Chapter 7

Hyperlocal Media and Social Networks: Methodology and Findings

While the primary motive of this study has been to make sense of the relationship between offline community engagement and online hyperlocal media, considering the nature of online engagement can also be very instructive in understanding all the qualities of communication at play in Columbia Heights. What differentiates the hyperlocal media studied here from the local newspapers of past ages is the ability for interaction built into the media platforms themselves and the social relationships captured by Web 2.0 tools. Chapter 4 discussed the merits of the networked public sphere as an inclusive communicative space and the role of weak ties in expanding participation in public discourse. An examination of these phenomena in Columbia Heights can illuminate the extent to which one real-life example diverges from or aligns with the theoretical ideals of the networked public sphere. As the content analysis in Chapter 5 detailed, some hyperlocal media have developed faster than others in Columbia Heights. This chapter will focus on Columbia Heights’ presence on Twitter, which is not yet the most frequently used medium for hyperlocal communication, but has enough activity to merit a closer look. Although previous chapters showed a strong correlation between community engagement and hyperlocal media in general, this chapter will show that relationships within the neighborhood’s network of Twitter users are somewhat weaker.

Methodology

Different hyperlocal media channels have the ability to unite individuals around offline causes or issues, but if these hyperlocal media are not connected to one another, they become just another example of the fragmentation of the public sphere. On the other hand, the creation of ties among hyperlocal media can also be viewed as something that evolves over time, and there are still compelling arguments to be made for maintaining independence among media so that each serves a unique role. By understanding the online relationships among hyperlocal media, we can gain a better sense of the context within which to make these kinds of judgments. With that in mind, this chapter will use social network analysis techniques to address the following research question:

RQ3: To what extent do social media demonstrate online interaction among Columbia Heights’ users?

There are a number of measures and terms from the social network analysis field that will be applied in this chapter. As a methodology, social network analysis relies upon understanding the social ties that bind individuals or organizations to one another. These connections are depicted in a social network graph, which consists of vertices (also known as nodes) and edges (also known as ties). A vertex or node—which can be depicted as a circle, label, or other shape in a graph—represents an individual or organization. In this study, a vertex represents a Twitter account. An edge—depicted as a line connecting two nodes—represents a relationship between two individuals or organizations. Edges can represent any kind of relationship the researcher seeks to understand. In this study edges are used to represent actions or relationships on Twitter, including “mentions,” the act of following another, the act of being followed, or one Twitter user’s “retweet” of another’s tweet. Because these behaviors are not always reciprocated between a pair of Twitter users, they are represented by an arrow. A graph such as this in which ties are not always mutual is a “directed graph.” The density of the graph refers to the percentage of possible ties between every node that are actually present. The number of edges radiating outward from or pointing inward towards a vertex is known as the degree and measures the number of connections a vertex has in a graph. By following the edges from one vertex to another, one can create a “path,” which measures the steps from one node to the next through a chain of linked vertices. One measure employed in this study that involves path length is betweenness centrality, which indicates how often a given vertex lies on a path between two other vertices. Another measure used in this study is Eigenvector centrality, which reflects how well-connected a node’s connections are. In other words, if the nodes to which node A is connected each have a high degree and the nodes to which node B is connected each have a low degree, node A would have higher Eigenvector centrality.

These techniques will be employed to test the following hypotheses:

H10: The Columbia Heights’ Twitter network is connected by relationships of attention more often than relationships of communication.

H0: The Columbia Heights’ Twitter network is connected by more relationships of communication than relationships of attention.

H11: Twitter feeds associated with other forms of community engagement (offline or through online counterparts in other media) are more likely to occupy an influential position in the network than those associated with businesses.

H0: Twitter feeds associated with businesses are more likely to occupy an influential position in the network than those associated with other forms of community engagement (offline or through online counterparts in other media).

These hypotheses utilize several different concepts endemic to this study. First, the term “relationships of attention” is used to refer to ties between Twitter users who follow one another; these can be directed (i.e., a one-way tie between users in which one follows the other) or undirected (i.e., a two-way tie in which both Twitter users follow one another). “Relationships of communication” refers to directed or undirected ties between Twitter users who mention (indicated by an “@” symbol preceding the other’s username) or reply to one another (indicated by a “RT” prefix before the other user’s tweet). Attention relationships imply a weaker tie, as they indicate a passive connection, while communicative relationships indicate stronger and more active ties between users who engage directly with one another.  In H11 “Twitter feeds associated with other forms of community engagement” refers to users that represent offline actors, such as ANCs or individual public figures, or those that supplement another online medium, such as a hyperlocal blog or website. Influence will be measured by Eigenvector centrality,[1] so vertices occupying influential positions are those with high Eigenvector centrality. This method of measuring influence is suitable because Eigenvector centrality is based on the connectedness of the other vertices to which a vertex is connected.

