship. And the comparison of those reports to one another has provided fertile ground for social scientists to assess the quality of the data they have the ability to obtain.
So, to ask the question simply—when two people are reporting about the presence (or absence) of a relationship between themselves, how often do they agree? That is, in Figure 5.2, if respondent R says they have connection C1 with alter A1, and data are available from A1 as a respondent as well, how often are these 2 reports on the same relationship’s state concordant? As you might expect, the answer is not so simple, and often varies across the type of relationship that is being examined. As a general trend those relationships that are more personally salient are more reliably agreed upon by both parties than those that are less so (e.g., sexual relationships are consistently matched in their reports between pairs more often than are friendships, than are acquaintances, etc. (see e.g., Helleringer et al. 2011).
The illustrative example in Figure 5.2 conveys several things about potential data and comparison availability. As described in Chapter 2 respondent R may be asked to report on their contacts (labeled ‘C’) to their own alters (A). In name interpreters, they may also report on the partnerships (P) among their alters, and even the indirect connections (I) they have to other nodes (N) that their alters (A) are connected to but they are not themselves. These labels are only meaningful from the perspective of the selected respondent. If the node labeled A1 is also among the respondents in the sample,102 the labels would have different alignments from their perspective; and are only applied for clarity of explaining examples here. Importantly though, for the undirected relationships represented in this hypothetical example, if we used the methods described for ego networks to gather data only from any one node in this image, and generated error-free data, any one of them could produce the same information about all of the relationships represented in this sub-graph.
Suppose I want to assess the communication network within an organization. I could construct a set of name generators that ask respondents to report on a variety of types of interactions with other members of that organization. The multiplex network in Figure 5.2 portrays two potential relationships simultaneously among the same nodeset (represented by solid or dashed lines). As an example of a study of this sort, examining performance within a school of management, researchers asked respondents to enumerate their digital chat, face-to-face, and phone interactions (Grippa et al. 2006).103 In this scenario, we could reasonably assess how readily one person’s account of their chat partners matched up to the reports of those partners.104 Moreover, we could evaluate whether agreement differed by mode of chat—e.g., in Figure 5.2 the dashed lines could represent digital communication compared to the solid lines, face-to-face interactions).
|Reporting Type||Digital Chat||F2F Chat||Phone|
NOTE: ‘Y’ denotes tie presence, `N’ tie absence, according to the row-defined reporting rule. () E-mail is a meaningfully directed* relationship, and union/intersection would erase that directionality.
As an example, imagine that in our hypothetical study of organizational communication, that Figure 5.2 represents the ground truth of actual relationships among this group of people. Suppose that person A1 reported that yesterday they talked to person A2 in person (solid line) and via digital chat (dashed line), but conversely on the corresponding survey questions, person A2 reported that they talked to Figure 5.2, because the phone lines were actually down yesterday in this hypothetical example). These reports are summarized in Table 5.2. In this case, A1 and A2 only agree that they spoke face-to-face yesterday (typically referred to as the intersection of their reports), while disagreeing on whether they spoke on the phone or via digital chat.105 For the moment, I will only note these potential discrepancies, and describe studies that have estimated the prevalence of such disagreements. In later sections of this chapter, I turn to strategies for dealing with discrepant reports. But, in principle we could code ties as present only when both parties agree it is (intersection), or as long as at least one of them does (union set).106 Unfortunately, if you compare the results in Table 5.2 from any of the potential reporting or data cleaning strategies applied in this contrived example to the hypothetical ground truth in Figure 5.2, you’ll note that none of these strategies would perfectly match up to the actual ties we intended to capture.107