The goal of network analysis is to discover relationships (links) among individuals or organizations (nodes). A network can be examined at one point in time, illuminating the current status of these relationships, or at consecutive points in time, highlighting changes in the network. Networks can be social/interpersonal, organizational, or both. Community networks often contain both kinds of relationships as individuals interact concurrently as representatives of organizations and as purveyors of personal interests.
A typical network evaluation commences with the collection of data from the target population, most likely in the form of a survey or a structured interview where respondents rate their connections with other respondents. Responses can be dichotomous (connection/no connection) or ranked on a Likert-type scale of connection strength. Although a variety of methodological and technical tools exist for portraying network results, they usually looks something like this:
The above image has a couple of advantages: the relative density of the network is easy to see at first glance, and the centrality of specific participants is mostly evident. Centrality refers to a person or organizations’ status within the network. Respondents with more connections are placed at the center of the network, while those with fewer connections fall to the outside. In some analyses, participants and/or organizations cluster into dyads or triads, revealing groupings of relationships. Both centrality and clusters can lead to revelations of power in networks – powerful “nodes” have more connections and powerful clusters have more members. These data are actionable, in that collaborative communities can identify less-connected and/or weaker members and work to pull them in (assuming confidentiality is not an issue). However, the above data representation also has limitations that make it not actionable; as the sample size increases, the network becomes increasingly difficult to view and it is hard to see changes over time—side by side network graphs can make even very young evaluators reach for their glasses! In addition, the network links are binomial in nature and do not show connection strength. Layers of variables are hard to integrate, for example, the image cannot show categorizations of affiliations or community sectors, whether the “node” is an individual or an organization, or if a tie is personal or professional.
Fortunately, selected methods of network graphing can make connection data more actionable. Consider the following image, a representation of the same data presented above:
This image places network respondents around the edge of a circle and draws curved lines, or Beziers, among them. The lines represent connections, as expected, but the thickness of the lines varies according to the rating of the connection provided by the respondent. In this case, connections were rated from zero (no line) to five. An additional dataset was incorporated into the results, and selected participants were affiliated with their community roles (respondents shown circled or boxed). This representation has the actionable advantages of highlighting both connection strength and sector or organizational membership (or any other categorical depiction). If desired, links can be color-coded or dashed to represent types of connections – professional, personal, or both. Furthermore, community stakeholders can more easily see changes over time. Before and after graphs are easier to view side by side, as the nodes are not rearranged. A simple glace reveals new and stronger connections. Finally, a graph can be produced that shows only the changes in connections. This graph (shown below) elicits a truly actionable response from those interested in the results, as it is a stark illustration of the effects of time and/or intervention on the network.
Network analysis is progressing. However, it still has some drawbacks that, at times, make the results less than actionable. Primarily, network data is highly dependent on complete samples and is therefore extremely sensitive to missing data. Most types of illustrative network images require a complete matrix of responses to produce accurate representations. Surveys rarely have response rates that come even close to 100%. Data sets with missing data (the majority) will not necessarily provide a “true” picture of the network. Often survey respondents are worried about confidentiality, yet network graphs are certainly less actionable when the participants are not identified. Without any identifying information on participants, stakeholders are less able to take action to further integrate members of the network. Including community roles, organizational ties, and sector affiliations can help – but in small communities this information can unwittingly identify participants against their wishes. It is essential to be cautious. Finally, network analysis is most actionable when samples are small. Large samples produce images that are unwieldy at best.
Actionable takeaway: To increase the chances that your network data are actionable, consider the following five tips:
a) Apply network analysis to a relatively small sample – a small community or a collaborative. Samples of less than 30 respondents result in the best “viewing”.
b) Maximize your sample size as much as possible with the resources you have. Follow-up with respondents and connect in person or by phone.
c) Consider how you will manage confidentiality. Confidentiality may increase your response rate, but may also make your data less actionable. List the pros and cons.
d) Select a network representation that provides as much information about your network as possible. Consider centrality, clusters, power, connection strength, connection type, and respondent roles. Decide what visual cues are the most meaningful (and therefore actionable) to your inquiry.
e) That said, do not include visual representations that are not related to your research questions or that are extraneous to what you want to learn about the network. A clean, easy to read, illustration is more actionable than one that is dense and confusing.