A variety of Twitter accounts were selected for inclusion in the social network graph. Twitter accounts were identified for inclusion by the list of usernames returned after performing a search for “Columbia Heights” on Twitter. Only two included were not specific to the Columbia Heights neighborhood: @bryanweaverdc, the Twitter account of Bryan Weaver, a candidate for D.C. Council in 2011, who was included because of his former history as a Columbia Heights resident and ongoing involvement in organizations based in the adjacent neighborhood of Adams Morgan; and @PoPville, the Twitter account associated with the Prince of Petworth blog, which has a large Columbia Heights readership and was included in this study’s content analysis. By choosing a range of Twitter accounts tied to hyperlocal media, community organizations or political bodies, and neighborhood restaurants, we can look for relationships among users who are tied not by function but by shared geographic interest, a means of testing to what extent the offline community translates to the unbounded online world. The table below shows the users included:

Table 7.1: Twitter Users Included in Social Network Analysis[2]

Twitter Username Category Description
@826DC Offline community organization Associated with non-profit writing center 826DC, located in Columbia Heights
@ANC1A Offline political organization Associated with ANC 1A, which includes North Columbia Heights
@BloomBars Offline community organization Associated with nonprofit arts and cultural center BloomBars, located in Columbia Heights
@BryanWeaverDC Individual Twitter account of local community figure/political candidate Bryan Weaver
@ColHeightsDay Offline community organization Associated with an annual neighborhood event, Columbia Heights Day
@ColumbiaHts Hyperlocal Twitter account Unique account created to share news on Columbia Heights; inactive since Aug. 2010
@CommonWealthPub Local business Associated with Columbia Heights restaurant that closed in Feb. 2010
@HeightsDC Local business Associated with Columbia Heights restaurant The Heights
@JimGraham_Ward1 Individual Account of the City Council member for Ward 1, Jim Graham
@LousCityBar Local business Associated with Columbia Heights restaurant Lou’s City Bar
@newcolumbiahts Hyperlocal media counterpart Associated with hyperlocal blog New Columbia Heights
@ourcoheigh Hyperlocal media counterpart Associated with Our Columbia Heights blog/video project; only one tweet since Nov. 2010
@petes_newhaven Local business Associated with Pete’s New Haven pizzeria, which has locations in Columbia Heights and two other neighborhoods
@PoPville Hyperlocal media counterpart Associated with Prince of Petworth blog
@RedRocksDC Local business Associated with Red Rocks restaurant in Columbia Heights
@room11dc Local business Associated with Room 11 restaurant in Columbia Heights
@TheWonderlandDC Local business Associated with The Wonderland Ballroom neighborhood and beer garden in Columbia Heights

Preparing the Data

Data about the networks of each Twitter account were simultaneously captured on March 24, 2011, using the NodeXL template for Microsoft Office Excel.[3] First, I created a list of the users in Table 7.1 titled “Columbia Heights” on my own personal Twitter account. Next, I imported data to NodeXL from that list.[4] The search scoured the latest tweet for each account for a reply or mention, in addition to cataloguing the accounts’ “following” relationships and gathering statistics about number of followers, number followed, number of tweets, and other statistics not used in this analysis. To prepare the data, I merged duplicate edges, which created thicker ties between nodes that were connected by more than one of the behaviors mentioned above. The thickness of the tie is referred to as its “edgeweight.”

To create the graph, I specified several statistical qualities to define the graph’s aesthetics. First, the edges are directed, so that they indicate whether the relationship between vertices is reciprocated or not. An arrow pointing away from a node indicates a “following” relationship, while an arrow pointing towards a node indicates a follower. Second, the edgeweights are reflected in the thickness of the edges: An edgeweight of 1 (thinner edges) indicates one kind of relationship between the two vertices (eg, retweet), and an edgeweight of 2 (thicker edges) indicates two kinds of relationships (eg, follower and mention). Third, edge colors were categorized by relationship: Communicative relationships are shown with blue edges and attentive relationships with orange. All of the ties with edgeweights of two represent both an attentive and a communicative relationship and are shown in blue. Thinner orange edges represent following relationships only. Vertex size, which appears here as the size of the box around the node’s name, was calculated in proportion to the number of followers each node (i.e., Twitter user) has.[5] Vertex label color (i.e., the color of the box around the node’s name) represents the Eigenvector centrality on a spectrum from pink (low) to red (high).

Findings

The graph created using the data imported from Twitter (Fig. 7.1) showed a striking level of support for H10. With only three edges designated as communicative relationships out of 124 total edges, we can reject the null hypothesis that the Columbia Heights Twitter network is connected by more relationships of communication than of attention in support of the following hypothesis:

H10: The Columbia Heights’ Twitter network is connected by relationships of attention more often than relationships of communication.

