Leadership Networks

In the preceding sections of the paper we have demonstrated how SNA can be used to analyze four types of leadership networks. The use of SNA is not, however, without risks. Careful consideration needs to be given to these issues by anyone who uses SNA as an evaluation tool. We highlight four of them here:

• Lack of privacy and related ethical issues
• Making evaluations from incomplete data
• Oversimplification and misreading
• Misuse of network measures

Our categorization of issues and risks is similar to that of Bender-deMoll (2008). Below we briefly elaborate on each category. For each one, we emphasize its implications for leadership networks and compare our perspective to Bender-deMoll's.

Lack of privacy and related ethical issues

Borgatti and Molina (2005) discuss ethical guidelines for using SNA to evaluate leadership networks. We follow their guidelines in our own work and devote special attention to privacy. In the table below, we highlight three distinct ways that network surveys lack privacy compared to traditional surveys:Here we focus our discussion primarily on the third issue, visibility; then we briefly remark on the other two issues. We include Figure 11 to provide context. It shows two maps: the advice and trust networks of a single organization studied by Krackhardt and Hanson (1993). Consider the advice network (a). Someone like Swinney (far left) might prefer that others not notice that his advice is sought by no one in the network, while Calder (center), who is perhaps overwhelmed by the number of people seeking his advice, might wish to be invisible so that others would not seek him out.Two factors exacerbate the risk of exposing people like Swinney and Calder. First, they have no way of previewing what others have reported about them before those reports are published (a matter we will revisit in the "Future Research" section). Second, they may assume-incorrectly-that not responding will keep them out of the survey results. For example, suppose Calder chooses not to participate in the next survey; then those results will not show that Calder seeks advice from Leers and Harris, but they will still show who reports that they go to Calder for advice (and there will probably still be many such people). Calder's ability to remove himself from the network map depends on the survey administrator, who must be clear that "opting out" and "not participating" are two entirely different things.

The above risks faced by participants in a network survey can be mitigated with the following steps. The first step is to educate people about the value of network data, as it benefits both each individual and the network as a whole. The second step is to explain clearly who will see the network data and what will be done with the data. The third step is to design the survey to be consistent with its intended use. For example, asking "whom do you trust"-as mapped in Figure 11 (b)-would probably be counter-productive if the survey results were to be shared openly with network members, but would be extremely valuable if the survey results were shown only to a trusted advisor who is not herself in the map.

The overall goal of the above three steps is to provide network members the ability to exercise informed consent. Clarity and transparency increase participation in the survey and acceptance of the results. Figure 12 shows how we put these steps into practice; we introduce a network survey with language similar to the following:There are also steps that can be taken to mitigate the other two privacy risks of network maps listed in Figure 10. The specificity of network survey results can be masked so that individuals' names cannot be inferred from the presented maps. This approach is quite practical when results are presented as an anonymized case study (i.e., the audience does not know what specific network is being displayed); however, this kind of network anonymity is extremely difficult to insure when the results are shared with the network members themselves.

Finally, we consider that each respondent to a network survey is asked to report information about others by name, rather than reporting information purely about himself. When trust among network members is in doubt, any question designed in this way can be difficult to ask. In such a situation, we recommend survey questions that elicit purely first-person information. The resulting data can then be used to create a network map of the group based on structural equivalence (as in Figures 4 and 9).

Making evaluations from incomplete data

Network survey results are much more sensitive to data omissions than other kinds of surveys. In order to produce a network map that provides network members with accurate pictures of bridging and bonding, a survey response rate of at least 75% is typically required (Borgatti et al, 2006). Smaller population samples can be surveyed in some situations, but evaluators usually cannot assess a large network by surveying small randomized samples in the same way they can with traditional non-network surveys.

Oversimplification and misreading

We caution people who use network maps to look for multiple interpretations of the data. The work of McGrath and Blythe (2004) illustrates why. They showed subjects the two organizational advice networks in Figure 13 and asked, "All other things being equal, which organization is more adept at change?"

Responses were mixed: some thought the less hierarchical left group (A) would be better at change, because of the wealth of informal connections. Others thought the more hierarchical right group (B) would be better at change, because of the influence of the central authority figure. Very few came up with the correct answer: that networks (A) and (B) are identical.

We agree with Bender-deMoll (2008): "Viewers are not used to thinking critically about network images. Like any statistical graphics, they can be easily manipulated to convey a viewpoint that would not hold up well to rigorous analysis."

One helpful rule of thumb is to rely on network maps more for raising questions than for answering them. For example, it is easy to jump to negative conclusions about peripheral members of a network, such as Swinney in Figure 11 (a). It is important to withhold premature judgment and instead ask: Why is Swinney at the periphery of the map? Possible answers include: Swinney is new; he is disengaged; or he is a vital source of expertise and innovation who bridges to a group not drawn on the map. Network data has the potential to be misused if it is not presented and discussed by skilled analysts who encourage critical thinking.

Misuse of network measures

Some network metrics are prone to misuse. One of the most common mistakes we observe in the field of SNA is the misuse of density, which is a seemingly intuitive metric that is in fact very easily misinterpreted. Density is especially prone to misinterpretation when comparing networks of different sizes. For example, the three networks of Figure 14 all have exactly the same density, even though the maps indicate how connectivity differs significantly between them. We recommend links per node as a measure of network connectivity that behaves much more intuitively than density.Anderson et al (1999) explain that many network metrics, in addition to density, interact "powerfully and subtly" with network size. Leadership networks are often changing in size or being compared to other networks of different sizes. Therefore, it is critically important that practitioners account for the interaction of network size with other network measures.

Bender-deMoll (2008) emphasizes another misuse of network measures: applying a measure designed for one kind of network to a set of data involving a different kind of network. For example, centrality means something different in an affiliation network than it does in a communication network.

< Prev		Next >