The three mention relationships were all one-way: @ColumbiaHts to @ColHeightsDay, @theheightslife to @ColHeightsDay, and @ANC1A to @newcolumbiahts.[6] However, the latest tweets (i.e., the ones imported for this data set) from @ANC1A and @ColumbiaHts were published in November 2010 and August 2010, respectively, so it is difficult to characterize even these as active communicative relationships. The most recent tweet from @theheightslife is actually a retweet, which is a less active form of engagement than if the poster had written new content directed towards the account being mentioned. Additionally, all of the communicative relationships are also attentive ones, so when communication does take place between Columbia Heights’ Twitter users, it is among those who already share a tie.

Figure 7.1: Columbia Heights Users on Twitter



Because the attentive relationships were directed, we can also look at the density of the graph to determine how well-connected these Twitter users are with one another. With 124 edges and 18 nodes, the graph has a density of 40.5%, which means that the majority of possible ties between vertices are not present (or not reciprocal, since the graph is directed). The highest in-degree—which reflects the number of users in the graph that mention or follow the node in question and is represented by inward-facing arrows—belonged to @newcolumbiahts and measured 14. The lowest, at 0, belonged to @BryanWeaverDC. Since @BryanWeaverDC was a former member of the Columbia Heights community, it makes sense that his role in this network would be that of one who is attentive to media regarding his former and now nearby neighborhood. (@BryanWeaverDC had an out-degree of 3, meaning he follows three sources in this network.) The out-degree measurements indicate how many nodes the user in question follows or has mentioned and are indicated by an outward-pointing arrow. The highest measure was 15, shared by @ColHeightsDay and @BloomBars, and the lowest belonged to @JimGraham_Ward1 and @room11dc, both of whom do not follow or mention any of the users in this network. These findings suggest that attentive relationships among Columbia Heights-based users on Twitter, though more common than communicative relationships, are still not a priority for the medium.

Another implication of the graph’s connectedness—or lack thereof—comes via an analysis of the influence of certain nodes in the graph. The second hypothesis in this chapter proposed the following:

H11: Twitter feeds associated with other forms of community engagement (offline or through online counterparts in other media) are more likely to occupy an influential position in the network than those associated with businesses.

Twitter accounts with high Eigenvector centrality are more influential in this network, because in this sparsely connected network, within two steps, they can reach the most other Twitter users. Although the analysis of communicative versus attentive relationships showed little action in retweeting or mentioning other users, that analysis did only study the most recent tweet for each source. When nodes in this graph do retweet or mention other users, their followers will potentially be privy to information from a wider variety of sources if the node in question has a high Eigenvector centrality, thereby facilitating the flow of communication throughout the network. The mean Eigenvector centrality was 0.055, so to test the hypothesis, I looked at the categorization of Twitter users’ with an Eigenvector centrality that was above average. The findings, as summarized in Table 7.2, show that of the 11 nodes with above-average Eigenvector centrality, 5 belonged to local businesses and 6 were associated with other individuals, community organizations, or media. Therefore, by a slight majority, the findings allow us to reject the null hypothesis and in support of the hypothesis that Twitter feeds associated with other forms of community engagement are more influential in the social network.

Table 7.2: Graph Metrics for Twitter Social Network[7]

Twitter Username Eigenvector centrality Betweenness centrality
@BloomBars 0.081 23.780
@ColHeightsDay 0.079 13.447
@theheightslife 0.079 13.447
@newcolumbiahts 0.077 36.263
@HeightsDC 0.065 10.187
@petes_newhaven 0.065 2.741
@RedRocksDC 0.065 1.975
@PoPville 0.059 10.560
@ColumbiaHts 0.058 2.587
@JimGraham_Ward1 0.057 9.915
@CommonWealthPub 0.056 0.722
@TheWonderlandDC 0.054 1.253
@room11dc 0.051 0.472
@ANC1A 0.039 0
@ourcoheigh 0.039 0.286
@LousCityBar 0.039 0.182
@826DC 0.020 0
@BryanWeaverDC 0.017 0.182
Mean: 0.055 7.11

There are a few additional explanations in support of the hypothesis. While the difference between Eigenvector centrality in business Twitter users and the rest in the sample was not striking, differences in betweenness centrality are more pronounced. As a measure of the frequency with which a node appears on the shortest path between two other nodes, betweenness centrality offers us a way of understanding which Twitter users are in a position to be frequent brokers of information between groups, which could be another way of understanding influence. The mean for betweenness centrality was 7.11, and of the 7 cases above the mean, only one represented a local business Twitter user. Therefore, this alternative definition of influence offers additional support for the hypothesis. Nonetheless, it is strategic for business Twitter users to have a high Eigenvector centrality, as this implies that they are following popular users.

Analysis

The predominance of attentive relationships in the Twitter network may have something to do with the technology itself. Unlike the more communicative Yahoo! Groups, which create a bounded space for discussion among a discrete list of members, Twitter users publish content in an online space that is much more amorphous and among social networks that are defined more by personal relationships or shared interests than by geography. As a newer technology, Twitter might still be awaiting its full maturation and the attendant cultural norms of use that now accompany e-mail and list-servs and give them a daily role in the lives of most individuals and organizations. At the social level, ties between Twitter users in the neighborhood would be weak at best, unless bolstered by extrinsic connections, which could be why the nodes in Figure 7.1 that were situated in the densest parts of the network were those that also operated in other communication-based situations, both online and offline.

At the discursive level, it is also evident that Twitter users in Columbia Heights are not primarily employing the tools to engage with others in the Twitter-sphere. As the content analysis showed, tweets were sparse across the study days, and unlike with some of the other media, this was not anomalous within the broader timeline of those media. One Twitter user that was active during the study days was @theheightslife. Although The Heights Life blog had no posts on the content analysis days, its authors did tweet brief comments about some of the news of day, such as the closing of the CommonWealth pub discussed on February 16, 2011 (@theheightslife 2011). Many other tweets directed readers to other media. For example, D.C. Councilman Jim Graham (D-Ward 1) tweeted the following on February 16: “Holding Public Oversight Hearing on the Department of Youth Rehabilitation Services. Watch live on Ch13” (@JimGraham_Ward1 2011).

Dan Silverman of The Prince of Petworth and its affiliated @PoPville account has this specific purpose in mind in his use of Twitter.  Silverman frequently tweets are used to publicize his posts. His communication is most often one-way because he recognizes that readers follow him for more information. “If I replied to every person who asks me something, it’s going to overwhelm the other tweets [from my account],” he explained. Instead, he only replies to things he deems significant (D. Silverman, pers. comm.). This mirrors his use of Twitter as a follower. He follows a limited number of users (approximately 75 at the time of this study) and does not appreciate it when other Twitter users publish content that strays from the main subject matter of their account (ibid.). For Silverman, Twitter is an efficient way to keep up with breaking news, as he explained that on Twitter, information is “all in one place. It’s very succinct….In Twitter, it’s direct, to the point. Things that I’m interested in that I’ll hear about on Twitter are crime, development, local businesses….It’s very easy to get overwhelmed if you follow too many people, so I’ve been very intentional [in choosing whom to follow]” (ibid.). This sort of attitude helps to explain the motives for attention-based relationships among Columbia Heights’ Twitter users; if everyone has different expectations for how the medium should be used, interaction within the virtual community becomes more complicated.

In conclusion, social network analysis shows that Twitter users in Columbia Heights are primarily connected by attentive relationships, rather than communicative ones. Even before the advent of social media, people turned to their social networks for local political information, as Mondak’s (1995) study of local news coverage in the 1992 election cycle showed. Therefore, this information-seeking behavior through Twitter is in line with an understanding of how people find local information. By looking at a sample that included Twitter users from local businesses as well as those associated with hyperlocal media, neighborhood politics, or community associations, we saw that it is the latter group that occupy the most influential positions within this network. Nonetheless, the network is not very dense, which suggests that hyperlocal media in Columbia Heights do not necessarily represent a well-connected online community. This finding is significant in that it suggests that hyperlocal media serve unique roles within their neighborhood context. But how sustainable is a technology such as Twitter when applied to a hyperlocal context? The next and final chapter will discuss the future of hyperlocal media—in Columbia Heights and beyond—and their broader implications in society.


[1] Eigenvector centrality reflects the connectivity of a node’s connections. Therefore, a node with high Eigenvector centrality is connected to nodes that have many ties to other nodes (i.e., a high degree).

[2] Please refer to Methodological Appendix D for URLs for each Twitter user.

[3] The NodeXL template is available free of charge at http://nodexl.codeplex.com. Accessed 24 March 2011.

[4] I checked options in the NodeXL import dialog box to include the following data: “Follows relationship,” “‘Replies-to’” relationship in latest tweet, “‘Mentions’ relationship in latest tweet,” “Add a Latest Tweet column to the Vertices worksheet,” and “Add statistics columns to the Vertices worksheet.”

[5] These were specified in the “Autofill columns” tool to range in size from 2.0 to 7.0, so that the node with least followers has a size of 2.0 and the one with the most, 7.0.

[6] The tweets themselves can be found in the Methodological Appendix D, Figure D.1.

[7] In the table, Twitter accounts associated with business are italicized; those falling into categories that align with other forms of community engagement, as discussed above in the methodology section, appear in regular typeface.

Be Sociable, Share!